SQLite

Changes On Branch lsm-vtab
Login

Many hyperlinks are disabled.
Use anonymous login to enable hyperlinks.

Changes In Branch lsm-vtab Excluding Merge-Ins

This is equivalent to a diff from 9afd7a2f to 680cc064

2017-06-29
13:19
Add the LSM1 extension. (check-in: 824e8327 user: drh tags: trunk)
12:54
Implement xUnlink, xShmMap, and xShmUnmap for lsm1 on Win32. (Closed-Leaf check-in: 680cc064 user: mistachkin tags: lsm-vtab)
00:20
Implement xRemap for lsm1 on Win32. Also, zero file handle when closing it. (check-in: 93c9aa7d user: mistachkin tags: lsm-vtab)
2017-06-15
16:45
Fix typo and improve the wording of the description of "Metadata" in the output of the sqlite3_analyzer tool. (check-in: ca1ff707 user: drh tags: trunk)
15:06
Merge the latest changes from trunk. (check-in: 60105c22 user: drh tags: lsm-vtab)
13:07
Fix harmless compiler warnings in the shell.c file (check-in: 9afd7a2f user: drh tags: trunk)
12:50
Improvements to the ".tables" command in the command-line shell so that it shows the name of all schemas if the name is anything other than "main". (check-in: c7f778b7 user: drh tags: trunk)

Changes to Makefile.msc.

2238
2239
2240
2241
2242
2243
2244



2245
2246
2247
2248
2249
2250
2251
dbselftest.exe:	$(TOP)\test\dbselftest.c $(SQLITE3C) $(SQLITE3H)
	$(LTLINK) $(NO_WARN) $(DBSELFTEST_COMPILE_OPTS) $(TOP)\test\dbselftest.c $(SQLITE3C)

rbu.exe:	$(TOP)\ext\rbu\rbu.c $(TOP)\ext\rbu\sqlite3rbu.c $(SQLITE3C) $(SQLITE3H)
	$(LTLINK) $(NO_WARN) -DSQLITE_ENABLE_RBU \
		$(TOP)\ext\rbu\rbu.c $(SQLITE3C) /link $(LDFLAGS) $(LTLINKOPTS)




moreclean:	clean
	del /Q $(SQLITE3C) $(SQLITE3H) 2>NUL
# <</mark>>

clean:
	del /Q *.exp *.lo *.ilk *.lib *.obj *.ncb *.pdb *.sdf *.suo 2>NUL
	del /Q *.bsc *.def *.cod *.da *.bb *.bbg *.vc gmon.out 2>NUL







>
>
>







2238
2239
2240
2241
2242
2243
2244
2245
2246
2247
2248
2249
2250
2251
2252
2253
2254
dbselftest.exe:	$(TOP)\test\dbselftest.c $(SQLITE3C) $(SQLITE3H)
	$(LTLINK) $(NO_WARN) $(DBSELFTEST_COMPILE_OPTS) $(TOP)\test\dbselftest.c $(SQLITE3C)

rbu.exe:	$(TOP)\ext\rbu\rbu.c $(TOP)\ext\rbu\sqlite3rbu.c $(SQLITE3C) $(SQLITE3H)
	$(LTLINK) $(NO_WARN) -DSQLITE_ENABLE_RBU \
		$(TOP)\ext\rbu\rbu.c $(SQLITE3C) /link $(LDFLAGS) $(LTLINKOPTS)

LSMDIR=$(TOP)\ext\lsm1
!INCLUDE $(LSMDIR)\Makefile.msc

moreclean:	clean
	del /Q $(SQLITE3C) $(SQLITE3H) 2>NUL
# <</mark>>

clean:
	del /Q *.exp *.lo *.ilk *.lib *.obj *.ncb *.pdb *.sdf *.suo 2>NUL
	del /Q *.bsc *.def *.cod *.da *.bb *.bbg *.vc gmon.out 2>NUL

Added ext/lsm1/Makefile.

















































































































>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
#
# This Makefile is designed for use with main.mk in the root directory of
# this project. After including main.mk, the users makefile should contain:
#
#    LSMDIR=$(TOP)/ext/lsm1/
#    include $(LSMDIR)/Makefile
#
# The most useful targets are [lsmtest] and [lsm.so].
#

LSMOBJ    = \
  lsm_ckpt.o \
  lsm_file.o \
  lsm_log.o \
  lsm_main.o \
  lsm_mem.o \
  lsm_mutex.o \
  lsm_shared.o \
  lsm_sorted.o \
  lsm_str.o \
  lsm_tree.o \
  lsm_unix.o \
  lsm_win32.o \
  lsm_varint.o \
  lsm_vtab.o

LSMHDR   = \
  $(LSMDIR)/lsm.h \
  $(LSMDIR)/lsmInt.h

LSMTESTSRC = $(LSMDIR)/lsm-test/lsmtest1.c $(LSMDIR)/lsm-test/lsmtest2.c     \
             $(LSMDIR)/lsm-test/lsmtest3.c $(LSMDIR)/lsm-test/lsmtest4.c     \
             $(LSMDIR)/lsm-test/lsmtest5.c $(LSMDIR)/lsm-test/lsmtest6.c     \
             $(LSMDIR)/lsm-test/lsmtest7.c $(LSMDIR)/lsm-test/lsmtest8.c     \
             $(LSMDIR)/lsm-test/lsmtest9.c                                   \
             $(LSMDIR)/lsm-test/lsmtest_datasource.c \
             $(LSMDIR)/lsm-test/lsmtest_func.c $(LSMDIR)/lsm-test/lsmtest_io.c  \
             $(LSMDIR)/lsm-test/lsmtest_main.c $(LSMDIR)/lsm-test/lsmtest_mem.c \
             $(LSMDIR)/lsm-test/lsmtest_tdb.c $(LSMDIR)/lsm-test/lsmtest_tdb3.c \
             $(LSMDIR)/lsm-test/lsmtest_util.c $(LSMDIR)/lsm-test/lsmtest_win32.c


# all: lsm.so

LSMOPTS = -DLSM_MUTEX_PTHREADS=1 -I$(LSMDIR)

lsm.so:	$(LSMOBJ)
	$(TCCX) -shared -o lsm.so $(LSMOBJ)

%.o:	$(LSMDIR)/%.c $(LSMHDR) sqlite3.h
	$(TCCX) $(LSMOPTS) -c $<
	
lsmtest$(EXE): $(LSMOBJ) $(LSMTESTSRC) $(LSMTESTHDR) sqlite3.o
	# $(TCPPX) -c $(TOP)/lsm-test/lsmtest_tdb2.cc
	$(TCCX) $(LSMOPTS) $(LSMTESTSRC) $(LSMOBJ) sqlite3.o -o lsmtest$(EXE) $(THREADLIB) 

Added ext/lsm1/Makefile.msc.

























































































































































































>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
#
# This Makefile is designed for use with main.mk in the root directory of
# this project. After including main.mk, the users makefile should contain:
#
#    LSMDIR=$(TOP)\ext\lsm1\
#    include $(LSMDIR)\Makefile.msc
#
# The most useful targets are [lsmtest.exe] and [lsm.dll].
#

LSMOBJ    = \
  lsm_ckpt.lo \
  lsm_file.lo \
  lsm_log.lo \
  lsm_main.lo \
  lsm_mem.lo \
  lsm_mutex.lo \
  lsm_shared.lo \
  lsm_sorted.lo \
  lsm_str.lo \
  lsm_tree.lo \
  lsm_unix.lo \
  lsm_win32.lo \
  lsm_varint.lo \
  lsm_vtab.lo

LSMHDR   = \
  $(LSMDIR)\lsm.h \
  $(LSMDIR)\lsmInt.h

LSMTESTSRC = $(LSMDIR)\lsm-test\lsmtest1.c $(LSMDIR)\lsm-test\lsmtest2.c     \
             $(LSMDIR)\lsm-test\lsmtest3.c $(LSMDIR)\lsm-test\lsmtest4.c     \
             $(LSMDIR)\lsm-test\lsmtest5.c $(LSMDIR)\lsm-test\lsmtest6.c     \
             $(LSMDIR)\lsm-test\lsmtest7.c $(LSMDIR)\lsm-test\lsmtest8.c     \
             $(LSMDIR)\lsm-test\lsmtest9.c                                   \
             $(LSMDIR)\lsm-test\lsmtest_datasource.c \
             $(LSMDIR)\lsm-test\lsmtest_func.c $(LSMDIR)\lsm-test\lsmtest_io.c  \
             $(LSMDIR)\lsm-test\lsmtest_main.c $(LSMDIR)\lsm-test\lsmtest_mem.c \
             $(LSMDIR)\lsm-test\lsmtest_tdb.c $(LSMDIR)\lsm-test\lsmtest_tdb3.c \
             $(LSMDIR)\lsm-test\lsmtest_util.c $(LSMDIR)\lsm-test\lsmtest_win32.c

# all: lsm.dll

LSMOPTS = -DLSM_MUTEX_WIN32=1 -I$(LSMDIR)

lsm_ckpt.lo:	$(LSMDIR)\lsm_ckpt.c $(LSMHDR) $(SQLITE3H)
	$(LTCOMPILE) $(LSMOPTS) -c $(LSMDIR)\lsm_ckpt.c

lsm_file.lo:	$(LSMDIR)\lsm_file.c $(LSMHDR) $(SQLITE3H)
	$(LTCOMPILE) $(LSMOPTS) -c $(LSMDIR)\lsm_file.c

lsm_log.lo:	$(LSMDIR)\lsm_log.c $(LSMHDR) $(SQLITE3H)
	$(LTCOMPILE) $(LSMOPTS) -c $(LSMDIR)\lsm_log.c

lsm_main.lo:	$(LSMDIR)\lsm_main.c $(LSMHDR) $(SQLITE3H)
	$(LTCOMPILE) $(LSMOPTS) -c $(LSMDIR)\lsm_main.c

lsm_mem.lo:	$(LSMDIR)\lsm_mem.c $(LSMHDR) $(SQLITE3H)
	$(LTCOMPILE) $(LSMOPTS) -c $(LSMDIR)\lsm_mem.c

lsm_mutex.lo:	$(LSMDIR)\lsm_mutex.c $(LSMHDR) $(SQLITE3H)
	$(LTCOMPILE) $(LSMOPTS) -c $(LSMDIR)\lsm_mutex.c

lsm_shared.lo:	$(LSMDIR)\lsm_shared.c $(LSMHDR) $(SQLITE3H)
	$(LTCOMPILE) $(LSMOPTS) -c $(LSMDIR)\lsm_shared.c

lsm_sorted.lo:	$(LSMDIR)\lsm_sorted.c $(LSMHDR) $(SQLITE3H)
	$(LTCOMPILE) $(LSMOPTS) -c $(LSMDIR)\lsm_sorted.c

lsm_str.lo:	$(LSMDIR)\lsm_str.c $(LSMHDR) $(SQLITE3H)
	$(LTCOMPILE) $(LSMOPTS) -c $(LSMDIR)\lsm_str.c

lsm_tree.lo:	$(LSMDIR)\lsm_tree.c $(LSMHDR) $(SQLITE3H)
	$(LTCOMPILE) $(LSMOPTS) -c $(LSMDIR)\lsm_tree.c

lsm_unix.lo:	$(LSMDIR)\lsm_unix.c $(LSMHDR) $(SQLITE3H)
	$(LTCOMPILE) $(LSMOPTS) -c $(LSMDIR)\lsm_unix.c

lsm_win32.lo:	$(LSMDIR)\lsm_win32.c $(LSMHDR) $(SQLITE3H)
	$(LTCOMPILE) $(LSMOPTS) -c $(LSMDIR)\lsm_win32.c

lsm_varint.lo:	$(LSMDIR)\lsm_varint.c $(LSMHDR) $(SQLITE3H)
	$(LTCOMPILE) $(LSMOPTS) -c $(LSMDIR)\lsm_varint.c

lsm_vtab.lo:	$(LSMDIR)\lsm_vtab.c $(LSMHDR) $(SQLITE3H)
	$(LTCOMPILE) $(LSMOPTS) -c $(LSMDIR)\lsm_vtab.c

lsm.dll:	$(LSMOBJ)
	$(LD) $(LDFLAGS) $(LTLINKOPTS) $(LTLIBPATHS) /DLL /OUT:$@ $(LSMOBJ)

lsmtest.exe: $(LSMOBJ) $(LSMTESTSRC) $(LSMTESTHDR) $(LIBOBJS1)
	$(LTLINK) $(LSMOPTS) $(LSMTESTSRC) /link $(LSMOBJ) $(LIBOBJS1)

Added ext/lsm1/lsm-test/README.

















































































>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40


Organization of test case files:

  lsmtest1.c: Data tests. Tests that perform many inserts and deletes on a 
              database file, then verify that the contents of the database can
              be queried.

  lsmtest2.c: Crash tests. Tests that attempt to verify that the database 
              recovers correctly following an application or system crash.

  lsmtest3.c: Rollback tests. Tests that focus on the explicit rollback of
              transactions and sub-transactions.

  lsmtest4.c: Multi-client tests.

  lsmtest5.c: Multi-client tests with a different thread for each client.

  lsmtest6.c: OOM injection tests.

  lsmtest7.c: API tests.

  lsmtest8.c: Writer crash tests. Tests in this file attempt to verify that
              the system recovers and other clients proceed unaffected if
              a process fails in the middle of a write transaction.

              The difference from lsmtest2.c is that this file tests
              live-recovery (recovery from a failure that occurs while other
              clients are still running) whereas lsmtest2.c tests recovery
              from a system or power failure.

  lsmtest9.c: More data tests. These focus on testing that calling
              lsm_work(nMerge=1) to compact the database does not corrupt it.
              In other words, that databases containing block-redirects
              can be read and written.





Added ext/lsm1/lsm-test/lsmtest.h.





















































































































































































































































































































































































































































































































































































>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282

#ifndef __WRAPPER_INT_H_
#define __WRAPPER_INT_H_

#include "lsmtest_tdb.h"
#include "sqlite3.h"
#include "lsm.h"

#include <assert.h>
#include <stdlib.h>
#include <string.h>
#include <stdio.h>

#ifdef __cplusplus
extern "C" {
#endif

#ifdef _WIN32
# include "windows.h"
# define gettimeofday win32GetTimeOfDay
# define F_OK  (0)
# define sleep(sec) Sleep(1000 * (sec))
# define usleep(usec) Sleep((usec) / 1000)
# ifdef _MSC_VER
#  include <io.h>
#  define snprintf _snprintf
#  define fsync(fd) FlushFileBuffers((HANDLE)_get_osfhandle((fd)))
#  define fdatasync(fd) FlushFileBuffers((HANDLE)_get_osfhandle((fd)))
#  define __va_copy(dst,src) ((dst) = (src))
#  define ftruncate(fd,sz) (_chsize_s((fd), (sz))==0)
# else
#  error Unsupported C compiler for Windows.
# endif
int win32GetTimeOfDay(struct timeval *, void *);
#endif

#ifndef _LSM_INT_H
typedef unsigned int  u32;
typedef unsigned char u8;
typedef long long int i64;
typedef unsigned long long int u64;
#endif


#define ArraySize(x) ((int)(sizeof(x) / sizeof((x)[0])))

#define MIN(x,y) ((x)<(y) ? (x) : (y))
#define MAX(x,y) ((x)>(y) ? (x) : (y))

#define unused_parameter(x) (void)(x)

#define TESTDB_DEFAULT_PAGE_SIZE   4096
#define TESTDB_DEFAULT_CACHE_SIZE  2048

/*
** Ideally, these should be in wrapper.c. But they are here instead so that 
** they can be used by the C++ database wrappers in wrapper2.cc.
*/
typedef struct DatabaseMethods DatabaseMethods;
struct TestDb {
  DatabaseMethods const *pMethods;          /* Database methods */
  const char *zLibrary;                     /* Library name for tdb_open() */
};
struct DatabaseMethods {
  int (*xClose)(TestDb *);
  int (*xWrite)(TestDb *, void *, int , void *, int);
  int (*xDelete)(TestDb *, void *, int);
  int (*xDeleteRange)(TestDb *, void *, int, void *, int);
  int (*xFetch)(TestDb *, void *, int, void **, int *);
  int (*xScan)(TestDb *, void *, int, void *, int, void *, int,
    void (*)(void *, void *, int , void *, int)
  );
  int (*xBegin)(TestDb *, int);
  int (*xCommit)(TestDb *, int);
  int (*xRollback)(TestDb *, int);
};

/* 
** Functions in wrapper2.cc (a C++ source file). wrapper2.cc contains the
** wrapper for Kyoto Cabinet. Kyoto cabinet has a C API, but
** the primary interface is the C++ API.
*/
int test_kc_open(const char*, const char *zFilename, int bClear, TestDb **ppDb);
int test_kc_close(TestDb *);
int test_kc_write(TestDb *, void *, int , void *, int);
int test_kc_delete(TestDb *, void *, int);
int test_kc_delete_range(TestDb *, void *, int, void *, int);
int test_kc_fetch(TestDb *, void *, int, void **, int *);
int test_kc_scan(TestDb *, void *, int, void *, int, void *, int,
  void (*)(void *, void *, int , void *, int)
);

int test_mdb_open(const char*, const char *zFile, int bClear, TestDb **ppDb);
int test_mdb_close(TestDb *);
int test_mdb_write(TestDb *, void *, int , void *, int);
int test_mdb_delete(TestDb *, void *, int);
int test_mdb_fetch(TestDb *, void *, int, void **, int *);
int test_mdb_scan(TestDb *, void *, int, void *, int, void *, int,
  void (*)(void *, void *, int , void *, int)
);

/* 
** Functions in wrapper3.c. This file contains the tdb wrapper for lsm.
** The wrapper for lsm is a bit more involved than the others, as it 
** includes code for a couple of different lsm configurations, and for
** various types of fault injection and robustness testing.
*/
int test_lsm_open(const char*, const char *zFile, int bClear, TestDb **ppDb);
int test_lsm_lomem_open(const char*, const char*, int bClear, TestDb **ppDb);
int test_lsm_zip_open(const char*, const char*, int bClear, TestDb **ppDb);
int test_lsm_small_open(const char*, const char*, int bClear, TestDb **ppDb);
int test_lsm_mt2(const char*, const char *zFile, int bClear, TestDb **ppDb);
int test_lsm_mt3(const char*, const char *zFile, int bClear, TestDb **ppDb);

int tdb_lsm_configure(lsm_db *, const char *);

/* Functions in lsmtest_tdb4.c */
int test_bt_open(const char*, const char *zFile, int bClear, TestDb **ppDb);
int test_fbt_open(const char*, const char *zFile, int bClear, TestDb **ppDb);
int test_fbts_open(const char*, const char *zFile, int bClear, TestDb **ppDb);


/* Functions in testutil.c. */
int  testPrngInit(void);
u32  testPrngValue(u32 iVal);
void testPrngArray(u32 iVal, u32 *aOut, int nOut);
void testPrngString(u32 iVal, char *aOut, int nOut);

void testErrorInit(int argc, char **);
void testPrintError(const char *zFormat, ...);
void testPrintUsage(const char *zArgs);
void testPrintFUsage(const char *zFormat, ...);
void testTimeInit(void);
int  testTimeGet(void);

/* Functions in testmem.c. */
void testMallocInstall(lsm_env *pEnv);
void testMallocUninstall(lsm_env *pEnv);
void testMallocCheck(lsm_env *pEnv, int *, int *, FILE *);
void testMallocOom(lsm_env *pEnv, int, int, void(*)(void*), void *);
void testMallocOomEnable(lsm_env *pEnv, int);

/* lsmtest.c */
TestDb *testOpen(const char *zSystem, int, int *pRc);
void testReopen(TestDb **ppDb, int *pRc);
void testClose(TestDb **ppDb);

void testFetch(TestDb *, void *, int, void *, int, int *);
void testWrite(TestDb *, void *, int, void *, int, int *);
void testDelete(TestDb *, void *, int, int *);
void testDeleteRange(TestDb *, void *, int, void *, int, int *);
void testWriteStr(TestDb *, const char *, const char *zVal, int *pRc);
void testFetchStr(TestDb *, const char *, const char *, int *pRc);

void testBegin(TestDb *pDb, int iTrans, int *pRc);
void testCommit(TestDb *pDb, int iTrans, int *pRc);

void test_failed(void);

char *testMallocPrintf(const char *zFormat, ...);
char *testMallocVPrintf(const char *zFormat, va_list ap);
int testGlobMatch(const char *zPattern, const char *zStr);

void testScanCompare(TestDb *, TestDb *, int, void *, int, void *, int, int *);
void testFetchCompare(TestDb *, TestDb *, void *, int, int *);

void *testMalloc(int);
void *testMallocCopy(void *pCopy, int nByte);
void *testRealloc(void *, int);
void testFree(void *);

/* lsmtest_bt.c */
int do_bt(int nArg, char **azArg);

/* testio.c */
int testVfsConfigureDb(TestDb *pDb);

/* testfunc.c */
int do_show(int nArg, char **azArg);
int do_work(int nArg, char **azArg);

/* testio.c */
int do_io(int nArg, char **azArg);

/* lsmtest2.c */
void do_crash_test(const char *zPattern, int *pRc);
int do_rollback_test(int nArg, char **azArg);

/* test3.c */
void test_rollback(const char *zSystem, const char *zPattern, int *pRc);

/* test4.c */
void test_mc(const char *zSystem, const char *zPattern, int *pRc);

/* test5.c */
void test_mt(const char *zSystem, const char *zPattern, int *pRc);

/* lsmtest6.c */
void test_oom(const char *zPattern, int *pRc);
void testDeleteLsmdb(const char *zFile);

void testSaveDb(const char *zFile, const char *zAuxExt);
void testRestoreDb(const char *zFile, const char *zAuxExt);
void testCopyLsmdb(const char *zFrom, const char *zTo);

/* lsmtest7.c */
void test_api(const char *zPattern, int *pRc);

/* lsmtest8.c */
void do_writer_crash_test(const char *zPattern, int *pRc);

/*************************************************************************
** Interface to functionality in test_datasource.c.
*/
typedef struct Datasource Datasource;
typedef struct DatasourceDefn DatasourceDefn;

struct DatasourceDefn {
  int eType;                      /* A TEST_DATASOURCE_* value */
  int nMinKey;                    /* Minimum key size */
  int nMaxKey;                    /* Maximum key size */
  int nMinVal;                    /* Minimum value size */
  int nMaxVal;                    /* Maximum value size */
};

#define TEST_DATASOURCE_RANDOM    1
#define TEST_DATASOURCE_SEQUENCE  2

char *testDatasourceName(const DatasourceDefn *);
Datasource *testDatasourceNew(const DatasourceDefn *);
void testDatasourceFree(Datasource *);
void testDatasourceEntry(Datasource *, int, void **, int *, void **, int *);
/* End of test_datasource.c interface.
*************************************************************************/

void testWriteDatasource(TestDb *, Datasource *, int, int *);
void testWriteDatasourceRange(TestDb *, Datasource *, int, int, int *);
void testDeleteDatasource(TestDb *, Datasource *, int, int *);
void testDeleteDatasourceRange(TestDb *, Datasource *, int, int, int *);


/* test1.c */
void test_data_1(const char *, const char *, int *pRc);
void test_data_2(const char *, const char *, int *pRc);
void test_data_3(const char *, const char *, int *pRc);
void testDbContents(TestDb *, Datasource *, int, int, int, int, int, int *);
void testCaseProgress(int, int, int, int *);
int testCaseNDot(void);

void testCompareDb(Datasource *, int, int, TestDb *, TestDb *, int *);
int testControlDb(TestDb **ppDb);

typedef struct CksumDb CksumDb;
CksumDb *testCksumArrayNew(Datasource *, int, int, int);
char *testCksumArrayGet(CksumDb *, int);
void testCksumArrayFree(CksumDb *);
void testCaseStart(int *pRc, char *zFmt, ...);
void testCaseFinish(int rc);
void testCaseSkip(void);
int testCaseBegin(int *, const char *, const char *, ...);

#define TEST_CKSUM_BYTES 29
int testCksumDatabase(TestDb *pDb, char *zOut);
int testCountDatabase(TestDb *pDb);
void testCompareInt(int, int, int *);
void testCompareStr(const char *z1, const char *z2, int *pRc);

/* lsmtest9.c */
void test_data_4(const char *, const char *, int *pRc);


/*
** Similar to the Tcl_GetIndexFromObjStruct() Tcl library function.
*/
#define testArgSelect(w,x,y,z) testArgSelectX(w,x,sizeof(w[0]),y,z)
int testArgSelectX(void *, const char *, int, const char *, int *);

#ifdef __cplusplus
}  /* End of the 'extern "C"' block */
#endif

#endif

Added ext/lsm1/lsm-test/lsmtest1.c.



























































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621

#include "lsmtest.h"

#define DATA_SEQUENTIAL TEST_DATASOURCE_SEQUENCE
#define DATA_RANDOM     TEST_DATASOURCE_RANDOM

typedef struct Datatest1 Datatest1;
typedef struct Datatest2 Datatest2;

/*
** An instance of the following structure contains parameters used to
** customize the test function in this file. Test procedure:
**
**   1. Create a data-source based on the "datasource definition" vars.
**
**   2. Insert nRow key value pairs into the database.
**
**   3. Delete all keys from the database. Deletes are done in the same 
**      order as the inserts.
**
** During steps 2 and 3 above, after each Datatest1.nVerify inserts or
** deletes, the following:
**
**   a. Run Datasource.nTest key lookups and check the results are as expected.
**
**   b. If Datasource.bTestScan is true, run a handful (8) of range
**      queries (scanning forwards and backwards). Check that the results
**      are as expected.
**
**   c. Close and reopen the database. Then run (a) and (b) again.
*/
struct Datatest1 {
  /* Datasource definition */
  DatasourceDefn defn;

  /* Test procedure parameters */
  int nRow;                       /* Number of rows to insert then delete */
  int nVerify;                    /* How often to verify the db contents */
  int nTest;                      /* Number of keys to test (0==all) */
  int bTestScan;                  /* True to do scan tests */
};

/*
** An instance of the following data structure is used to describe the
** second type of test case in this file. The chief difference between 
** these tests and those described by Datatest1 is that these tests also
** experiment with range-delete operations. Tests proceed as follows:
**
**     1. Open the datasource described by Datatest2.defn. 
**
**     2. Open a connection on an empty database.
**
**     3. Do this Datatest2.nIter times:
**
**        a) Insert Datatest2.nWrite key-value pairs from the datasource.
**
**        b) Select two pseudo-random keys and use them as the start
**           and end points of a range-delete operation.
**
**        c) Verify that the contents of the database are as expected (see
**           below for details).
**
**        d) Close and then reopen the database handle.
**
**        e) Verify that the contents of the database are still as expected.
**
** The inserts and range deletes are run twice - once on the database being
** tested and once using a control system (sqlite3, kc etc. - something that 
** works). In order to verify that the contents of the db being tested are
** correct, the test runs a bunch of scans and lookups on both the test and
** control databases. If the results are the same, the test passes.
*/
struct Datatest2 {
  DatasourceDefn defn;
  int nRange;
  int nWrite;                     /* Number of writes per iteration */
  int nIter;                      /* Total number of iterations to run */
};

/*
** Generate a unique name for the test case pTest with database system
** zSystem.
*/
static char *getName(const char *zSystem, Datatest1 *pTest){
  char *zRet;
  char *zData;
  zData = testDatasourceName(&pTest->defn);
  zRet = testMallocPrintf("data.%s.%s.%d.%d", 
      zSystem, zData, pTest->nRow, pTest->nVerify
  );
  testFree(zData);
  return zRet;
}

int testControlDb(TestDb **ppDb){
#ifdef HAVE_KYOTOCABINET
  return tdb_open("kyotocabinet", "tmp.db", 1, ppDb);
#else
  return tdb_open("sqlite3", ":memory:", 1, ppDb);
#endif
}

void testDatasourceFetch(
  TestDb *pDb,                    /* Database handle */
  Datasource *pData,
  int iKey,
  int *pRc                        /* IN/OUT: Error code */
){
  void *pKey; int nKey;           /* Database key to query for */
  void *pVal; int nVal;           /* Expected result of query */

  testDatasourceEntry(pData, iKey, &pKey, &nKey, &pVal, &nVal);
  testFetch(pDb, pKey, nKey, pVal, nVal, pRc);
}

/*
** This function is called to test that the contents of database pDb
** are as expected. In this case, expected is defined as containing
** key-value pairs iFirst through iLast, inclusive, from data source 
** pData. In other words, a loop like the following could be used to
** construct a database with identical contents from scratch.
**
**   for(i=iFirst; i<=iLast; i++){
**     testDatasourceEntry(pData, i, &pKey, &nKey, &pVal, &nVal);
**     // insert (pKey, nKey) -> (pVal, nVal) into database
**   }
**
** The key domain consists of keys 0 to (nRow-1), inclusive, from
** data source pData. For both scan and lookup tests, keys are selected
** pseudo-randomly from within this set.
**
** This function runs nLookupTest lookup tests and nScanTest scan tests.
**
** A lookup test consists of selecting a key from the domain and querying
** pDb for it. The test fails if the presence of the key and, if present,
** the associated value do not match the expectations defined above.
**
** A scan test involves selecting a key from the domain and running
** the following queries:
**
**   1. Scan all keys equal to or greater than the key, in ascending order.
**   2. Scan all keys equal to or smaller than the key, in descending order.
**
** Additionally, if nLookupTest is greater than zero, the following are
** run once:
**
**   1. Scan all keys in the db, in ascending order.
**   2. Scan all keys in the db, in descending order.
**
** As you would assume, the test fails if the returned values do not match
** expectations.
*/
void testDbContents(
  TestDb *pDb,                    /* Database handle being tested */
  Datasource *pData,              /* pDb contains data from here */
  int nRow,                       /* Size of key domain */
  int iFirst,                     /* Index of first key from pData in pDb */
  int iLast,                      /* Index of last key from pData in pDb */
  int nLookupTest,                /* Number of lookup tests to run */
  int nScanTest,                  /* Number of scan tests to run */
  int *pRc                        /* IN/OUT: Error code */
){
  int j;
  int rc = *pRc;

  if( rc==0 && nScanTest ){
    TestDb *pDb2 = 0;

    /* Open a control db (i.e. one that we assume works) */
    rc = testControlDb(&pDb2);

    for(j=iFirst; rc==0 && j<=iLast; j++){
      void *pKey; int nKey;         /* Database key to insert */
      void *pVal; int nVal;         /* Database value to insert */
      testDatasourceEntry(pData, j, &pKey, &nKey, &pVal, &nVal);
      rc = tdb_write(pDb2, pKey, nKey, pVal, nVal);
    }

    if( rc==0 ){
      int iKey1;
      int iKey2;
      void *pKey1; int nKey1;       /* Start key */
      void *pKey2; int nKey2;       /* Final key */

      iKey1 = testPrngValue((iFirst<<8) + (iLast<<16)) % nRow;
      iKey2 = testPrngValue((iLast<<8) + (iFirst<<16)) % nRow;
      testDatasourceEntry(pData, iKey1, &pKey2, &nKey1, 0, 0);
      pKey1 = testMalloc(nKey1+1);
      memcpy(pKey1, pKey2, nKey1+1);
      testDatasourceEntry(pData, iKey2, &pKey2, &nKey2, 0, 0);

      testScanCompare(pDb2, pDb, 0, 0, 0,         0, 0,         &rc);
      testScanCompare(pDb2, pDb, 0, 0, 0,         pKey2, nKey2, &rc);
      testScanCompare(pDb2, pDb, 0, pKey1, nKey1, 0, 0,         &rc);
      testScanCompare(pDb2, pDb, 0, pKey1, nKey1, pKey2, nKey2, &rc);
      testScanCompare(pDb2, pDb, 1, 0, 0,         0, 0,         &rc);
      testScanCompare(pDb2, pDb, 1, 0, 0,         pKey2, nKey2, &rc);
      testScanCompare(pDb2, pDb, 1, pKey1, nKey1, 0, 0,         &rc);
      testScanCompare(pDb2, pDb, 1, pKey1, nKey1, pKey2, nKey2, &rc);
      testFree(pKey1);
    }
    tdb_close(pDb2);
  }

  /* Test some lookups. */
  for(j=0; rc==0 && j<nLookupTest; j++){
    int iKey;                     /* Datasource key to test */
    void *pKey; int nKey;         /* Database key to query for */
    void *pVal; int nVal;         /* Expected result of query */

    if( nLookupTest>=nRow ){
      iKey = j;
    }else{
      iKey = testPrngValue(j + (iFirst<<8) + (iLast<<16)) % nRow;
    }

    testDatasourceEntry(pData, iKey, &pKey, &nKey, &pVal, &nVal);
    if( iFirst>iKey || iKey>iLast ){
      pVal = 0;
      nVal = -1;
    }

    testFetch(pDb, pKey, nKey, pVal, nVal, &rc);
  }

  *pRc = rc;
}

/*
** This function should be called during long running test cases to output
** the progress dots (...) to stdout.
*/
void testCaseProgress(int i, int n, int nDot, int *piDot){
  int iDot = *piDot;
  while( iDot < ( ((nDot*2+1) * i) / (n*2) ) ){
    printf(".");
    fflush(stdout);
    iDot++;
  }
  *piDot = iDot;
}

int testCaseNDot(void){ return 20; }

#if 0
static void printScanCb(
    void *pCtx, void *pKey, int nKey, void *pVal, int nVal
){
  printf("%s\n", (char *)pKey);
  fflush(stdout);
}
#endif

static void doDataTest1(
  const char *zSystem,            /* Database system to test */
  Datatest1 *p,                   /* Structure containing test parameters */
  int *pRc                        /* OUT: Error code */
){
  int i;
  int iDot;
  int rc = LSM_OK;
  Datasource *pData;
  TestDb *pDb;

  /* Start the test case, open a database and allocate the datasource. */
  pDb = testOpen(zSystem, 1, &rc);
  pData = testDatasourceNew(&p->defn);

  i = 0;
  iDot = 0;
  while( rc==LSM_OK && i<p->nRow ){

    /* Insert some data */
    testWriteDatasourceRange(pDb, pData, i, p->nVerify, &rc);
    i += p->nVerify;

    /* Check that the db content is correct. */
    testDbContents(pDb, pData, p->nRow, 0, i-1, p->nTest, p->bTestScan, &rc);

    /* Close and reopen the database. */
    testReopen(&pDb, &rc);

    /* Check that the db content is still correct. */
    testDbContents(pDb, pData, p->nRow, 0, i-1, p->nTest, p->bTestScan, &rc);

    /* Update the progress dots... */
    testCaseProgress(i, p->nRow, testCaseNDot()/2, &iDot);
  }

  i = 0;
  iDot = 0;
  while( rc==LSM_OK && i<p->nRow ){

    /* Delete some entries */
    testDeleteDatasourceRange(pDb, pData, i, p->nVerify, &rc);
    i += p->nVerify;

    /* Check that the db content is correct. */
    testDbContents(pDb, pData, p->nRow, i, p->nRow-1,p->nTest,p->bTestScan,&rc);

    /* Close and reopen the database. */
    testReopen(&pDb, &rc);

    /* Check that the db content is still correct. */
    testDbContents(pDb, pData, p->nRow, i, p->nRow-1,p->nTest,p->bTestScan,&rc);

    /* Update the progress dots... */
    testCaseProgress(i, p->nRow, testCaseNDot()/2, &iDot);
  }

  /* Free the datasource, close the database and finish the test case. */
  testDatasourceFree(pData);
  tdb_close(pDb);
  testCaseFinish(rc);
  *pRc = rc;
}


void test_data_1(
  const char *zSystem,            /* Database system name */
  const char *zPattern,           /* Run test cases that match this pattern */
  int *pRc                        /* IN/OUT: Error code */
){
  Datatest1 aTest[] = {
    { {DATA_RANDOM,     500,600,   1000,2000},     1000,  100,  10,  0},
    { {DATA_RANDOM,     20,25,     100,200},       1000,  250, 1000, 1},
    { {DATA_RANDOM,     8,10,      100,200},       1000,  250, 1000, 1},
    { {DATA_RANDOM,     8,10,      10,20},         1000,  250, 1000, 1},
    { {DATA_RANDOM,     8,10,      1000,2000},     1000,  250, 1000, 1},
    { {DATA_RANDOM,     8,100,     10000,20000},    100,   25,  100, 1},
    { {DATA_RANDOM,     80,100,    10,20},         1000,  250, 1000, 1},
    { {DATA_RANDOM,     5000,6000, 10,20},          100,   25,  100, 1},
    { {DATA_SEQUENTIAL, 5,10,      10,20},         1000,  250, 1000, 1},
    { {DATA_SEQUENTIAL, 5,10,      100,200},       1000,  250, 1000, 1},
    { {DATA_SEQUENTIAL, 5,10,      1000,2000},     1000,  250, 1000, 1},
    { {DATA_SEQUENTIAL, 5,100,     10000,20000},    100,   25,  100, 1},
    { {DATA_RANDOM,     10,10,     100,100},     100000, 1000,  100, 0},
    { {DATA_SEQUENTIAL, 10,10,     100,100},     100000, 1000,  100, 0},
  };

  int i;

  for(i=0; *pRc==LSM_OK && i<ArraySize(aTest); i++){
    char *zName = getName(zSystem, &aTest[i]);
    if( testCaseBegin(pRc, zPattern, "%s", zName) ){
      doDataTest1(zSystem, &aTest[i], pRc);
    }
    testFree(zName);
  }
}

void testCompareDb(
  Datasource *pData,
  int nData,
  int iSeed,
  TestDb *pControl,
  TestDb *pDb,
  int *pRc
){
  int i;

  static int nCall = 0;
  nCall++;

  testScanCompare(pControl, pDb, 0, 0, 0,         0, 0,         pRc);
  testScanCompare(pControl, pDb, 1, 0, 0,         0, 0,         pRc);

  if( *pRc==0 ){
    int iKey1;
    int iKey2;
    void *pKey1; int nKey1;       /* Start key */
    void *pKey2; int nKey2;       /* Final key */

    iKey1 = testPrngValue(iSeed) % nData;
    iKey2 = testPrngValue(iSeed+1) % nData;
    testDatasourceEntry(pData, iKey1, &pKey2, &nKey1, 0, 0);
    pKey1 = testMalloc(nKey1+1);
    memcpy(pKey1, pKey2, nKey1+1);
    testDatasourceEntry(pData, iKey2, &pKey2, &nKey2, 0, 0);

    testScanCompare(pControl, pDb, 0, 0, 0,         pKey2, nKey2, pRc);
    testScanCompare(pControl, pDb, 0, pKey1, nKey1, 0, 0,         pRc);
    testScanCompare(pControl, pDb, 0, pKey1, nKey1, pKey2, nKey2, pRc);
    testScanCompare(pControl, pDb, 1, 0, 0,         pKey2, nKey2, pRc);
    testScanCompare(pControl, pDb, 1, pKey1, nKey1, 0, 0,         pRc);
    testScanCompare(pControl, pDb, 1, pKey1, nKey1, pKey2, nKey2, pRc);
    testFree(pKey1);
  }

  for(i=0; i<nData && *pRc==0; i++){
    void *pKey; int nKey;
    testDatasourceEntry(pData, i, &pKey, &nKey, 0, 0);
    testFetchCompare(pControl, pDb, pKey, nKey, pRc);
  }
}

static void doDataTest2(
  const char *zSystem,            /* Database system to test */
  Datatest2 *p,                   /* Structure containing test parameters */
  int *pRc                        /* OUT: Error code */
){
  TestDb *pDb;
  TestDb *pControl;
  Datasource *pData;
  int i;
  int rc = LSM_OK;
  int iDot = 0;

  /* Start the test case, open a database and allocate the datasource. */
  pDb = testOpen(zSystem, 1, &rc);
  pData = testDatasourceNew(&p->defn);
  rc = testControlDb(&pControl);

  if( tdb_lsm(pDb) ){
    int nBuf = 32 * 1024 * 1024;
    lsm_config(tdb_lsm(pDb), LSM_CONFIG_AUTOFLUSH, &nBuf);
  }

  for(i=0; rc==0 && i<p->nIter; i++){
    void *pKey1; int nKey1;
    void *pKey2; int nKey2;
    int ii;
    int nRange = MIN(p->nIter*p->nWrite, p->nRange);

    for(ii=0; rc==0 && ii<p->nWrite; ii++){
      int iKey = (i*p->nWrite + ii) % p->nRange;
      testWriteDatasource(pControl, pData, iKey, &rc);
      testWriteDatasource(pDb, pData, iKey, &rc);
    }

    testDatasourceEntry(pData, i+1000000, &pKey1, &nKey1, 0, 0);
    pKey1 = testMallocCopy(pKey1, nKey1);
    testDatasourceEntry(pData, i+2000000, &pKey2, &nKey2, 0, 0);

    testDeleteRange(pDb, pKey1, nKey1, pKey2, nKey2, &rc);
    testDeleteRange(pControl, pKey1, nKey1, pKey2, nKey2, &rc);
    testFree(pKey1);

    testCompareDb(pData, nRange, i, pControl, pDb, &rc);
    testReopen(&pDb, &rc);
    testCompareDb(pData, nRange, i, pControl, pDb, &rc);

    /* Update the progress dots... */
    testCaseProgress(i, p->nIter, testCaseNDot(), &iDot);
  }

  testClose(&pDb);
  testClose(&pControl);
  testDatasourceFree(pData);
  testCaseFinish(rc);
  *pRc = rc;
}

static char *getName2(const char *zSystem, Datatest2 *pTest){
  char *zRet;
  char *zData;
  zData = testDatasourceName(&pTest->defn);
  zRet = testMallocPrintf("data2.%s.%s.%d.%d.%d", 
      zSystem, zData, pTest->nRange, pTest->nWrite, pTest->nIter
  );
  testFree(zData);
  return zRet;
}

void test_data_2(
  const char *zSystem,            /* Database system name */
  const char *zPattern,           /* Run test cases that match this pattern */
  int *pRc                        /* IN/OUT: Error code */
){
  Datatest2 aTest[] = {
      /* defn,                                 nRange, nWrite, nIter */
    { {DATA_RANDOM,     20,25,     100,200},   10000,  10,     50   },
    { {DATA_RANDOM,     20,25,     100,200},   10000,  200,    50   },
    { {DATA_RANDOM,     20,25,     100,200},   100,    10,     1000 },
    { {DATA_RANDOM,     20,25,     100,200},   100,    200,    50   },
  };

  int i;

  for(i=0; *pRc==LSM_OK && i<ArraySize(aTest); i++){
    char *zName = getName2(zSystem, &aTest[i]);
    if( testCaseBegin(pRc, zPattern, "%s", zName) ){
      doDataTest2(zSystem, &aTest[i], pRc);
    }
    testFree(zName);
  }
}

/*************************************************************************
** Test case data3.*
*/

typedef struct Datatest3 Datatest3;
struct Datatest3 {
  int nRange;                     /* Keys are between 1 and this value, incl. */
  int nIter;                      /* Number of iterations */
  int nWrite;                     /* Number of writes per iteration */
  int nDelete;                    /* Number of deletes per iteration */

  int nValMin;                    /* Minimum value size for writes */
  int nValMax;                    /* Maximum value size for writes */
};

void testPutU32(u8 *aBuf, u32 iVal){
  aBuf[0] = (iVal >> 24) & 0xFF;
  aBuf[1] = (iVal >> 16) & 0xFF;
  aBuf[2] = (iVal >>  8) & 0xFF;
  aBuf[3] = (iVal >>  0) & 0xFF;
}

void dt3PutKey(u8 *aBuf, int iKey){
  assert( iKey<100000 && iKey>=0 );
  sprintf((char *)aBuf, "%.5d", iKey);
}

static void doDataTest3(
  const char *zSystem,            /* Database system to test */
  Datatest3 *p,                   /* Structure containing test parameters */
  int *pRc                        /* OUT: Error code */
){
  int iDot = 0;
  int rc = *pRc;
  TestDb *pDb;
  u8 *abPresent;                  /* Array of boolean */
  char *aVal;                     /* Buffer to hold values */
  int i;
  u32 iSeq = 10;                  /* prng counter */

  abPresent = (u8 *)testMalloc(p->nRange+1);
  aVal = (char *)testMalloc(p->nValMax+1);
  pDb = testOpen(zSystem, 1, &rc);

  for(i=0; i<p->nIter && rc==0; i++){
    int ii;

    testCaseProgress(i, p->nIter, testCaseNDot(), &iDot);

    /* Perform nWrite inserts */
    for(ii=0; ii<p->nWrite; ii++){
      u8 aKey[6];
      u32 iKey;
      int nVal;

      iKey = (testPrngValue(iSeq++) % p->nRange) + 1;
      nVal = (testPrngValue(iSeq++) % (p->nValMax - p->nValMin)) + p->nValMin;
      testPrngString(testPrngValue(iSeq++), aVal, nVal);
      dt3PutKey(aKey, iKey);

      testWrite(pDb, aKey, sizeof(aKey)-1, aVal, nVal, &rc);
      abPresent[iKey] = 1;
    }

    /* Perform nDelete deletes */
    for(ii=0; ii<p->nDelete; ii++){
      u8 aKey1[6];
      u8 aKey2[6];
      u32 iKey;

      iKey = (testPrngValue(iSeq++) % p->nRange) + 1;
      dt3PutKey(aKey1, iKey-1);
      dt3PutKey(aKey2, iKey+1);

      testDeleteRange(pDb, aKey1, sizeof(aKey1)-1, aKey2, sizeof(aKey2)-1, &rc);
      abPresent[iKey] = 0;
    }

    testReopen(&pDb, &rc);

    for(ii=1; rc==0 && ii<=p->nRange; ii++){
      int nDbVal;
      void *pDbVal;
      u8 aKey[6];
      int dbrc;

      dt3PutKey(aKey, ii);
      dbrc = tdb_fetch(pDb, aKey, sizeof(aKey)-1, &pDbVal, &nDbVal);
      testCompareInt(0, dbrc, &rc);

      if( abPresent[ii] ){
        testCompareInt(1, (nDbVal>0), &rc);
      }else{
        testCompareInt(1, (nDbVal<0), &rc);
      }
    }
  }

  testClose(&pDb);
  testCaseFinish(rc);
  *pRc = rc;
}

static char *getName3(const char *zSystem, Datatest3 *p){
  return testMallocPrintf("data3.%s.%d.%d.%d.%d.(%d..%d)",
      zSystem, p->nRange, p->nIter, p->nWrite, p->nDelete, 
      p->nValMin, p->nValMax
  );
}

void test_data_3(
  const char *zSystem,            /* Database system name */
  const char *zPattern,           /* Run test cases that match this pattern */
  int *pRc                        /* IN/OUT: Error code */
){
  Datatest3 aTest[] = {
    /* nRange, nIter, nWrite, nDelete, nValMin, nValMax */
    {  100,    1000,  5,      5,       50,      100 },
    {  100,    1000,  2,      2,        5,       10 },
  };

  int i;

  for(i=0; *pRc==LSM_OK && i<ArraySize(aTest); i++){
    char *zName = getName3(zSystem, &aTest[i]);
    if( testCaseBegin(pRc, zPattern, "%s", zName) ){
      doDataTest3(zSystem, &aTest[i], pRc);
    }
    testFree(zName);
  }
}


Added ext/lsm1/lsm-test/lsmtest2.c.















































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487

/*
** This file contains tests related to recovery following application 
** and system crashes (power failures) while writing to the database.
*/

#include "lsmtest.h"

/*
** Structure used by testCksumDatabase() to accumulate checksum values in.
*/
typedef struct Cksum Cksum;
struct Cksum {
  int nRow;
  int cksum1;
  int cksum2;
};

/*
** tdb_scan() callback used by testCksumDatabase()
*/
static void scanCksumDb(
  void *pCtx, 
  void *pKey, int nKey,
  void *pVal, int nVal
){
  Cksum *p = (Cksum *)pCtx;
  int i;

  p->nRow++;
  for(i=0; i<nKey; i++){
    p->cksum1 += ((u8 *)pKey)[i];
    p->cksum2 += p->cksum1;
  }
  for(i=0; i<nVal; i++){
    p->cksum1 += ((u8 *)pVal)[i];
    p->cksum2 += p->cksum1;
  }
}

/*
** tdb_scan() callback used by testCountDatabase()
*/
static void scanCountDb(
  void *pCtx, 
  void *pKey, int nKey,
  void *pVal, int nVal
){
  Cksum *p = (Cksum *)pCtx;
  p->nRow++;

  unused_parameter(pKey);
  unused_parameter(nKey);
  unused_parameter(pVal);
  unused_parameter(nVal);
}


/*
** Iterate through the entire contents of database pDb. Write a checksum
** string based on the db contents into buffer zOut before returning. A
** checksum string is at most 29 (TEST_CKSUM_BYTES) bytes in size:
**
**    * 32-bit integer (10 bytes)
**    * 1 space        (1 byte)
**    * 32-bit hex     (8 bytes)
**    * 1 space        (1 byte)
**    * 32-bit hex     (8 bytes)
**    * nul-terminator (1 byte)
**
** The number of entries in the database is returned.
*/
int testCksumDatabase(
  TestDb *pDb,                    /* Database handle */
  char *zOut                      /* Buffer to write checksum to */
){
  Cksum cksum;
  memset(&cksum, 0, sizeof(Cksum));
  tdb_scan(pDb, (void *)&cksum, 0, 0, 0, 0, 0, scanCksumDb);
  sprintf(zOut, "%d %x %x", 
      cksum.nRow, (u32)cksum.cksum1, (u32)cksum.cksum2
  );
  assert( strlen(zOut)<TEST_CKSUM_BYTES );
  return cksum.nRow;
}

int testCountDatabase(TestDb *pDb){
  Cksum cksum;
  memset(&cksum, 0, sizeof(Cksum));
  tdb_scan(pDb, (void *)&cksum, 0, 0, 0, 0, 0, scanCountDb);
  return cksum.nRow;
}

/*
** This function is a no-op if *pRc is not 0 when it is called.
**
** Otherwise, the two nul-terminated strings z1 and z1 are compared. If
** they are the same, the function returns without doing anything. Otherwise,
** an error message is printed, *pRc is set to 1 and the test_failed()
** function called.
*/
void testCompareStr(const char *z1, const char *z2, int *pRc){
  if( *pRc==0 ){
    if( strcmp(z1, z2) ){
      testPrintError("testCompareStr: \"%s\" != \"%s\"\n", z1, z2);
      *pRc = 1;
      test_failed();
    }
  }
}

/*
** This function is a no-op if *pRc is not 0 when it is called.
**
** Otherwise, the two integers i1 and i2 are compared. If they are equal,
** the function returns without doing anything. Otherwise, an error message 
** is printed, *pRc is set to 1 and the test_failed() function called.
*/
void testCompareInt(int i1, int i2, int *pRc){
  if( *pRc==0 && i1!=i2 ){
    testPrintError("testCompareInt: %d != %d\n", i1, i2);
    *pRc = 1;
    test_failed();
  }
}

void testCaseStart(int *pRc, char *zFmt, ...){
  va_list ap;
  va_start(ap, zFmt);
  vprintf(zFmt, ap);
  printf(" ...");
  va_end(ap);
  *pRc = 0;
  fflush(stdout);
}

/*
** This function is a no-op if *pRc is non-zero when it is called. Zero
** is returned in this case.
**
** Otherwise, the zFmt (a printf style format string) and following arguments 
** are used to create a test case name. If zPattern is NULL or a glob pattern
** that matches the test case name, 1 is returned and the test case started.
** Otherwise, zero is returned and the test case does not start.
*/
int testCaseBegin(int *pRc, const char *zPattern, const char *zFmt, ...){
  int res = 0;
  if( *pRc==0 ){
    char *zTest;
    va_list ap;

    va_start(ap, zFmt);
    zTest = testMallocVPrintf(zFmt, ap);
    va_end(ap);
    if( zPattern==0 || testGlobMatch(zPattern, zTest) ){
      printf("%-50s ...", zTest);
      res = 1;
    }
    testFree(zTest);
    fflush(stdout);
  }

  return res;
}

void testCaseFinish(int rc){
  if( rc==0 ){
    printf("Ok\n");
  }else{
    printf("FAILED\n");
  }
  fflush(stdout);
}

void testCaseSkip(){
  printf("Skipped\n");
}

void testSetupSavedLsmdb(
  const char *zCfg,
  const char *zFile,
  Datasource *pData,
  int nRow,
  int *pRc
){
  if( *pRc==0 ){
    int rc;
    TestDb *pDb;
    rc = tdb_lsm_open(zCfg, zFile, 1, &pDb);
    if( rc==0 ){
      testWriteDatasourceRange(pDb, pData, 0, nRow, &rc);
      testClose(&pDb);
      if( rc==0 ) testSaveDb(zFile, "log");
    }
    *pRc = rc;
  }
}

/*
** This function is a no-op if *pRc is non-zero when it is called.
**
** Open the LSM database identified by zFile and compute its checksum
** (a string, as returned by testCksumDatabase()). If the checksum is
** identical to zExpect1 or, if it is not NULL, zExpect2, the test passes.
** Otherwise, print an error message and set *pRc to 1.
*/
static void testCompareCksumLsmdb(
  const char *zFile,              /* Path to LSM database */
  int bCompress,                  /* True if db is compressed */
  const char *zExpect1,           /* Expected checksum 1 */
  const char *zExpect2,           /* Expected checksum 2 (or NULL) */
  int *pRc                        /* IN/OUT: Test case error code */
){
  if( *pRc==0 ){
    char zCksum[TEST_CKSUM_BYTES];
    TestDb *pDb;

    *pRc = tdb_lsm_open((bCompress?"compression=1 mmap=0":""), zFile, 0, &pDb);
    testCksumDatabase(pDb, zCksum);
    testClose(&pDb);

    if( *pRc==0 ){
      int r1 = 0;
      int r2 = -1;

      r1 = strcmp(zCksum, zExpect1);
      if( zExpect2 ) r2 = strcmp(zCksum, zExpect2);
      if( r1 && r2 ){
        if( zExpect2 ){
          testPrintError("testCompareCksumLsmdb: \"%s\" != (\"%s\" OR \"%s\")",
              zCksum, zExpect1, zExpect2
          );
        }else{
          testPrintError("testCompareCksumLsmdb: \"%s\" != \"%s\"",
              zCksum, zExpect1
          );
        }
        *pRc = 1;
        test_failed();
      }
    }
  }
}

static void testCompareCksumBtdb(
  const char *zFile,              /* Path to LSM database */
  const char *zExpect1,           /* Expected checksum 1 */
  const char *zExpect2,           /* Expected checksum 2 (or NULL) */
  int *pRc                        /* IN/OUT: Test case error code */
){
  if( *pRc==0 ){
    char zCksum[TEST_CKSUM_BYTES];
    TestDb *pDb;

    *pRc = tdb_open("bt", zFile, 0, &pDb);
    testCksumDatabase(pDb, zCksum);
    testClose(&pDb);

    if( *pRc==0 ){
      int r1 = 0;
      int r2 = -1;

      r1 = strcmp(zCksum, zExpect1);
      if( zExpect2 ) r2 = strcmp(zCksum, zExpect2);
      if( r1 && r2 ){
        if( zExpect2 ){
          testPrintError("testCompareCksumLsmdb: \"%s\" != (\"%s\" OR \"%s\")",
              zCksum, zExpect1, zExpect2
          );
        }else{
          testPrintError("testCompareCksumLsmdb: \"%s\" != \"%s\"",
              zCksum, zExpect1
          );
        }
        *pRc = 1;
        test_failed();
      }
    }
  }
}

/* Above this point are reusable test routines. Not clear that they
** should really be in this file.
*************************************************************************/

/*
** This test verifies that if a system crash occurs while doing merge work
** on the db, no data is lost.
*/
static void crash_test1(int bCompress, int *pRc){
  const char *DBNAME = "testdb.lsm";
  const DatasourceDefn defn = {TEST_DATASOURCE_RANDOM, 12, 16, 200, 200};

  const int nRow = 5000;          /* Database size */
  const int nIter = 200;          /* Number of test iterations */
  const int nWork = 20;           /* Maximum lsm_work() calls per iteration */
  const int nPage = 15;           /* Pages per lsm_work call */

  int i;
  int iDot = 0;
  Datasource *pData;
  CksumDb *pCksumDb;
  TestDb *pDb;
  char *zCfg;

  const char *azConfig[2] = {
    "page_size=1024 block_size=65536 autoflush=16384 safety=2 mmap=0", 
    "page_size=1024 block_size=65536 autoflush=16384 safety=2 "
    " compression=1 mmap=0"
  };
  assert( bCompress==0 || bCompress==1 );

  /* Allocate datasource. And calculate the expected checksums. */
  pData = testDatasourceNew(&defn);
  pCksumDb = testCksumArrayNew(pData, nRow, nRow, 1);

  /* Setup and save the initial database. */

  zCfg = testMallocPrintf("%s automerge=7", azConfig[bCompress]);
  testSetupSavedLsmdb(zCfg, DBNAME, pData, 5000, pRc);
  testFree(zCfg);

  for(i=0; i<nIter && *pRc==0; i++){
    int iWork;
    int testrc = 0;

    testCaseProgress(i, nIter, testCaseNDot(), &iDot);

    /* Restore and open the database. */
    testRestoreDb(DBNAME, "log");
    testrc = tdb_lsm_open(azConfig[bCompress], DBNAME, 0, &pDb);
    assert( testrc==0 );

    /* Call lsm_work() on the db */
    tdb_lsm_prepare_sync_crash(pDb, 1 + (i%(nWork*2)));
    for(iWork=0; testrc==0 && iWork<nWork; iWork++){
      int nWrite = 0;
      lsm_db *db = tdb_lsm(pDb);
      testrc = lsm_work(db, 0, nPage, &nWrite);
      /* assert( testrc!=0 || nWrite>0 ); */
      if( testrc==0 ) testrc = lsm_checkpoint(db, 0);
    }
    tdb_close(pDb);

    /* Check that the database content is still correct */
    testCompareCksumLsmdb(DBNAME, 
        bCompress, testCksumArrayGet(pCksumDb, nRow), 0, pRc);
  }

  testCksumArrayFree(pCksumDb);
  testDatasourceFree(pData);
}

/*
** This test verifies that if a system crash occurs while committing a
** transaction to the log file, no earlier transactions are lost or damaged.
*/
static void crash_test2(int bCompress, int *pRc){
  const char *DBNAME = "testdb.lsm";
  const DatasourceDefn defn = {TEST_DATASOURCE_RANDOM, 12, 16, 1000, 1000};

  const int nIter = 200;
  const int nInsert = 20;

  int i;
  int iDot = 0;
  Datasource *pData;
  CksumDb *pCksumDb;
  TestDb *pDb;

  /* Allocate datasource. And calculate the expected checksums. */
  pData = testDatasourceNew(&defn);
  pCksumDb = testCksumArrayNew(pData, 100, 100+nInsert, 1);

  /* Setup and save the initial database. */
  testSetupSavedLsmdb("", DBNAME, pData, 100, pRc);

  for(i=0; i<nIter && *pRc==0; i++){
    int iIns;
    int testrc = 0;

    testCaseProgress(i, nIter, testCaseNDot(), &iDot);

    /* Restore and open the database. */
    testRestoreDb(DBNAME, "log");
    testrc = tdb_lsm_open("safety=2", DBNAME, 0, &pDb);
    assert( testrc==0 );

    /* Insert nInsert records into the database. Crash midway through. */
    tdb_lsm_prepare_sync_crash(pDb, 1 + (i%(nInsert+2)));
    for(iIns=0; iIns<nInsert; iIns++){
      void *pKey; int nKey;
      void *pVal; int nVal;

      testDatasourceEntry(pData, 100+iIns, &pKey, &nKey, &pVal, &nVal);
      testrc = tdb_write(pDb, pKey, nKey, pVal, nVal);
      if( testrc ) break;
    }
    tdb_close(pDb);

    /* Check that no data was lost when the system crashed. */
    testCompareCksumLsmdb(DBNAME, bCompress,
      testCksumArrayGet(pCksumDb, 100 + iIns),
      testCksumArrayGet(pCksumDb, 100 + iIns + 1),
      pRc
    );
  }

  testDatasourceFree(pData);
  testCksumArrayFree(pCksumDb);
}


/*
** This test verifies that if a system crash occurs when checkpointing
** the database, data is not lost (assuming that any writes not synced
** to the db have been synced into the log file).
*/
static void crash_test3(int bCompress, int *pRc){
  const char *DBNAME = "testdb.lsm";
  const int nIter = 100;
  const DatasourceDefn defn = {TEST_DATASOURCE_RANDOM, 12, 16, 1000, 1000};

  int i;
  int iDot = 0;
  Datasource *pData;
  CksumDb *pCksumDb;
  TestDb *pDb;

  /* Allocate datasource. And calculate the expected checksums. */
  pData = testDatasourceNew(&defn);
  pCksumDb = testCksumArrayNew(pData, 110, 150, 10);

  /* Setup and save the initial database. */
  testSetupSavedLsmdb("", DBNAME, pData, 100, pRc);

  for(i=0; i<nIter && *pRc==0; i++){
    int iOpen;
    testCaseProgress(i, nIter, testCaseNDot(), &iDot);
    testRestoreDb(DBNAME, "log");

    for(iOpen=0; iOpen<5; iOpen++){
      /* Open the database. Insert 10 more records. */
      pDb = testOpen("lsm", 0, pRc);
      testWriteDatasourceRange(pDb, pData, 100+iOpen*10, 10, pRc);

      /* Schedule a crash simulation then close the db. */
      tdb_lsm_prepare_sync_crash(pDb, 1 + (i%2));
      tdb_close(pDb);

      /* Open the database and check that the crash did not cause any
      ** data loss.  */
      testCompareCksumLsmdb(DBNAME, bCompress,
        testCksumArrayGet(pCksumDb, 110 + iOpen*10), 0,
        pRc
      );
    }
  }

  testDatasourceFree(pData);
  testCksumArrayFree(pCksumDb);
}

void do_crash_test(const char *zPattern, int *pRc){
  struct Test {
    const char *zTest;
    void (*x)(int, int *);
    int bCompress;
  } aTest [] = {
    { "crash.lsm.1",     crash_test1, 0 },
#ifdef HAVE_ZLIB
    { "crash.lsm_zip.1", crash_test1, 1 },
#endif
    { "crash.lsm.2",     crash_test2, 0 },
    { "crash.lsm.3",     crash_test3, 0 },
  };
  int i;

  for(i=0; *pRc==LSM_OK && i<ArraySize(aTest); i++){
    struct Test *p = &aTest[i];
    if( testCaseBegin(pRc, zPattern, "%s", p->zTest) ){
      p->x(p->bCompress, pRc);
      testCaseFinish(*pRc);
    }
  }
}

Added ext/lsm1/lsm-test/lsmtest3.c.





























































































































































































































































































































































































































































































>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238


/*
** This file contains tests related to the explicit rollback of database
** transactions and sub-transactions.
*/


/*
** Repeat 2000 times (until the db contains 100,000 entries):
**
**   1. Open a transaction and insert 500 rows, opening a nested 
**      sub-transaction each 100 rows.
**
**   2. Roll back to each sub-transaction savepoint. Check the database
**      checksum looks Ok.
**
**   3. Every second iteration, roll back the main transaction. Check the
**      db checksum is correct. Every other iteration, commit the main
**      transaction (increasing the size of the db by 100 rows).
*/


#include "lsmtest.h"

struct CksumDb {
  int nFirst;
  int nLast;
  int nStep;
  char **azCksum;
};

CksumDb *testCksumArrayNew(
  Datasource *pData, 
  int nFirst, 
  int nLast, 
  int nStep
){
  TestDb *pDb;
  CksumDb *pRet;
  int i;
  int nEntry;
  int rc = 0;

  assert( nLast>=nFirst && ((nLast-nFirst)%nStep)==0 );
 
  pRet = malloc(sizeof(CksumDb));
  memset(pRet, 0, sizeof(CksumDb));
  pRet->nFirst = nFirst;
  pRet->nLast = nLast;
  pRet->nStep = nStep;
  nEntry = 1 + ((nLast - nFirst) / nStep);

  /* Allocate space so that azCksum is an array of nEntry pointers to
  ** buffers each TEST_CKSUM_BYTES in size.  */
  pRet->azCksum = (char **)malloc(nEntry * (sizeof(char *) + TEST_CKSUM_BYTES));
  for(i=0; i<nEntry; i++){
    char *pStart = (char *)(&pRet->azCksum[nEntry]);
    pRet->azCksum[i] = &pStart[i * TEST_CKSUM_BYTES];
  }

  tdb_open("lsm", "tempdb.lsm", 1, &pDb);
  testWriteDatasourceRange(pDb, pData, 0, nFirst, &rc);
  for(i=0; i<nEntry; i++){
    testCksumDatabase(pDb, pRet->azCksum[i]);
    if( i==nEntry ) break;
    testWriteDatasourceRange(pDb, pData, nFirst+i*nStep, nStep, &rc);
  }

  tdb_close(pDb);

  return pRet;
}

char *testCksumArrayGet(CksumDb *p, int nRow){
  int i;
  assert( nRow>=p->nFirst );
  assert( nRow<=p->nLast );
  assert( ((nRow-p->nFirst) % p->nStep)==0 );

  i = (nRow - p->nFirst) / p->nStep;
  return p->azCksum[i];
}

void testCksumArrayFree(CksumDb *p){
  free(p->azCksum);
  memset(p, 0x55, sizeof(*p));
  free(p);
}

/* End of CksumDb code.
**************************************************************************/

/*
** Test utility function. Write key-value pair $i from datasource pData 
** into database pDb.
*/
void testWriteDatasource(TestDb *pDb, Datasource *pData, int i, int *pRc){
  void *pKey; int nKey;
  void *pVal; int nVal;
  testDatasourceEntry(pData, i, &pKey, &nKey, &pVal, &nVal);
  testWrite(pDb, pKey, nKey, pVal, nVal, pRc);
}

/*
** Test utility function. Delete datasource pData key $i from database pDb.
*/
void testDeleteDatasource(TestDb *pDb, Datasource *pData, int i, int *pRc){
  void *pKey; int nKey;
  testDatasourceEntry(pData, i, &pKey, &nKey, 0, 0);
  testDelete(pDb, pKey, nKey, pRc);
}

/*
** This function inserts nWrite key/value pairs into database pDb - the
** nWrite key value pairs starting at iFirst from data source pData.
*/
void testWriteDatasourceRange(
  TestDb *pDb,                    /* Database to write to */
  Datasource *pData,              /* Data source to read values from */
  int iFirst,                     /* Index of first key/value pair */
  int nWrite,                     /* Number of key/value pairs to write */
  int *pRc                        /* IN/OUT: Error code */
){
  int i;
  for(i=0; i<nWrite; i++){
    testWriteDatasource(pDb, pData, iFirst+i, pRc);
  }
}

void testDeleteDatasourceRange(
  TestDb *pDb,                    /* Database to write to */
  Datasource *pData,              /* Data source to read keys from */
  int iFirst,                     /* Index of first key */
  int nWrite,                     /* Number of keys to delete */
  int *pRc                        /* IN/OUT: Error code */
){
  int i;
  for(i=0; i<nWrite; i++){
    testDeleteDatasource(pDb, pData, iFirst+i, pRc);
  }
}

static char *getName(const char *zSystem){ 
  char *zRet; 
  zRet = testMallocPrintf("rollback.%s", zSystem);
  return zRet;
}

static int rollback_test_1(
  const char *zSystem,
  Datasource *pData
){
  const int nRepeat = 100;

  TestDb *pDb;
  int rc;
  int i;
  CksumDb *pCksum;
  char *zName;

  zName = getName(zSystem);
  testCaseStart(&rc, zName);
  testFree(zName);

  pCksum = testCksumArrayNew(pData, 0, nRepeat*100, 100);
  pDb = 0;
  rc = tdb_open(zSystem, 0, 1, &pDb);
  if( pDb && tdb_transaction_support(pDb)==0 ){
    testCaseSkip();
    goto skip_rollback_test;
  }

  for(i=0; i<nRepeat && rc==0; i++){
    char zCksum[TEST_CKSUM_BYTES];
    int nCurrent = (((i+1)/2) * 100);
    int nDbRow;
    int iTrans;

    /* Check that the database is the expected size. */
    nDbRow = testCountDatabase(pDb);
    testCompareInt(nCurrent, nDbRow, &rc);

    for(iTrans=2; iTrans<=6 && rc==0; iTrans++){
      tdb_begin(pDb, iTrans);
      testWriteDatasourceRange(pDb, pData, nCurrent, 100, &rc);
      nCurrent += 100;
    }

    testCksumDatabase(pDb, zCksum);
    testCompareStr(zCksum, testCksumArrayGet(pCksum, nCurrent), &rc);

    for(iTrans=6; iTrans>2 && rc==0; iTrans--){
      tdb_rollback(pDb, iTrans);
      nCurrent -= 100;
      testCksumDatabase(pDb, zCksum);
      testCompareStr(zCksum, testCksumArrayGet(pCksum, nCurrent), &rc);
    }

    if( i%2 ){
      tdb_rollback(pDb, 0);
      nCurrent -= 100;
      testCksumDatabase(pDb, zCksum);
      testCompareStr(zCksum, testCksumArrayGet(pCksum, nCurrent), &rc);
    }else{
      tdb_commit(pDb, 0);
    }
  }
  testCaseFinish(rc);

 skip_rollback_test:
  tdb_close(pDb);
  testCksumArrayFree(pCksum);
  return rc;
}

void test_rollback(
  const char *zSystem, 
  const char *zPattern, 
  int *pRc
){
  if( *pRc==0 ){
    int bRun = 1;

    if( zPattern ){
      char *zName = getName(zSystem);
      bRun = testGlobMatch(zPattern, zName);
      testFree(zName);
    }

    if( bRun ){
      DatasourceDefn defn = { TEST_DATASOURCE_RANDOM, 10, 15, 50, 100 };
      Datasource *pData = testDatasourceNew(&defn);
      *pRc = rollback_test_1(zSystem, pData);
      testDatasourceFree(pData);
    }
  }
}

Added ext/lsm1/lsm-test/lsmtest4.c.































































































































































































































































>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127

/*
** This file contains test cases involving multiple database clients.
*/

#include "lsmtest.h"

/*
** The following code implements test cases "mc1.*".
**
** This test case uses one writer and $nReader readers. All connections
** are driven by a single thread. All connections are opened at the start
** of the test and remain open until the test is finished.
**
** The test consists of $nStep steps. Each step the following is performed:
**
**   1. The writer inserts $nWriteStep records into the db.
**
**   2. The writer checks that the contents of the db are as expected.
**
**   3. Each reader that currently has an open read transaction also checks
**      that the contents of the db are as expected (according to the snapshot
**      the read transaction is reading - see below).
**
** After step 1, reader 1 opens a read transaction. After step 2, reader
** 2 opens a read transaction, and so on. At step ($nReader+1), reader 1
** closes the current read transaction and opens a new one. And so on.
** The result is that at step N (for N > $nReader), there exists a reader
** with an open read transaction reading the snapshot committed following
** steps (N-$nReader-1) to N. 
*/
typedef struct Mctest Mctest;
struct Mctest {
  DatasourceDefn defn;            /* Datasource to use */
  int nStep;                      /* Total number of steps in test */
  int nWriteStep;                 /* Number of rows to insert each step */
  int nReader;                    /* Number of read connections */
};
static void do_mc_test(
  const char *zSystem,            /* Database system to test */
  Mctest *pTest,
  int *pRc                        /* IN/OUT: return code */
){
  const int nDomain = pTest->nStep * pTest->nWriteStep;
  Datasource *pData;              /* Source of data */
  TestDb *pDb;                    /* First database connection (writer) */
  int iReader;                    /* Used to iterate through aReader */
  int iStep;                      /* Current step in test */
  int iDot = 0;                   /* Current step in test */

  /* Array of reader connections */
  struct Reader {
    TestDb *pDb;                  /* Connection handle */
    int iLast;                    /* Current snapshot contains keys 0..iLast */
  } *aReader;

  /* Create a data source */
  pData = testDatasourceNew(&pTest->defn);

  /* Open the writer connection */
  pDb = testOpen(zSystem, 1, pRc);

  /* Allocate aReader */
  aReader = (struct Reader *)testMalloc(sizeof(aReader[0]) * pTest->nReader);
  for(iReader=0; iReader<pTest->nReader; iReader++){
    aReader[iReader].pDb = testOpen(zSystem, 0, pRc);
  }

  for(iStep=0; iStep<pTest->nStep; iStep++){
    int iLast;
    int iBegin;                   /* Start read trans using aReader[iBegin] */

    /* Insert nWriteStep more records into the database */
    int iFirst = iStep*pTest->nWriteStep;
    testWriteDatasourceRange(pDb, pData, iFirst, pTest->nWriteStep, pRc);

    /* Check that the db is Ok according to the writer */
    iLast = (iStep+1) * pTest->nWriteStep - 1;
    testDbContents(pDb, pData, nDomain, 0, iLast, iLast, 1, pRc);

    /* Have reader (iStep % nReader) open a read transaction here. */
    iBegin = (iStep % pTest->nReader);
    if( iBegin<iStep ) tdb_commit(aReader[iBegin].pDb, 0);
    tdb_begin(aReader[iBegin].pDb, 1);
    aReader[iBegin].iLast = iLast;

    /* Check that the db is Ok for each open reader */
    for(iReader=0; iReader<pTest->nReader && aReader[iReader].iLast; iReader++){
      iLast = aReader[iReader].iLast;
      testDbContents(
          aReader[iReader].pDb, pData, nDomain, 0, iLast, iLast, 1, pRc
      );
    }

    /* Report progress */
    testCaseProgress(iStep, pTest->nStep, testCaseNDot(), &iDot);
  }

  /* Close all readers */
  for(iReader=0; iReader<pTest->nReader; iReader++){
    testClose(&aReader[iReader].pDb);
  }
  testFree(aReader);

  /* Close the writer-connection and free the datasource */
  testClose(&pDb);
  testDatasourceFree(pData);
}


void test_mc(
  const char *zSystem,            /* Database system name */
  const char *zPattern,           /* Run test cases that match this pattern */
  int *pRc                        /* IN/OUT: Error code */
){
  int i;
  Mctest aTest[] = {
    { { TEST_DATASOURCE_RANDOM, 10,10, 100,100 }, 100, 10, 5 },
  };

  for(i=0; i<ArraySize(aTest); i++){
    if( testCaseBegin(pRc, zPattern, "mc1.%s.%d", zSystem, i) ){
      do_mc_test(zSystem, &aTest[i], pRc);
      testCaseFinish(*pRc);
    }
  }
}

Added ext/lsm1/lsm-test/lsmtest5.c.



















































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633

/*
** This file is broken into three semi-autonomous parts:
**
**   1. The database functions.
**   2. The thread wrappers.
**   3. The implementation of the mt1.* tests.
*/

/*************************************************************************
** DATABASE CONTENTS:
**
**   The database contains up to N key/value pairs, where N is some large 
**   number (say 10,000,000). Keys are integer values between 0 and (N-1).
**   The value associated with each key is a pseudo-random blob of data.
**
**   Key/value pair keys are encoded as the two bytes "k." followed by a 
**   10-digit decimal number. i.e. key 45 -> "k.0000000045".
**
**   As well as the key/value pairs, the database also contains checksum 
**   entries. The checksums form a hierarchy - for every F key/value
**   entries there is one level 1 checksum. And for each F level 1 checksums
**   there is one level 2 checksum. And so on.
**
**   Checksum keys are encoded as the two byte "c." followed by the 
**   checksum level, followed by a 10 digit decimal number containing
**   the value of the first key that contributes to the checksum value.
**   For example, assuming F==10, the level 1 checksum that spans keys
**   10 to 19 is "c.1.0000000010".
**
**   Clients may perform one of two operations on the database: a read
**   or a write.
** 
** READ OPERATIONS:
**
**   A read operation scans a range of F key/value pairs. It computes
**   the expected checksum and then compares the computed value to the
**   actual value stored in the level 1 checksum entry. It then scans 
**   the group of F level 1 checksums, and compares the computed checksum 
**   to the associated level 2 checksum value, and so on until the 
**   highest level checksum value has been verified.
**
**   If a checksum ever fails to match the expected value, the test 
**   has failed.
**
** WRITE OPERATIONS:
**
**   A write operation involves writing (possibly clobbering) a single
**   key/value pair. The associated level 1 checksum is then recalculated
**   updated. Then the level 2 checksum, and so on until the highest
**   level checksum has been modified.
**
**   All updates occur inside a single transaction.
**
** INTERFACE:
**
**   The interface used by test cases to read and write the db consists
**   of type DbParameters and the following functions:
**
**       dbReadOperation()
**       dbWriteOperation()
*/

#include "lsmtest.h"

typedef struct DbParameters DbParameters;
struct DbParameters {
  int nFanout;                    /* Checksum fanout (F) */
  int nKey;                       /* Size of key space (N) */
};

#define DB_KEY_BYTES          (2+5+10+1)

/*
** Argument aBuf[] must point to a buffer at least DB_KEY_BYTES in size.
** This function populates the buffer with a nul-terminated key string 
** corresponding to key iKey.
*/
static void dbFormatKey(
  DbParameters *pParam,
  int iLevel,
  int iKey,                       /* Key value */
  char *aBuf                      /* Write key string here */
){
  if( iLevel==0 ){
    snprintf(aBuf, DB_KEY_BYTES, "k.%.10d", iKey);
  }else{
    int f = 1;
    int i;
    for(i=0; i<iLevel; i++) f = f * pParam->nFanout;
    snprintf(aBuf, DB_KEY_BYTES, "c.%d.%.10d", iLevel, f*(iKey/f));
  }
}

/*
** Argument aBuf[] must point to a buffer at least DB_KEY_BYTES in size.
** This function populates the buffer with the string representation of
** checksum value iVal.
*/
static void dbFormatCksumValue(u32 iVal, char *aBuf){
  snprintf(aBuf, DB_KEY_BYTES, "%.10u", iVal);
}

/*
** Return the highest level of checksum in the database described
** by *pParam.
*/
static int dbMaxLevel(DbParameters *pParam){
  int iMax;
  int n = 1;
  for(iMax=0; n<pParam->nKey; iMax++){
    n = n * pParam->nFanout;
  }
  return iMax;
}

static void dbCksum(
  void *pCtx,                     /* IN/OUT: Pointer to u32 containing cksum */
  void *pKey, int nKey,           /* Database key. Unused. */
  void *pVal, int nVal            /* Database value. Checksum this. */
){
  u8 *aVal = (u8 *)pVal;
  u32 *pCksum = (u32 *)pCtx;
  u32 cksum = *pCksum;
  int i;

  unused_parameter(pKey);
  unused_parameter(nKey);

  for(i=0; i<nVal; i++){
    cksum += (cksum<<3) + (int)aVal[i];
  }

  *pCksum = cksum;
}

/*
** Compute the value of the checksum stored on level iLevel that contains
** data from key iKey by scanning the pParam->nFanout entries at level 
** iLevel-1.
*/
static u32 dbComputeCksum(
  DbParameters *pParam,           /* Database parameters */
  TestDb *pDb,                    /* Database connection handle */
  int iLevel,                     /* Level of checksum to compute */
  int iKey,                       /* Compute checksum for this key */
  int *pRc                        /* IN/OUT: Error code */
){
  u32 cksum = 0;
  if( *pRc==0 ){
    int nFirst;
    int nLast;
    int iFirst = 0;
    int iLast = 0;
    int i;
    int f = 1;
    char zFirst[DB_KEY_BYTES];
    char zLast[DB_KEY_BYTES];

    assert( iLevel>=1 );
    for(i=0; i<iLevel; i++) f = f * pParam->nFanout;

    iFirst = f*(iKey/f);
    iLast = iFirst + f - 1;
    dbFormatKey(pParam, iLevel-1, iFirst, zFirst);
    dbFormatKey(pParam, iLevel-1, iLast, zLast);
    nFirst = strlen(zFirst);
    nLast = strlen(zLast);

    *pRc = tdb_scan(pDb, (u32*)&cksum, 0, zFirst, nFirst, zLast, nLast,dbCksum);
  }

  return cksum;
}

static void dbReadOperation(
  DbParameters *pParam,           /* Database parameters */
  TestDb *pDb,                    /* Database connection handle */
  void (*xDelay)(void *),
  void *pDelayCtx,
  int iKey,                       /* Key to read */
  int *pRc                        /* IN/OUT: Error code */
){
  const int iMax = dbMaxLevel(pParam);
  int i;

  if( tdb_transaction_support(pDb) ) testBegin(pDb, 1, pRc);
  for(i=1; *pRc==0 && i<=iMax; i++){
    char zCksum[DB_KEY_BYTES];
    char zKey[DB_KEY_BYTES];
    u32 iCksum = 0;

    iCksum = dbComputeCksum(pParam, pDb, i, iKey, pRc);
    if( iCksum ){
      if( xDelay && i==1 ) xDelay(pDelayCtx);
      dbFormatCksumValue(iCksum, zCksum);
      dbFormatKey(pParam, i, iKey, zKey);
      testFetchStr(pDb, zKey, zCksum, pRc);
    }
  }
  if( tdb_transaction_support(pDb) ) testCommit(pDb, 0, pRc);
}

static int dbWriteOperation(
  DbParameters *pParam,           /* Database parameters */
  TestDb *pDb,                    /* Database connection handle */
  int iKey,                       /* Key to write to */
  const char *zValue,             /* Nul-terminated value to write */
  int *pRc                        /* IN/OUT: Error code */
){
  const int iMax = dbMaxLevel(pParam);
  char zKey[DB_KEY_BYTES];
  int i;
  int rc;

  assert( iKey>=0 && iKey<pParam->nKey );
  dbFormatKey(pParam, 0, iKey, zKey);

  /* Open a write transaction. This may fail - SQLITE4_BUSY */
  if( *pRc==0 && tdb_transaction_support(pDb) ){
    rc = tdb_begin(pDb, 2);
    if( rc==5 ) return 0;
    *pRc = rc;
  }

  testWriteStr(pDb, zKey, zValue, pRc);
  for(i=1; i<=iMax; i++){
    char zCksum[DB_KEY_BYTES];
    u32 iCksum = 0;

    iCksum = dbComputeCksum(pParam, pDb, i, iKey, pRc);
    dbFormatCksumValue(iCksum, zCksum);
    dbFormatKey(pParam, i, iKey, zKey);
    testWriteStr(pDb, zKey, zCksum, pRc);
  }
  if( tdb_transaction_support(pDb) ) testCommit(pDb, 0, pRc);
  return 1;
}

/*************************************************************************
** The following block contains testXXX() functions that implement a
** wrapper around the systems native multi-thread support. There are no
** synchronization primitives - just functions to launch and join 
** threads. Wrapper functions are:
**
**    testThreadSupport()
**
**    testThreadInit()
**    testThreadShutdown()
**    testThreadLaunch()
**    testThreadWait()
**
**    testThreadSetHalt()
**    testThreadGetHalt()
**    testThreadSetResult()
**    testThreadGetResult()
**
**    testThreadEnterMutex()
**    testThreadLeaveMutex()
*/
typedef struct ThreadSet ThreadSet;
#ifdef LSM_MUTEX_PTHREADS

#include <pthread.h>
#include <unistd.h>

typedef struct Thread Thread;
struct Thread {
  int rc;
  char *zMsg;
  pthread_t id;
  void (*xMain)(ThreadSet *, int, void *);
  void *pCtx;
  ThreadSet *pThreadSet;
};

struct ThreadSet {
  int bHalt;                      /* Halt flag */
  int nThread;                    /* Number of threads */
  Thread *aThread;                /* Array of Thread structures */
  pthread_mutex_t mutex;          /* Mutex used for cheating */
};

/*
** Return true if this build supports threads, or false otherwise. If
** this function returns false, no other testThreadXXX() functions should
** be called.
*/
static int testThreadSupport(){ return 1; }

/*
** Allocate and return a thread-set handle with enough space allocated
** to handle up to nMax threads. Each call to this function should be
** matched by a call to testThreadShutdown() to delete the object.
*/
static ThreadSet *testThreadInit(int nMax){
  int nByte;                      /* Total space to allocate */
  ThreadSet *p;                   /* Return value */

  nByte = sizeof(ThreadSet) + sizeof(struct Thread) * nMax;
  p = (ThreadSet *)testMalloc(nByte);
  p->nThread = nMax;
  p->aThread = (Thread *)&p[1];
  pthread_mutex_init(&p->mutex, 0);

  return p;
}

/*
** Delete a thread-set object and release all resources held by it.
*/
static void testThreadShutdown(ThreadSet *p){
  int i;
  for(i=0; i<p->nThread; i++){
    testFree(p->aThread[i].zMsg);
  }
  pthread_mutex_destroy(&p->mutex);
  testFree(p);
}

static void *ttMain(void *pArg){
  Thread *pThread = (Thread *)pArg;
  int iThread;
  iThread = (pThread - pThread->pThreadSet->aThread);
  pThread->xMain(pThread->pThreadSet, iThread, pThread->pCtx);
  return 0;
}

/*
** Launch a new thread.
*/
static int testThreadLaunch(
  ThreadSet *p,
  int iThread,
  void (*xMain)(ThreadSet *, int, void *),
  void *pCtx
){
  int rc;
  Thread *pThread;

  assert( iThread>=0 && iThread<p->nThread );

  pThread = &p->aThread[iThread];
  assert( pThread->pThreadSet==0 );
  pThread->xMain = xMain;
  pThread->pCtx = pCtx;
  pThread->pThreadSet = p;
  rc = pthread_create(&pThread->id, 0, ttMain, (void *)pThread);

  return rc;
}

/*
** Set the thread-set "halt" flag.
*/
static void testThreadSetHalt(ThreadSet *pThreadSet){
  pThreadSet->bHalt = 1;
}

/*
** Return the current value of the thread-set "halt" flag.
*/
static int testThreadGetHalt(ThreadSet *pThreadSet){
  return pThreadSet->bHalt;
}

static void testThreadSleep(ThreadSet *pThreadSet, int nMs){
  int nRem = nMs;
  while( nRem>0 && testThreadGetHalt(pThreadSet)==0 ){
    usleep(50000);
    nRem -= 50;
  }
}

/*
** Wait for all threads launched to finish before returning. If nMs
** is greater than zero, set the "halt" flag to tell all threads
** to halt after waiting nMs milliseconds.
*/
static void testThreadWait(ThreadSet *pThreadSet, int nMs){
  int i;

  testThreadSleep(pThreadSet, nMs);
  testThreadSetHalt(pThreadSet);
  for(i=0; i<pThreadSet->nThread; i++){
    Thread *pThread = &pThreadSet->aThread[i];
    if( pThread->xMain ){
      pthread_join(pThread->id, 0);
    }
  }
}

/*
** Set the result for thread iThread. 
*/
static void testThreadSetResult(
  ThreadSet *pThreadSet,          /* Thread-set handle */
  int iThread,                    /* Set result for this thread */
  int rc,                         /* Result error code */
  char *zFmt,                     /* Result string format */
  ...                             /* Result string formatting args... */
){
  va_list ap;

  testFree(pThreadSet->aThread[iThread].zMsg);
  pThreadSet->aThread[iThread].rc = rc;
  pThreadSet->aThread[iThread].zMsg = 0;
  if( zFmt ){
    va_start(ap, zFmt);
    pThreadSet->aThread[iThread].zMsg = testMallocVPrintf(zFmt, ap);
    va_end(ap);
  }
}

/*
** Retrieve the result for thread iThread. 
*/
static int testThreadGetResult(
  ThreadSet *pThreadSet,          /* Thread-set handle */
  int iThread,                    /* Get result for this thread */
  const char **pzRes              /* OUT: Pointer to result string */
){
  if( pzRes ) *pzRes = pThreadSet->aThread[iThread].zMsg;
  return pThreadSet->aThread[iThread].rc;
}

/*
** Enter and leave the test case mutex.
*/
#if 0
static void testThreadEnterMutex(ThreadSet *p){
  pthread_mutex_lock(&p->mutex);
}
static void testThreadLeaveMutex(ThreadSet *p){
  pthread_mutex_unlock(&p->mutex);
}
#endif
#endif

#if !defined(LSM_MUTEX_PTHREADS)
static int testThreadSupport(){ return 0; }

#define testThreadInit(a) 0
#define testThreadShutdown(a)
#define testThreadLaunch(a,b,c,d) 0
#define testThreadWait(a,b)
#define testThreadSetHalt(a)
#define testThreadGetHalt(a) 0
#define testThreadGetResult(a,b,c) 0
#define testThreadSleep(a,b) 0

static void testThreadSetResult(ThreadSet *a, int b, int c, char *d, ...){
  unused_parameter(a);
  unused_parameter(b);
  unused_parameter(c);
  unused_parameter(d);
}
#endif
/* End of threads wrapper.
*************************************************************************/

/*************************************************************************
** Below this point is the third part of this file - the implementation
** of the mt1.* tests.
*/
typedef struct Mt1Test Mt1Test;
struct Mt1Test {
  DbParameters param;             /* Description of database to read/write */
  int nReadwrite;                 /* Number of read/write threads */
  int nFastReader;                /* Number of fast reader threads */
  int nSlowReader;                /* Number of slow reader threads */
  int nMs;                        /* How long to run for */
  const char *zSystem;            /* Database system to test */
};

typedef struct Mt1DelayCtx Mt1DelayCtx;
struct Mt1DelayCtx {
  ThreadSet *pSet;                /* Threadset to sleep within */
  int nMs;                        /* Sleep in ms */
};

static void xMt1Delay(void *pCtx){
  Mt1DelayCtx *p = (Mt1DelayCtx *)pCtx;
  testThreadSleep(p->pSet, p->nMs);
}

#define MT1_THREAD_RDWR 0
#define MT1_THREAD_SLOW 1
#define MT1_THREAD_FAST 2

static void xMt1Work(lsm_db *pDb, void *pCtx){
#if 0
  char *z = 0;
  lsm_info(pDb, LSM_INFO_DB_STRUCTURE, &z);
  printf("%s\n", z);
  fflush(stdout);
#endif
}

/*
** This is the main() proc for all threads in test case "mt1".
*/
static void mt1Main(ThreadSet *pThreadSet, int iThread, void *pCtx){
  Mt1Test *p = (Mt1Test *)pCtx;   /* Test parameters */
  Mt1DelayCtx delay;
  int nRead = 0;                  /* Number of calls to dbReadOperation() */
  int nWrite = 0;                 /* Number of completed database writes */
  int rc = 0;                     /* Error code */
  int iPrng;                      /* Prng argument variable */
  TestDb *pDb;                    /* Database handle */
  int eType;

  delay.pSet = pThreadSet;
  delay.nMs = 0;
  if( iThread<p->nReadwrite ){
    eType = MT1_THREAD_RDWR;
  }else if( iThread<(p->nReadwrite+p->nFastReader) ){
    eType = MT1_THREAD_FAST;
  }else{
    eType = MT1_THREAD_SLOW;
    delay.nMs = (p->nMs / 20);
  }

  /* Open a new database connection. Initialize the pseudo-random number
  ** argument based on the thread number.  */
  iPrng = testPrngValue(iThread);
  pDb = testOpen(p->zSystem, 0, &rc);

  if( rc==0 ){
    tdb_lsm_config_work_hook(pDb, xMt1Work, 0);
  }

  /* Loop until either an error occurs or some other thread sets the
  ** halt flag.  */
  while( rc==0 && testThreadGetHalt(pThreadSet)==0 ){
    int iKey;

    /* Perform a read operation on an arbitrarily selected key. */
    iKey = (testPrngValue(iPrng++) % p->param.nKey);
    dbReadOperation(&p->param, pDb, xMt1Delay, (void *)&delay, iKey, &rc);
    if( rc ) continue;
    nRead++;

    /* Attempt to write an arbitrary key value pair (and update the associated
    ** checksum entries). dbWriteOperation() returns 1 if the write is
    ** successful, or 0 if it failed with an LSM_BUSY error.  */
    if( eType==MT1_THREAD_RDWR ){
      char aValue[50];
      char aRnd[25];

      iKey = (testPrngValue(iPrng++) % p->param.nKey);
      testPrngString(iPrng, aRnd, sizeof(aRnd));
      iPrng += sizeof(aRnd);
      snprintf(aValue, sizeof(aValue), "%d.%s", iThread, aRnd);
      nWrite += dbWriteOperation(&p->param, pDb, iKey, aValue, &rc);
    }
  }
  testClose(&pDb);

  /* If an error has occured, set the thread error code and the threadset 
  ** halt flag to tell the other test threads to halt. Otherwise, set the
  ** thread error code to 0 and post a message with the number of read
  ** and write operations completed.  */
  if( rc ){
    testThreadSetResult(pThreadSet, iThread, rc, 0);
    testThreadSetHalt(pThreadSet);
  }else{
    testThreadSetResult(pThreadSet, iThread, 0, "r/w: %d/%d", nRead, nWrite);
  }
}

static void do_test_mt1(
  const char *zSystem,            /* Database system name */
  const char *zPattern,           /* Run test cases that match this pattern */
  int *pRc                        /* IN/OUT: Error code */
){
  Mt1Test aTest[] = {
    /* param, nReadwrite, nFastReader, nSlowReader, nMs, zSystem */
    { {10, 1000},     4, 0, 0,   10000,   0 },
    { {10, 1000},     4, 4, 2,   100000,  0 },
    { {10, 100000},   4, 0, 0,   10000,   0 },
    { {10, 100000},   4, 4, 2,   100000,  0 },
  };
  int i;

  for(i=0; *pRc==0 && i<ArraySize(aTest); i++){
    Mt1Test *p = &aTest[i];
    int bRun = testCaseBegin(pRc, zPattern, 
        "mt1.%s.db=%d,%d.ms=%d.rdwr=%d.fast=%d.slow=%d", 
        zSystem, p->param.nFanout, p->param.nKey, 
        p->nMs, p->nReadwrite, p->nFastReader, p->nSlowReader
    );
    if( bRun ){
      TestDb *pDb;
      ThreadSet *pSet;
      int iThread;
      int nThread;

      p->zSystem = zSystem;
      pDb = testOpen(zSystem, 1, pRc);

      nThread = p->nReadwrite + p->nFastReader + p->nSlowReader;
      pSet = testThreadInit(nThread);
      for(iThread=0; *pRc==0 && iThread<nThread; iThread++){
        testThreadLaunch(pSet, iThread, mt1Main, (void *)p);
      }

      testThreadWait(pSet, p->nMs);
      for(iThread=0; *pRc==0 && iThread<nThread; iThread++){
        *pRc = testThreadGetResult(pSet, iThread, 0);
      }
      testCaseFinish(*pRc);

      for(iThread=0; *pRc==0 && iThread<nThread; iThread++){
        const char *zMsg = 0;
        *pRc = testThreadGetResult(pSet, iThread, &zMsg);
        printf("  Info: thread %d (%d): %s\n", iThread, *pRc, zMsg);
      }

      testThreadShutdown(pSet);
      testClose(&pDb);
    }
  }
}

void test_mt(
  const char *zSystem,            /* Database system name */
  const char *zPattern,           /* Run test cases that match this pattern */
  int *pRc                        /* IN/OUT: Error code */
){
  if( testThreadSupport()==0 ) return;
  do_test_mt1(zSystem, zPattern, pRc);
}

Added ext/lsm1/lsm-test/lsmtest6.c.













































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662

#include "lsmtest.h"

typedef struct OomTest OomTest;
struct OomTest {
  lsm_env *pEnv;
  int iNext;                      /* Next value to pass to testMallocOom() */
  int nFail;                      /* Number of OOM events injected */
  int bEnable;
  int rc;                         /* Test case error code */
};

static void testOomStart(OomTest *p){
  memset(p, 0, sizeof(OomTest));
  p->iNext = 1;
  p->bEnable = 1;
  p->nFail = 1;
  p->pEnv = tdb_lsm_env();
}

static void xOomHook(OomTest *p){
  p->nFail++;
}

static int testOomContinue(OomTest *p){
  if( p->rc!=0 || (p->iNext>1 && p->nFail==0) ){
    return 0;
  }
  p->nFail = 0;
  testMallocOom(p->pEnv, p->iNext, 0, (void (*)(void*))xOomHook, (void *)p);
  return 1;
}

static void testOomEnable(OomTest *p, int bEnable){
  p->bEnable = bEnable;
  testMallocOomEnable(p->pEnv, bEnable);
}

static void testOomNext(OomTest *p){
  p->iNext++;
}

static int testOomHit(OomTest *p){
  return (p->nFail>0);
}

static int testOomFinish(OomTest *p){
  return p->rc;
}

static void testOomAssert(OomTest *p, int bVal){
  if( bVal==0 ){
    test_failed();
    p->rc = 1;
  }
}

/*
** Test that the error code matches the state of the OomTest object passed
** as the first argument. Specifically, check that rc is LSM_NOMEM if an 
** OOM error has already been injected, or LSM_OK if not.
*/
static void testOomAssertRc(OomTest *p, int rc){
  testOomAssert(p, rc==LSM_OK || rc==LSM_NOMEM);
  testOomAssert(p, testOomHit(p)==(rc==LSM_NOMEM) || p->bEnable==0 );
}

static void testOomOpen(
  OomTest *pOom,
  const char *zName,
  lsm_db **ppDb,
  int *pRc
){
  if( *pRc==LSM_OK ){
    int rc;
    rc = lsm_new(tdb_lsm_env(), ppDb);
    if( rc==LSM_OK ) rc = lsm_open(*ppDb, zName);
    testOomAssertRc(pOom, rc);
    *pRc = rc;
  }
}

static void testOomFetch(
  OomTest *pOom,
  lsm_db *pDb,
  void *pKey, int nKey,
  void *pVal, int nVal,
  int *pRc
){
  testOomAssertRc(pOom, *pRc);
  if( *pRc==LSM_OK ){
    lsm_cursor *pCsr;
    int rc;

    rc = lsm_csr_open(pDb, &pCsr);
    if( rc==LSM_OK ) rc = lsm_csr_seek(pCsr, pKey, nKey, 0);
    testOomAssertRc(pOom, rc);

    if( rc==LSM_OK ){
      const void *p; int n;
      testOomAssert(pOom, lsm_csr_valid(pCsr));

      rc = lsm_csr_key(pCsr, &p, &n);
      testOomAssertRc(pOom, rc);
      testOomAssert(pOom, rc!=LSM_OK || (n==nKey && memcmp(pKey, p, nKey)==0) );
    }

    if( rc==LSM_OK ){
      const void *p; int n;
      testOomAssert(pOom, lsm_csr_valid(pCsr));

      rc = lsm_csr_value(pCsr, &p, &n);
      testOomAssertRc(pOom, rc);
      testOomAssert(pOom, rc!=LSM_OK || (n==nVal && memcmp(pVal, p, nVal)==0) );
    }

    lsm_csr_close(pCsr);
    *pRc = rc;
  }
}

static void testOomWrite(
  OomTest *pOom,
  lsm_db *pDb,
  void *pKey, int nKey,
  void *pVal, int nVal,
  int *pRc
){
  testOomAssertRc(pOom, *pRc);
  if( *pRc==LSM_OK ){
    int rc;

    rc = lsm_insert(pDb, pKey, nKey, pVal, nVal);
    testOomAssertRc(pOom, rc);

    *pRc = rc;
  }
}


static void testOomFetchStr(
  OomTest *pOom,
  lsm_db *pDb,
  const char *zKey,
  const char *zVal,
  int *pRc
){
  int nKey = strlen(zKey);
  int nVal = strlen(zVal);
  testOomFetch(pOom, pDb, (void *)zKey, nKey, (void *)zVal, nVal, pRc);
}

static void testOomFetchData(
  OomTest *pOom,
  lsm_db *pDb,
  Datasource *pData,
  int iKey,
  int *pRc
){
  void *pKey; int nKey;
  void *pVal; int nVal;
  testDatasourceEntry(pData, iKey, &pKey, &nKey, &pVal, &nVal);
  testOomFetch(pOom, pDb, pKey, nKey, pVal, nVal, pRc);
}

static void testOomWriteStr(
  OomTest *pOom,
  lsm_db *pDb,
  const char *zKey,
  const char *zVal,
  int *pRc
){
  int nKey = strlen(zKey);
  int nVal = strlen(zVal);
  testOomWrite(pOom, pDb, (void *)zKey, nKey, (void *)zVal, nVal, pRc);
}

static void testOomWriteData(
  OomTest *pOom,
  lsm_db *pDb,
  Datasource *pData,
  int iKey,
  int *pRc
){
  void *pKey; int nKey;
  void *pVal; int nVal;
  testDatasourceEntry(pData, iKey, &pKey, &nKey, &pVal, &nVal);
  testOomWrite(pOom, pDb, pKey, nKey, pVal, nVal, pRc);
}

static void testOomScan(
  OomTest *pOom, 
  lsm_db *pDb, 
  int bReverse,
  const void *pKey, int nKey,
  int nScan,
  int *pRc
){
  if( *pRc==0 ){
    int rc;
    int iScan = 0;
    lsm_cursor *pCsr;
    int (*xAdvance)(lsm_cursor *);
    

    rc = lsm_csr_open(pDb, &pCsr);
    testOomAssertRc(pOom, rc);

    if( rc==LSM_OK ){
      if( bReverse ){
        rc = lsm_csr_seek(pCsr, pKey, nKey, LSM_SEEK_LE);
        xAdvance = lsm_csr_prev;
      }else{
        rc = lsm_csr_seek(pCsr, pKey, nKey, LSM_SEEK_GE);
        xAdvance = lsm_csr_next;
      }
    }
    testOomAssertRc(pOom, rc);

    while( rc==LSM_OK && lsm_csr_valid(pCsr) && iScan<nScan ){
      const void *p; int n;

      rc = lsm_csr_key(pCsr, &p, &n);
      testOomAssertRc(pOom, rc);
      if( rc==LSM_OK ){
        rc = lsm_csr_value(pCsr, &p, &n);
        testOomAssertRc(pOom, rc);
      }
      if( rc==LSM_OK ){
        rc = xAdvance(pCsr);
        testOomAssertRc(pOom, rc);
      }
      iScan++;
    }

    lsm_csr_close(pCsr);
    *pRc = rc;
  }
}

#define LSMTEST6_TESTDB "testdb.lsm" 

#ifndef _WIN32
# include <unistd.h>
#endif
#include <sys/types.h>
#include <sys/stat.h>
#include <fcntl.h>

void testDeleteLsmdb(const char *zFile){
  char *zLog = testMallocPrintf("%s-log", zFile);
  char *zShm = testMallocPrintf("%s-shm", zFile);
  unlink(zFile);
  unlink(zLog);
  unlink(zShm);
  testFree(zLog);
  testFree(zShm);
}

static void copy_file(const char *zFrom, const char *zTo){

  if( access(zFrom, F_OK) ){
    unlink(zTo);
  }else{
    int fd1;
    int fd2;
    off_t sz;
    off_t i;
    struct stat buf;
    u8 *aBuf;

    fd1 = open(zFrom, O_RDONLY, 0644);
    fd2 = open(zTo, O_RDWR | O_CREAT, 0644);

    fstat(fd1, &buf);
    sz = buf.st_size;
    ftruncate(fd2, sz);

    aBuf = testMalloc(4096);
    for(i=0; i<sz; i+=4096){
      int nByte = MIN(4096, sz - i);
      read(fd1, aBuf, nByte);
      write(fd2, aBuf, nByte);
    }
    testFree(aBuf);

    close(fd1);
    close(fd2);
  }
}

void testCopyLsmdb(const char *zFrom, const char *zTo){
  char *zLog1 = testMallocPrintf("%s-log", zFrom);
  char *zLog2 = testMallocPrintf("%s-log", zTo);
  char *zShm1 = testMallocPrintf("%s-shm", zFrom);
  char *zShm2 = testMallocPrintf("%s-shm", zTo);

  unlink(zShm2);
  unlink(zLog2);
  unlink(zTo);
  copy_file(zFrom, zTo);
  copy_file(zLog1, zLog2);
  copy_file(zShm1, zShm2);

  testFree(zLog1); testFree(zLog2); testFree(zShm1); testFree(zShm2);
}

/*
** File zFile is the path to a database. This function makes backups
** of the database file and its log as follows:
**
**     cp $(zFile)         $(zFile)-save
**     cp $(zFile)-$(zAux) $(zFile)-save-$(zAux)
**
** Function testRestoreDb() can be used to copy the files back in the
** other direction.
*/
void testSaveDb(const char *zFile, const char *zAux){
  char *zLog = testMallocPrintf("%s-%s", zFile, zAux);
  char *zFileSave = testMallocPrintf("%s-save", zFile);
  char *zLogSave = testMallocPrintf("%s-%s-save", zFile, zAux);

  unlink(zFileSave);
  unlink(zLogSave);
  copy_file(zFile, zFileSave);
  copy_file(zLog, zLogSave);

  testFree(zLog); testFree(zFileSave); testFree(zLogSave);
}

/*
** File zFile is the path to a database. This function restores
** a backup of the database made by a previous call to testSaveDb().
** Specifically, it does the equivalent of:
**
**     cp $(zFile)-save         $(zFile)
**     cp $(zFile)-save-$(zAux) $(zFile)-$(zAux)
*/
void testRestoreDb(const char *zFile, const char *zAux){
  char *zLog = testMallocPrintf("%s-%s", zFile, zAux);
  char *zFileSave = testMallocPrintf("%s-save", zFile);
  char *zLogSave = testMallocPrintf("%s-%s-save", zFile, zAux);

  copy_file(zFileSave, zFile);
  copy_file(zLogSave, zLog);

  testFree(zLog); testFree(zFileSave); testFree(zLogSave);
}


static int lsmWriteStr(lsm_db *pDb, const char *zKey, const char *zVal){
  int nKey = strlen(zKey);
  int nVal = strlen(zVal);
  return lsm_insert(pDb, (void *)zKey, nKey, (void *)zVal, nVal);
}

static void setup_delete_db(){
  testDeleteLsmdb(LSMTEST6_TESTDB);
}

/*
** Create a small database. With the following content:
**
**    "one"   -> "one"
**    "two"   -> "four"
**    "three" -> "nine"
**    "four"  -> "sixteen"
**    "five"  -> "twentyfive"
**    "six"   -> "thirtysix"
**    "seven" -> "fourtynine"
**    "eight" -> "sixtyfour"
*/
static void setup_populate_db(){
  const char *azStr[] = {
    "one",   "one",
    "two",   "four",
    "three", "nine",
    "four",  "sixteen",
    "five",  "twentyfive",
    "six",   "thirtysix",
    "seven", "fourtynine",
    "eight", "sixtyfour",
  };
  int rc;
  int ii;
  lsm_db *pDb;

  testDeleteLsmdb(LSMTEST6_TESTDB);

  rc = lsm_new(tdb_lsm_env(), &pDb);
  if( rc==LSM_OK ) rc = lsm_open(pDb, LSMTEST6_TESTDB);

  for(ii=0; rc==LSM_OK && ii<ArraySize(azStr); ii+=2){
    rc = lsmWriteStr(pDb, azStr[ii], azStr[ii+1]);
  }
  lsm_close(pDb);

  testSaveDb(LSMTEST6_TESTDB, "log");
  assert( rc==LSM_OK );
}

static Datasource *getDatasource(void){
  const DatasourceDefn defn = { TEST_DATASOURCE_RANDOM, 10, 15, 200, 250 };
  return testDatasourceNew(&defn);
}

/*
** Set up a database file with the following properties:
**
**   * Page size is 1024 bytes.
**   * Block size is 64 KB.
**   * Contains 5000 key-value pairs starting at 0 from the
**     datasource returned getDatasource().
*/
static void setup_populate_db2(){
  Datasource *pData;
  int ii;
  int rc;
  int nBlocksize = 64*1024;
  int nPagesize = 1024;
  int nWritebuffer = 4*1024;
  lsm_db *pDb;

  testDeleteLsmdb(LSMTEST6_TESTDB);
  rc = lsm_new(tdb_lsm_env(), &pDb);
  if( rc==LSM_OK ) rc = lsm_open(pDb, LSMTEST6_TESTDB);

  lsm_config(pDb, LSM_CONFIG_BLOCK_SIZE, &nBlocksize); 
  lsm_config(pDb, LSM_CONFIG_PAGE_SIZE, &nPagesize); 
  lsm_config(pDb, LSM_CONFIG_AUTOFLUSH, &nWritebuffer); 

  pData = getDatasource();
  for(ii=0; rc==LSM_OK && ii<5000; ii++){
    void *pKey; int nKey;
    void *pVal; int nVal;
    testDatasourceEntry(pData, ii, &pKey, &nKey, &pVal, &nVal);
    lsm_insert(pDb, pKey, nKey, pVal, nVal);
  }
  testDatasourceFree(pData);
  lsm_close(pDb);

  testSaveDb(LSMTEST6_TESTDB, "log");
  assert( rc==LSM_OK );
}

/*
** Test the results of OOM conditions in lsm_new().
*/
static void simple_oom_1(OomTest *pOom){
  int rc;
  lsm_db *pDb;

  rc = lsm_new(tdb_lsm_env(), &pDb);
  testOomAssertRc(pOom, rc);

  lsm_close(pDb);
}

/*
** Test the results of OOM conditions in lsm_open().
*/
static void simple_oom_2(OomTest *pOom){
  int rc;
  lsm_db *pDb;

  rc = lsm_new(tdb_lsm_env(), &pDb);
  if( rc==LSM_OK ){
    rc = lsm_open(pDb, "testdb.lsm");
  }
  testOomAssertRc(pOom, rc);

  lsm_close(pDb);
}

/*
** Test the results of OOM conditions in simple fetch operations.
*/
static void simple_oom_3(OomTest *pOom){
  int rc = LSM_OK;
  lsm_db *pDb;

  testOomOpen(pOom, LSMTEST6_TESTDB, &pDb, &rc);

  testOomFetchStr(pOom, pDb, "four",  "sixteen",    &rc);
  testOomFetchStr(pOom, pDb, "seven", "fourtynine", &rc);
  testOomFetchStr(pOom, pDb, "one",   "one",        &rc);
  testOomFetchStr(pOom, pDb, "eight", "sixtyfour",  &rc);

  lsm_close(pDb);
}

/*
** Test the results of OOM conditions in simple write operations.
*/
static void simple_oom_4(OomTest *pOom){
  int rc = LSM_OK;
  lsm_db *pDb;

  testDeleteLsmdb(LSMTEST6_TESTDB);
  testOomOpen(pOom, LSMTEST6_TESTDB, &pDb, &rc);

  testOomWriteStr(pOom, pDb, "123", "onetwothree", &rc);
  testOomWriteStr(pOom, pDb, "456", "fourfivesix", &rc);
  testOomWriteStr(pOom, pDb, "789", "seveneightnine", &rc);
  testOomWriteStr(pOom, pDb, "123", "teneleventwelve", &rc);
  testOomWriteStr(pOom, pDb, "456", "fourteenfifteensixteen", &rc);

  lsm_close(pDb);
}

static void simple_oom_5(OomTest *pOom){
  Datasource *pData = getDatasource();
  int rc = LSM_OK;
  lsm_db *pDb;

  testRestoreDb(LSMTEST6_TESTDB, "log");
  testOomOpen(pOom, LSMTEST6_TESTDB, &pDb, &rc);

  testOomFetchData(pOom, pDb, pData, 3333, &rc);
  testOomFetchData(pOom, pDb, pData, 0, &rc);
  testOomFetchData(pOom, pDb, pData, 4999, &rc);

  lsm_close(pDb);
  testDatasourceFree(pData);
}

static void simple_oom_6(OomTest *pOom){
  Datasource *pData = getDatasource();
  int rc = LSM_OK;
  lsm_db *pDb;

  testRestoreDb(LSMTEST6_TESTDB, "log");
  testOomOpen(pOom, LSMTEST6_TESTDB, &pDb, &rc);

  testOomWriteData(pOom, pDb, pData, 5000, &rc);
  testOomWriteData(pOom, pDb, pData, 5001, &rc);
  testOomWriteData(pOom, pDb, pData, 5002, &rc);
  testOomFetchData(pOom, pDb, pData, 5001, &rc);
  testOomFetchData(pOom, pDb, pData, 1234, &rc);

  lsm_close(pDb);
  testDatasourceFree(pData);
}

static void simple_oom_7(OomTest *pOom){
  Datasource *pData = getDatasource();
  int rc = LSM_OK;
  lsm_db *pDb;

  testRestoreDb(LSMTEST6_TESTDB, "log");
  testOomOpen(pOom, LSMTEST6_TESTDB, &pDb, &rc);
  testOomScan(pOom, pDb, 0, "abc", 3, 20, &rc);
  lsm_close(pDb);
  testDatasourceFree(pData);
}

static void simple_oom_8(OomTest *pOom){
  Datasource *pData = getDatasource();
  int rc = LSM_OK;
  lsm_db *pDb;
  testRestoreDb(LSMTEST6_TESTDB, "log");
  testOomOpen(pOom, LSMTEST6_TESTDB, &pDb, &rc);
  testOomScan(pOom, pDb, 1, "xyz", 3, 20, &rc);
  lsm_close(pDb);
  testDatasourceFree(pData);
}

/*
** This test case has two clients connected to a database. The first client
** hits an OOM while writing to the database. Check that the second 
** connection is still able to query the db following the OOM.
*/
static void simple_oom2_1(OomTest *pOom){
  const int nRecord = 100;        /* Number of records initially in db */
  const int nIns = 10;            /* Number of records inserted with OOM */

  Datasource *pData = getDatasource();
  int rc = LSM_OK;
  lsm_db *pDb1;
  lsm_db *pDb2;
  int i;

  testDeleteLsmdb(LSMTEST6_TESTDB);

  /* Open the two connections. Initialize the in-memory tree so that it
  ** contains 100 records. Do all this with OOM injection disabled. */
  testOomEnable(pOom, 0);
  testOomOpen(pOom, LSMTEST6_TESTDB, &pDb1, &rc);
  testOomOpen(pOom, LSMTEST6_TESTDB, &pDb2, &rc);
  for(i=0; i<nRecord; i++){
    testOomWriteData(pOom, pDb1, pData, i, &rc);
  }
  testOomEnable(pOom, 1);
  assert( rc==0 );

  /* Insert 10 more records using pDb1. Stop when an OOM is encountered. */
  for(i=nRecord; i<nRecord+nIns; i++){
    testOomWriteData(pOom, pDb1, pData, i, &rc);
    if( rc ) break;
  }
  testOomAssertRc(pOom, rc);

  /* Switch off OOM injection. Write a few rows using pDb2. Then check
  ** that the database may be successfully queried.  */
  testOomEnable(pOom, 0);
  rc = 0;
  for(; i<nRecord+nIns && rc==0; i++){
    testOomWriteData(pOom, pDb2, pData, i, &rc);
  }
  for(i=0; i<nRecord+nIns; i++) testOomFetchData(pOom, pDb2, pData, i, &rc);
  testOomEnable(pOom, 1);

  lsm_close(pDb1);
  lsm_close(pDb2);
  testDatasourceFree(pData);
}


static void do_test_oom1(const char *zPattern, int *pRc){
  struct SimpleOom {
    const char *zName;
    void (*xSetup)(void);
    void (*xFunc)(OomTest *);
  } aSimple[] = {
    { "oom1.lsm.1", setup_delete_db,    simple_oom_1 },
    { "oom1.lsm.2", setup_delete_db,    simple_oom_2 },
    { "oom1.lsm.3", setup_populate_db,  simple_oom_3 },
    { "oom1.lsm.4", setup_delete_db,    simple_oom_4 },
    { "oom1.lsm.5", setup_populate_db2, simple_oom_5 },
    { "oom1.lsm.6", setup_populate_db2, simple_oom_6 },
    { "oom1.lsm.7", setup_populate_db2, simple_oom_7 },
    { "oom1.lsm.8", setup_populate_db2, simple_oom_8 },

    { "oom2.lsm.1", setup_delete_db,    simple_oom2_1 },
  };
  int i;

  for(i=0; i<ArraySize(aSimple); i++){
    if( *pRc==0 && testCaseBegin(pRc, zPattern, "%s", aSimple[i].zName) ){
      OomTest t;

      if( aSimple[i].xSetup ){
        aSimple[i].xSetup();
      }

      for(testOomStart(&t); testOomContinue(&t); testOomNext(&t)){
        aSimple[i].xFunc(&t);
      }

      printf("(%d injections).", t.iNext-2);
      testCaseFinish( (*pRc = testOomFinish(&t)) );
      testMallocOom(tdb_lsm_env(), 0, 0, 0, 0);
    }
  }
}

void test_oom(
  const char *zPattern,           /* Run test cases that match this pattern */
  int *pRc                        /* IN/OUT: Error code */
){
  do_test_oom1(zPattern, pRc);
}

Added ext/lsm1/lsm-test/lsmtest7.c.





























































































































































































































































































































































































































>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206


#include "lsmtest.h"


/*
** Test that the rules for when lsm_csr_next() and lsm_csr_prev() are
** enforced. Specifically:
**
**   * Both functions always return LSM_MISUSE if the cursor is at EOF
**     when they are called.
**
**   * lsm_csr_next() may only be used after lsm_csr_seek(LSM_SEEK_GE) or 
**     lsm_csr_first(). 
**
**   * lsm_csr_prev() may only be used after lsm_csr_seek(LSM_SEEK_LE) or 
**     lsm_csr_last().
*/
static void do_test_api1_lsm(lsm_db *pDb, int *pRc){
  int ret;
  lsm_cursor *pCsr;
  lsm_cursor *pCsr2;
  int nKey;
  const void *pKey;

  ret = lsm_csr_open(pDb, &pCsr);
  testCompareInt(LSM_OK, ret, pRc);

  ret = lsm_csr_next(pCsr);
  testCompareInt(LSM_MISUSE, ret, pRc);
  ret = lsm_csr_prev(pCsr);
  testCompareInt(LSM_MISUSE, ret, pRc);

  ret = lsm_csr_seek(pCsr, "jjj", 3, LSM_SEEK_GE);
  testCompareInt(LSM_OK, ret, pRc);
  ret = lsm_csr_next(pCsr);
  testCompareInt(LSM_OK, ret, pRc);
  ret = lsm_csr_prev(pCsr);
  testCompareInt(LSM_MISUSE, ret, pRc);

  ret = lsm_csr_seek(pCsr, "jjj", 3, LSM_SEEK_LE);
  testCompareInt(LSM_OK, ret, pRc);
  ret = lsm_csr_next(pCsr);
  testCompareInt(LSM_MISUSE, ret, pRc);
  ret = lsm_csr_prev(pCsr);
  testCompareInt(LSM_OK, ret, pRc);

  ret = lsm_csr_seek(pCsr, "jjj", 3, LSM_SEEK_LEFAST);
  testCompareInt(LSM_OK, ret, pRc);
  ret = lsm_csr_next(pCsr);
  testCompareInt(LSM_MISUSE, ret, pRc);
  ret = lsm_csr_prev(pCsr);
  testCompareInt(LSM_MISUSE, ret, pRc);

  ret = lsm_csr_key(pCsr, &pKey, &nKey);
  testCompareInt(LSM_OK, ret, pRc);

  ret = lsm_csr_open(pDb, &pCsr2);
  testCompareInt(LSM_OK, ret, pRc);

  ret = lsm_csr_seek(pCsr2, pKey, nKey, LSM_SEEK_EQ);
  testCompareInt(LSM_OK, ret, pRc);
  testCompareInt(1, lsm_csr_valid(pCsr2), pRc);
  ret = lsm_csr_next(pCsr2);
  testCompareInt(LSM_MISUSE, ret, pRc);
  ret = lsm_csr_prev(pCsr2);
  testCompareInt(LSM_MISUSE, ret, pRc);

  lsm_csr_close(pCsr2);

  ret = lsm_csr_first(pCsr);
  testCompareInt(LSM_OK, ret, pRc);
  ret = lsm_csr_next(pCsr);
  testCompareInt(LSM_OK, ret, pRc);
  ret = lsm_csr_prev(pCsr);
  testCompareInt(LSM_MISUSE, ret, pRc);

  ret = lsm_csr_last(pCsr);
  testCompareInt(LSM_OK, ret, pRc);
  ret = lsm_csr_prev(pCsr);
  testCompareInt(LSM_OK, ret, pRc);
  ret = lsm_csr_next(pCsr);
  testCompareInt(LSM_MISUSE, ret, pRc);

  ret = lsm_csr_first(pCsr);
  while( lsm_csr_valid(pCsr) ){
    ret = lsm_csr_next(pCsr);
    testCompareInt(LSM_OK, ret, pRc);
  }
  ret = lsm_csr_next(pCsr);
  testCompareInt(LSM_OK, ret, pRc);
  ret = lsm_csr_prev(pCsr);
  testCompareInt(LSM_MISUSE, ret, pRc);

  ret = lsm_csr_last(pCsr);
  while( lsm_csr_valid(pCsr) ){
    ret = lsm_csr_prev(pCsr);
    testCompareInt(LSM_OK, ret, pRc);
  }
  ret = lsm_csr_prev(pCsr);
  testCompareInt(LSM_OK, ret, pRc);
  ret = lsm_csr_next(pCsr);
  testCompareInt(LSM_MISUSE, ret, pRc);

  lsm_csr_close(pCsr);
}

static void do_test_api1(const char *zPattern, int *pRc){
  if( testCaseBegin(pRc, zPattern, "api1.lsm") ){
    const DatasourceDefn defn = { TEST_DATASOURCE_RANDOM, 10, 15, 200, 250 };
    Datasource *pData;
    TestDb *pDb;
    int rc = 0;

    pDb = testOpen("lsm_lomem", 1, &rc);
    pData = testDatasourceNew(&defn);
    testWriteDatasourceRange(pDb, pData, 0, 1000, pRc);

    do_test_api1_lsm(tdb_lsm(pDb), pRc);

    testDatasourceFree(pData);
    testClose(&pDb);

    testCaseFinish(*pRc);
  }
}

static lsm_db *newLsmConnection(
  const char *zDb, 
  int nPgsz, 
  int nBlksz,
  int *pRc
){
  lsm_db *db = 0;
  if( *pRc==0 ){
    int n1 = nPgsz;
    int n2 = nBlksz;
    *pRc = lsm_new(tdb_lsm_env(), &db);
    if( *pRc==0 ){
      if( n1 ) lsm_config(db, LSM_CONFIG_PAGE_SIZE, &n1);
      if( n2 ) lsm_config(db, LSM_CONFIG_BLOCK_SIZE, &n2);
      *pRc = lsm_open(db, "testdb.lsm");
    }
  }
  return db;
}

static void testPagesize(lsm_db *db, int nPgsz, int nBlksz, int *pRc){
  if( *pRc==0 ){
    int n1 = 0;
    int n2 = 0;

    lsm_config(db, LSM_CONFIG_PAGE_SIZE, &n1);
    lsm_config(db, LSM_CONFIG_BLOCK_SIZE, &n2);

    testCompareInt(n1, nPgsz, pRc);
    testCompareInt(n2, nBlksz, pRc);
  }
}

/*
** Test case "api2" tests that the default page and block sizes of a 
** database may only be modified before lsm_open() is called. And that
** after lsm_open() is called lsm_config() may be used to read the 
** actual page and block size of the db.
*/
static void do_test_api2(const char *zPattern, int *pRc){
  if( *pRc==0 && testCaseBegin(pRc, zPattern, "api2.lsm") ){
    lsm_db *db1 = 0;
    lsm_db *db2 = 0;

    testDeleteLsmdb("testdb.lsm");
    db1 = newLsmConnection("testdb.lsm", 0, 0, pRc);
    testPagesize(db1, 4096, 1024, pRc);
    db2 = newLsmConnection("testdb.lsm", 1024, 64*1024, pRc);
    testPagesize(db2, 4096, 1024, pRc);
    lsm_close(db1);
    lsm_close(db2);

    testDeleteLsmdb("testdb.lsm");
    db1 = newLsmConnection("testdb.lsm", 1024, 64*1024, pRc);
    testPagesize(db1, 1024, 64*1024, pRc);
    db2 = newLsmConnection("testdb.lsm", 0, 0, pRc);
    testPagesize(db2, 1024, 64*1024, pRc);
    lsm_close(db1);
    lsm_close(db2);

    testDeleteLsmdb("testdb.lsm");
    db1 = newLsmConnection("testdb.lsm", 8192, 2*1024, pRc);
    testPagesize(db1, 8192, 2*1024, pRc);
    db2 = newLsmConnection("testdb.lsm", 1024, 64*1024, pRc);
    testPagesize(db2, 8192, 2*1024, pRc);
    lsm_close(db1);
    lsm_close(db2);

    testCaseFinish(*pRc);
  }
}

void test_api(
  const char *zPattern,           /* Run test cases that match this pattern */
  int *pRc                        /* IN/OUT: Error code */
){
  do_test_api1(zPattern, pRc);
  do_test_api2(zPattern, pRc);
}

Added ext/lsm1/lsm-test/lsmtest8.c.





































































































































































































































































































































































































































































































































































































































































>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322

/*
** This file contains test cases to verify that "live-recovery" following
** a mid-transaction failure of a writer process.
*/


/* 
** This test file includes lsmInt.h to get access to the definition of the
** ShmHeader structure. This is required to cause strategic damage to the
** shared memory header as part of recovery testing.
*/
#include "lsmInt.h"

#include "lsmtest.h"

typedef struct SetupStep SetupStep;
struct SetupStep {
  int bFlush;                     /* Flush to disk and checkpoint */
  int iInsStart;                  /* First key-value from ds to insert */
  int nIns;                       /* Number of rows to insert */
  int iDelStart;                  /* First key from ds to delete */
  int nDel;                       /* Number of rows to delete */
};

static void doSetupStep(
  TestDb *pDb, 
  Datasource *pData, 
  const SetupStep *pStep, 
  int *pRc
){
  testWriteDatasourceRange(pDb, pData, pStep->iInsStart, pStep->nIns, pRc);
  testDeleteDatasourceRange(pDb, pData, pStep->iDelStart, pStep->nDel, pRc);
  if( *pRc==0 ){
    int nSave = -1;
    int nBuf = 64;
    lsm_db *db = tdb_lsm(pDb);

    lsm_config(db, LSM_CONFIG_AUTOFLUSH, &nSave);
    lsm_config(db, LSM_CONFIG_AUTOFLUSH, &nBuf);
    lsm_begin(db, 1);
    lsm_commit(db, 0);
    lsm_config(db, LSM_CONFIG_AUTOFLUSH, &nSave);

    *pRc = lsm_work(db, 0, 0, 0);
    if( *pRc==0 ){
      *pRc = lsm_checkpoint(db, 0);
    }
  }
}

static void doSetupStepArray(
  TestDb *pDb, 
  Datasource *pData, 
  const SetupStep *aStep, 
  int nStep
){
  int i;
  for(i=0; i<nStep; i++){
    int rc = 0;
    doSetupStep(pDb, pData, &aStep[i], &rc);
    assert( rc==0 );
  }
}

static void setupDatabase1(TestDb *pDb, Datasource **ppData){
  const SetupStep aStep[] = {
    { 0,                                  1,     2000, 0, 0 },
    { 1,                                  0,     0, 0, 0 },
    { 0,                                  10001, 1000, 0, 0 },
  };
  const DatasourceDefn defn = {TEST_DATASOURCE_RANDOM, 12, 16, 100, 500};
  Datasource *pData;

  pData = testDatasourceNew(&defn);
  doSetupStepArray(pDb, pData, aStep, ArraySize(aStep));
  if( ppData ){
    *ppData = pData;
  }else{
    testDatasourceFree(pData);
  }
}

#include <stdio.h>
void testReadFile(const char *zFile, int iOff, void *pOut, int nByte, int *pRc){
  if( *pRc==0 ){
    FILE *fd;
    fd = fopen(zFile, "r");
    if( fd==0 ){
      *pRc = 1;
    }else{
      if( 0!=fseek(fd, iOff, SEEK_SET) ){
        *pRc = 1;
      }else{
        if( nByte!=fread(pOut, 1, nByte, fd) ){
          *pRc = 1;
        }
      }
      fclose(fd);
    }
  }
}

void testWriteFile(
  const char *zFile, 
  int iOff, 
  void *pOut, 
  int nByte, 
  int *pRc
){
  if( *pRc==0 ){
    FILE *fd;
    fd = fopen(zFile, "r+");
    if( fd==0 ){
      *pRc = 1;
    }else{
      if( 0!=fseek(fd, iOff, SEEK_SET) ){
        *pRc = 1;
      }else{
        if( nByte!=fwrite(pOut, 1, nByte, fd) ){
          *pRc = 1;
        }
      }
      fclose(fd);
    }
  }
}

static ShmHeader *getShmHeader(const char *zDb){
  int rc = 0;
  char *zShm = testMallocPrintf("%s-shm", zDb);
  ShmHeader *pHdr;

  pHdr = testMalloc(sizeof(ShmHeader));
  testReadFile(zShm, 0, (void *)pHdr, sizeof(ShmHeader), &rc);
  assert( rc==0 );

  return pHdr;
}

/*
** This function makes a copy of the three files associated with LSM 
** database zDb (i.e. if zDb is "test.db", it makes copies of "test.db",
** "test.db-log" and "test.db-shm").
**
** It then opens a new database connection to the copy with the xLock() call
** instrumented so that it appears that some other process already connected
** to the db (holding a shared lock on DMS2). This prevents recovery from
** running. Then:
**
**    1) Check that the checksum of the database is zCksum. 
**    2) Write a few keys to the database. Then delete the same keys. 
**    3) Check that the checksum is zCksum.
**    4) Flush the db to disk and run a checkpoint. 
**    5) Check once more that the checksum is still zCksum.
*/
static void doLiveRecovery(const char *zDb, const char *zCksum, int *pRc){
  const DatasourceDefn defn = {TEST_DATASOURCE_RANDOM, 20, 25, 100, 500};
  Datasource *pData;
  const char *zCopy = "testcopy.lsm";
  char zCksum2[TEST_CKSUM_BYTES];
  TestDb *pDb = 0;
  int rc;

  pData = testDatasourceNew(&defn);

  testCopyLsmdb(zDb, zCopy);
  rc = tdb_lsm_open("test_no_recovery=1", zCopy, 0, &pDb);
  if( rc==0 ){
    ShmHeader *pHdr;
    lsm_db *db;
    testCksumDatabase(pDb, zCksum2);
    testCompareStr(zCksum, zCksum2, &rc);

    testWriteDatasourceRange(pDb, pData, 1, 10, &rc);
    testDeleteDatasourceRange(pDb, pData, 1, 10, &rc);

    /* Test that the two tree-headers are now consistent. */
    pHdr = getShmHeader(zCopy);
    if( rc==0 && memcmp(&pHdr->hdr1, &pHdr->hdr2, sizeof(pHdr->hdr1)) ){
      rc = 1;
    }
    testFree(pHdr);

    if( rc==0 ){
      int nBuf = 64;
      db = tdb_lsm(pDb);
      lsm_config(db, LSM_CONFIG_AUTOFLUSH, &nBuf);
      lsm_begin(db, 1);
      lsm_commit(db, 0);
      rc = lsm_work(db, 0, 0, 0);
    }

    testCksumDatabase(pDb, zCksum2);
    testCompareStr(zCksum, zCksum2, &rc);
  }

  testDatasourceFree(pData);
  testClose(&pDb);
  testDeleteLsmdb(zCopy);
  *pRc = rc;
}

static void doWriterCrash1(int *pRc){
  const int nWrite = 2000;
  const int nStep = 10;
  const int iWriteStart = 20000;
  int rc = 0;
  TestDb *pDb = 0;
  Datasource *pData = 0;

  rc = tdb_lsm_open("autowork=0", "testdb.lsm", 1, &pDb);
  if( rc==0 ){
    int iDot = 0;
    char zCksum[TEST_CKSUM_BYTES];
    int i;
    setupDatabase1(pDb, &pData);
    testCksumDatabase(pDb, zCksum);
    testBegin(pDb, 2, &rc);
    for(i=0; rc==0 && i<nWrite; i+=nStep){
      testCaseProgress(i, nWrite, testCaseNDot(), &iDot);
      testWriteDatasourceRange(pDb, pData, iWriteStart+i, nStep, &rc);
      doLiveRecovery("testdb.lsm", zCksum, &rc);
    }
  }
  testCommit(pDb, 0, &rc);
  testClose(&pDb);
  testDatasourceFree(pData);
  *pRc = rc;
}

/*
** This test case verifies that inconsistent tree-headers in shared-memory
** are resolved correctly. 
*/
static void doWriterCrash2(int *pRc){
  int rc = 0;
  TestDb *pDb = 0;
  Datasource *pData = 0;

  rc = tdb_lsm_open("autowork=0", "testdb.lsm", 1, &pDb);
  if( rc==0 ){
    ShmHeader *pHdr1;
    ShmHeader *pHdr2;
    char zCksum1[TEST_CKSUM_BYTES];
    char zCksum2[TEST_CKSUM_BYTES];

    pHdr1 = testMalloc(sizeof(ShmHeader));
    pHdr2 = testMalloc(sizeof(ShmHeader));
    setupDatabase1(pDb, &pData);

    /* Grab a copy of the shared-memory header. And the db checksum */
    testReadFile("testdb.lsm-shm", 0, (void *)pHdr1, sizeof(ShmHeader), &rc);
    testCksumDatabase(pDb, zCksum1);

    /* Modify the database */
    testBegin(pDb, 2, &rc);
    testWriteDatasourceRange(pDb, pData, 30000, 200, &rc);
    testCommit(pDb, 0, &rc);

    /* Grab a second copy of the shared-memory header. And the db checksum */
    testReadFile("testdb.lsm-shm", 0, (void *)pHdr2, sizeof(ShmHeader), &rc);
    testCksumDatabase(pDb, zCksum2);
    doLiveRecovery("testdb.lsm", zCksum2, &rc);

    /* If both tree-headers are valid, tree-header-1 is used. */
    memcpy(&pHdr2->hdr1, &pHdr1->hdr1, sizeof(pHdr1->hdr1));
    pHdr2->bWriter = 1;
    testWriteFile("testdb.lsm-shm", 0, (void *)pHdr2, sizeof(ShmHeader), &rc);
    doLiveRecovery("testdb.lsm", zCksum1, &rc);

    /* If both tree-headers are valid, tree-header-1 is used. */
    memcpy(&pHdr2->hdr1, &pHdr2->hdr2, sizeof(pHdr1->hdr1));
    memcpy(&pHdr2->hdr2, &pHdr1->hdr1, sizeof(pHdr1->hdr1));
    pHdr2->bWriter = 1;
    testWriteFile("testdb.lsm-shm", 0, (void *)pHdr2, sizeof(ShmHeader), &rc);
    doLiveRecovery("testdb.lsm", zCksum2, &rc);

    /* If tree-header 1 is invalid, tree-header-2 is used */
    memcpy(&pHdr2->hdr2, &pHdr2->hdr1, sizeof(pHdr1->hdr1));
    pHdr2->hdr1.aCksum[0] = 5;
    pHdr2->hdr1.aCksum[0] = 6;
    pHdr2->bWriter = 1;
    testWriteFile("testdb.lsm-shm", 0, (void *)pHdr2, sizeof(ShmHeader), &rc);
    doLiveRecovery("testdb.lsm", zCksum2, &rc);

    /* If tree-header 2 is invalid, tree-header-1 is used */
    memcpy(&pHdr2->hdr1, &pHdr2->hdr2, sizeof(pHdr1->hdr1));
    pHdr2->hdr2.aCksum[0] = 5;
    pHdr2->hdr2.aCksum[0] = 6;
    pHdr2->bWriter = 1;
    testWriteFile("testdb.lsm-shm", 0, (void *)pHdr2, sizeof(ShmHeader), &rc);
    doLiveRecovery("testdb.lsm", zCksum2, &rc);

    testFree(pHdr1);
    testFree(pHdr2);
    testClose(&pDb);
  }

  *pRc = rc;
}

void do_writer_crash_test(const char *zPattern, int *pRc){
  struct Test {
    const char *zName;
    void (*xFunc)(int *);
  } aTest[] = {
    { "writercrash1.lsm", doWriterCrash1 },
    { "writercrash2.lsm", doWriterCrash2 },
  };
  int i;
  for(i=0; i<ArraySize(aTest); i++){
    struct Test *p = &aTest[i];
    if( testCaseBegin(pRc, zPattern, p->zName) ){
      p->xFunc(pRc);
      testCaseFinish(*pRc);
    }
  }

}


Added ext/lsm1/lsm-test/lsmtest9.c.





















































































































































































































































































>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138

#include "lsmtest.h"

#define DATA_SEQUENTIAL TEST_DATASOURCE_SEQUENCE
#define DATA_RANDOM     TEST_DATASOURCE_RANDOM

typedef struct Datatest4 Datatest4;

/*
** Test overview:
**
**   1. Insert (Datatest4.nRec) records into a database.
**
**   2. Repeat (Datatest4.nRepeat) times:
**
**      2a. Delete 2/3 of the records in the database.
**
**      2b. Run lsm_work(nMerge=1).
**
**      2c. Insert as many records as were deleted in 2a.
**
**      2d. Check database content is as expected.
**
**      2e. If (Datatest4.bReopen) is true, close and reopen the database.
*/
struct Datatest4 {
  /* Datasource definition */
  DatasourceDefn defn;

  int nRec;
  int nRepeat;
  int bReopen;
};

static void doDataTest4(
  const char *zSystem,            /* Database system to test */
  Datatest4 *p,                   /* Structure containing test parameters */
  int *pRc                        /* OUT: Error code */
){
  lsm_db *db = 0;
  TestDb *pDb;
  TestDb *pControl;
  Datasource *pData;
  int i;
  int rc = 0;
  int iDot = 0;

  int nRecOn3 = (p->nRec / 3);
  int iData = 0;

  /* Start the test case, open a database and allocate the datasource. */
  rc = testControlDb(&pControl);
  pDb = testOpen(zSystem, 1, &rc);
  pData = testDatasourceNew(&p->defn);
  if( rc==0 ) db = tdb_lsm(pDb);

  testWriteDatasourceRange(pControl, pData, iData, nRecOn3*3, &rc);
  testWriteDatasourceRange(pDb,      pData, iData, nRecOn3*3, &rc);

  for(i=0; rc==0 && i<p->nRepeat; i++){

    testDeleteDatasourceRange(pControl, pData, iData, nRecOn3*2, &rc);
    testDeleteDatasourceRange(pDb,      pData, iData, nRecOn3*2, &rc);

    if( db ){
      int nDone;
#if 0
      fprintf(stderr, "lsm_work() start...\n"); fflush(stderr);
#endif
      do {
        nDone = 0;
        rc = lsm_work(db, 1, (1<<30), &nDone);
      }while( rc==0 && nDone>0 );
#if 0 
      fprintf(stderr, "lsm_work() done...\n"); fflush(stderr);
#endif
    }

if( i+1<p->nRepeat ){
    iData += (nRecOn3*2);
    testWriteDatasourceRange(pControl, pData, iData+nRecOn3, nRecOn3*2, &rc);
    testWriteDatasourceRange(pDb,      pData, iData+nRecOn3, nRecOn3*2, &rc);

    testCompareDb(pData, nRecOn3*3, iData, pControl, pDb, &rc);

    /* If Datatest4.bReopen is true, close and reopen the database */
    if( p->bReopen ){
      testReopen(&pDb, &rc);
      if( rc==0 ) db = tdb_lsm(pDb);
    }
}

    /* Update the progress dots... */
    testCaseProgress(i, p->nRepeat, testCaseNDot(), &iDot);
  }

  testClose(&pDb);
  testClose(&pControl);
  testDatasourceFree(pData);
  testCaseFinish(rc);
  *pRc = rc;
}

static char *getName4(const char *zSystem, Datatest4 *pTest){
  char *zRet;
  char *zData;
  zData = testDatasourceName(&pTest->defn);
  zRet = testMallocPrintf("data4.%s.%s.%d.%d.%d", 
      zSystem, zData, pTest->nRec, pTest->nRepeat, pTest->bReopen
  );
  testFree(zData);
  return zRet;
}

void test_data_4(
  const char *zSystem,            /* Database system name */
  const char *zPattern,           /* Run test cases that match this pattern */
  int *pRc                        /* IN/OUT: Error code */
){
  Datatest4 aTest[] = {
      /* defn,                                 nRec, nRepeat, bReopen */
    { {DATA_RANDOM,     20,25,     500,600}, 10000,      10,       0   },
    { {DATA_RANDOM,     20,25,     500,600}, 10000,      10,       1   },
  };

  int i;

  for(i=0; *pRc==LSM_OK && i<ArraySize(aTest); i++){
    char *zName = getName4(zSystem, &aTest[i]);
    if( testCaseBegin(pRc, zPattern, "%s", zName) ){
      doDataTest4(zSystem, &aTest[i], pRc);
    }
    testFree(zName);
  }
}



Added ext/lsm1/lsm-test/lsmtest_bt.c.























































































































































>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75

#include "lsmtest.h"
#include "bt.h"

int do_bt(int nArg, char **azArg){
  struct Option {
    const char *zName;
    int bPgno;
    int eOpt;
  } aOpt [] = { 
    { "dbhdr",          0, BT_INFO_HDRDUMP },
    { "filename",       0, BT_INFO_FILENAME },
    { "block_freelist", 0, BT_INFO_BLOCK_FREELIST },
    { "page_freelist",  0, BT_INFO_PAGE_FREELIST },
    { "filename",       0, BT_INFO_FILENAME },
    { "page",           1, BT_INFO_PAGEDUMP },
    { "page_ascii",     1, BT_INFO_PAGEDUMP_ASCII },
    { "leaks",          0, BT_INFO_PAGE_LEAKS },
    { 0, 0 } 
  };
  int iOpt;
  int rc;
  bt_info buf;
  char *zOpt;
  char *zFile;

  bt_db *db = 0;

  if( nArg<2 ){
    testPrintUsage("FILENAME OPTION ...");
    return -1;
  }
  zFile = azArg[0];
  zOpt = azArg[1];

  rc = testArgSelect(aOpt, "option", zOpt, &iOpt);
  if( rc!=0 ) return rc;
  if( nArg!=2+aOpt[iOpt].bPgno ){
    testPrintFUsage("FILENAME %s %s", zOpt, aOpt[iOpt].bPgno ? "PGNO" : "");
    return -4;
  }

  rc = sqlite4BtNew(sqlite4_env_default(), 0, &db);
  if( rc!=SQLITE4_OK ){
    testPrintError("sqlite4BtNew() failed: %d", rc);
    return -2;
  }
  rc = sqlite4BtOpen(db, zFile);
  if( rc!=SQLITE4_OK ){
    testPrintError("sqlite4BtOpen() failed: %d", rc);
    return -3;
  }

  buf.eType = aOpt[iOpt].eOpt;
  buf.pgno = 0;
  sqlite4_buffer_init(&buf.output, 0);

  if( aOpt[iOpt].bPgno ){
    buf.pgno = (u32)atoi(azArg[2]);
  }

  rc = sqlite4BtControl(db, BT_CONTROL_INFO, &buf);
  if( rc!=SQLITE4_OK ){
    testPrintError("sqlite4BtControl() failed: %d\n", rc);
    return -4;
  }

  printf("%s\n", (char*)buf.output.p);
  sqlite4_buffer_clear(&buf.output);
  return 0;
}




Added ext/lsm1/lsm-test/lsmtest_datasource.c.

































































































































































































>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96


#include "lsmtest.h"

struct Datasource {
  int eType;

  int nMinKey;
  int nMaxKey;
  int nMinVal;
  int nMaxVal;

  char *aKey;
  char *aVal;
};

void testDatasourceEntry(
  Datasource *p, 
  int iData, 
  void **ppKey, int *pnKey,
  void **ppVal, int *pnVal
){
  assert( (ppKey==0)==(pnKey==0) );
  assert( (ppVal==0)==(pnVal==0) );

  if( ppKey ){
    int nKey = 0;
    switch( p->eType ){
      case TEST_DATASOURCE_RANDOM: {
        int nRange = (1 + p->nMaxKey - p->nMinKey);
        nKey = (int)( testPrngValue((u32)iData) % nRange ) + p->nMinKey; 
        testPrngString((u32)iData, p->aKey, nKey);
        break;
      }
      case TEST_DATASOURCE_SEQUENCE:
        nKey = sprintf(p->aKey, "%012d", iData);
        break;
    }
    *ppKey = p->aKey;
    *pnKey = nKey;
  }
  if( ppVal ){
    u32 nVal = testPrngValue((u32)iData)%(1+p->nMaxVal-p->nMinVal)+p->nMinVal;
    testPrngString((u32)~iData, p->aVal, (int)nVal);
    *ppVal = p->aVal;
    *pnVal = (int)nVal;
  }
}

void testDatasourceFree(Datasource *p){
  testFree(p);
}

/*
** Return a pointer to a nul-terminated string that corresponds to the
** contents of the datasource-definition passed as the first argument.
** The caller should eventually free the returned pointer using testFree().
*/
char *testDatasourceName(const DatasourceDefn *p){
  char *zRet;
  zRet = testMallocPrintf("%s.(%d-%d).(%d-%d)",
      (p->eType==TEST_DATASOURCE_SEQUENCE ? "seq" : "rnd"),
      p->nMinKey, p->nMaxKey,
      p->nMinVal, p->nMaxVal
  );
  return zRet;
}

Datasource *testDatasourceNew(const DatasourceDefn *pDefn){
  Datasource *p;
  int nMinKey; 
  int nMaxKey;
  int nMinVal;
  int nMaxVal; 

  if( pDefn->eType==TEST_DATASOURCE_SEQUENCE ){
    nMinKey = 128;
    nMaxKey = 128;
  }else{
    nMinKey = MAX(0, pDefn->nMinKey);
    nMaxKey = MAX(nMinKey, pDefn->nMaxKey);
  }
  nMinVal = MAX(0, pDefn->nMinVal);
  nMaxVal = MAX(nMinVal, pDefn->nMaxVal);

  p = (Datasource *)testMalloc(sizeof(Datasource) + nMaxKey + nMaxVal + 1);
  p->eType = pDefn->eType;
  p->nMinKey = nMinKey;
  p->nMinVal = nMinVal;
  p->nMaxKey = nMaxKey;
  p->nMaxVal = nMaxVal;
  
  p->aKey = (char *)&p[1];
  p->aVal = &p->aKey[nMaxKey];
  return p;
};

Added ext/lsm1/lsm-test/lsmtest_func.c.



































































































































































































































































































































































>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177

#include "lsmtest.h"


int do_work(int nArg, char **azArg){
  struct Option {
    const char *zName;
  } aOpt [] = {
    { "-nmerge" },
    { "-nkb" },
    { 0 }
  };

  lsm_db *pDb;
  int rc;
  int i;
  const char *zDb;
  int nMerge = 1;
  int nKB = (1<<30);

  if( nArg==0 ) goto usage;
  zDb = azArg[nArg-1];
  for(i=0; i<(nArg-1); i++){
    int iSel;
    rc = testArgSelect(aOpt, "option", azArg[i], &iSel);
    if( rc ) return rc;
    switch( iSel ){
      case 0:
        i++;
        if( i==(nArg-1) ) goto usage;
        nMerge = atoi(azArg[i]);
        break;
      case 1:
        i++;
        if( i==(nArg-1) ) goto usage;
        nKB = atoi(azArg[i]);
        break;
    }
  }

  rc = lsm_new(0, &pDb);
  if( rc!=LSM_OK ){
    testPrintError("lsm_open(): rc=%d\n", rc);
  }else{
    rc = lsm_open(pDb, zDb);
    if( rc!=LSM_OK ){
      testPrintError("lsm_open(): rc=%d\n", rc);
    }else{
      int n = -1;
      lsm_config(pDb, LSM_CONFIG_BLOCK_SIZE, &n);
      n = n*2;
      lsm_config(pDb, LSM_CONFIG_AUTOCHECKPOINT, &n);

      rc = lsm_work(pDb, nMerge, nKB, 0);
      if( rc!=LSM_OK ){
        testPrintError("lsm_work(): rc=%d\n", rc);
      }
    }
  }
  if( rc==LSM_OK ){
    rc = lsm_checkpoint(pDb, 0);
  }

  lsm_close(pDb);
  return rc;

 usage:
  testPrintUsage("?-optimize? ?-n N? DATABASE");
  return -1;
}


/*
**   lsmtest show ?-config LSM-CONFIG? DATABASE ?COMMAND ?PGNO??
*/
int do_show(int nArg, char **azArg){
  lsm_db *pDb;
  int rc;
  const char *zDb;

  int eOpt = LSM_INFO_DB_STRUCTURE;
  unsigned int iPg = 0;
  int bConfig = 0;
  const char *zConfig = "";

  struct Option {
    const char *zName;
    int bConfig;
    int eOpt;
  } aOpt [] = { 
    { "array",       0, LSM_INFO_ARRAY_STRUCTURE },
    { "array-pages", 0, LSM_INFO_ARRAY_PAGES },
    { "blocksize",   1, LSM_CONFIG_BLOCK_SIZE },
    { "pagesize",    1, LSM_CONFIG_PAGE_SIZE },
    { "freelist",    0, LSM_INFO_FREELIST },
    { "page-ascii",  0, LSM_INFO_PAGE_ASCII_DUMP },
    { "page-hex",    0, LSM_INFO_PAGE_HEX_DUMP },
    { 0, 0 } 
  };

  char *z = 0; 
  int iDb = 0;                    /* Index of DATABASE in azArg[] */

  /* Check if there is a "-config" option: */
  if( nArg>2 && strlen(azArg[0])>1 
   && memcmp(azArg[0], "-config", strlen(azArg[0]))==0
  ){
    zConfig = azArg[1];
    iDb = 2;
  }
  if( nArg<(iDb+1) ) goto usage;

  if( nArg>(iDb+1) ){
    rc = testArgSelect(aOpt, "option", azArg[iDb+1], &eOpt);
    if( rc!=0 ) return rc;
    bConfig = aOpt[eOpt].bConfig;
    eOpt = aOpt[eOpt].eOpt;
    if( (bConfig==0 && eOpt==LSM_INFO_FREELIST)
     || (bConfig==1 && eOpt==LSM_CONFIG_BLOCK_SIZE)
     || (bConfig==1 && eOpt==LSM_CONFIG_PAGE_SIZE)
    ){
      if( nArg!=(iDb+2) ) goto usage;
    }else{
      if( nArg!=(iDb+3) ) goto usage;
      iPg = atoi(azArg[iDb+2]);
    }
  }
  zDb = azArg[iDb];

  rc = lsm_new(0, &pDb);
  tdb_lsm_configure(pDb, zConfig);
  if( rc!=LSM_OK ){
    testPrintError("lsm_new(): rc=%d\n", rc);
  }else{
    rc = lsm_open(pDb, zDb);
    if( rc!=LSM_OK ){
      testPrintError("lsm_open(): rc=%d\n", rc);
    }
  }

  if( rc==LSM_OK ){
    if( bConfig==0 ){
      switch( eOpt ){
        case LSM_INFO_DB_STRUCTURE:
        case LSM_INFO_FREELIST:
          rc = lsm_info(pDb, eOpt, &z);
          break;
        case LSM_INFO_ARRAY_STRUCTURE:
        case LSM_INFO_ARRAY_PAGES:
        case LSM_INFO_PAGE_ASCII_DUMP:
        case LSM_INFO_PAGE_HEX_DUMP:
          rc = lsm_info(pDb, eOpt, iPg, &z);
          break;
        default:
          assert( !"no chance" );
      }

      if( rc==LSM_OK ){
        printf("%s\n", z ? z : "");
        fflush(stdout);
      }
      lsm_free(lsm_get_env(pDb), z);
    }else{
      int iRes = -1;
      lsm_config(pDb, eOpt, &iRes);
      printf("%d\n", iRes);
      fflush(stdout);
    }
  }

  lsm_close(pDb);
  return rc;

 usage:
  testPrintUsage("DATABASE ?array|page-ascii|page-hex PGNO?");
  return -1;
}

Added ext/lsm1/lsm-test/lsmtest_io.c.

































































































































































































































































































































































































































































































































>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256

/*
** SUMMARY
**
**   This file implements the 'io' subcommand of the test program. It is used
**   for testing the performance of various combinations of write() and fsync()
**   system calls. All operations occur on a single file, which may or may not
**   exist when a test is started.
**
**   A test consists of a series of commands. Each command is either a write
**   or an fsync. A write is specified as "<amount>@<offset>", where <amount>
**   is the amount of data written, and <offset> is the offset of the file
**   to write to. An <amount> or an <offset> is specified as an integer number
**   of bytes. Or, if postfixed with a "K", "M" or "G", an integer number of
**   KB, MB or GB, respectively. An fsync is simply "S". All commands are
**   case-insensitive.
**
**   Example test program:
**
**        2M@6M 1492K@4M S 4096@4K S
**
**   This program writes 2 MB of data starting at the offset 6MB offset of
**   the file, followed by 1492 KB of data written at the 4MB offset of the
**   file, followed by a call to fsync(), a write of 4KB of data at byte
**   offset 4096, and finally another call to fsync().
**
**   Commands may either be specified on the command line (one command per
**   command line argument) or read from stdin. Commands read from stdin
**   must be separated by white-space.
**
** COMMAND LINE INVOCATION
**
**   The sub-command implemented in this file must be invoked with at least
**   two arguments - the path to the file to write to and the page-size to
**   use for writing. If there are more than two arguments, then each
**   subsequent argument is assumed to be a test command. If there are exactly
**   two arguments, the test commands are read from stdin.
**
**   A write command does not result in a single call to system call write().
**   Instead, the specified region is written sequentially using one or
**   more calls to write(), each of which writes not more than one page of
**   data. For example, if the page-size is 4KB, the command "2M@6M" results
**   in 512 calls to write(), each of which writes 4KB of data.
**
** EXAMPLES
**
**   Two equivalent examples:
**
**     $ lsmtest io testfile.db 4KB 2M@6M 1492K@4M S 4096@4K S
**     3544K written in 129 ms
**     $ echo "2M@6M 1492K@4M S 4096@4K S" | lsmtest io testfile.db 4096 
**     3544K written in 127 ms
**
*/

#include "lsmtest.h"

#include <sys/types.h>
#include <sys/stat.h>
#include <fcntl.h>
#ifndef _WIN32
# include <unistd.h>
#endif
#include <ctype.h>

typedef struct IoContext IoContext;

struct IoContext {
  int fd;
  int nWrite;
};

/*
** As isspace(3)
*/
static int safe_isspace(char c){
  if( c&0x80) return 0;
  return isspace(c);
}

/*
** As isdigit(3)
*/
static int safe_isdigit(char c){
  if( c&0x80) return 0;
  return isdigit(c);
}

static i64 getNextSize(char *zIn, char **pzOut, int *pRc){
  i64 iRet = 0;
  if( *pRc==0 ){
    char *z = zIn;

    if( !safe_isdigit(*z) ){
      *pRc = 1;
      return 0;
    }

    /* Process digits */
    while( safe_isdigit(*z) ){
      iRet = iRet*10 + (*z - '0');
      z++;
    }

    /* Process suffix */
    switch( *z ){
      case 'k': case 'K':
        iRet = iRet * 1024;
        z++;
        break;

      case 'm': case 'M':
        iRet = iRet * 1024 * 1024;
        z++;
        break;

      case 'g': case 'G':
        iRet = iRet * 1024 * 1024 * 1024;
        z++;
        break;
    }

    if( pzOut ) *pzOut = z;
  }
  return iRet;
}

static int doOneCmd(
  IoContext *pCtx,
  u8 *aData,
  int pgsz,
  char *zCmd,
  char **pzOut
){
  char c;
  char *z = zCmd;

  while( safe_isspace(*z) ) z++;
  c = *z;

  if( c==0 ){
    if( pzOut ) *pzOut = z;
    return 0;
  }

  if( c=='s' || c=='S' ){
    if( pzOut ) *pzOut = &z[1];
    return fdatasync(pCtx->fd);
  }

  if( safe_isdigit(c) ){
    i64 iOff = 0;
    int nByte = 0;
    int rc = 0;
    int nPg;
    int iPg;

    nByte = getNextSize(z, &z, &rc);
    if( rc || *z!='@' ) goto bad_command;
    z++;
    iOff = getNextSize(z, &z, &rc);
    if( rc || (safe_isspace(*z)==0 && *z!='\0') ) goto bad_command;
    if( pzOut ) *pzOut = z;

    nPg = (nByte+pgsz-1) / pgsz;
    lseek(pCtx->fd, iOff, SEEK_SET);
    for(iPg=0; iPg<nPg; iPg++){
      write(pCtx->fd, aData, pgsz);
    }
    pCtx->nWrite += nByte/1024;

    return 0;
  }

 bad_command:
  testPrintError("unrecognized command: %s", zCmd);
  return 1;
}

static int readStdin(char **pzOut){
  int nAlloc = 128;
  char *zOut = 0;
  int nOut = 0;

  while( !feof(stdin) ){
    int nRead;

    nAlloc = nAlloc*2;
    zOut = realloc(zOut, nAlloc);
    nRead = fread(&zOut[nOut], 1, nAlloc-nOut-1, stdin);

    if( nRead==0 ) break;
    nOut += nRead;
    zOut[nOut] = '\0';
  }

  *pzOut = zOut;
  return 0;
}

int do_io(int nArg, char **azArg){
  IoContext ctx;
  int pgsz;
  char *zFile;
  char *zPgsz;
  int i;
  int rc = 0;

  char *zStdin = 0;
  char *z;

  u8 *aData;

  memset(&ctx, 0, sizeof(IoContext));
  if( nArg<2 ){
    testPrintUsage("FILE PGSZ ?CMD-1 ...?");
    return -1;
  }
  zFile = azArg[0];
  zPgsz = azArg[1];

  pgsz = getNextSize(zPgsz, 0, &rc);
  if( pgsz<=0 ){
    testPrintError("Ridiculous page size: %d", pgsz);
    return -1;
  }
  aData = malloc(pgsz);
  memset(aData, 0x77, pgsz);

  ctx.fd = open(zFile, O_RDWR|O_CREAT, 0644);
  if( ctx.fd<0 ){
    perror("open: ");
    return -1;
  }

  if( nArg==2 ){
    readStdin(&zStdin);
    testTimeInit();
    z = zStdin;
    while( *z && rc==0 ){
      rc = doOneCmd(&ctx, aData, pgsz, z, &z);
    }
  }else{
    testTimeInit();
    for(i=2; i<nArg; i++){
      rc = doOneCmd(&ctx, aData, pgsz, azArg[i], 0);
    }
  }

  printf("%dK written in %d ms\n", ctx.nWrite, testTimeGet());

  free(zStdin);
  close(ctx.fd);

  return 0;
}

Added ext/lsm1/lsm-test/lsmtest_main.c.







































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
1388
1389
1390
1391
1392
1393
1394
1395
1396
1397
1398
1399
1400
1401
1402
1403
1404
1405
1406
1407
1408
1409
1410
1411
1412
1413
1414
1415
1416
1417
1418
1419
1420
1421
1422
1423
1424
1425
1426
1427
1428
1429
1430
1431
1432
1433
1434
1435
1436
1437
1438
1439
1440
1441
1442
1443
1444
1445
1446
1447
1448
1449
1450
1451
1452
1453
1454
1455
1456
1457
1458
1459
1460
1461
1462
1463
1464
1465
1466
1467
1468
1469
1470
1471
1472
1473
1474
1475
1476
1477
1478
1479
1480
1481
1482
1483
1484
1485
1486
1487
1488
1489
1490
1491
1492
1493
1494
1495
1496
1497
1498
1499
1500
1501
1502
1503
1504
1505
1506
1507
1508
1509
1510
1511
1512
1513
1514
1515
1516
1517
1518
1519
1520
1521
1522
1523
1524
1525
1526
1527
1528
1529
1530
1531
1532
1533
1534
1535
1536
1537
1538
1539
1540
1541
1542
1543
1544
1545
1546
1547
1548
1549
1550
1551
1552
1553
1554
1555

#include "stdarg.h"
#include "lsmtest.h"
#include "stdio.h"
#include "assert.h"
#include "string.h"
#include "stdlib.h"

#include <sqlite3.h>

#ifndef _WIN32
# include <unistd.h>
#endif
#include <sys/types.h>
#include <sys/stat.h>
#include <fcntl.h>
#include <errno.h>


void test_failed(){ 
  assert( 0 );
  return; 
}

#define testSetError(rc) testSetErrorFunc(rc, pRc, __FILE__, __LINE__)
static void testSetErrorFunc(int rc, int *pRc, const char *zFile, int iLine){
  if( rc ){
    *pRc = rc;
    printf("FAILED (%s:%d) rc=%d ", zFile, iLine, rc);
    test_failed();
  }
}

static int lsm_memcmp(u8 *a, u8 *b, int c){
  int i;
  for(i=0; i<c; i++){
    if( a[i]!=b[i] ) return a[i] - b[i];
  }
  return 0;
}

/*
** A test utility function.
*/
void testFetch(
  TestDb *pDb,                    /* Database handle */
  void *pKey, int nKey,           /* Key to query database for */
  void *pVal, int nVal,           /* Expected value */
  int *pRc                        /* IN/OUT: Error code */
){
  if( *pRc==0 ){
    void *pDbVal;
    int nDbVal;
    int rc;

    static int nCall = 0; nCall++;

    rc = tdb_fetch(pDb, pKey, nKey, &pDbVal, &nDbVal);
    testSetError(rc);
    if( rc==0 && (nVal!=nDbVal || (nVal>0 && lsm_memcmp(pVal, pDbVal, nVal))) ){
      testSetError(1);
    }
  }
}

void testWrite(
  TestDb *pDb,                    /* Database handle */
  void *pKey, int nKey,           /* Key to query database for */
  void *pVal, int nVal,           /* Value to write */
  int *pRc                        /* IN/OUT: Error code */
){
  if( *pRc==0 ){
    int rc;
static int nCall = 0;
nCall++;
    rc = tdb_write(pDb, pKey, nKey, pVal, nVal);
    testSetError(rc);
  }
}
void testDelete(
  TestDb *pDb,                    /* Database handle */
  void *pKey, int nKey,           /* Key to query database for */
  int *pRc                        /* IN/OUT: Error code */
){
  if( *pRc==0 ){
    int rc;
    *pRc = rc = tdb_delete(pDb, pKey, nKey);
    testSetError(rc);
  }
}
void testDeleteRange(
  TestDb *pDb,                    /* Database handle */
  void *pKey1, int nKey1,
  void *pKey2, int nKey2,
  int *pRc                        /* IN/OUT: Error code */
){
  if( *pRc==0 ){
    int rc;
    *pRc = rc = tdb_delete_range(pDb, pKey1, nKey1, pKey2, nKey2);
    testSetError(rc);
  }
}

void testBegin(TestDb *pDb, int iTrans, int *pRc){
  if( *pRc==0 ){
    int rc;
    rc = tdb_begin(pDb, iTrans);
    testSetError(rc);
  }
}
void testCommit(TestDb *pDb, int iTrans, int *pRc){
  if( *pRc==0 ){
    int rc;
    rc = tdb_commit(pDb, iTrans);
    testSetError(rc);
  }
}
static void testRollback(TestDb *pDb, int iTrans, int *pRc){
  if( *pRc==0 ){
    int rc;
    rc = tdb_rollback(pDb, iTrans);
    testSetError(rc);
  }
}

void testWriteStr(
  TestDb *pDb,                    /* Database handle */
  const char *zKey,               /* Key to query database for */
  const char *zVal,               /* Value to write */
  int *pRc                        /* IN/OUT: Error code */
){
  int nVal = (zVal ? strlen(zVal) : 0);
  testWrite(pDb, (void *)zKey, strlen(zKey), (void *)zVal, nVal, pRc);
}

static void testDeleteStr(TestDb *pDb, const char *zKey, int *pRc){
  testDelete(pDb, (void *)zKey, strlen(zKey), pRc);
}

void testFetchStr(
  TestDb *pDb,                    /* Database handle */
  const char *zKey,               /* Key to query database for */
  const char *zVal,               /* Value to write */
  int *pRc                        /* IN/OUT: Error code */
){
  int nVal = (zVal ? strlen(zVal) : 0);
  testFetch(pDb, (void *)zKey, strlen(zKey), (void *)zVal, nVal, pRc);
}

void testFetchCompare(
  TestDb *pControl, 
  TestDb *pDb, 
  void *pKey, int nKey, 
  int *pRc
){
  int rc;
  void *pDbVal1;
  void *pDbVal2;
  int nDbVal1;
  int nDbVal2;

  static int nCall = 0;
  nCall++;

  rc = tdb_fetch(pControl, pKey, nKey, &pDbVal1, &nDbVal1);
  testSetError(rc);

  rc = tdb_fetch(pDb, pKey, nKey, &pDbVal2, &nDbVal2);
  testSetError(rc);

  if( *pRc==0 
   && (nDbVal1!=nDbVal2 || (nDbVal1>0 && memcmp(pDbVal1, pDbVal2, nDbVal1)))
  ){
    testSetError(1);
  }
}

typedef struct ScanResult ScanResult;
struct ScanResult {
  TestDb *pDb;

  int nRow;
  u32 cksum1;
  u32 cksum2;
  void *pKey1; int nKey1;
  void *pKey2; int nKey2;

  int bReverse;
  int nPrevKey;
  u8 aPrevKey[256];
};

static int keyCompare(void *pKey1, int nKey1, void *pKey2, int nKey2){
  int res;
  res = memcmp(pKey1, pKey2, MIN(nKey1, nKey2));
  if( res==0 ){
    res = nKey1 - nKey2;
  }
  return res;
}

int test_scan_debug = 0;

static void scanCompareCb(
  void *pCtx, 
  void *pKey, int nKey,
  void *pVal, int nVal
){
  ScanResult *p = (ScanResult *)pCtx;
  u8 *aKey = (u8 *)pKey;
  u8 *aVal = (u8 *)pVal;
  int i;

  if( test_scan_debug ){
    printf("%d: %.*s\n", p->nRow, nKey, (char *)pKey);
    fflush(stdout);
  }
#if 0
  if( test_scan_debug ) printf("%.20s\n", (char *)pVal);
#endif

#if 0
  /* Check tdb_fetch() matches */
  int rc = 0;
  testFetch(p->pDb, pKey, nKey, pVal, nVal, &rc);
  assert( rc==0 );
#endif

  /* Update the checksum data */
  p->nRow++;
  for(i=0; i<nKey; i++){
    p->cksum1 += ((int)aKey[i] << (i&0x0F));
    p->cksum2 += p->cksum1;
  }
  for(i=0; i<nVal; i++){
    p->cksum1 += ((int)aVal[i] << (i&0x0F));
    p->cksum2 += p->cksum1;
  }

  /* Check that the delivered row is not out of order. */
  if( nKey<(int)sizeof(p->aPrevKey) ){
    if( p->nPrevKey ){
      int res = keyCompare(p->aPrevKey, p->nPrevKey, pKey, nKey);
      if( (res<0 && p->bReverse) || (res>0 && p->bReverse==0) ){
        testPrintError("Returned key out of order at %s:%d\n", 
            __FILE__, __LINE__
        );
      }
    }

    p->nPrevKey = nKey;
    memcpy(p->aPrevKey, pKey, MIN(p->nPrevKey, nKey));
  }

  /* Check that the delivered row is within range. */
  if( p->pKey1 && (
      (memcmp(p->pKey1, pKey, MIN(p->nKey1, nKey))>0)
   || (memcmp(p->pKey1, pKey, MIN(p->nKey1, nKey))==0 && p->nKey1>nKey)
  )){
    testPrintError("Returned key too small at %s:%d\n", __FILE__, __LINE__);
  }
  if( p->pKey2 && (
      (memcmp(p->pKey2, pKey, MIN(p->nKey2, nKey))<0)
   || (memcmp(p->pKey2, pKey, MIN(p->nKey2, nKey))==0 && p->nKey2<nKey)
  )){
    testPrintError("Returned key too large at %s:%d\n", __FILE__, __LINE__);
  }

}

/*
** Scan the contents of the two databases. Check that they match.
*/
void testScanCompare(
  TestDb *pDb1,                   /* Control (trusted) database */
  TestDb *pDb2,                   /* Database being tested */
  int bReverse,
  void *pKey1, int nKey1, 
  void *pKey2, int nKey2, 
  int *pRc
){
  static int nCall = 0; nCall++;
  if( *pRc==0 ){
    ScanResult res1;
    ScanResult res2;
    void *pRes1 = (void *)&res1;
    void *pRes2 = (void *)&res2;

    memset(&res1, 0, sizeof(ScanResult));
    memset(&res2, 0, sizeof(ScanResult));

    res1.pDb = pDb1;
    res1.nKey1 = nKey1; res1.pKey1 = pKey1;
    res1.nKey2 = nKey2; res1.pKey2 = pKey2;
    res1.bReverse = bReverse;
    res2.pDb = pDb2;
    res2.nKey1 = nKey1; res2.pKey1 = pKey1;
    res2.nKey2 = nKey2; res2.pKey2 = pKey2;
    res2.bReverse = bReverse;

    tdb_scan(pDb1, pRes1, bReverse, pKey1, nKey1, pKey2, nKey2, scanCompareCb);
if( test_scan_debug ) printf("\n\n\n");
    tdb_scan(pDb2, pRes2, bReverse, pKey1, nKey1, pKey2, nKey2, scanCompareCb);
if( test_scan_debug ) printf("\n\n\n");

    if( res1.nRow!=res2.nRow 
     || res1.cksum1!=res2.cksum1 
     || res1.cksum2!=res2.cksum2
    ){
      printf("expected: %d %X %X\n", res1.nRow, res1.cksum1, res1.cksum2);
      printf("got:      %d %X %X\n", res2.nRow, res2.cksum1, res2.cksum2);
      testSetError(1);
      *pRc = 1;
    }
  }
}

void testClose(TestDb **ppDb){
  tdb_close(*ppDb);
  *ppDb = 0;
}

TestDb *testOpen(const char *zSystem, int bClear, int *pRc){
  TestDb *pDb = 0;
  if( *pRc==0 ){
    int rc;
    rc = tdb_open(zSystem, 0, bClear, &pDb);
    if( rc!=0 ){
      testSetError(rc);
      *pRc = rc;
    }
  }
  return pDb;
}

void testReopen(TestDb **ppDb, int *pRc){
  if( *pRc==0 ){
    const char *zLib;
    zLib = tdb_library_name(*ppDb);
    testClose(ppDb);
    *pRc = tdb_open(zLib, 0, 0, ppDb);
  }
}


static void testSystemSelect(const char *zSys, int *piSel, int *pRc){
  if( *pRc==0 ){
    struct SysName { const char *zName; } *aName;
    int nSys;
    int i;

    for(nSys=0; tdb_system_name(nSys); nSys++);
    aName = malloc(sizeof(struct SysName) * (nSys+1));
    for(i=0; i<=nSys; i++){
      aName[i].zName = tdb_system_name(i);
    }

    *pRc = testArgSelect(aName, "db", zSys, piSel);
    free(aName);
  }
}

char *testMallocVPrintf(const char *zFormat, va_list ap){
  int nByte;
  va_list copy;
  char *zRet;

  __va_copy(copy, ap);
  nByte = vsnprintf(0, 0, zFormat, copy);
  va_end(copy);

  assert( nByte>=0 );
  zRet = (char *)testMalloc(nByte+1);
  vsnprintf(zRet, nByte+1, zFormat, ap);
  return zRet;
}

char *testMallocPrintf(const char *zFormat, ...){
  va_list ap;
  char *zRet;

  va_start(ap, zFormat);
  zRet = testMallocVPrintf(zFormat, ap);
  va_end(ap);

  return zRet;
}


/*
** A wrapper around malloc(3).
**
** This function should be used for all allocations made by test procedures.
** It has the following properties:
**
**   * Test code may assume that allocations may not fail.
**   * Returned memory is always zeroed.
**
** Allocations made using testMalloc() should be freed using testFree().
*/
void *testMalloc(int n){
  u8 *p = (u8*)malloc(n + 8);
  memset(p, 0, n+8);
  *(int*)p = n;
  return (void*)&p[8];
}

void *testMallocCopy(void *pCopy, int nByte){
  void *pRet = testMalloc(nByte);
  memcpy(pRet, pCopy, nByte);
  return pRet;
}

void *testRealloc(void *ptr, int n){
  if( ptr ){
    u8 *p = (u8*)ptr - 8;
    int nOrig =  *(int*)p;
    p = (u8*)realloc(p, n+8);
    if( nOrig<n ){
      memset(&p[8+nOrig], 0, n-nOrig);
    }
    *(int*)p = n;
    return (void*)&p[8];
  }
  return testMalloc(n);
}

/*
** Free an allocation made by an earlier call to testMalloc().
*/
void testFree(void *ptr){
  if( ptr ){
    u8 *p = (u8*)ptr - 8;
    memset(p, 0x55, *(int*)p + 8);
    free(p);
  }
}

/*
** String zPattern contains a glob pattern. Return true if zStr matches 
** the pattern, or false if it does not.
*/
int testGlobMatch(const char *zPattern, const char *zStr){
  int i = 0;
  int j = 0;

  while( zPattern[i] ){
    char p = zPattern[i];

    if( p=='*' || p=='%' ){
      do {
        if( testGlobMatch(&zPattern[i+1], &zStr[j]) ) return 1;
      }while( zStr[j++] );
      return 0;
    }

    if( zStr[j]==0 || (p!='?' && p!=zStr[j]) ){
      /* Match failed. */
      return 0;
    }

    j++;
    i++;
  }

  return (zPattern[i]==0 && zStr[j]==0);
}

/* 
** End of test utilities 
**************************************************************************/

int do_test(int nArg, char **azArg){
  int j;
  int rc;
  int nFail = 0;
  const char *zPattern = 0;

  if( nArg>1 ){
    testPrintError("Usage: test ?PATTERN?\n");
    return 1;
  }
  if( nArg==1 ){
    zPattern = azArg[0];
  }

  for(j=0; tdb_system_name(j); j++){
    rc = 0;

    test_data_1(tdb_system_name(j), zPattern, &rc);
    test_data_2(tdb_system_name(j), zPattern, &rc);
    test_data_3(tdb_system_name(j), zPattern, &rc);
    test_data_4(tdb_system_name(j), zPattern, &rc);
    test_rollback(tdb_system_name(j), zPattern, &rc);
    test_mc(tdb_system_name(j), zPattern, &rc);
    test_mt(tdb_system_name(j), zPattern, &rc);

    if( rc ) nFail++;
  }

  rc = 0;
  test_oom(zPattern, &rc);
  if( rc ) nFail++;

  rc = 0;
  test_api(zPattern, &rc);
  if( rc ) nFail++;

  rc = 0;
  do_crash_test(zPattern, &rc);
  if( rc ) nFail++;

  rc = 0;
  do_writer_crash_test(zPattern, &rc);
  if( rc ) nFail++;

  return (nFail!=0);
}

static lsm_db *configure_lsm_db(TestDb *pDb){
  lsm_db *pLsm;
  pLsm = tdb_lsm(pDb);
  if( pLsm ){
    tdb_lsm_config_str(pDb, "mmap=1 autowork=1 automerge=4 worker_automerge=4");
  }
  return pLsm;
}

typedef struct WriteHookEvent WriteHookEvent;
struct WriteHookEvent {
  i64 iOff;
  int nData;
  int nUs;
};
WriteHookEvent prev = {0, 0, 0};

static void flushPrev(FILE *pOut){
  if( prev.nData ){
    fprintf(pOut, "w %s %lld %d %d\n", "d", prev.iOff, prev.nData, prev.nUs);
    prev.nData = 0;
  }
}

static void do_speed_write_hook2(
  void *pCtx,
  int bLog,
  i64 iOff,
  int nData,
  int nUs
){
  FILE *pOut = (FILE *)pCtx;
  if( bLog ) return;

  if( prev.nData && nData && iOff==prev.iOff+prev.nData ){
    prev.nData += nData;
    prev.nUs += nUs;
  }else{
    flushPrev(pOut);
    if( nData==0 ){
      fprintf(pOut, "s %s 0 0 %d\n", (bLog ? "l" : "d"), nUs);
    }else{
      prev.iOff = iOff;
      prev.nData = nData;
      prev.nUs = nUs;
    }
  }
}

#define ST_REPEAT  0
#define ST_WRITE   1
#define ST_PAUSE   2
#define ST_FETCH   3
#define ST_SCAN    4
#define ST_NSCAN   5
#define ST_KEYSIZE 6
#define ST_VALSIZE 7
#define ST_TRANS   8


static void print_speed_test_help(){
  printf(
"\n"
"Repeat the following $repeat times:\n"
"  1. Insert $write key-value pairs. One transaction for each write op.\n"
"  2. Pause for $pause ms.\n"
"  3. Perform $fetch queries on the database.\n"
"\n"
"  Keys are $keysize bytes in size. Values are $valsize bytes in size\n"
"  Both keys and values are pseudo-randomly generated\n"
"\n"
"Options are:\n"
"  -repeat  $repeat                 (default value 10)\n"
"  -write   $write                  (default value 10000)\n"
"  -pause   $pause                  (default value 0)\n"
"  -fetch   $fetch                  (default value 0)\n"
"  -keysize $keysize                (default value 12)\n"
"  -valsize $valsize                (default value 100)\n"
"  -system  $system                 (default value \"lsm\")\n"
"  -trans   $trans                  (default value 0)\n"
"\n"
);
}

int do_speed_test2(int nArg, char **azArg){
  struct Option {
    const char *zOpt;
    int eVal;
    int iDefault;
  } aOpt[] = {
    { "-repeat",  ST_REPEAT,    10},
    { "-write",   ST_WRITE,  10000},
    { "-pause",   ST_PAUSE,      0},
    { "-fetch",   ST_FETCH,      0},
    { "-scan",    ST_SCAN,       0},
    { "-nscan",   ST_NSCAN,      0},
    { "-keysize", ST_KEYSIZE,   12},
    { "-valsize", ST_VALSIZE,  100},
    { "-trans",   ST_TRANS,      0},
    { "-system",  -1,            0},
    { "help",     -2,            0},
    {0, 0, 0}
  };
  int i;
  int aParam[9];
  int rc = 0;
  int bReadonly = 0;
  int nContent = 0;

  TestDb *pDb;
  Datasource *pData;
  DatasourceDefn defn = { TEST_DATASOURCE_RANDOM, 0, 0, 0, 0 };
  char *zSystem = "";
  int bLsm = 1;
  FILE *pLog = 0;

#ifdef NDEBUG
  /* If NDEBUG is defined, disable the dynamic memory related checks in
  ** lsmtest_mem.c. They slow things down.  */
  testMallocUninstall(tdb_lsm_env());
#endif

  /* Initialize aParam[] with default values. */
  for(i=0; i<ArraySize(aOpt); i++){
    if( aOpt[i].zOpt ) aParam[aOpt[i].eVal] = aOpt[i].iDefault;
  }

  /* Process the command line switches. */
  for(i=0; i<nArg; i+=2){
    int iSel;
    rc = testArgSelect(aOpt, "switch", azArg[i], &iSel);
    if( rc ){
      return rc;
    }
    if( aOpt[iSel].eVal==-2 ){
      print_speed_test_help();
      return 0;
    }
    if( i+1==nArg ){
      testPrintError("option %s requires an argument\n", aOpt[iSel].zOpt);
      return 1;
    }
    if( aOpt[iSel].eVal>=0 ){
      aParam[aOpt[iSel].eVal] = atoi(azArg[i+1]);
    }else{
      int j;
      zSystem = azArg[i+1];
      bLsm = 0;
#if 0
      for(j=0; zSystem[j]; j++){
        if( zSystem[j]=='=' ) bLsm = 1;
      }
#endif
    }
  }
  
  printf("#");
  for(i=0; i<ArraySize(aOpt); i++){
    if( aOpt[i].zOpt ){
      if( aOpt[i].eVal>=0 ){
        printf(" %s=%d", &aOpt[i].zOpt[1], aParam[aOpt[i].eVal]);
      }else if( aOpt[i].eVal==-1 ){
        printf(" %s=\"%s\"", &aOpt[i].zOpt[1], zSystem);
      }
    }
  }
  printf("\n");

  defn.nMinKey = defn.nMaxKey = aParam[ST_KEYSIZE];
  defn.nMinVal = defn.nMaxVal = aParam[ST_VALSIZE];
  pData = testDatasourceNew(&defn);

  if( aParam[ST_WRITE]==0 ){
    bReadonly = 1;
  }

  if( bLsm ){
    rc = tdb_lsm_open(zSystem, "testdb.lsm", !bReadonly, &pDb);
  }else{
    pDb = testOpen(zSystem, !bReadonly, &rc);
  }
  if( rc!=0 ) return rc;
  if( bReadonly ){
    nContent = testCountDatabase(pDb);
  }

#if 0
  pLog = fopen("/tmp/speed.log", "w");
  tdb_lsm_write_hook(pDb, do_speed_write_hook2, (void *)pLog);
#endif

  for(i=0; i<aParam[ST_REPEAT] && rc==0; i++){
    int msWrite, msFetch;
    int iFetch;
    int nWrite = aParam[ST_WRITE];

    if( bReadonly ){
      msWrite = 0;
    }else{
      testTimeInit();

      if( aParam[ST_TRANS] ) testBegin(pDb, 2, &rc);
      testWriteDatasourceRange(pDb, pData, i*nWrite, nWrite, &rc);
      if( aParam[ST_TRANS] ) testCommit(pDb, 0, &rc);

      msWrite = testTimeGet();
      nContent += nWrite;
    }

    if( aParam[ST_PAUSE] ){
      if( aParam[ST_PAUSE]/1000 ) sleep(aParam[ST_PAUSE]/1000);
      if( aParam[ST_PAUSE]%1000 ) usleep(1000 * (aParam[ST_PAUSE]%1000));
    }

    if( aParam[ST_FETCH] ){
      testTimeInit();
      if( aParam[ST_TRANS] ) testBegin(pDb, 1, &rc);
      for(iFetch=0; iFetch<aParam[ST_FETCH]; iFetch++){
        int iKey = testPrngValue(i*nWrite+iFetch) % nContent;
#ifndef NDEBUG
        testDatasourceFetch(pDb, pData, iKey, &rc);
#else
        void *pKey; int nKey;           /* Database key to query for */
        void *pVal; int nVal;           /* Result of query */

        testDatasourceEntry(pData, iKey, &pKey, &nKey, 0, 0);
        rc = tdb_fetch(pDb, pKey, nKey, &pVal, &nVal);
        if( rc==0 && nVal<0 ) rc = 1;
        if( rc ) break;
#endif
      }
      if( aParam[ST_TRANS] ) testCommit(pDb, 0, &rc);
      msFetch = testTimeGet();
    }else{
      msFetch = 0;
    }

    if( i==(aParam[ST_REPEAT]-1) ){
      testTimeInit();
      testClose(&pDb);
      msWrite += testTimeGet();
    }

    printf("%d %d %d\n", i, msWrite, msFetch);
    fflush(stdout);
  }

  testClose(&pDb);
  testDatasourceFree(pData);

  if( pLog ){
    flushPrev(pLog);
    fclose(pLog);
  }
  return rc;
}

int do_speed_tests(int nArg, char **azArg){

  struct DbSystem {
    const char *zLibrary;
    const char *zColor;
  } aSys[] = {
    { "sqlite3",      "black" },
    { "leveldb",      "blue" },
    { "lsm",          "red" },
    { "lsm_mt2",      "orange" },
    { "lsm_mt3",      "purple" },
    { "kyotocabinet", "green" },
    {0, 0}
  };

  int i;
  int j;
  int rc;
  int nSleep = 0;                 /* ms of rest allowed between INSERT tests */
  int nRow = 0;                   /* Number of rows to insert into database */
  int nStep;                      /* Measure INSERT time after this many rows */
  int nSelStep;                   /* Measure SELECT time after this many rows */
  int nSelTest;                   /* Number of SELECTs to run for timing */
  int doReadTest = 1;
  int doWriteTest = 1;

  int *aTime;                     /* INSERT timing data */
  int *aWrite;                    /* Writes per nStep inserts */
  int *aSelTime;                  /* SELECT timing data */
  int isFirst = 1;
  int bSleep = 0;

  /* File to write gnuplot script to. */
  const char *zOut = "lsmtest_speed.gnuplot";

  u32 sys_mask = 0;

  testMallocUninstall(tdb_lsm_env());

  for(i=0; i<nArg; i++){
    struct Opt { 
      const char *zOpt; 
      int isSwitch;
    } aOpt[] = {
      { "sqlite3" , 0},
      { "leveldb" , 0},
      { "lsm" , 0},
      { "lsm_mt2" , 0},
      { "lsm_mt3" , 0},
      { "kyotocabinet" , 0},
      { "-rows"     , 1},
      { "-sleep"    , 2},
      { "-testmode" , 3},
      { "-out"      , 4},
      { 0, 0}
    };
    int iSel;

    rc = testArgSelect(aOpt, "argument", azArg[i], &iSel);
    if( rc ) return rc;

    if( aOpt[iSel].isSwitch ){
      i++;

      if( i>=nArg ){
        testPrintError("option %s requires an argument\n", aOpt[iSel].zOpt);
        return 1;
      }
      if( aOpt[iSel].isSwitch==1 ){
        nRow = atoi(azArg[i]);
      }
      if( aOpt[iSel].isSwitch==2 ){
        nSleep = atoi(azArg[i]);
      }
      if( aOpt[iSel].isSwitch==3 ){
        struct Mode {
          const char *zMode;
          int doReadTest;
          int doWriteTest;
        } aMode[] = {{"ro", 1, 0} , {"rw", 1, 1}, {"wo", 0, 1}, {0, 0, 0}};
        int iMode;
        rc = testArgSelect(aMode, "option", azArg[i], &iMode);
        if( rc ) return rc;
        doReadTest = aMode[iMode].doReadTest;
        doWriteTest = aMode[iMode].doWriteTest;
      }
      if( aOpt[iSel].isSwitch==4 ){
        /* The "-out FILE" switch. This option is used to specify a file to
        ** write the gnuplot script to. */
        zOut = azArg[i];
      }
    }else{
      /* A db name */
      rc = testArgSelect(aOpt, "system", azArg[i], &iSel);
      if( rc ) return rc;
      sys_mask |= (1<<iSel);
    }
  }

  if( sys_mask==0 ) sys_mask = (1 << 0) | (1 << 1) | (1 << 2) | (1 << 3);
  nRow = MAX(nRow, 100000);
  nStep = nRow/100;
  nSelStep = nRow/10;
  nSelTest = (nSelStep > 100000) ? 100000 : nSelStep;

  aTime = malloc(sizeof(int) * ArraySize(aSys) * nRow/nStep);
  aWrite = malloc(sizeof(int) * nRow/nStep);
  aSelTime = malloc(sizeof(int) * ArraySize(aSys) * nRow/nSelStep);

  /* This loop collects the INSERT speed data. */
  if( doWriteTest ){
    printf("Writing output to file \"%s\".\n",  zOut);

    for(j=0; aSys[j].zLibrary; j++){
      FILE *pLog = 0;
      TestDb *pDb;                  /* Database being tested */
      lsm_db *pLsm;
      int iDot = 0;
  
      if( ((1<<j)&sys_mask)==0 ) continue;
      if( bSleep && nSleep ) sqlite3_sleep(nSleep);
      bSleep = 1;

      testCaseBegin(&rc, 0, "speed.insert.%s", aSys[j].zLibrary);

      rc = tdb_open(aSys[j].zLibrary, 0, 1, &pDb);
      if( rc ) return rc;

      pLsm = configure_lsm_db(pDb);
#if 0
      pLog = fopen("/tmp/speed.log", "w");
      tdb_lsm_write_hook(pDb, do_speed_write_hook2, (void *)pLog);
#endif
  
      testTimeInit();
      for(i=0; i<nRow; i+=nStep){
        int iStep;
        int nWrite1, nWrite2;
        testCaseProgress(i, nRow, testCaseNDot(), &iDot);
        if( pLsm ) lsm_info(pLsm, LSM_INFO_NWRITE, &nWrite1);
        for(iStep=0; iStep<nStep; iStep++){
          u32 aKey[4];                  /* 16-byte key */
          u32 aVal[25];                 /* 100 byte value */
          testPrngArray(i+iStep, aKey, ArraySize(aKey));
          testPrngArray(i+iStep, aVal, ArraySize(aVal));
          rc = tdb_write(pDb, aKey, sizeof(aKey), aVal, sizeof(aVal));
        }
        aTime[(j*nRow+i)/nStep] = testTimeGet();
        if( pLsm ) lsm_info(pLsm, LSM_INFO_NWRITE, &nWrite2);
        aWrite[i/nStep] = nWrite2 - nWrite1;
      }

      tdb_close(pDb);
      if( pLog ) fclose(pLog);
      testCaseFinish(rc);
    }
  }

  /* This loop collects the SELECT speed data. */
  if( doReadTest ){
    for(j=0; aSys[j].zLibrary; j++){
      int iDot = 0;
      TestDb *pDb;                  /* Database being tested */

      if( ((1<<j)&sys_mask)==0 ) continue;
      if( bSleep && nSleep ) sqlite3_sleep(nSleep);
      bSleep = 1;

      testCaseBegin(&rc, 0, "speed.select.%s", aSys[j].zLibrary);

      if( doWriteTest ){
        rc = tdb_open(aSys[j].zLibrary, 0, 1, &pDb);
        if( rc ) return rc;
        configure_lsm_db(pDb);

        for(i=0; i<nRow; i+=nSelStep){
          int iStep;
          int iSel;
          testCaseProgress(i, nRow, testCaseNDot(), &iDot);
          for(iStep=0; iStep<nSelStep; iStep++){
            u32 aKey[4];                  /* 16-byte key */
            u32 aVal[25];                 /* 100 byte value */
            testPrngArray(i+iStep, aKey, ArraySize(aKey));
            testPrngArray(i+iStep, aVal, ArraySize(aVal));
            rc = tdb_write(pDb, aKey, sizeof(aKey), aVal, sizeof(aVal));
          }
    
          testTimeInit();
          for(iSel=0; iSel<nSelTest; iSel++){
            void *pDummy;
            int nDummy;
            u32 iKey;
            u32 aKey[4];                  /* 16-byte key */
    
            iKey = testPrngValue(iSel) % (i+nSelStep);
            testPrngArray(iKey, aKey, ArraySize(aKey));
            rc = tdb_fetch(pDb, aKey, sizeof(aKey), &pDummy, &nDummy);
          }
          aSelTime[(j*nRow+i)/nSelStep] = testTimeGet();
          tdb_fetch(pDb, 0, 0, 0, 0);
        }
      }else{
        int t;
        int iSel;

        rc = tdb_open(aSys[j].zLibrary, 0, 0, &pDb);
        configure_lsm_db(pDb);

        testTimeInit();
        for(iSel=0; rc==LSM_OK && iSel<nSelTest; iSel++){
          void *pDummy;
          int nDummy;
          u32 iKey;
          u32 aKey[4];                  /* 16-byte key */

          testCaseProgress(iSel, nSelTest, testCaseNDot(), &iDot);
    
          iKey = testPrngValue(iSel) % nRow;
          testPrngArray(iKey, aKey, ArraySize(aKey));
          rc = tdb_fetch(pDb, aKey, sizeof(aKey), &pDummy, &nDummy);

#ifndef NDEBUG
          u32 aVal[25];                 /* 100 byte value */
          testPrngArray(iKey, aVal, ArraySize(aVal));
          assert( nDummy==100 && memcmp(aVal, pDummy, 100)==0 );
#endif
        }
        if( rc!=LSM_OK ) return rc;

        t = testTimeGet();
        tdb_fetch(pDb, 0, 0, 0, 0);

        printf("%s: %d selects/second\n", 
            aSys[j].zLibrary, (int)((double)nSelTest*1000.0/t)
        );
      }

      tdb_close(pDb);
      testCaseFinish(rc);
    }
  }


  if( doWriteTest ){
    FILE *pOut = fopen(zOut, "w");
    if( !pOut ){
      printf("fopen(\"%s\", \"w\"): %s\n", zOut, strerror(errno));
      return 1;
    }

    fprintf(pOut, "set xlabel \"Rows Inserted\"\n");
    fprintf(pOut, "set ylabel \"Inserts per second\"\n");
    if( doReadTest ){
      fprintf(pOut, "set y2label \"Selects per second\"\n");
    }else if( sys_mask==(1<<2) ){
      fprintf(pOut, "set y2label \"Page writes per insert\"\n");
    }
    fprintf(pOut, "set yrange [0:*]\n");
    fprintf(pOut, "set y2range [0:*]\n");
    fprintf(pOut, "set xrange [%d:*]\n", MAX(nStep, nRow/20) );
    fprintf(pOut, "set ytics nomirror\n");
    fprintf(pOut, "set y2tics nomirror\n");
    fprintf(pOut, "set key box lw 0.01\n");
    fprintf(pOut, "plot ");
  
    for(j=0; aSys[j].zLibrary; j++){
      if( (1<<j)&sys_mask ){
        const char *zLib = aSys[j].zLibrary;
        fprintf(pOut, "%s\"-\" ti \"%s INSERT\" with lines lc rgb \"%s\" ", 
            (isFirst?"":", "), zLib, aSys[j].zColor
        );
        if( doReadTest ){
          fprintf(pOut, ", \"-\" ti \"%s SELECT\" "
                 "axis x1y2 with points lw 3 lc rgb \"%s\""
              , zLib, aSys[j].zColor
          );
        }
        isFirst = 0;
      }
    }

    assert( strcmp(aSys[2].zLibrary, "lsm")==0 );
    if( sys_mask==(1<<2) && !doReadTest ){
      fprintf(pOut, ", \"-\" ti \"lsm pages written\" "
        "axis x1y2 with boxes lw 1 lc rgb \"grey\""
      );
    }
  
    fprintf(pOut, "\n");
  
    for(j=0; aSys[j].zLibrary; j++){
      if( ((1<<j)&sys_mask)==0 ) continue;
      fprintf(pOut, "# Rows    Inserts per second\n");
      for(i=0; i<nRow; i+=nStep){
        int iTime = aTime[(j*nRow+i)/nStep];
        int ips = (int)((i+nStep)*1000.0 / (double)iTime);
        fprintf(pOut, "%d %d\n", i+nStep, ips);
      }
      fprintf(pOut, "end\n");
  
      if( doReadTest ){
        fprintf(pOut, "# Rows    Selects per second\n");
        for(i=0; i<nRow; i+=nSelStep){
          int sps = (int)(nSelTest*1000.0/(double)aSelTime[(j*nRow+i)/nSelStep]);
          fprintf(pOut, "%d %d\n", i+nSelStep, sps);
        }
        fprintf(pOut, "end\n");
      }else if( sys_mask==(1<<2) ){
        for(i=0; i<(nRow/nStep); i++){
          fprintf(pOut, "%d %f\n", i*nStep, (double)aWrite[i] / (double)nStep);
        }
        fprintf(pOut, "end\n");
      }
    }
  
    fprintf(pOut, "pause -1\n");
    fclose(pOut);
  }

  free(aTime);
  free(aSelTime);
  free(aWrite);
  testMallocInstall(tdb_lsm_env());
  return 0;
}

/*
** Usage: lsmtest random ?N?
**
** This command prints a sequence of zero or more numbers from the PRNG
** system to stdout. If the "N" argument is missing, values the first 10
** values (i=0, i=1, ... i=9) are printed. Otherwise, the first N.
**
** This was added to verify that the PRNG values do not change between
** runs of the lsmtest program.
*/
int do_random_tests(int nArg, char **azArg){
  int i;
  int nRand;
  if( nArg==0 ){
    nRand = 10;
  }else if( nArg==1 ){
    nRand = atoi(azArg[0]);
  }else{
    testPrintError("Usage: random ?N?\n");
    return -1;
  }
  for(i=0; i<nRand; i++){
    printf("0x%x\n", testPrngValue(i));
  }
  return 0;
}

static int testFormatSize(char *aBuf, int nBuf, i64 nByte){
  int res;
  if( nByte<(1<<10) ){
    res = snprintf(aBuf, nBuf, "%d byte", (int)nByte);
  }else if( nByte<(1<<20) ){
    res = snprintf(aBuf, nBuf, "%dK", (int)(nByte/(1<<10)));
  }else{
    res = snprintf(aBuf, nBuf, "%dM", (int)(nByte/(1<<20)));
  }
  return res;
}

static i64 testReadSize(char *z){
  int n = strlen(z);
  char c = z[n-1];
  i64 nMul = 1;

  switch( c ){
    case 'g': case 'G':
      nMul = (1<<30);
      break;

    case 'm': case 'M':
      nMul = (1<<20);
      break;

    case 'k': case 'K':
      nMul = (1<<10);
      break;

    default:
      nMul = 1;
  }

  return nMul * (i64)atoi(z);
} 

/*
** Usage: lsmtest writespeed FILESIZE BLOCKSIZE SYNCSIZE
*/
static int do_writer_test(int nArg, char **azArg){
  int nBlock;
  int nSize;
  int i;
  int fd;
  int ms;
  char aFilesize[32];
  char aBlockSize[32];

  char *aPage;
  int *aOrder;
  int nSync;

  i64 filesize;
  i64 blocksize;
  i64 syncsize;
  int nPage = 4096;

  /* How long to sleep before running a trial (in ms). */
#if 0
  const int nSleep = 10000;
#endif
  const int nSleep = 0;

  if( nArg!=3 ){
    testPrintUsage("FILESIZE BLOCKSIZE SYNCSIZE");
    return -1;
  }

  filesize = testReadSize(azArg[0]);
  blocksize = testReadSize(azArg[1]);
  syncsize = testReadSize(azArg[2]);

  nBlock = (int)(filesize / blocksize);
  nSize = (int)blocksize;
  nSync = (int)(syncsize / blocksize);

  aPage = (char *)malloc(4096);
  aOrder = (int *)malloc(nBlock * sizeof(int));
  for(i=0; i<nBlock; i++) aOrder[i] = i;
  for(i=0; i<(nBlock*25); i++){
    int tmp;
    u32 a = testPrngValue(i);
    u32 b = testPrngValue(a);
    a = a % nBlock;
    b = b % nBlock;
    tmp = aOrder[a];
    aOrder[a] = aOrder[b];
    aOrder[b] = tmp;
  }

  testFormatSize(aFilesize, sizeof(aFilesize), (i64)nBlock * (i64)nSize);
  testFormatSize(aBlockSize, sizeof(aFilesize), nSize);

  printf("Testing writing a %s file using %s blocks. ", aFilesize, aBlockSize);
  if( nSync==1 ){
    printf("Sync after each block.\n");
  }else{
    printf("Sync after each %d blocks.\n", nSync);
  }

  printf("Preparing file... ");
  fflush(stdout);
  unlink("writer.out");
  fd = open("writer.out", O_RDWR|O_CREAT, 0664);
  if( fd<0 ){
    testPrintError("open(): %d - %s\n", errno, strerror(errno));
    return -1;
  }
  testTimeInit();
  for(i=0; i<nBlock; i++){
    int iPg;
    memset(aPage, i&0xFF, nPage);
    for(iPg=0; iPg<(nSize/nPage); iPg++){
      write(fd, aPage, nPage);
    }
  }
  fsync(fd);
  printf("ok (%d ms)\n", testTimeGet());

  for(i=0; i<5; i++){
    int j;

    sqlite3_sleep(nSleep);
    printf("Now writing sequentially...  ");
    fflush(stdout);

    lseek(fd, 0, SEEK_SET);
    testTimeInit();
    for(j=0; j<nBlock; j++){
      int iPg;
      if( ((j+1)%nSync)==0 ) fdatasync(fd);
      memset(aPage, j&0xFF, nPage);
      for(iPg=0; iPg<(nSize/nPage); iPg++){
        write(fd, aPage, nPage);
      }
    }
    fdatasync(fd);
    ms = testTimeGet();
    printf("%d ms\n", ms);
    sqlite3_sleep(nSleep);
    printf("Now in an arbitrary order... ");

    fflush(stdout);
    testTimeInit();
    for(j=0; j<nBlock; j++){
      int iPg;
      if( ((j+1)%nSync)==0 ) fdatasync(fd);
      lseek(fd, aOrder[j]*nSize, SEEK_SET);
      memset(aPage, j&0xFF, nPage);
      for(iPg=0; iPg<(nSize/nPage); iPg++){
        write(fd, aPage, nPage);
      }
    }
    fdatasync(fd);
    ms = testTimeGet();
    printf("%d ms\n", ms);
  }

  close(fd);
  free(aPage);
  free(aOrder);

  return 0;
}

static void do_insert_work_hook(lsm_db *db, void *p){
  char *z = 0;
  lsm_info(db, LSM_INFO_DB_STRUCTURE, &z);
  if( z ){
    printf("%s\n", z);
    fflush(stdout);
    lsm_free(lsm_get_env(db), z);
  }

  unused_parameter(p);
}

typedef struct InsertWriteHook InsertWriteHook;
struct InsertWriteHook {
  FILE *pOut;
  int bLog;
  i64 iOff;
  int nData;
};

static void flushHook(InsertWriteHook *pHook){
  if( pHook->nData ){
    fprintf(pHook->pOut, "write %s %d %d\n", 
        (pHook->bLog ? "log" : "db"), (int)pHook->iOff, pHook->nData
    );
    pHook->nData = 0;
    fflush(pHook->pOut);
  }
}

static void do_insert_write_hook(
  void *pCtx,
  int bLog,
  i64 iOff,
  int nData,
  int nUs
){
  InsertWriteHook *pHook = (InsertWriteHook *)pCtx;
  if( bLog ) return;

  if( nData==0 ){
    flushHook(pHook);
    fprintf(pHook->pOut, "sync %s\n", (bLog ? "log" : "db"));
  }else if( pHook->nData 
         && bLog==pHook->bLog 
         && iOff==(pHook->iOff+pHook->nData) 
  ){
    pHook->nData += nData;
  }else{
    flushHook(pHook);
    pHook->bLog = bLog;
    pHook->iOff = iOff;
    pHook->nData = nData;
  }
}

static int do_replay(int nArg, char **azArg){
  char aBuf[4096];
  FILE *pInput;
  FILE *pClose = 0;
  const char *zDb;

  lsm_env *pEnv;
  lsm_file *pOut;
  int rc;

  if( nArg!=2 ){
    testPrintError("Usage: replay WRITELOG FILE\n");
    return 1;
  }

  if( strcmp(azArg[0], "-")==0 ){
    pInput = stdin;
  }else{
    pClose = pInput = fopen(azArg[0], "r");
  }
  zDb = azArg[1];
  pEnv = tdb_lsm_env();
  rc = pEnv->xOpen(pEnv, zDb, 0, &pOut);
  if( rc!=LSM_OK ) return rc;

  while( feof(pInput)==0 ){
    char zLine[80];
    fgets(zLine, sizeof(zLine)-1, pInput);
    zLine[sizeof(zLine)-1] = '\0';

    if( 0==memcmp("sync db", zLine, 7) ){
      rc = pEnv->xSync(pOut);
      if( rc!=0 ) break;
    }else{
      int iOff;
      int nData;
      int nMatch;
      nMatch = sscanf(zLine, "write db %d %d", &iOff, &nData);
      if( nMatch==2 ){
        int i;
        for(i=0; i<nData; i+=sizeof(aBuf)){
          memset(aBuf, i&0xFF, sizeof(aBuf));
          rc = pEnv->xWrite(pOut, iOff+i, aBuf, sizeof(aBuf));
          if( rc!=0 ) break;
        }
      }
    }
  }
  if( pClose ) fclose(pClose);
  pEnv->xClose(pOut);

  return rc;
}

static int do_insert(int nArg, char **azArg){
  const char *zDb = "lsm";
  TestDb *pDb = 0;
  int i;
  int rc;
  const int nRow = 1 * 1000 * 1000;

  DatasourceDefn defn = { TEST_DATASOURCE_RANDOM, 8, 15, 80, 150 };
  Datasource *pData = 0;

  if( nArg>1 ){
    testPrintError("Usage: insert ?DATABASE?\n");
    return 1;
  }
  if( nArg==1 ){ zDb = azArg[0]; }

  testMallocUninstall(tdb_lsm_env());
  for(i=0; zDb[i] && zDb[i]!='='; i++);
  if( zDb[i] ){
    rc = tdb_lsm_open(zDb, "testdb.lsm", 1, &pDb);
  }else{
    rc = tdb_open(zDb, 0, 1, &pDb);
  }

  if( rc!=0 ){
    testPrintError("Error opening db \"%s\": %d\n", zDb, rc);
  }else{
    InsertWriteHook hook;
    memset(&hook, 0, sizeof(hook));
    hook.pOut = fopen("writelog.txt", "w");

    pData = testDatasourceNew(&defn);
    tdb_lsm_config_work_hook(pDb, do_insert_work_hook, 0);
    tdb_lsm_write_hook(pDb, do_insert_write_hook, (void *)&hook);

    if( rc==0 ){
      for(i=0; i<nRow; i++){
        void *pKey; int nKey;     /* Database key to insert */
        void *pVal; int nVal;     /* Database value to insert */
        testDatasourceEntry(pData, i, &pKey, &nKey, &pVal, &nVal);
        tdb_write(pDb, pKey, nKey, pVal, nVal);
      }
    }

    testDatasourceFree(pData);
    tdb_close(pDb);
    flushHook(&hook);
    fclose(hook.pOut);
  }
  testMallocInstall(tdb_lsm_env());

  return rc;
}

static int st_do_show(int a, char **b)      { return do_show(a, b); }
static int st_do_work(int a, char **b)      { return do_work(a, b); }
static int st_do_io(int a, char **b)        { return do_io(a, b); }

#ifdef __linux__
#include <sys/time.h>
#include <sys/resource.h>

static void lsmtest_rusage_report(void){
  struct rusage r;
  memset(&r, 0, sizeof(r));

  getrusage(RUSAGE_SELF, &r);
  printf("# getrusage: { ru_maxrss %d ru_oublock %d ru_inblock %d }\n", 
      (int)r.ru_maxrss, (int)r.ru_oublock, (int)r.ru_inblock
  );
}
#else
static void lsmtest_rusage_report(void){
  /* no-op */
}
#endif

int main(int argc, char **argv){
  struct TestFunc {
    const char *zName;
    int bRusageReport;
    int (*xFunc)(int, char **);
  } aTest[] = {
    {"random",      1, do_random_tests},
    {"writespeed",  1, do_writer_test},
    {"io",          1, st_do_io},

    {"insert",      1, do_insert},
    {"replay",      1, do_replay},

    {"speed",       1, do_speed_tests},
    {"speed2",      1, do_speed_test2},
    {"show",        0, st_do_show},
    {"work",        1, st_do_work},
    {"test",        1, do_test},

    {0, 0}
  };
  int rc;                         /* Return Code */
  int iFunc;                      /* Index into aTest[] */

  int nLeakAlloc = 0;             /* Allocations leaked by lsm */
  int nLeakByte = 0;              /* Bytes leaked by lsm */

#ifdef LSM_DEBUG_MEM
  FILE *pReport = 0;              /* lsm malloc() report file */
  const char *zReport = "malloc.txt generated";
#else
  const char *zReport = "malloc.txt NOT generated";
#endif

  testMallocInstall(tdb_lsm_env());

  if( argc<2 ){
    testPrintError("Usage: %s sub-command ?args...?\n", argv[0]);
    return -1;
  }

  /* Initialize error reporting */
  testErrorInit(argc, argv);

  /* Initialize PRNG system */
  testPrngInit();

  rc = testArgSelect(aTest, "sub-command", argv[1], &iFunc);
  if( rc==0 ){
    rc = aTest[iFunc].xFunc(argc-2, &argv[2]);
  }

#ifdef LSM_DEBUG_MEM
  pReport = fopen("malloc.txt", "w");
  testMallocCheck(tdb_lsm_env(), &nLeakAlloc, &nLeakByte, pReport);
  fclose(pReport);
#else
  testMallocCheck(tdb_lsm_env(), &nLeakAlloc, &nLeakByte, 0);
#endif

  if( nLeakAlloc ){
    testPrintError("Leaked %d bytes in %d allocations (%s)\n", 
        nLeakByte, nLeakAlloc, zReport
    );
    if( rc==0 ) rc = -1;
  }
  testMallocUninstall(tdb_lsm_env());

  if( aTest[iFunc].bRusageReport ){
    lsmtest_rusage_report();
  }
  return rc;
}

Added ext/lsm1/lsm-test/lsmtest_mem.c.



















































































































































































































































































































































































































































































































































































































































































































































































































































>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409

#include <stdio.h>
#include <assert.h>
#include <string.h>

#define ArraySize(x) ((int)(sizeof(x) / sizeof((x)[0])))

#define MIN(x,y) ((x)<(y) ? (x) : (y))

typedef unsigned int  u32;
typedef unsigned char u8;
typedef long long int i64;
typedef unsigned long long int u64;

#if defined(__GLIBC__) && defined(LSM_DEBUG_MEM)
  extern int backtrace(void**,int);
  extern void backtrace_symbols_fd(void*const*,int,int);
# define TM_BACKTRACE 12
#else
# define backtrace(A,B) 1
# define backtrace_symbols_fd(A,B,C)
#endif


typedef struct TmBlockHdr TmBlockHdr;
typedef struct TmAgg TmAgg;
typedef struct TmGlobal TmGlobal;

struct TmGlobal {
  /* Linked list of all currently outstanding allocations. And a table of
  ** all allocations, past and present, indexed by backtrace() info.  */
  TmBlockHdr *pFirst;
#ifdef TM_BACKTRACE
  TmAgg *aHash[10000];
#endif

  /* Underlying malloc/realloc/free functions */
  void *(*xMalloc)(int);          /* underlying malloc(3) function */
  void *(*xRealloc)(void *, int); /* underlying realloc(3) function */
  void (*xFree)(void *);          /* underlying free(3) function */

  /* Mutex to protect pFirst and aHash */
  void (*xEnterMutex)(TmGlobal*); /* Call this to enter the mutex */
  void (*xLeaveMutex)(TmGlobal*); /* Call this to leave mutex */
  void (*xDelMutex)(TmGlobal*);   /* Call this to delete mutex */
  void *pMutex;                   /* Mutex handle */

  void *xSaveMalloc;
  void *xSaveRealloc;
  void *xSaveFree;

  /* OOM injection scheduling. If nCountdown is greater than zero when a 
  ** malloc attempt is made, it is decremented. If this means nCountdown 
  ** transitions from 1 to 0, then the allocation fails. If bPersist is true 
  ** when this happens, nCountdown is then incremented back to 1 (so that the 
  ** next attempt fails too).  
  */
  int nCountdown;
  int bPersist;
  int bEnable;
  void (*xHook)(void *);
  void *pHookCtx;
};

struct TmBlockHdr {
  TmBlockHdr *pNext;
  TmBlockHdr *pPrev;
  int nByte;
#ifdef TM_BACKTRACE
  TmAgg *pAgg;
#endif
  u32 iForeGuard;
};

#ifdef TM_BACKTRACE
struct TmAgg {
  int nAlloc;                     /* Number of allocations at this path */
  int nByte;                      /* Total number of bytes allocated */
  int nOutAlloc;                  /* Number of outstanding allocations */
  int nOutByte;                   /* Number of outstanding bytes */
  void *aFrame[TM_BACKTRACE];     /* backtrace() output */
  TmAgg *pNext;                   /* Next object in hash-table collision */
};
#endif

#define FOREGUARD 0x80F5E153
#define REARGUARD 0xE4676B53
static const u32 rearguard = REARGUARD;

#define ROUND8(x) (((x)+7)&~7)

#define BLOCK_HDR_SIZE (ROUND8( sizeof(TmBlockHdr) ))

static void lsmtest_oom_error(void){
  static int nErr = 0;
  nErr++;
}

static void tmEnterMutex(TmGlobal *pTm){
  pTm->xEnterMutex(pTm);
}
static void tmLeaveMutex(TmGlobal *pTm){
  pTm->xLeaveMutex(pTm);
}

static void *tmMalloc(TmGlobal *pTm, int nByte){
  TmBlockHdr *pNew;               /* New allocation header block */
  u8 *pUser;                      /* Return value */
  int nReq;                       /* Total number of bytes requested */

  assert( sizeof(rearguard)==4 );
  nReq = BLOCK_HDR_SIZE + nByte + 4;
  pNew = (TmBlockHdr *)pTm->xMalloc(nReq);
  memset(pNew, 0, sizeof(TmBlockHdr));

  tmEnterMutex(pTm);
  assert( pTm->nCountdown>=0 );
  assert( pTm->bPersist==0 || pTm->bPersist==1 );

  if( pTm->bEnable && pTm->nCountdown==1 ){
    /* Simulate an OOM error. */
    lsmtest_oom_error();
    pTm->xFree(pNew);
    pTm->nCountdown = pTm->bPersist;
    if( pTm->xHook ) pTm->xHook(pTm->pHookCtx);
    pUser = 0;
  }else{
    if( pTm->bEnable && pTm->nCountdown ) pTm->nCountdown--;

    pNew->iForeGuard = FOREGUARD;
    pNew->nByte = nByte;
    pNew->pNext = pTm->pFirst;

    if( pTm->pFirst ){
      pTm->pFirst->pPrev = pNew;
    }
    pTm->pFirst = pNew;

    pUser = &((u8 *)pNew)[BLOCK_HDR_SIZE];
    memset(pUser, 0x56, nByte);
    memcpy(&pUser[nByte], &rearguard, 4);

#ifdef TM_BACKTRACE
    {
      TmAgg *pAgg;
      int i;
      u32 iHash = 0;
      void *aFrame[TM_BACKTRACE];
      memset(aFrame, 0, sizeof(aFrame));
      backtrace(aFrame, TM_BACKTRACE);

      for(i=0; i<ArraySize(aFrame); i++){
        iHash += (u64)(aFrame[i]) + (iHash<<3);
      }
      iHash = iHash % ArraySize(pTm->aHash);

      for(pAgg=pTm->aHash[iHash]; pAgg; pAgg=pAgg->pNext){
        if( memcmp(pAgg->aFrame, aFrame, sizeof(aFrame))==0 ) break;
      }
      if( !pAgg ){
        pAgg = (TmAgg *)pTm->xMalloc(sizeof(TmAgg));
        memset(pAgg, 0, sizeof(TmAgg));
        memcpy(pAgg->aFrame, aFrame, sizeof(aFrame));
        pAgg->pNext = pTm->aHash[iHash];
        pTm->aHash[iHash] = pAgg;
      }
      pAgg->nAlloc++;
      pAgg->nByte += nByte;
      pAgg->nOutAlloc++;
      pAgg->nOutByte += nByte;
      pNew->pAgg = pAgg;
    }
#endif
  }

  tmLeaveMutex(pTm);
  return pUser;
}

static void tmFree(TmGlobal *pTm, void *p){
  if( p ){
    TmBlockHdr *pHdr;
    u8 *pUser = (u8 *)p;

    tmEnterMutex(pTm);
    pHdr = (TmBlockHdr *)&pUser[BLOCK_HDR_SIZE * -1];
    assert( pHdr->iForeGuard==FOREGUARD );
    assert( 0==memcmp(&pUser[pHdr->nByte], &rearguard, 4) );

    if( pHdr->pPrev ){
      assert( pHdr->pPrev->pNext==pHdr );
      pHdr->pPrev->pNext = pHdr->pNext;
    }else{
      assert( pHdr==pTm->pFirst );
      pTm->pFirst = pHdr->pNext;
    }
    if( pHdr->pNext ){
      assert( pHdr->pNext->pPrev==pHdr );
      pHdr->pNext->pPrev = pHdr->pPrev;
    }

#ifdef TM_BACKTRACE
    pHdr->pAgg->nOutAlloc--;
    pHdr->pAgg->nOutByte -= pHdr->nByte;
#endif

    tmLeaveMutex(pTm);
    memset(pUser, 0x58, pHdr->nByte);
    memset(pHdr, 0x57, sizeof(TmBlockHdr));
    pTm->xFree(pHdr);
  }
}

static void *tmRealloc(TmGlobal *pTm, void *p, int nByte){
  void *pNew;

  pNew = tmMalloc(pTm, nByte);
  if( pNew && p ){
    TmBlockHdr *pHdr;
    u8 *pUser = (u8 *)p;
    pHdr = (TmBlockHdr *)&pUser[BLOCK_HDR_SIZE * -1];
    memcpy(pNew, p, MIN(nByte, pHdr->nByte));
    tmFree(pTm, p);
  }
  return pNew;
}

static void tmMallocOom(
  TmGlobal *pTm, 
  int nCountdown, 
  int bPersist,
  void (*xHook)(void *),
  void *pHookCtx
){
  assert( nCountdown>=0 );
  assert( bPersist==0 || bPersist==1 );
  pTm->nCountdown = nCountdown;
  pTm->bPersist = bPersist;
  pTm->xHook = xHook;
  pTm->pHookCtx = pHookCtx;
  pTm->bEnable = 1;
}

static void tmMallocOomEnable(
  TmGlobal *pTm, 
  int bEnable
){
  pTm->bEnable = bEnable;
}

static void tmMallocCheck(
  TmGlobal *pTm,
  int *pnLeakAlloc,
  int *pnLeakByte,
  FILE *pFile
){
  TmBlockHdr *pHdr;
  int nLeak = 0;
  int nByte = 0;

  if( pTm==0 ) return;

  for(pHdr=pTm->pFirst; pHdr; pHdr=pHdr->pNext){
    nLeak++; 
    nByte += pHdr->nByte;
  }
  if( pnLeakAlloc ) *pnLeakAlloc = nLeak;
  if( pnLeakByte ) *pnLeakByte = nByte;

#ifdef TM_BACKTRACE
  if( pFile ){
    int i;
    fprintf(pFile, "LEAKS\n");
    for(i=0; i<ArraySize(pTm->aHash); i++){
      TmAgg *pAgg;
      for(pAgg=pTm->aHash[i]; pAgg; pAgg=pAgg->pNext){
        if( pAgg->nOutAlloc ){
          int j;
          fprintf(pFile, "%d %d ", pAgg->nOutByte, pAgg->nOutAlloc);
          for(j=0; j<TM_BACKTRACE; j++){
            fprintf(pFile, "%p ", pAgg->aFrame[j]);
          }
          fprintf(pFile, "\n");
        }
      }
    }
    fprintf(pFile, "\nALLOCATIONS\n");
    for(i=0; i<ArraySize(pTm->aHash); i++){
      TmAgg *pAgg;
      for(pAgg=pTm->aHash[i]; pAgg; pAgg=pAgg->pNext){
        int j;
        fprintf(pFile, "%d %d ", pAgg->nByte, pAgg->nAlloc);
        for(j=0; j<TM_BACKTRACE; j++) fprintf(pFile, "%p ", pAgg->aFrame[j]);
        fprintf(pFile, "\n");
      }
    }
  }
#else
  (void)pFile;
#endif
}


#include "lsm.h"
#include "stdlib.h"

typedef struct LsmMutex LsmMutex;
struct LsmMutex {
  lsm_env *pEnv;
  lsm_mutex *pMutex;
};

static void tmLsmMutexEnter(TmGlobal *pTm){
  LsmMutex *p = (LsmMutex *)pTm->pMutex;
  p->pEnv->xMutexEnter(p->pMutex);
}
static void tmLsmMutexLeave(TmGlobal *pTm){
  LsmMutex *p = (LsmMutex *)(pTm->pMutex);
  p->pEnv->xMutexLeave(p->pMutex);
}
static void tmLsmMutexDel(TmGlobal *pTm){
  LsmMutex *p = (LsmMutex *)pTm->pMutex;
  pTm->xFree(p);
}
static void *tmLsmMalloc(int n){ return malloc(n); }
static void tmLsmFree(void *ptr){ free(ptr); }
static void *tmLsmRealloc(void *ptr, int n){ return realloc(ptr, n); }

static void *tmLsmEnvMalloc(lsm_env *p, int n){ 
  return tmMalloc((TmGlobal *)(p->pMemCtx), n); 
}
static void tmLsmEnvFree(lsm_env *p, void *ptr){ 
  tmFree((TmGlobal *)(p->pMemCtx), ptr); 
}
static void *tmLsmEnvRealloc(lsm_env *p, void *ptr, int n){ 
  return tmRealloc((TmGlobal *)(p->pMemCtx), ptr, n);
}

void testMallocInstall(lsm_env *pEnv){
  TmGlobal *pGlobal;
  LsmMutex *pMutex;
  assert( pEnv->pMemCtx==0 );

  /* Allocate and populate a TmGlobal structure. */
  pGlobal = (TmGlobal *)tmLsmMalloc(sizeof(TmGlobal));
  memset(pGlobal, 0, sizeof(TmGlobal));
  pGlobal->xMalloc = tmLsmMalloc;
  pGlobal->xRealloc = tmLsmRealloc;
  pGlobal->xFree = tmLsmFree;
  pMutex = (LsmMutex *)pGlobal->xMalloc(sizeof(LsmMutex));
  pMutex->pEnv = pEnv;
  pEnv->xMutexStatic(pEnv, LSM_MUTEX_HEAP, &pMutex->pMutex);
  pGlobal->xEnterMutex = tmLsmMutexEnter;
  pGlobal->xLeaveMutex = tmLsmMutexLeave;
  pGlobal->xDelMutex = tmLsmMutexDel;
  pGlobal->pMutex = (void *)pMutex;

  pGlobal->xSaveMalloc = (void *)pEnv->xMalloc;
  pGlobal->xSaveRealloc = (void *)pEnv->xRealloc;
  pGlobal->xSaveFree = (void *)pEnv->xFree;

  /* Set up pEnv to the use the new TmGlobal */
  pEnv->pMemCtx = (void *)pGlobal;
  pEnv->xMalloc = tmLsmEnvMalloc;
  pEnv->xRealloc = tmLsmEnvRealloc;
  pEnv->xFree = tmLsmEnvFree;
}

void testMallocUninstall(lsm_env *pEnv){
  TmGlobal *p = (TmGlobal *)pEnv->pMemCtx;
  pEnv->pMemCtx = 0;
  if( p ){
    pEnv->xMalloc = (void *(*)(lsm_env*, int))(p->xSaveMalloc);
    pEnv->xRealloc = (void *(*)(lsm_env*, void*, int))(p->xSaveRealloc);
    pEnv->xFree = (void (*)(lsm_env*, void*))(p->xSaveFree);
    p->xDelMutex(p);
    tmLsmFree(p);
  }
}

void testMallocCheck(
  lsm_env *pEnv,
  int *pnLeakAlloc,
  int *pnLeakByte,
  FILE *pFile
){
  if( pEnv->pMemCtx==0 ){
    *pnLeakAlloc = 0;
    *pnLeakByte = 0;
  }else{
    tmMallocCheck((TmGlobal *)(pEnv->pMemCtx), pnLeakAlloc, pnLeakByte, pFile);
  }
}

void testMallocOom(
  lsm_env *pEnv, 
  int nCountdown, 
  int bPersist,
  void (*xHook)(void *),
  void *pHookCtx
){
  TmGlobal *pTm = (TmGlobal *)(pEnv->pMemCtx);
  tmMallocOom(pTm, nCountdown, bPersist, xHook, pHookCtx);
}

void testMallocOomEnable(lsm_env *pEnv, int bEnable){
  TmGlobal *pTm = (TmGlobal *)(pEnv->pMemCtx);
  tmMallocOomEnable(pTm, bEnable);
}

Added ext/lsm1/lsm-test/lsmtest_tdb.c.









































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836

/*
** This program attempts to test the correctness of some facets of the 
** LSM database library. Specifically, that the contents of the database
** are maintained correctly during a series of inserts and deletes.
*/


#include "lsmtest_tdb.h"
#include "lsm.h"

#include "lsmtest.h"

#include <stdlib.h>
#include <string.h>
#include <assert.h>
#ifndef _WIN32
# include <unistd.h>
#endif
#include <stdio.h>


typedef struct SqlDb SqlDb;

static int error_transaction_function(TestDb *p, int iLevel){ 
  unused_parameter(p);
  unused_parameter(iLevel);
  return -1; 
}


/*************************************************************************
** Begin wrapper for LevelDB.
*/
#ifdef HAVE_LEVELDB

#include <leveldb/c.h>

typedef struct LevelDb LevelDb;
struct LevelDb {
  TestDb base;
  leveldb_t *db;
  leveldb_options_t *pOpt;
  leveldb_writeoptions_t *pWriteOpt;
  leveldb_readoptions_t *pReadOpt;

  char *pVal;
};

static int test_leveldb_close(TestDb *pTestDb){
  LevelDb *pDb = (LevelDb *)pTestDb;

  leveldb_close(pDb->db);
  leveldb_writeoptions_destroy(pDb->pWriteOpt);
  leveldb_readoptions_destroy(pDb->pReadOpt);
  leveldb_options_destroy(pDb->pOpt);
  free(pDb->pVal);
  free(pDb);

  return 0;
}

static int test_leveldb_write(
  TestDb *pTestDb, 
  void *pKey, 
  int nKey, 
  void *pVal, 
  int nVal
){
  LevelDb *pDb = (LevelDb *)pTestDb;
  char *zErr = 0;
  leveldb_put(pDb->db, pDb->pWriteOpt, pKey, nKey, pVal, nVal, &zErr);
  return (zErr!=0);
}

static int test_leveldb_delete(TestDb *pTestDb, void *pKey, int nKey){
  LevelDb *pDb = (LevelDb *)pTestDb;
  char *zErr = 0;
  leveldb_delete(pDb->db, pDb->pWriteOpt, pKey, nKey, &zErr);
  return (zErr!=0);
}

static int test_leveldb_fetch(
  TestDb *pTestDb, 
  void *pKey, 
  int nKey, 
  void **ppVal, 
  int *pnVal
){
  LevelDb *pDb = (LevelDb *)pTestDb;
  char *zErr = 0;
  size_t nVal = 0;

  if( pKey==0 ) return 0;
  free(pDb->pVal);
  pDb->pVal = leveldb_get(pDb->db, pDb->pReadOpt, pKey, nKey, &nVal, &zErr);
  *ppVal = (void *)(pDb->pVal);
  if( pDb->pVal==0 ){
    *pnVal = -1;
  }else{
    *pnVal = (int)nVal;
  }

  return (zErr!=0);
}

static int test_leveldb_scan(
  TestDb *pTestDb,
  void *pCtx,
  int bReverse,
  void *pKey1, int nKey1,         /* Start of search */
  void *pKey2, int nKey2,         /* End of search */
  void (*xCallback)(void *, void *, int , void *, int)
){
  LevelDb *pDb = (LevelDb *)pTestDb;
  leveldb_iterator_t *iter;

  iter = leveldb_create_iterator(pDb->db, pDb->pReadOpt);

  if( bReverse==0 ){
    if( pKey1 ){
      leveldb_iter_seek(iter, pKey1, nKey1);
    }else{
      leveldb_iter_seek_to_first(iter);
    }
  }else{
    if( pKey2 ){
      leveldb_iter_seek(iter, pKey2, nKey2);

      if( leveldb_iter_valid(iter)==0 ){
        leveldb_iter_seek_to_last(iter);
      }else{
        const char *k; size_t n;
        int res;
        k = leveldb_iter_key(iter, &n);
        res = memcmp(k, pKey2, MIN(n, nKey2));
        if( res==0 ) res = n - nKey2;
        assert( res>=0 );
        if( res>0 ){
          leveldb_iter_prev(iter);
        }
      }
    }else{
      leveldb_iter_seek_to_last(iter);
    }
  }


  while( leveldb_iter_valid(iter) ){
    const char *k; size_t n;
    const char *v; size_t n2;
    int res;

    k = leveldb_iter_key(iter, &n);
    if( bReverse==0 && pKey2 ){
      res = memcmp(k, pKey2, MIN(n, nKey2));
      if( res==0 ) res = n - nKey2;
      if( res>0 ) break;
    }
    if( bReverse!=0 && pKey1 ){
      res = memcmp(k, pKey1, MIN(n, nKey1));
      if( res==0 ) res = n - nKey1;
      if( res<0 ) break;
    }

    v = leveldb_iter_value(iter, &n2);

    xCallback(pCtx, (void *)k, n, (void *)v, n2);

    if( bReverse==0 ){
      leveldb_iter_next(iter);
    }else{
      leveldb_iter_prev(iter);
    }
  }

  leveldb_iter_destroy(iter);
  return 0;
}

static int test_leveldb_open(
  const char *zSpec, 
  const char *zFilename, 
  int bClear, 
  TestDb **ppDb
){
  static const DatabaseMethods LeveldbMethods = {
    test_leveldb_close,
    test_leveldb_write,
    test_leveldb_delete,
    0,
    test_leveldb_fetch,
    test_leveldb_scan,
    error_transaction_function,
    error_transaction_function,
    error_transaction_function
  };

  LevelDb *pLevelDb;
  char *zErr = 0;

  if( bClear ){
    char *zCmd = sqlite3_mprintf("rm -rf %s\n", zFilename);
    system(zCmd);
    sqlite3_free(zCmd);
  }

  pLevelDb = (LevelDb *)malloc(sizeof(LevelDb));
  memset(pLevelDb, 0, sizeof(LevelDb));

  pLevelDb->pOpt = leveldb_options_create();
  leveldb_options_set_create_if_missing(pLevelDb->pOpt, 1);
  pLevelDb->pWriteOpt = leveldb_writeoptions_create();
  pLevelDb->pReadOpt = leveldb_readoptions_create();

  pLevelDb->db = leveldb_open(pLevelDb->pOpt, zFilename, &zErr);

  if( zErr ){
    test_leveldb_close((TestDb *)pLevelDb);
    *ppDb = 0;
    return 1;
  }

  *ppDb = (TestDb *)pLevelDb;
  pLevelDb->base.pMethods = &LeveldbMethods;
  return 0;
}
#endif  /* HAVE_LEVELDB */
/* 
** End wrapper for LevelDB.
*************************************************************************/

#ifdef HAVE_KYOTOCABINET
static int kc_close(TestDb *pTestDb){
  return test_kc_close(pTestDb);
}

static int kc_write(
  TestDb *pTestDb, 
  void *pKey, 
  int nKey, 
  void *pVal, 
  int nVal
){
  return test_kc_write(pTestDb, pKey, nKey, pVal, nVal);
}

static int kc_delete(TestDb *pTestDb, void *pKey, int nKey){
  return test_kc_delete(pTestDb, pKey, nKey);
}

static int kc_delete_range(
  TestDb *pTestDb, 
  void *pKey1, int nKey1,
  void *pKey2, int nKey2
){
  return test_kc_delete_range(pTestDb, pKey1, nKey1, pKey2, nKey2);
}

static int kc_fetch(
  TestDb *pTestDb, 
  void *pKey, 
  int nKey, 
  void **ppVal, 
  int *pnVal
){
  if( pKey==0 ) return LSM_OK;
  return test_kc_fetch(pTestDb, pKey, nKey, ppVal, pnVal);
}

static int kc_scan(
  TestDb *pTestDb,
  void *pCtx,
  int bReverse,
  void *pFirst, int nFirst,
  void *pLast, int nLast,
  void (*xCallback)(void *, void *, int , void *, int)
){
  return test_kc_scan(
      pTestDb, pCtx, bReverse, pFirst, nFirst, pLast, nLast, xCallback
  );
}

static int kc_open(
  const char *zSpec, 
  const char *zFilename, 
  int bClear, 
  TestDb **ppDb
){
  static const DatabaseMethods KcdbMethods = {
    kc_close,
    kc_write,
    kc_delete,
    kc_delete_range,
    kc_fetch,
    kc_scan,
    error_transaction_function,
    error_transaction_function,
    error_transaction_function
  };

  int rc;
  TestDb *pTestDb = 0;

  rc = test_kc_open(zFilename, bClear, &pTestDb);
  if( rc!=0 ){
    *ppDb = 0;
    return rc;
  }
  pTestDb->pMethods = &KcdbMethods;
  *ppDb = pTestDb;
  return 0;
}
#endif /* HAVE_KYOTOCABINET */
/* 
** End wrapper for Kyoto cabinet.
*************************************************************************/

#ifdef HAVE_MDB
static int mdb_close(TestDb *pTestDb){
  return test_mdb_close(pTestDb);
}

static int mdb_write(
  TestDb *pTestDb, 
  void *pKey, 
  int nKey, 
  void *pVal, 
  int nVal
){
  return test_mdb_write(pTestDb, pKey, nKey, pVal, nVal);
}

static int mdb_delete(TestDb *pTestDb, void *pKey, int nKey){
  return test_mdb_delete(pTestDb, pKey, nKey);
}

static int mdb_fetch(
  TestDb *pTestDb, 
  void *pKey, 
  int nKey, 
  void **ppVal, 
  int *pnVal
){
  if( pKey==0 ) return LSM_OK;
  return test_mdb_fetch(pTestDb, pKey, nKey, ppVal, pnVal);
}

static int mdb_scan(
  TestDb *pTestDb,
  void *pCtx,
  int bReverse,
  void *pFirst, int nFirst,
  void *pLast, int nLast,
  void (*xCallback)(void *, void *, int , void *, int)
){
  return test_mdb_scan(
      pTestDb, pCtx, bReverse, pFirst, nFirst, pLast, nLast, xCallback
  );
}

static int mdb_open(
  const char *zSpec, 
  const char *zFilename, 
  int bClear, 
  TestDb **ppDb
){
  static const DatabaseMethods KcdbMethods = {
    mdb_close,
    mdb_write,
    mdb_delete,
    0,
    mdb_fetch,
    mdb_scan,
    error_transaction_function,
    error_transaction_function,
    error_transaction_function
  };

  int rc;
  TestDb *pTestDb = 0;

  rc = test_mdb_open(zSpec, zFilename, bClear, &pTestDb);
  if( rc!=0 ){
    *ppDb = 0;
    return rc;
  }
  pTestDb->pMethods = &KcdbMethods;
  *ppDb = pTestDb;
  return 0;
}
#endif /* HAVE_MDB */

/*************************************************************************
** Begin wrapper for SQLite.
*/

/*
** nOpenTrans:
**   The number of open nested transactions, in the same sense as used
**   by the tdb_begin/commit/rollback and SQLite 4 KV interfaces. If this
**   value is 0, there are no transactions open at all. If it is 1, then
**   there is a read transaction. If it is 2 or greater, then there are
**   (nOpenTrans-1) nested write transactions open.
*/
struct SqlDb {
  TestDb base;
  sqlite3 *db;
  sqlite3_stmt *pInsert;
  sqlite3_stmt *pDelete;
  sqlite3_stmt *pDeleteRange;
  sqlite3_stmt *pFetch;
  sqlite3_stmt *apScan[8];

  int nOpenTrans;

  /* Used by sql_fetch() to allocate space for results */
  int nAlloc;
  u8 *aAlloc;
};

static int sql_close(TestDb *pTestDb){
  SqlDb *pDb = (SqlDb *)pTestDb;
  sqlite3_finalize(pDb->pInsert);
  sqlite3_finalize(pDb->pDelete);
  sqlite3_finalize(pDb->pDeleteRange);
  sqlite3_finalize(pDb->pFetch);
  sqlite3_finalize(pDb->apScan[0]);
  sqlite3_finalize(pDb->apScan[1]);
  sqlite3_finalize(pDb->apScan[2]);
  sqlite3_finalize(pDb->apScan[3]);
  sqlite3_finalize(pDb->apScan[4]);
  sqlite3_finalize(pDb->apScan[5]);
  sqlite3_finalize(pDb->apScan[6]);
  sqlite3_finalize(pDb->apScan[7]);
  sqlite3_close(pDb->db);
  free((char *)pDb->aAlloc);
  free((char *)pDb);
  return SQLITE_OK;
}

static int sql_write(
  TestDb *pTestDb, 
  void *pKey, 
  int nKey, 
  void *pVal, 
  int nVal
){
  SqlDb *pDb = (SqlDb *)pTestDb;
  sqlite3_bind_blob(pDb->pInsert, 1, pKey, nKey, SQLITE_STATIC);
  sqlite3_bind_blob(pDb->pInsert, 2, pVal, nVal, SQLITE_STATIC);
  sqlite3_step(pDb->pInsert);
  return sqlite3_reset(pDb->pInsert);
}

static int sql_delete(TestDb *pTestDb, void *pKey, int nKey){
  SqlDb *pDb = (SqlDb *)pTestDb;
  sqlite3_bind_blob(pDb->pDelete, 1, pKey, nKey, SQLITE_STATIC);
  sqlite3_step(pDb->pDelete);
  return sqlite3_reset(pDb->pDelete);
}

static int sql_delete_range(
  TestDb *pTestDb, 
  void *pKey1, int nKey1,
  void *pKey2, int nKey2
){
  SqlDb *pDb = (SqlDb *)pTestDb;
  sqlite3_bind_blob(pDb->pDeleteRange, 1, pKey1, nKey1, SQLITE_STATIC);
  sqlite3_bind_blob(pDb->pDeleteRange, 2, pKey2, nKey2, SQLITE_STATIC);
  sqlite3_step(pDb->pDeleteRange);
  return sqlite3_reset(pDb->pDeleteRange);
}

static int sql_fetch(
  TestDb *pTestDb, 
  void *pKey, 
  int nKey, 
  void **ppVal, 
  int *pnVal
){
  SqlDb *pDb = (SqlDb *)pTestDb;
  int rc;

  sqlite3_reset(pDb->pFetch);
  if( pKey==0 ){
    assert( ppVal==0 );
    assert( pnVal==0 );
    return LSM_OK;
  }

  sqlite3_bind_blob(pDb->pFetch, 1, pKey, nKey, SQLITE_STATIC);
  rc = sqlite3_step(pDb->pFetch);
  if( rc==SQLITE_ROW ){
    int nVal = sqlite3_column_bytes(pDb->pFetch, 0);
    u8 *aVal = (void *)sqlite3_column_blob(pDb->pFetch, 0);

    if( nVal>pDb->nAlloc ){
      free(pDb->aAlloc);
      pDb->aAlloc = (u8 *)malloc(nVal*2);
      pDb->nAlloc = nVal*2;
    }
    memcpy(pDb->aAlloc, aVal, nVal);
    *pnVal = nVal;
    *ppVal = (void *)pDb->aAlloc;
  }else{
    *pnVal = -1;
    *ppVal = 0;
  }

  rc = sqlite3_reset(pDb->pFetch);
  return rc;
}

static int sql_scan(
  TestDb *pTestDb,
  void *pCtx,
  int bReverse,
  void *pFirst, int nFirst,
  void *pLast, int nLast,
  void (*xCallback)(void *, void *, int , void *, int)
){
  SqlDb *pDb = (SqlDb *)pTestDb;
  sqlite3_stmt *pScan;

  assert( bReverse==1 || bReverse==0 );
  pScan = pDb->apScan[(pFirst==0) + (pLast==0)*2 + bReverse*4];

  if( pFirst ) sqlite3_bind_blob(pScan, 1, pFirst, nFirst, SQLITE_STATIC);
  if( pLast ) sqlite3_bind_blob(pScan, 2, pLast, nLast, SQLITE_STATIC);

  while( SQLITE_ROW==sqlite3_step(pScan) ){
    void *pKey; int nKey;
    void *pVal; int nVal;

    nKey = sqlite3_column_bytes(pScan, 0);
    pKey = (void *)sqlite3_column_blob(pScan, 0);
    nVal = sqlite3_column_bytes(pScan, 1);
    pVal = (void *)sqlite3_column_blob(pScan, 1);

    xCallback(pCtx, pKey, nKey, pVal, nVal);
  }
  return sqlite3_reset(pScan);
}

static int sql_begin(TestDb *pTestDb, int iLevel){
  int i;
  SqlDb *pDb = (SqlDb *)pTestDb;

  /* iLevel==0 is a no-op */
  if( iLevel==0 ) return 0;

  /* If there are no transactions at all open, open a read transaction. */
  if( pDb->nOpenTrans==0 ){
    int rc = sqlite3_exec(pDb->db, 
        "BEGIN; SELECT * FROM sqlite_master LIMIT 1;" , 0, 0, 0
    );
    if( rc!=0 ) return rc;
    pDb->nOpenTrans = 1;
  }

  /* Open any required write transactions */
  for(i=pDb->nOpenTrans; i<iLevel; i++){
    char *zSql = sqlite3_mprintf("SAVEPOINT x%d", i);
    int rc = sqlite3_exec(pDb->db, zSql, 0, 0, 0);
    sqlite3_free(zSql);
    if( rc!=SQLITE_OK ) return rc;
  }

  pDb->nOpenTrans = iLevel;
  return 0;
}

static int sql_commit(TestDb *pTestDb, int iLevel){
  SqlDb *pDb = (SqlDb *)pTestDb;
  assert( iLevel>=0 );

  /* Close the read transaction if requested. */
  if( pDb->nOpenTrans>=1 && iLevel==0 ){
    int rc = sqlite3_exec(pDb->db, "COMMIT", 0, 0, 0);
    if( rc!=0 ) return rc;
    pDb->nOpenTrans = 0;
  }

  /* Close write transactions as required */
  if( pDb->nOpenTrans>iLevel ){
    char *zSql = sqlite3_mprintf("RELEASE x%d", iLevel);
    int rc = sqlite3_exec(pDb->db, zSql, 0, 0, 0);
    sqlite3_free(zSql);
    if( rc!=0 ) return rc;
  }

  pDb->nOpenTrans = iLevel;
  return 0;
}

static int sql_rollback(TestDb *pTestDb, int iLevel){
  SqlDb *pDb = (SqlDb *)pTestDb;
  assert( iLevel>=0 );

  if( pDb->nOpenTrans>=1 && iLevel==0 ){
    /* Close the read transaction if requested. */
    int rc = sqlite3_exec(pDb->db, "ROLLBACK", 0, 0, 0);
    if( rc!=0 ) return rc;
  }else if( pDb->nOpenTrans>1 && iLevel==1 ){
    /* Or, rollback and close the top-level write transaction */
    int rc = sqlite3_exec(pDb->db, "ROLLBACK TO x1; RELEASE x1;", 0, 0, 0);
    if( rc!=0 ) return rc;
  }else{
    /* Or, just roll back some nested transactions */
    char *zSql = sqlite3_mprintf("ROLLBACK TO x%d", iLevel-1);
    int rc = sqlite3_exec(pDb->db, zSql, 0, 0, 0);
    sqlite3_free(zSql);
    if( rc!=0 ) return rc;
  }

  pDb->nOpenTrans = iLevel;
  return 0;
}

static int sql_open(
  const char *zSpec, 
  const char *zFilename, 
  int bClear, 
  TestDb **ppDb
){
  static const DatabaseMethods SqlMethods = {
    sql_close,
    sql_write,
    sql_delete,
    sql_delete_range,
    sql_fetch,
    sql_scan,
    sql_begin,
    sql_commit,
    sql_rollback
  };
  const char *zCreate = "CREATE TABLE IF NOT EXISTS t1(k PRIMARY KEY, v)";
  const char *zInsert = "REPLACE INTO t1 VALUES(?, ?)";
  const char *zDelete = "DELETE FROM t1 WHERE k = ?";
  const char *zRange = "DELETE FROM t1 WHERE k>? AND k<?";
  const char *zFetch  = "SELECT v FROM t1 WHERE k = ?";

  const char *zScan0  = "SELECT * FROM t1 WHERE k BETWEEN ?1 AND ?2 ORDER BY k";
  const char *zScan1  = "SELECT * FROM t1 WHERE k <= ?2 ORDER BY k";
  const char *zScan2  = "SELECT * FROM t1 WHERE k >= ?1 ORDER BY k";
  const char *zScan3  = "SELECT * FROM t1 ORDER BY k";

  const char *zScan4  = 
    "SELECT * FROM t1 WHERE k BETWEEN ?1 AND ?2 ORDER BY k DESC";
  const char *zScan5  = "SELECT * FROM t1 WHERE k <= ?2 ORDER BY k DESC";
  const char *zScan6  = "SELECT * FROM t1 WHERE k >= ?1 ORDER BY k DESC";
  const char *zScan7  = "SELECT * FROM t1 ORDER BY k DESC";

  int rc;
  SqlDb *pDb;
  char *zPragma;

  if( bClear && zFilename && zFilename[0] ){
    unlink(zFilename);
  }

  pDb = (SqlDb *)malloc(sizeof(SqlDb));
  memset(pDb, 0, sizeof(SqlDb));
  pDb->base.pMethods = &SqlMethods;

  if( 0!=(rc = sqlite3_open(zFilename, &pDb->db))
   || 0!=(rc = sqlite3_exec(pDb->db, zCreate, 0, 0, 0))
   || 0!=(rc = sqlite3_prepare_v2(pDb->db, zInsert, -1, &pDb->pInsert, 0))
   || 0!=(rc = sqlite3_prepare_v2(pDb->db, zDelete, -1, &pDb->pDelete, 0))
   || 0!=(rc = sqlite3_prepare_v2(pDb->db, zRange, -1, &pDb->pDeleteRange, 0))
   || 0!=(rc = sqlite3_prepare_v2(pDb->db, zFetch, -1, &pDb->pFetch, 0))
   || 0!=(rc = sqlite3_prepare_v2(pDb->db, zScan0, -1, &pDb->apScan[0], 0))
   || 0!=(rc = sqlite3_prepare_v2(pDb->db, zScan1, -1, &pDb->apScan[1], 0))
   || 0!=(rc = sqlite3_prepare_v2(pDb->db, zScan2, -1, &pDb->apScan[2], 0))
   || 0!=(rc = sqlite3_prepare_v2(pDb->db, zScan3, -1, &pDb->apScan[3], 0))
   || 0!=(rc = sqlite3_prepare_v2(pDb->db, zScan4, -1, &pDb->apScan[4], 0))
   || 0!=(rc = sqlite3_prepare_v2(pDb->db, zScan5, -1, &pDb->apScan[5], 0))
   || 0!=(rc = sqlite3_prepare_v2(pDb->db, zScan6, -1, &pDb->apScan[6], 0))
   || 0!=(rc = sqlite3_prepare_v2(pDb->db, zScan7, -1, &pDb->apScan[7], 0))
  ){
    *ppDb = 0;
    sql_close((TestDb *)pDb);
    return rc;
  }

  zPragma = sqlite3_mprintf("PRAGMA page_size=%d", TESTDB_DEFAULT_PAGE_SIZE);
  sqlite3_exec(pDb->db, zPragma, 0, 0, 0);
  sqlite3_free(zPragma);
  zPragma = sqlite3_mprintf("PRAGMA cache_size=%d", TESTDB_DEFAULT_CACHE_SIZE);
  sqlite3_exec(pDb->db, zPragma, 0, 0, 0);
  sqlite3_free(zPragma);

  /* sqlite3_exec(pDb->db, "PRAGMA locking_mode=EXCLUSIVE", 0, 0, 0); */
  sqlite3_exec(pDb->db, "PRAGMA synchronous=OFF", 0, 0, 0);
  sqlite3_exec(pDb->db, "PRAGMA journal_mode=WAL", 0, 0, 0);
  sqlite3_exec(pDb->db, "PRAGMA wal_autocheckpoint=4096", 0, 0, 0);
  if( zSpec ){
    rc = sqlite3_exec(pDb->db, zSpec, 0, 0, 0);
    if( rc!=SQLITE_OK ){
      sql_close((TestDb *)pDb);
      return rc;
    }
  }

  *ppDb = (TestDb *)pDb;
  return 0;
}
/* 
** End wrapper for SQLite.
*************************************************************************/

/*************************************************************************
** Begin exported functions.
*/
static struct Lib {
  const char *zName;
  const char *zDefaultDb;
  int (*xOpen)(const char *, const char *zFilename, int bClear, TestDb **ppDb);
} aLib[] = {
  { "sqlite3",      "testdb.sqlite",    sql_open },
  { "lsm_small",    "testdb.lsm_small", test_lsm_small_open },
  { "lsm_lomem",    "testdb.lsm_lomem", test_lsm_lomem_open },
#ifdef HAVE_ZLIB
  { "lsm_zip",      "testdb.lsm_zip",   test_lsm_zip_open },
#endif
  { "lsm",          "testdb.lsm",       test_lsm_open },
#ifdef LSM_MUTEX_PTHREADS
  { "lsm_mt2",      "testdb.lsm_mt2",   test_lsm_mt2 },
  { "lsm_mt3",      "testdb.lsm_mt3",   test_lsm_mt3 },
#endif
#ifdef HAVE_LEVELDB
  { "leveldb",      "testdb.leveldb",   test_leveldb_open },
#endif
#ifdef HAVE_KYOTOCABINET
  { "kyotocabinet", "testdb.kc",        kc_open },
#endif
#ifdef HAVE_MDB
  { "mdb", "./testdb.mdb",        mdb_open }
#endif
};

const char *tdb_system_name(int i){
  if( i<0 || i>=ArraySize(aLib) ) return 0;
  return aLib[i].zName;
}

int tdb_open(const char *zLib, const char *zDb, int bClear, TestDb **ppDb){
  int i;
  int rc = 1;
  const char *zSpec = 0;

  int nLib = 0;
  while( zLib[nLib] && zLib[nLib]!=' ' ){
    nLib++;
  }
  zSpec = &zLib[nLib];
  while( *zSpec==' ' ) zSpec++;
  if( *zSpec=='\0' ) zSpec = 0;

  for(i=0; i<ArraySize(aLib); i++){
    if( strlen(aLib[i].zName)==nLib && 0==memcmp(zLib, aLib[i].zName, nLib) ){
      rc = aLib[i].xOpen(zSpec, (zDb ? zDb : aLib[i].zDefaultDb), bClear, ppDb);
      if( rc==0 ){
        (*ppDb)->zLibrary = aLib[i].zName;
      }
      break;
    }
  }

  if( rc ){
    /* Failed to find the requested database library. Return an error. */
    *ppDb = 0;
  }
  return rc;
}

int tdb_close(TestDb *pDb){
  if( pDb ){
    return pDb->pMethods->xClose(pDb);
  }
  return 0;
}

int tdb_write(TestDb *pDb, void *pKey, int nKey, void *pVal, int nVal){
  return pDb->pMethods->xWrite(pDb, pKey, nKey, pVal, nVal);
}

int tdb_delete(TestDb *pDb, void *pKey, int nKey){
  return pDb->pMethods->xDelete(pDb, pKey, nKey);
}

int tdb_delete_range(
    TestDb *pDb, void *pKey1, int nKey1, void *pKey2, int nKey2
){
  return pDb->pMethods->xDeleteRange(pDb, pKey1, nKey1, pKey2, nKey2);
}

int tdb_fetch(TestDb *pDb, void *pKey, int nKey, void **ppVal, int *pnVal){
  return pDb->pMethods->xFetch(pDb, pKey, nKey, ppVal, pnVal);
}

int tdb_scan(
  TestDb *pDb,                    /* Database handle */
  void *pCtx,                     /* Context pointer to pass to xCallback */
  int bReverse,                   /* True to scan in reverse order */
  void *pKey1, int nKey1,         /* Start of search */
  void *pKey2, int nKey2,         /* End of search */
  void (*xCallback)(void *pCtx, void *pKey, int nKey, void *pVal, int nVal)
){
  return pDb->pMethods->xScan(
      pDb, pCtx, bReverse, pKey1, nKey1, pKey2, nKey2, xCallback
  );
}

int tdb_begin(TestDb *pDb, int iLevel){
  return pDb->pMethods->xBegin(pDb, iLevel);
}
int tdb_commit(TestDb *pDb, int iLevel){
  return pDb->pMethods->xCommit(pDb, iLevel);
}
int tdb_rollback(TestDb *pDb, int iLevel){
  return pDb->pMethods->xRollback(pDb, iLevel);
}

int tdb_transaction_support(TestDb *pDb){
  return (pDb->pMethods->xBegin != error_transaction_function);
}

const char *tdb_library_name(TestDb *pDb){
  return pDb->zLibrary;
}

/* 
** End exported functions.
*************************************************************************/

Added ext/lsm1/lsm-test/lsmtest_tdb.h.















































































































































































































































































































































>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167

/*
** This file is the interface to a very simple database library used for
** testing. The interface is similar to that of the LSM. The main virtue 
** of this library is that the same API may be used to access a key-value
** store implemented by LSM, SQLite or another database system. Which 
** makes it easy to use for correctness and performance tests.
*/

#ifndef __WRAPPER_H_
#define __WRAPPER_H_

#ifdef __cplusplus
extern "C" {
#endif

#include "lsm.h"

typedef struct TestDb TestDb;

/*
** Open a new database connection. The first argument is the name of the
** database library to use. e.g. something like:
**
**     "sqlite3"
**     "lsm"
**
** See function tdb_system_name() for a list of available database systems.
**
** The second argument is the name of the database to open (e.g. a filename).
**
** If the third parameter is non-zero, then any existing database by the
** name of zDb is removed before opening a new one. If it is zero, then an
** existing database may be opened.
*/
int tdb_open(const char *zLibrary, const char *zDb, int bClear, TestDb **ppDb);

/*
** Close a database handle.
*/
int tdb_close(TestDb *pDb);

/*
** Write a new key/value into the database.
*/
int tdb_write(TestDb *pDb, void *pKey, int nKey, void *pVal, int nVal);

/*
** Delete a key from the database.
*/
int tdb_delete(TestDb *pDb, void *pKey, int nKey);

/*
** Delete a range of keys from the database.
*/
int tdb_delete_range(TestDb *, void *pKey1, int nKey1, void *pKey2, int nKey2);

/*
** Query the database for key (pKey/nKey). If no entry is found, set *ppVal
** to 0 and *pnVal to -1 before returning. Otherwise, set *ppVal and *pnVal
** to a pointer to and size of the value associated with (pKey/nKey).
*/
int tdb_fetch(TestDb *pDb, void *pKey, int nKey, void **ppVal, int *pnVal);

/*
** Open and close nested transactions. Currently, these functions only 
** work for SQLite3 and LSM systems. Use the tdb_transaction_support() 
** function to determine if a given TestDb handle supports these methods.
**
** These functions and the iLevel parameter follow the same conventions as
** the SQLite 4 transaction interface. Note that this is slightly different
** from the way LSM does things. As follows:
**
** tdb_begin():
**   A successful call to tdb_begin() with (iLevel>1) guarantees that 
**   there are at least (iLevel-1) write transactions open. If iLevel==1,
**   then it guarantees that at least a read-transaction is open. Calling
**   tdb_begin() with iLevel==0 is a no-op.
**
** tdb_commit():
**   A successful call to tdb_commit() with (iLevel>1) guarantees that 
**   there are at most (iLevel-1) write transactions open. If iLevel==1,
**   then it guarantees that there are no write transactions open (although
**   a read-transaction may remain open).  Calling tdb_commit() with 
**   iLevel==0 ensures that all transactions, read or write, have been 
**   closed and committed.
**
** tdb_rollback():
**   This call is similar to tdb_commit(), except that instead of committing
**   transactions, it reverts them. For example, calling tdb_rollback() with
**   iLevel==2 ensures that there is at most one write transaction open, and
**   restores the database to the state that it was in when that transaction
**   was opened.
**
**   In other words, tdb_commit() just closes transactions - tdb_rollback()
**   closes transactions and then restores the database to the state it
**   was in before those transactions were even opened.
*/
int tdb_begin(TestDb *pDb, int iLevel);
int tdb_commit(TestDb *pDb, int iLevel);
int tdb_rollback(TestDb *pDb, int iLevel);

/*
** Return true if transactions are supported, or false otherwise.
*/
int tdb_transaction_support(TestDb *pDb);

/*
** Return the name of the database library (as passed to tdb_open()) used
** by the handled passed as the first argument.
*/
const char *tdb_library_name(TestDb *pDb);

/*
** Scan a range of database keys. Invoke the callback function for each
** key visited.
*/
int tdb_scan(
  TestDb *pDb,                    /* Database handle */
  void *pCtx,                     /* Context pointer to pass to xCallback */
  int bReverse,                   /* True to scan in reverse order */
  void *pKey1, int nKey1,         /* Start of search */
  void *pKey2, int nKey2,         /* End of search */
  void (*xCallback)(void *pCtx, void *pKey, int nKey, void *pVal, int nVal)
);

const char *tdb_system_name(int i);

int tdb_lsm_open(const char *zCfg, const char *zDb, int bClear, TestDb **ppDb);

/*
** If the TestDb handle passed as an argument is a wrapper around an LSM
** database, return the LSM handle. Otherwise, if the argument is some other
** database system, return NULL.
*/
lsm_db *tdb_lsm(TestDb *pDb);

/*
** Return a pointer to the lsm_env object used by all lsm database
** connections initialized as a copy of the object returned by 
** lsm_default_env(). It may be modified (e.g. to override functions)
** if the caller can guarantee that it is not already in use.
*/
lsm_env *tdb_lsm_env(void);

/*
** The following functions only work with LSM database handles. It is
** illegal to call them with any other type of database handle specified
** as an argument.
*/
void tdb_lsm_enable_log(TestDb *pDb, int bEnable);
void tdb_lsm_application_crash(TestDb *pDb);
void tdb_lsm_prepare_system_crash(TestDb *pDb);
void tdb_lsm_system_crash(TestDb *pDb);
void tdb_lsm_prepare_sync_crash(TestDb *pDb, int iSync);


void tdb_lsm_safety(TestDb *pDb, int eMode);
void tdb_lsm_config_work_hook(TestDb *pDb, void (*)(lsm_db *, void *), void *);
void tdb_lsm_write_hook(TestDb *, void(*)(void*,int,lsm_i64,int,int), void*);
int tdb_lsm_config_str(TestDb *pDb, const char *zStr);

#ifdef __cplusplus
}  /* End of the 'extern "C"' block */
#endif

#endif

Added ext/lsm1/lsm-test/lsmtest_tdb2.cc.





































































































































































































































































































































































































































































































































































































































































































































































>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370


#include "lsmtest.h"
#include <stdlib.h>

#ifdef HAVE_KYOTOCABINET
#include "kcpolydb.h"
extern "C" {
  struct KcDb {
    TestDb base;
    kyotocabinet::TreeDB* db;
    char *pVal;
  };
}

int test_kc_open(const char *zFilename, int bClear, TestDb **ppDb){
  KcDb *pKcDb;
  int ok;
  int rc = 0;

  if( bClear ){
    char *zCmd = sqlite3_mprintf("rm -rf %s\n", zFilename);
    system(zCmd);
    sqlite3_free(zCmd);
  }

  pKcDb = (KcDb *)malloc(sizeof(KcDb));
  memset(pKcDb, 0, sizeof(KcDb));


  pKcDb->db = new kyotocabinet::TreeDB();
  pKcDb->db->tune_page(TESTDB_DEFAULT_PAGE_SIZE);
  pKcDb->db->tune_page_cache(
      TESTDB_DEFAULT_PAGE_SIZE * TESTDB_DEFAULT_CACHE_SIZE
  );
  ok = pKcDb->db->open(zFilename,
      kyotocabinet::PolyDB::OWRITER | kyotocabinet::PolyDB::OCREATE
  );
  if( ok==0 ){
    free(pKcDb);
    pKcDb = 0;
    rc = 1;
  }

  *ppDb = (TestDb *)pKcDb;
  return rc;
}

int test_kc_close(TestDb *pDb){
  KcDb *pKcDb = (KcDb *)pDb;
  if( pKcDb->pVal ){
    delete [] pKcDb->pVal;
  }
  pKcDb->db->close();
  delete pKcDb->db;
  free(pKcDb);
  return 0;
}

int test_kc_write(TestDb *pDb, void *pKey, int nKey, void *pVal, int nVal){
  KcDb *pKcDb = (KcDb *)pDb;
  int ok;

  ok = pKcDb->db->set((const char *)pKey, nKey, (const char *)pVal, nVal);
  return (ok ? 0 : 1);
}

int test_kc_delete(TestDb *pDb, void *pKey, int nKey){
  KcDb *pKcDb = (KcDb *)pDb;
  int ok;

  ok = pKcDb->db->remove((const char *)pKey, nKey);
  return (ok ? 0 : 1);
}

int test_kc_delete_range(
  TestDb *pDb, 
  void *pKey1, int nKey1,
  void *pKey2, int nKey2
){
  int res;
  KcDb *pKcDb = (KcDb *)pDb;
  kyotocabinet::DB::Cursor* pCur = pKcDb->db->cursor();

  if( pKey1 ){
    res = pCur->jump((const char *)pKey1, nKey1);
  }else{
    res = pCur->jump();
  }

  while( 1 ){
    const char *pKey; size_t nKey;
    const char *pVal; size_t nVal;

    pKey = pCur->get(&nKey, &pVal, &nVal);
    if( pKey==0 ) break;

#ifndef NDEBUG
    if( pKey1 ){
      res = memcmp(pKey, pKey1, MIN((size_t)nKey1, nKey));
      assert( res>0 || (res==0 && nKey>nKey1) );
    }
#endif

    if( pKey2 ){
      res = memcmp(pKey, pKey2, MIN((size_t)nKey2, nKey));
      if( res>0 || (res==0 && (size_t)nKey2<nKey) ){
        delete [] pKey;
        break;
      }
    }
    pCur->remove();
    delete [] pKey;
  }

  delete pCur;
  return 0;
}

int test_kc_fetch(
  TestDb *pDb, 
  void *pKey, 
  int nKey, 
  void **ppVal,
  int *pnVal
){
  KcDb *pKcDb = (KcDb *)pDb;
  size_t nVal;

  if( pKcDb->pVal ){
    delete [] pKcDb->pVal;
    pKcDb->pVal = 0;
  }

  pKcDb->pVal = pKcDb->db->get((const char *)pKey, nKey, &nVal);
  if( pKcDb->pVal ){
    *ppVal = pKcDb->pVal;
    *pnVal = nVal;
  }else{
    *ppVal = 0;
    *pnVal = -1;
  }

  return 0;
}

int test_kc_scan(
  TestDb *pDb,                    /* Database handle */
  void *pCtx,                     /* Context pointer to pass to xCallback */
  int bReverse,                   /* True for a reverse order scan */
  void *pKey1, int nKey1,         /* Start of search */
  void *pKey2, int nKey2,         /* End of search */
  void (*xCallback)(void *pCtx, void *pKey, int nKey, void *pVal, int nVal)
){
  KcDb *pKcDb = (KcDb *)pDb;
  kyotocabinet::DB::Cursor* pCur = pKcDb->db->cursor();
  int res;

  if( bReverse==0 ){
    if( pKey1 ){
      res = pCur->jump((const char *)pKey1, nKey1);
    }else{
      res = pCur->jump();
    }
  }else{
    if( pKey2 ){
      res = pCur->jump_back((const char *)pKey2, nKey2);
    }else{
      res = pCur->jump_back();
    }
  }

  while( res ){
    const char *pKey; size_t nKey;
    const char *pVal; size_t nVal;
    pKey = pCur->get(&nKey, &pVal, &nVal);

    if( bReverse==0 && pKey2 ){
      res = memcmp(pKey, pKey2, MIN((size_t)nKey2, nKey));
      if( res>0 || (res==0 && (size_t)nKey2<nKey) ){
        delete [] pKey;
        break;
      }
    }else if( bReverse!=0 && pKey1 ){
      res = memcmp(pKey, pKey1, MIN((size_t)nKey1, nKey));
      if( res<0 || (res==0 && (size_t)nKey1>nKey) ){
        delete [] pKey;
        break;
      }
    }

    xCallback(pCtx, (void *)pKey, (int)nKey, (void *)pVal, (int)nVal);
    delete [] pKey;

    if( bReverse ){
      res = pCur->step_back();
    }else{
      res = pCur->step();
    }
  }

  delete pCur;
  return 0;
}
#endif /* HAVE_KYOTOCABINET */

#ifdef HAVE_MDB 
#include "lmdb.h"

extern "C" {
  struct MdbDb {
    TestDb base;
    MDB_env *env;
    MDB_dbi dbi;
  };
}

int test_mdb_open(
  const char *zSpec, 
  const char *zFilename, 
  int bClear, 
  TestDb **ppDb
){
  MDB_txn *txn;
  MdbDb *pMdb;
  int rc;

  if( bClear ){
    char *zCmd = sqlite3_mprintf("rm -rf %s\n", zFilename);
    system(zCmd);
    sqlite3_free(zCmd);
  }

  pMdb = (MdbDb *)malloc(sizeof(MdbDb));
  memset(pMdb, 0, sizeof(MdbDb));

  rc = mdb_env_create(&pMdb->env);
  if( rc==0 ) rc = mdb_env_set_mapsize(pMdb->env, 1*1024*1024*1024);
  if( rc==0 ) rc = mdb_env_open(pMdb->env, zFilename, MDB_NOSYNC|MDB_NOSUBDIR, 0600);
  if( rc==0 ) rc = mdb_txn_begin(pMdb->env, NULL, 0, &txn);
  if( rc==0 ){
    rc = mdb_open(txn, NULL, 0, &pMdb->dbi);
    mdb_txn_commit(txn);
  }

  *ppDb = (TestDb *)pMdb;
  return rc;
}

int test_mdb_close(TestDb *pDb){
  MdbDb *pMdb = (MdbDb *)pDb;

  mdb_close(pMdb->env, pMdb->dbi);
  mdb_env_close(pMdb->env);
  free(pMdb);
  return 0;
}

int test_mdb_write(TestDb *pDb, void *pKey, int nKey, void *pVal, int nVal){
  int rc;
  MdbDb *pMdb = (MdbDb *)pDb;
  MDB_val val;
  MDB_val key;
  MDB_txn *txn;

  val.mv_size = nVal; 
  val.mv_data = pVal;
  key.mv_size = nKey; 
  key.mv_data = pKey;

  rc = mdb_txn_begin(pMdb->env, NULL, 0, &txn);
  if( rc==0 ){
    rc = mdb_put(txn, pMdb->dbi, &key, &val, 0);
    if( rc==0 ){
      rc = mdb_txn_commit(txn);
    }else{
      mdb_txn_abort(txn);
    }
  }
  
  return rc;
}

int test_mdb_delete(TestDb *pDb, void *pKey, int nKey){
  int rc;
  MdbDb *pMdb = (MdbDb *)pDb;
  MDB_val key;
  MDB_txn *txn;

  key.mv_size = nKey; 
  key.mv_data = pKey;
  rc = mdb_txn_begin(pMdb->env, NULL, 0, &txn);
  if( rc==0 ){
    rc = mdb_del(txn, pMdb->dbi, &key, 0);
    if( rc==0 ){
      rc = mdb_txn_commit(txn);
    }else{
      mdb_txn_abort(txn);
    }
  }
  
  return rc;
}

int test_mdb_fetch(
  TestDb *pDb, 
  void *pKey, 
  int nKey, 
  void **ppVal,
  int *pnVal
){
  int rc;
  MdbDb *pMdb = (MdbDb *)pDb;
  MDB_val key;
  MDB_txn *txn;

  key.mv_size = nKey;
  key.mv_data = pKey;

  rc = mdb_txn_begin(pMdb->env, NULL, MDB_RDONLY, &txn);
  if( rc==0 ){
    MDB_val val = {0, 0};
    rc = mdb_get(txn, pMdb->dbi, &key, &val);
    if( rc==MDB_NOTFOUND ){
      rc = 0;
      *ppVal = 0;
      *pnVal = -1;
    }else{
      *ppVal = val.mv_data;
      *pnVal = val.mv_size;
    }
    mdb_txn_commit(txn);
  }

  return rc;
}

int test_mdb_scan(
  TestDb *pDb,                    /* Database handle */
  void *pCtx,                     /* Context pointer to pass to xCallback */
  int bReverse,                   /* True for a reverse order scan */
  void *pKey1, int nKey1,         /* Start of search */
  void *pKey2, int nKey2,         /* End of search */
  void (*xCallback)(void *pCtx, void *pKey, int nKey, void *pVal, int nVal)
){
  MdbDb *pMdb = (MdbDb *)pDb;
  int rc;
  MDB_cursor_op op = bReverse ? MDB_PREV : MDB_NEXT;
  MDB_txn *txn;

  rc = mdb_txn_begin(pMdb->env, NULL, MDB_RDONLY, &txn);
  if( rc==0 ){
    MDB_cursor *csr;
    MDB_val key = {0, 0};
    MDB_val val = {0, 0};

    rc = mdb_cursor_open(txn, pMdb->dbi, &csr);
    if( rc==0 ){
      while( mdb_cursor_get(csr, &key, &val, op)==0 ){
        xCallback(pCtx, key.mv_data, key.mv_size, val.mv_data, val.mv_size);
      }
      mdb_cursor_close(csr);
    }
  }

  return rc;
}

#endif /* HAVE_MDB */

Added ext/lsm1/lsm-test/lsmtest_tdb3.c.







































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379

#include "lsmtest_tdb.h"
#include "lsm.h"
#include "lsmtest.h"

#include <stdlib.h>
#include <string.h>
#include <assert.h>
#ifndef _WIN32
# include <unistd.h>
#endif
#include <stdio.h>

#ifndef _WIN32
# include <sys/time.h>
#endif

typedef struct LsmDb LsmDb;
typedef struct LsmWorker LsmWorker;
typedef struct LsmFile LsmFile;

#define LSMTEST_DFLT_MT_MAX_CKPT (8*1024)
#define LSMTEST_DFLT_MT_MIN_CKPT (2*1024)

#ifdef LSM_MUTEX_PTHREADS
#include <pthread.h>

#define LSMTEST_THREAD_CKPT      1
#define LSMTEST_THREAD_WORKER    2
#define LSMTEST_THREAD_WORKER_AC 3

/*
** There are several different types of worker threads that run in different
** test configurations, depending on the value of LsmWorker.eType.
**
**   1. Checkpointer.
**   2. Worker with auto-checkpoint.
**   3. Worker without auto-checkpoint.
*/
struct LsmWorker {
  LsmDb *pDb;                     /* Main database structure */
  lsm_db *pWorker;                /* Worker database handle */
  pthread_t worker_thread;        /* Worker thread */
  pthread_cond_t worker_cond;     /* Condition var the worker waits on */
  pthread_mutex_t worker_mutex;   /* Mutex used with worker_cond */
  int bDoWork;                    /* Set to true by client when there is work */
  int worker_rc;                  /* Store error code here */
  int eType;                      /* LSMTEST_THREAD_XXX constant */
  int bBlock;
};
#else
struct LsmWorker { int worker_rc; int bBlock; };
#endif

static void mt_shutdown(LsmDb *);

lsm_env *tdb_lsm_env(void){
  static int bInit = 0;
  static lsm_env env;
  if( bInit==0 ){
    memcpy(&env, lsm_default_env(), sizeof(env));
    bInit = 1;
  }
  return &env;
}

typedef struct FileSector FileSector;
typedef struct FileData FileData;

struct FileSector {
  u8 *aOld;                       /* Old data for this sector */
};

struct FileData {
  int nSector;                    /* Allocated size of apSector[] array */
  FileSector *aSector;            /* Array of file sectors */
};

/*
** bPrepareCrash:
**   If non-zero, the file wrappers maintain enough in-memory data to
**   simulate the effect of a power-failure on the file-system (i.e. that
**   unsynced sectors may be written, not written, or overwritten with
**   arbitrary data when the crash occurs).
**
** bCrashed:
**   Set to true after a crash is simulated. Once this variable is true, all
**   VFS methods other than xClose() return LSM_IOERR as soon as they are
**   called (without affecting the contents of the file-system).
**
** env:
**   The environment object used by all lsm_db* handles opened by this
**   object (i.e. LsmDb.db plus any worker connections). Variable env.pVfsCtx
**   always points to the containing LsmDb structure.
*/
struct LsmDb {
  TestDb base;                    /* Base class - methods table */
  lsm_env env;                    /* Environment used by connection db */
  char *zName;                    /* Database file name */
  lsm_db *db;                     /* LSM database handle */

  lsm_cursor *pCsr;               /* Cursor held open during read transaction */
  void *pBuf;                     /* Buffer for tdb_fetch() output */
  int nBuf;                       /* Allocated (not used) size of pBuf */

  /* Crash testing related state */
  int bCrashed;                   /* True once a crash has occurred */
  int nAutoCrash;                 /* Number of syncs until a crash */
  int bPrepareCrash;              /* True to store writes in memory */

  /* Unsynced data (while crash testing) */
  int szSector;                   /* Assumed size of disk sectors (512B) */
  FileData aFile[2];              /* Database and log file data */

  /* Other test instrumentation */
  int bNoRecovery;                /* If true, assume DMS2 is locked */

  /* Work hook redirection */
  void (*xWork)(lsm_db *, void *);
  void *pWorkCtx;

  /* IO logging hook */
  void (*xWriteHook)(void *, int, lsm_i64, int, int);
  void *pWriteCtx;
  
  /* Worker threads (for lsm_mt) */
  int nMtMinCkpt;
  int nMtMaxCkpt;
  int eMode;
  int nWorker;
  LsmWorker *aWorker;
};

#define LSMTEST_MODE_SINGLETHREAD    1
#define LSMTEST_MODE_BACKGROUND_CKPT 2
#define LSMTEST_MODE_BACKGROUND_WORK 3
#define LSMTEST_MODE_BACKGROUND_BOTH 4

/*************************************************************************
**************************************************************************
** Begin test VFS code.
*/

struct LsmFile {
  lsm_file *pReal;                /* Real underlying file */
  int bLog;                       /* True for log file. False for db file */
  LsmDb *pDb;                     /* Database handle that uses this file */
};

static int testEnvFullpath(
  lsm_env *pEnv,                  /* Environment for current LsmDb */
  const char *zFile,              /* Relative path name */
  char *zOut,                     /* Output buffer */
  int *pnOut                      /* IN/OUT: Size of output buffer */
){
  lsm_env *pRealEnv = tdb_lsm_env();
  return pRealEnv->xFullpath(pRealEnv, zFile, zOut, pnOut);
}

static int testEnvOpen(
  lsm_env *pEnv,                  /* Environment for current LsmDb */
  const char *zFile,              /* Name of file to open */
  int flags,
  lsm_file **ppFile               /* OUT: New file handle object */
){
  lsm_env *pRealEnv = tdb_lsm_env();
  LsmDb *pDb = (LsmDb *)pEnv->pVfsCtx;
  int rc;                         /* Return Code */
  LsmFile *pRet;                  /* The new file handle */
  int nFile;                      /* Length of string zFile in bytes */

  nFile = strlen(zFile);
  pRet = (LsmFile *)testMalloc(sizeof(LsmFile));
  pRet->pDb = pDb;
  pRet->bLog = (nFile > 4 && 0==memcmp("-log", &zFile[nFile-4], 4));

  rc = pRealEnv->xOpen(pRealEnv, zFile, flags, &pRet->pReal);
  if( rc!=LSM_OK ){
    testFree(pRet);
    pRet = 0;
  }

  *ppFile = (lsm_file *)pRet;
  return rc;
}

static int testEnvRead(lsm_file *pFile, lsm_i64 iOff, void *pData, int nData){
  lsm_env *pRealEnv = tdb_lsm_env();
  LsmFile *p = (LsmFile *)pFile;
  if( p->pDb->bCrashed ) return LSM_IOERR;
  return pRealEnv->xRead(p->pReal, iOff, pData, nData);
}

static int testEnvWrite(lsm_file *pFile, lsm_i64 iOff, void *pData, int nData){
  lsm_env *pRealEnv = tdb_lsm_env();
  LsmFile *p = (LsmFile *)pFile;
  LsmDb *pDb = p->pDb;

  if( pDb->bCrashed ) return LSM_IOERR;

  if( pDb->bPrepareCrash ){
    FileData *pData = &pDb->aFile[p->bLog];
    int iFirst;                 
    int iLast;
    int iSector;

    iFirst = (iOff / pDb->szSector);
    iLast =  ((iOff + nData - 1) / pDb->szSector);

    if( pData->nSector<(iLast+1) ){
      int nNew = ( ((iLast + 1) + 63) / 64 ) * 64;
      assert( nNew>iLast );
      pData->aSector = (FileSector *)testRealloc(
          pData->aSector, nNew*sizeof(FileSector)
      );
      memset(&pData->aSector[pData->nSector], 
          0, (nNew - pData->nSector) * sizeof(FileSector)
      );
      pData->nSector = nNew;
    }

    for(iSector=iFirst; iSector<=iLast; iSector++){
      if( pData->aSector[iSector].aOld==0 ){
        u8 *aOld = (u8 *)testMalloc(pDb->szSector);
        pRealEnv->xRead(
            p->pReal, (lsm_i64)iSector*pDb->szSector, aOld, pDb->szSector
        );
        pData->aSector[iSector].aOld = aOld;
      }
    }
  }

  if( pDb->xWriteHook ){
    int rc;
    int nUs;
    struct timeval t1;
    struct timeval t2;

    gettimeofday(&t1, 0);
    assert( nData>0 );
    rc = pRealEnv->xWrite(p->pReal, iOff, pData, nData);
    gettimeofday(&t2, 0);

    nUs = (t2.tv_sec - t1.tv_sec) * 1000000 + (t2.tv_usec - t1.tv_usec);
    pDb->xWriteHook(pDb->pWriteCtx, p->bLog, iOff, nData, nUs);
    return rc;
  }

  return pRealEnv->xWrite(p->pReal, iOff, pData, nData);
}

static void doSystemCrash(LsmDb *pDb);

static int testEnvSync(lsm_file *pFile){
  lsm_env *pRealEnv = tdb_lsm_env();
  LsmFile *p = (LsmFile *)pFile;
  LsmDb *pDb = p->pDb;
  FileData *pData = &pDb->aFile[p->bLog];
  int i;

  if( pDb->bCrashed ) return LSM_IOERR;

  if( pDb->nAutoCrash ){
    pDb->nAutoCrash--;
    if( pDb->nAutoCrash==0 ){
      doSystemCrash(pDb);
      pDb->bCrashed = 1;
      return LSM_IOERR;
    }
  }

  if( pDb->bPrepareCrash ){
    for(i=0; i<pData->nSector; i++){
      testFree(pData->aSector[i].aOld);
      pData->aSector[i].aOld = 0;
    }
  }

  if( pDb->xWriteHook ){
    int rc;
    int nUs;
    struct timeval t1;
    struct timeval t2;

    gettimeofday(&t1, 0);
    rc = pRealEnv->xSync(p->pReal);
    gettimeofday(&t2, 0);

    nUs = (t2.tv_sec - t1.tv_sec) * 1000000 + (t2.tv_usec - t1.tv_usec);
    pDb->xWriteHook(pDb->pWriteCtx, p->bLog, 0, 0, nUs);
    return rc;
  }

  return pRealEnv->xSync(p->pReal);
}

static int testEnvTruncate(lsm_file *pFile, lsm_i64 iOff){
  lsm_env *pRealEnv = tdb_lsm_env();
  LsmFile *p = (LsmFile *)pFile;
  if( p->pDb->bCrashed ) return LSM_IOERR;
  return pRealEnv->xTruncate(p->pReal, iOff);
}

static int testEnvSectorSize(lsm_file *pFile){
  lsm_env *pRealEnv = tdb_lsm_env();
  LsmFile *p = (LsmFile *)pFile;
  return pRealEnv->xSectorSize(p->pReal);
}

static int testEnvRemap(
  lsm_file *pFile, 
  lsm_i64 iMin, 
  void **ppOut,
  lsm_i64 *pnOut
){
  lsm_env *pRealEnv = tdb_lsm_env();
  LsmFile *p = (LsmFile *)pFile;
  return pRealEnv->xRemap(p->pReal, iMin, ppOut, pnOut);
}

static int testEnvFileid(
  lsm_file *pFile, 
  void *ppOut,
  int *pnOut
){
  lsm_env *pRealEnv = tdb_lsm_env();
  LsmFile *p = (LsmFile *)pFile;
  return pRealEnv->xFileid(p->pReal, ppOut, pnOut);
}

static int testEnvClose(lsm_file *pFile){
  lsm_env *pRealEnv = tdb_lsm_env();
  LsmFile *p = (LsmFile *)pFile;

  pRealEnv->xClose(p->pReal);
  testFree(p);
  return LSM_OK;
}

static int testEnvUnlink(lsm_env *pEnv, const char *zFile){
  lsm_env *pRealEnv = tdb_lsm_env();
  unused_parameter(pEnv);
  return pRealEnv->xUnlink(pRealEnv, zFile);
}

static int testEnvLock(lsm_file *pFile, int iLock, int eType){
  LsmFile *p = (LsmFile *)pFile;
  lsm_env *pRealEnv = tdb_lsm_env();

  if( iLock==2 && eType==LSM_LOCK_EXCL && p->pDb->bNoRecovery ){
    return LSM_BUSY;
  }
  return pRealEnv->xLock(p->pReal, iLock, eType);
}

static int testEnvTestLock(lsm_file *pFile, int iLock, int nLock, int eType){
  LsmFile *p = (LsmFile *)pFile;
  lsm_env *pRealEnv = tdb_lsm_env();

  if( iLock==2 && eType==LSM_LOCK_EXCL && p->pDb->bNoRecovery ){
    return LSM_BUSY;
  }
  return pRealEnv->xTestLock(p->pReal, iLock, nLock, eType);
}

static int testEnvShmMap(lsm_file *pFile, int iRegion, int sz, void **pp){
  LsmFile *p = (LsmFile *)pFile;
  lsm_env *pRealEnv = tdb_lsm_env();
  return pRealEnv->xShmMap(p->pReal, iRegion, sz, pp);
}

static void testEnvShmBarrier(void){
}

static int testEnvShmUnmap(lsm_file *pFile, int bDel){
  LsmFile *p = (LsmFile *)pFile;
  lsm_env *pRealEnv = tdb_lsm_env();
  return pRealEnv->xShmUnmap(p->pReal, bDel);
}

static int testEnvSleep(lsm_env *pEnv, int us){
  lsm_env *pRealEnv = tdb_lsm_env();
  return pRealEnv->xSleep(pRealEnv, us);
}

static void doSystemCrash(LsmDb *pDb){
  lsm_env *pEnv = tdb_lsm_env();
  int iFile;
  int iSeed = pDb->aFile[0].nSector + pDb->aFile[1].nSector;

  char *zFile = pDb->zName;
  char *zFree = 0;

  for(iFile=0; iFile<2; iFile++){
    lsm_file *pFile = 0;
    int i;

    pEnv->xOpen(pEnv, zFile, 0, &pFile);
    for(i=0; i<pDb->aFile[iFile].nSector; i++){
      u8 *aOld = pDb->aFile[iFile].aSector[i].aOld;
      if( aOld ){
        int iOpt = testPrngValue(iSeed++) % 3;
        switch( iOpt ){
          case 0:
            break;

          case 1:
            testPrngArray(iSeed++, (u32 *)aOld, pDb->szSector/4);
            /* Fall-through */

          case 2:
            pEnv->xWrite(
                pFile, (lsm_i64)i * pDb->szSector, aOld, pDb->szSector
            );
            break;
        }
        testFree(aOld);
        pDb->aFile[iFile].aSector[i].aOld = 0;
      }
    }
    pEnv->xClose(pFile);
    zFree = zFile = sqlite3_mprintf("%s-log", pDb->zName);
  }

  sqlite3_free(zFree);
}
/*
** End test VFS code.
**************************************************************************
*************************************************************************/

/*************************************************************************
**************************************************************************
** Begin test compression hooks.
*/

#ifdef HAVE_ZLIB
#include <zlib.h>

static int testZipBound(void *pCtx, int nSrc){
  return compressBound(nSrc);
}

static int testZipCompress(
  void *pCtx,                     /* Context pointer */
  char *aOut, int *pnOut,         /* OUT: Buffer containing compressed data */
  const char *aIn, int nIn        /* Buffer containing input data */
){
  uLongf n = *pnOut;              /* In/out buffer size for compress() */
  int rc;                         /* compress() return code */
 
  rc = compress((Bytef*)aOut, &n, (Bytef*)aIn, nIn);
  *pnOut = n;
  return (rc==Z_OK ? 0 : LSM_ERROR);
}

static int testZipUncompress(
  void *pCtx,                     /* Context pointer */
  char *aOut, int *pnOut,         /* OUT: Buffer containing uncompressed data */
  const char *aIn, int nIn        /* Buffer containing input data */
){
  uLongf n = *pnOut;              /* In/out buffer size for uncompress() */
  int rc;                         /* uncompress() return code */

  rc = uncompress((Bytef*)aOut, &n, (Bytef*)aIn, nIn);
  *pnOut = n;
  return (rc==Z_OK ? 0 : LSM_ERROR);
}

static int testConfigureCompression(lsm_db *pDb){
  static lsm_compress zip = {
    0,                            /* Context pointer (unused) */
    1,                            /* Id value */
    testZipBound,                 /* xBound method */
    testZipCompress,              /* xCompress method */
    testZipUncompress             /* xUncompress method */
  };
  return lsm_config(pDb, LSM_CONFIG_SET_COMPRESSION, &zip);
}
#endif /* ifdef HAVE_ZLIB */

/*
** End test compression hooks.
**************************************************************************
*************************************************************************/

static int test_lsm_close(TestDb *pTestDb){
  int i;
  int rc = LSM_OK;
  LsmDb *pDb = (LsmDb *)pTestDb;

  lsm_csr_close(pDb->pCsr);
  lsm_close(pDb->db);

  /* If this is a multi-threaded database, wait on the worker threads. */
  mt_shutdown(pDb);
  for(i=0; i<pDb->nWorker && rc==LSM_OK; i++){
    rc = pDb->aWorker[i].worker_rc;
  }

  for(i=0; i<pDb->aFile[0].nSector; i++){
    testFree(pDb->aFile[0].aSector[i].aOld);
  }
  testFree(pDb->aFile[0].aSector);
  for(i=0; i<pDb->aFile[1].nSector; i++){
    testFree(pDb->aFile[1].aSector[i].aOld);
  }
  testFree(pDb->aFile[1].aSector);

  memset(pDb, sizeof(LsmDb), 0x11);
  testFree((char *)pDb->pBuf);
  testFree((char *)pDb);
  return rc;
}

static void mt_signal_worker(LsmDb*, int);

static int waitOnCheckpointer(LsmDb *pDb, lsm_db *db){
  int nSleep = 0;
  int nKB;
  int rc;

  do {
    nKB = 0;
    rc = lsm_info(db, LSM_INFO_CHECKPOINT_SIZE, &nKB);
    if( rc!=LSM_OK || nKB<pDb->nMtMaxCkpt ) break;
#ifdef LSM_MUTEX_PTHREADS
    mt_signal_worker(pDb, 
        (pDb->eMode==LSMTEST_MODE_BACKGROUND_CKPT ? 0 : 1)
    );
#endif
    usleep(5000);
    nSleep += 5;
  }while( 1 );

#if 0
    if( nSleep ) printf("# waitOnCheckpointer(): nSleep=%d\n", nSleep);
#endif

  return rc;
}

static int waitOnWorker(LsmDb *pDb){
  int rc;
  int nLimit = -1;
  int nSleep = 0;

  rc = lsm_config(pDb->db, LSM_CONFIG_AUTOFLUSH, &nLimit);
  do {
    int nOld, nNew, rc;
    rc = lsm_info(pDb->db, LSM_INFO_TREE_SIZE, &nOld, &nNew);
    if( rc!=LSM_OK ) return rc;
    if( nOld==0 || nNew<(nLimit/2) ) break;
#ifdef LSM_MUTEX_PTHREADS
    mt_signal_worker(pDb, 0);
#endif
    usleep(5000);
    nSleep += 5;
  }while( 1 );

#if 0
  if( nSleep ) printf("# waitOnWorker(): nSleep=%d\n", nSleep);
#endif

  return rc;
}

static int test_lsm_write(
  TestDb *pTestDb, 
  void *pKey, 
  int nKey, 
  void *pVal,
  int nVal
){
  LsmDb *pDb = (LsmDb *)pTestDb;
  int rc = LSM_OK;

  if( pDb->eMode==LSMTEST_MODE_BACKGROUND_CKPT ){
    rc = waitOnCheckpointer(pDb, pDb->db);
  }else if( 
      pDb->eMode==LSMTEST_MODE_BACKGROUND_WORK
   || pDb->eMode==LSMTEST_MODE_BACKGROUND_BOTH 
  ){
    rc = waitOnWorker(pDb);
  }

  if( rc==LSM_OK ){
    rc = lsm_insert(pDb->db, pKey, nKey, pVal, nVal);
  }
  return rc;
}

static int test_lsm_delete(TestDb *pTestDb, void *pKey, int nKey){
  LsmDb *pDb = (LsmDb *)pTestDb;
  return lsm_delete(pDb->db, pKey, nKey);
}

static int test_lsm_delete_range(
  TestDb *pTestDb, 
  void *pKey1, int nKey1,
  void *pKey2, int nKey2
){
  LsmDb *pDb = (LsmDb *)pTestDb;
  return lsm_delete_range(pDb->db, pKey1, nKey1, pKey2, nKey2);
}

static int test_lsm_fetch(
  TestDb *pTestDb, 
  void *pKey, 
  int nKey, 
  void **ppVal, 
  int *pnVal
){
  int rc;
  LsmDb *pDb = (LsmDb *)pTestDb;
  lsm_cursor *csr;

  if( pKey==0 ) return LSM_OK;

  rc = lsm_csr_open(pDb->db, &csr);
  if( rc!=LSM_OK ) return rc;

  rc = lsm_csr_seek(csr, pKey, nKey, LSM_SEEK_EQ);
  if( rc==LSM_OK ){
    if( lsm_csr_valid(csr) ){
      const void *pVal; int nVal;
      rc = lsm_csr_value(csr, &pVal, &nVal);
      if( nVal>pDb->nBuf ){
        testFree(pDb->pBuf);
        pDb->pBuf = testMalloc(nVal*2);
        pDb->nBuf = nVal*2;
      }
      memcpy(pDb->pBuf, pVal, nVal);
      *ppVal = pDb->pBuf;
      *pnVal = nVal;
    }else{
      *ppVal = 0;
      *pnVal = -1;
    }
  }
  lsm_csr_close(csr);
  return rc;
}

static int test_lsm_scan(
  TestDb *pTestDb,
  void *pCtx,
  int bReverse,
  void *pFirst, int nFirst,
  void *pLast, int nLast,
  void (*xCallback)(void *, void *, int , void *, int)
){
  LsmDb *pDb = (LsmDb *)pTestDb;
  lsm_cursor *csr;
  int rc;

  rc = lsm_csr_open(pDb->db, &csr);
  if( rc!=LSM_OK ) return rc;

  if( bReverse ){
    if( pLast ){
      rc = lsm_csr_seek(csr, pLast, nLast, LSM_SEEK_LE);
    }else{
      rc = lsm_csr_last(csr);
    }
  }else{
    if( pFirst ){
      rc = lsm_csr_seek(csr, pFirst, nFirst, LSM_SEEK_GE);
    }else{
      rc = lsm_csr_first(csr);
    }
  }

  while( rc==LSM_OK && lsm_csr_valid(csr) ){
    const void *pKey; int nKey;
    const void *pVal; int nVal;
    int cmp;

    lsm_csr_key(csr, &pKey, &nKey);
    lsm_csr_value(csr, &pVal, &nVal);

    if( bReverse && pFirst ){
      cmp = memcmp(pFirst, pKey, MIN(nKey, nFirst));
      if( cmp>0 || (cmp==0 && nFirst>nKey) ) break;
    }else if( bReverse==0 && pLast ){
      cmp = memcmp(pLast, pKey, MIN(nKey, nLast));
      if( cmp<0 || (cmp==0 && nLast<nKey) ) break;
    }

    xCallback(pCtx, (void *)pKey, nKey, (void *)pVal, nVal);

    if( bReverse ){
      rc = lsm_csr_prev(csr);
    }else{
      rc = lsm_csr_next(csr);
    }
  }

  lsm_csr_close(csr);
  return rc;
}

static int test_lsm_begin(TestDb *pTestDb, int iLevel){
  int rc = LSM_OK;
  LsmDb *pDb = (LsmDb *)pTestDb;

  /* iLevel==0 is a no-op. */
  if( iLevel==0 ) return 0;

  if( pDb->pCsr==0 ) rc = lsm_csr_open(pDb->db, &pDb->pCsr);
  if( rc==LSM_OK && iLevel>1 ){
    rc = lsm_begin(pDb->db, iLevel-1);
  }

  return rc;
}
static int test_lsm_commit(TestDb *pTestDb, int iLevel){
  LsmDb *pDb = (LsmDb *)pTestDb;

  /* If iLevel==0, close any open read transaction */
  if( iLevel==0 && pDb->pCsr ){
    lsm_csr_close(pDb->pCsr);
    pDb->pCsr = 0;
  }

  /* If iLevel==0, close any open read transaction */
  return lsm_commit(pDb->db, MAX(0, iLevel-1));
}
static int test_lsm_rollback(TestDb *pTestDb, int iLevel){
  LsmDb *pDb = (LsmDb *)pTestDb;

  /* If iLevel==0, close any open read transaction */
  if( iLevel==0 && pDb->pCsr ){
    lsm_csr_close(pDb->pCsr);
    pDb->pCsr = 0;
  }

  return lsm_rollback(pDb->db, MAX(0, iLevel-1));
}

/*
** A log message callback registered with lsm connections. Prints all 
** messages to stderr.
*/
static void xLog(void *pCtx, int rc, const char *z){
  unused_parameter(rc);
  /* fprintf(stderr, "lsm: rc=%d \"%s\"\n", rc, z); */
  if( pCtx ) fprintf(stderr, "%s: ", (char *)pCtx);
  fprintf(stderr, "%s\n", z);
  fflush(stderr);
}

static void xWorkHook(lsm_db *db, void *pArg){
  LsmDb *p = (LsmDb *)pArg;
  if( p->xWork ) p->xWork(db, p->pWorkCtx);
}

#define TEST_NO_RECOVERY -1
#define TEST_COMPRESSION -3

#define TEST_MT_MODE     -2
#define TEST_MT_MIN_CKPT -4
#define TEST_MT_MAX_CKPT -5

int test_lsm_config_str(
  LsmDb *pLsm,
  lsm_db *db, 
  int bWorker,
  const char *zStr,
  int *pnThread
){
  struct CfgParam {
    const char *zParam;
    int bWorker;
    int eParam;
  } aParam[] = {
    { "autoflush",        0, LSM_CONFIG_AUTOFLUSH },
    { "page_size",        0, LSM_CONFIG_PAGE_SIZE },
    { "block_size",       0, LSM_CONFIG_BLOCK_SIZE },
    { "safety",           0, LSM_CONFIG_SAFETY },
    { "autowork",         0, LSM_CONFIG_AUTOWORK },
    { "autocheckpoint",   0, LSM_CONFIG_AUTOCHECKPOINT },
    { "mmap",             0, LSM_CONFIG_MMAP },
    { "use_log",          0, LSM_CONFIG_USE_LOG },
    { "automerge",        0, LSM_CONFIG_AUTOMERGE },
    { "max_freelist",     0, LSM_CONFIG_MAX_FREELIST },
    { "multi_proc",       0, LSM_CONFIG_MULTIPLE_PROCESSES },
    { "worker_automerge", 1, LSM_CONFIG_AUTOMERGE },
    { "test_no_recovery", 0, TEST_NO_RECOVERY },
    { "bg_min_ckpt",      0, TEST_NO_RECOVERY },

    { "mt_mode",          0, TEST_MT_MODE },
    { "mt_min_ckpt",      0, TEST_MT_MIN_CKPT },
    { "mt_max_ckpt",      0, TEST_MT_MAX_CKPT },

#ifdef HAVE_ZLIB
    { "compression",      0, TEST_COMPRESSION },
#endif
    { 0, 0 }
  };
  const char *z = zStr;
  int nThread = 1;

  if( zStr==0 ) return 0;

  assert( db );
  while( z[0] ){
    const char *zStart;

    /* Skip whitespace */
    while( *z==' ' ) z++;
    zStart = z;

    while( *z && *z!='=' ) z++;
    if( *z ){
      int eParam;
      int i;
      int iVal;
      int iMul = 1;
      int rc;
      char zParam[32];
      int nParam = z-zStart;
      if( nParam==0 || nParam>sizeof(zParam)-1 ) goto syntax_error;

      memcpy(zParam, zStart, nParam);
      zParam[nParam] = '\0';
      rc = testArgSelect(aParam, "param", zParam, &i);
      if( rc!=0 ) return rc;
      eParam = aParam[i].eParam;

      z++;
      zStart = z;
      while( *z>='0' && *z<='9' ) z++;
      if( *z=='k' || *z=='K' ){
        iMul = 1;
        z++;
      }else if( *z=='M' || *z=='M' ){
        iMul = 1024;
        z++;
      }
      nParam = z-zStart;
      if( nParam==0 || nParam>sizeof(zParam)-1 ) goto syntax_error;
      memcpy(zParam, zStart, nParam);
      zParam[nParam] = '\0';
      iVal = atoi(zParam) * iMul;

      if( eParam>0 ){
        if( bWorker || aParam[i].bWorker==0 ){
          lsm_config(db, eParam, &iVal);
        }
      }else{
        switch( eParam ){
          case TEST_NO_RECOVERY:
            if( pLsm ) pLsm->bNoRecovery = iVal;
            break;
          case TEST_MT_MODE:
            if( pLsm ) nThread = iVal;
            break;
          case TEST_MT_MIN_CKPT:
            if( pLsm && iVal>0 ) pLsm->nMtMinCkpt = iVal*1024;
            break;
          case TEST_MT_MAX_CKPT:
            if( pLsm && iVal>0 ) pLsm->nMtMaxCkpt = iVal*1024;
            break;
#ifdef HAVE_ZLIB
          case TEST_COMPRESSION:
            testConfigureCompression(db);
            break;
#endif
        }
      }
    }else if( z!=zStart ){
      goto syntax_error;
    }
  }

  if( pnThread ) *pnThread = nThread;
  if( pLsm && pLsm->nMtMaxCkpt < pLsm->nMtMinCkpt ){
    pLsm->nMtMinCkpt = pLsm->nMtMaxCkpt;
  }

  return 0;
 syntax_error:
  testPrintError("syntax error at: \"%s\"\n", z);
  return 1;
}

int tdb_lsm_config_str(TestDb *pDb, const char *zStr){
  int rc = 0;
  if( tdb_lsm(pDb) ){
    int i;
    LsmDb *pLsm = (LsmDb *)pDb;

    rc = test_lsm_config_str(pLsm, pLsm->db, 0, zStr, 0);
#ifdef LSM_MUTEX_PTHREADS
    for(i=0; rc==0 && i<pLsm->nWorker; i++){
      rc = test_lsm_config_str(0, pLsm->aWorker[i].pWorker, 1, zStr, 0);
    }
#endif
  }
  return rc;
}

int tdb_lsm_configure(lsm_db *db, const char *zConfig){
  return test_lsm_config_str(0, db, 0, zConfig, 0);
}

static int testLsmStartWorkers(LsmDb *, int, const char *, const char *);

static int testLsmOpen(
  const char *zCfg,
  const char *zFilename, 
  int bClear, 
  TestDb **ppDb
){
  static const DatabaseMethods LsmMethods = {
    test_lsm_close,
    test_lsm_write,
    test_lsm_delete,
    test_lsm_delete_range,
    test_lsm_fetch,
    test_lsm_scan,
    test_lsm_begin,
    test_lsm_commit,
    test_lsm_rollback
  };

  int rc;
  int nFilename;
  LsmDb *pDb;

  /* If the bClear flag is set, delete any existing database. */
  assert( zFilename);
  if( bClear ) testDeleteLsmdb(zFilename);
  nFilename = strlen(zFilename);

  pDb = (LsmDb *)testMalloc(sizeof(LsmDb) + nFilename + 1);
  memset(pDb, 0, sizeof(LsmDb));
  pDb->base.pMethods = &LsmMethods;
  pDb->zName = (char *)&pDb[1];
  memcpy(pDb->zName, zFilename, nFilename + 1);

  /* Default the sector size used for crash simulation to 512 bytes. 
  ** Todo: There should be an OS method to obtain this value - just as
  ** there is in SQLite. For now, LSM assumes that it is smaller than
  ** the page size (default 4KB).
  */
  pDb->szSector = 256;

  /* Default values for the mt_min_ckpt and mt_max_ckpt parameters. */
  pDb->nMtMinCkpt = LSMTEST_DFLT_MT_MIN_CKPT;
  pDb->nMtMaxCkpt = LSMTEST_DFLT_MT_MAX_CKPT;

  memcpy(&pDb->env, tdb_lsm_env(), sizeof(lsm_env));
  pDb->env.pVfsCtx = (void *)pDb;
  pDb->env.xFullpath = testEnvFullpath;
  pDb->env.xOpen = testEnvOpen;
  pDb->env.xRead = testEnvRead;
  pDb->env.xWrite = testEnvWrite;
  pDb->env.xTruncate = testEnvTruncate;
  pDb->env.xSync = testEnvSync;
  pDb->env.xSectorSize = testEnvSectorSize;
  pDb->env.xRemap = testEnvRemap;
  pDb->env.xFileid = testEnvFileid;
  pDb->env.xClose = testEnvClose;
  pDb->env.xUnlink = testEnvUnlink;
  pDb->env.xLock = testEnvLock;
  pDb->env.xTestLock = testEnvTestLock;
  pDb->env.xShmBarrier = testEnvShmBarrier;
  pDb->env.xShmMap = testEnvShmMap;
  pDb->env.xShmUnmap = testEnvShmUnmap;
  pDb->env.xSleep = testEnvSleep;

  rc = lsm_new(&pDb->env, &pDb->db);
  if( rc==LSM_OK ){
    int nThread = 1;
    lsm_config_log(pDb->db, xLog, 0);
    lsm_config_work_hook(pDb->db, xWorkHook, (void *)pDb);

    rc = test_lsm_config_str(pDb, pDb->db, 0, zCfg, &nThread);
    if( rc==LSM_OK ) rc = lsm_open(pDb->db, zFilename);

    pDb->eMode = nThread;
#ifdef LSM_MUTEX_PTHREADS
    if( rc==LSM_OK && nThread>1 ){
      testLsmStartWorkers(pDb, nThread, zFilename, zCfg);
    }
#endif

    if( rc!=LSM_OK ){
      test_lsm_close((TestDb *)pDb);
      pDb = 0;
    }
  }

  *ppDb = (TestDb *)pDb;
  return rc;
}

int test_lsm_open(
  const char *zSpec, 
  const char *zFilename, 
  int bClear, 
  TestDb **ppDb
){
  return testLsmOpen(zSpec, zFilename, bClear, ppDb);
}

int test_lsm_small_open(
  const char *zSpec, 
  const char *zFile, 
  int bClear, 
  TestDb **ppDb
){
  const char *zCfg = "page_size=256 block_size=64 mmap=1024";
  return testLsmOpen(zCfg, zFile, bClear, ppDb);
}

int test_lsm_lomem_open(
  const char *zSpec, 
  const char *zFilename, 
  int bClear, 
  TestDb **ppDb
){
    /* "max_freelist=4 autocheckpoint=32" */
  const char *zCfg = 
    "page_size=256 block_size=64 autoflush=16 "
    "autocheckpoint=32"
    "mmap=0 "
  ;
  return testLsmOpen(zCfg, zFilename, bClear, ppDb);
}

int test_lsm_zip_open(
  const char *zSpec, 
  const char *zFilename, 
  int bClear, 
  TestDb **ppDb
){
  const char *zCfg = 
    "page_size=256 block_size=64 autoflush=16 "
    "autocheckpoint=32 compression=1 mmap=0 "
  ;
  return testLsmOpen(zCfg, zFilename, bClear, ppDb);
}

lsm_db *tdb_lsm(TestDb *pDb){
  if( pDb->pMethods->xClose==test_lsm_close ){
    return ((LsmDb *)pDb)->db;
  }
  return 0;
}

void tdb_lsm_enable_log(TestDb *pDb, int bEnable){
  lsm_db *db = tdb_lsm(pDb);
  if( db ){
    lsm_config_log(db, (bEnable ? xLog : 0), (void *)"client");
  }
}

void tdb_lsm_application_crash(TestDb *pDb){
  if( tdb_lsm(pDb) ){
    LsmDb *p = (LsmDb *)pDb;
    p->bCrashed = 1;
  }
}

void tdb_lsm_prepare_system_crash(TestDb *pDb){
  if( tdb_lsm(pDb) ){
    LsmDb *p = (LsmDb *)pDb;
    p->bPrepareCrash = 1;
  }
}

void tdb_lsm_system_crash(TestDb *pDb){
  if( tdb_lsm(pDb) ){
    LsmDb *p = (LsmDb *)pDb;
    p->bCrashed = 1;
    doSystemCrash(p);
  }
}

void tdb_lsm_safety(TestDb *pDb, int eMode){
  assert( eMode==LSM_SAFETY_OFF 
       || eMode==LSM_SAFETY_NORMAL 
       || eMode==LSM_SAFETY_FULL 
  );
  if( tdb_lsm(pDb) ){
    int iParam = eMode;
    LsmDb *p = (LsmDb *)pDb;
    lsm_config(p->db, LSM_CONFIG_SAFETY, &iParam);
  }
}

void tdb_lsm_prepare_sync_crash(TestDb *pDb, int iSync){
  assert( iSync>0 );
  if( tdb_lsm(pDb) ){
    LsmDb *p = (LsmDb *)pDb;
    p->nAutoCrash = iSync;
    p->bPrepareCrash = 1;
  }
}

void tdb_lsm_config_work_hook(
  TestDb *pDb, 
  void (*xWork)(lsm_db *, void *), 
  void *pWorkCtx
){
  if( tdb_lsm(pDb) ){
    LsmDb *p = (LsmDb *)pDb;
    p->xWork = xWork;
    p->pWorkCtx = pWorkCtx;
  }
}

void tdb_lsm_write_hook(
  TestDb *pDb, 
  void (*xWrite)(void *, int, lsm_i64, int, int),
  void *pWriteCtx
){
  if( tdb_lsm(pDb) ){
    LsmDb *p = (LsmDb *)pDb;
    p->xWriteHook = xWrite;
    p->pWriteCtx = pWriteCtx;
  }
}

int tdb_lsm_open(const char *zCfg, const char *zDb, int bClear, TestDb **ppDb){
  return testLsmOpen(zCfg, zDb, bClear, ppDb);
}

#ifdef LSM_MUTEX_PTHREADS

/*
** Signal worker thread iWorker that there may be work to do.
*/
static void mt_signal_worker(LsmDb *pDb, int iWorker){
  LsmWorker *p = &pDb->aWorker[iWorker];
  pthread_mutex_lock(&p->worker_mutex);
  p->bDoWork = 1;
  pthread_cond_signal(&p->worker_cond);
  pthread_mutex_unlock(&p->worker_mutex);
}

/*
** This routine is used as the main() for all worker threads.
*/
static void *worker_main(void *pArg){
  LsmWorker *p = (LsmWorker *)pArg;
  lsm_db *pWorker;                /* Connection to access db through */

  pthread_mutex_lock(&p->worker_mutex);
  while( (pWorker = p->pWorker) ){
    int rc = LSM_OK;

    /* Do some work. If an error occurs, exit. */

    pthread_mutex_unlock(&p->worker_mutex);
    if( p->eType==LSMTEST_THREAD_CKPT ){
      int nKB = 0;
      rc = lsm_info(pWorker, LSM_INFO_CHECKPOINT_SIZE, &nKB);
      if( rc==LSM_OK && nKB>=p->pDb->nMtMinCkpt ){
        rc = lsm_checkpoint(pWorker, 0);
      }
    }else{
      int nWrite;
      do {

        if( p->eType==LSMTEST_THREAD_WORKER ){
          waitOnCheckpointer(p->pDb, pWorker);
        }

        nWrite = 0;
        rc = lsm_work(pWorker, 0, 256, &nWrite);

        if( p->eType==LSMTEST_THREAD_WORKER && nWrite ){
          mt_signal_worker(p->pDb, 1);
        }
      }while( nWrite && p->pWorker );
    }
    pthread_mutex_lock(&p->worker_mutex);

    if( rc!=LSM_OK && rc!=LSM_BUSY ){
      p->worker_rc = rc;
      break;
    }

    /* The thread will wake up when it is signaled either because another
    ** thread has created some work for this one or because the connection
    ** is being closed.  */
    if( p->pWorker && p->bDoWork==0 ){
      pthread_cond_wait(&p->worker_cond, &p->worker_mutex);
    }
    p->bDoWork = 0;
  }
  pthread_mutex_unlock(&p->worker_mutex);
  
  return 0;
}


static void mt_stop_worker(LsmDb *pDb, int iWorker){
  LsmWorker *p = &pDb->aWorker[iWorker];
  if( p->pWorker ){
    void *pDummy;
    lsm_db *pWorker;

    /* Signal the worker to stop */
    pthread_mutex_lock(&p->worker_mutex);
    pWorker = p->pWorker;
    p->pWorker = 0;
    pthread_cond_signal(&p->worker_cond);
    pthread_mutex_unlock(&p->worker_mutex);

    /* Join the worker thread. */
    pthread_join(p->worker_thread, &pDummy);

    /* Free resources allocated in mt_start_worker() */
    pthread_cond_destroy(&p->worker_cond);
    pthread_mutex_destroy(&p->worker_mutex);
    lsm_close(pWorker);
  }
}

static void mt_shutdown(LsmDb *pDb){
  int i;
  for(i=0; i<pDb->nWorker; i++){
    mt_stop_worker(pDb, i);
  }
}

/*
** This callback is invoked by LSM when the client database writes to
** the database file (i.e. to flush the contents of the in-memory tree).
** This implies there may be work to do on the database, so signal
** the worker threads.
*/
static void mt_client_work_hook(lsm_db *db, void *pArg){
  LsmDb *pDb = (LsmDb *)pArg;     /* LsmDb database handle */

  /* Invoke the user level work-hook, if any. */
  if( pDb->xWork ) pDb->xWork(db, pDb->pWorkCtx);

  /* Wake up worker thread 0. */
  mt_signal_worker(pDb, 0);
}

static void mt_worker_work_hook(lsm_db *db, void *pArg){
  LsmDb *pDb = (LsmDb *)pArg;     /* LsmDb database handle */

  /* Invoke the user level work-hook, if any. */
  if( pDb->xWork ) pDb->xWork(db, pDb->pWorkCtx);
}

/*
** Launch worker thread iWorker for database connection pDb.
*/
static int mt_start_worker(
  LsmDb *pDb,                     /* Main database structure */
  int iWorker,                    /* Worker number to start */
  const char *zFilename,          /* File name of database to open */
  const char *zCfg,               /* Connection configuration string */
  int eType                       /* Type of worker thread */
){
  int rc = 0;                     /* Return code */
  LsmWorker *p;                   /* Object to initialize */

  assert( iWorker<pDb->nWorker );
  assert( eType==LSMTEST_THREAD_CKPT 
       || eType==LSMTEST_THREAD_WORKER 
       || eType==LSMTEST_THREAD_WORKER_AC 
  );

  p = &pDb->aWorker[iWorker];
  p->eType = eType;
  p->pDb = pDb;

  /* Open the worker connection */
  if( rc==0 ) rc = lsm_new(&pDb->env, &p->pWorker);
  if( zCfg ){
    test_lsm_config_str(pDb, p->pWorker, 1, zCfg, 0);
  }
  if( rc==0 ) rc = lsm_open(p->pWorker, zFilename);
  lsm_config_log(p->pWorker, xLog, (void *)"worker");

  /* Configure the work-hook */
  if( rc==0 ){
    lsm_config_work_hook(p->pWorker, mt_worker_work_hook, (void *)pDb);
  }

  if( eType==LSMTEST_THREAD_WORKER ){
    test_lsm_config_str(0, p->pWorker, 1, "autocheckpoint=0", 0);
  }

  /* Kick off the worker thread. */
  if( rc==0 ) rc = pthread_cond_init(&p->worker_cond, 0);
  if( rc==0 ) rc = pthread_mutex_init(&p->worker_mutex, 0);
  if( rc==0 ) rc = pthread_create(&p->worker_thread, 0, worker_main, (void *)p);

  return rc;
}


static int testLsmStartWorkers(
  LsmDb *pDb, int eModel, const char *zFilename, const char *zCfg
){
  int rc;

  if( eModel<1 || eModel>4 ) return 1;
  if( eModel==1 ) return 0;

  /* Configure a work-hook for the client connection. Worker 0 is signalled
  ** every time the users connection writes to the database.  */
  lsm_config_work_hook(pDb->db, mt_client_work_hook, (void *)pDb);

  /* Allocate space for two worker connections. They may not both be
  ** used, but both are allocated.  */
  pDb->aWorker = (LsmWorker *)testMalloc(sizeof(LsmWorker) * 2);
  memset(pDb->aWorker, 0, sizeof(LsmWorker) * 2);

  switch( eModel ){
    case LSMTEST_MODE_BACKGROUND_CKPT:
      pDb->nWorker = 1;
      test_lsm_config_str(0, pDb->db, 0, "autocheckpoint=0", 0);
      rc = mt_start_worker(pDb, 0, zFilename, zCfg, LSMTEST_THREAD_CKPT);
      break;

    case LSMTEST_MODE_BACKGROUND_WORK:
      pDb->nWorker = 1;
      test_lsm_config_str(0, pDb->db, 0, "autowork=0", 0);
      rc = mt_start_worker(pDb, 0, zFilename, zCfg, LSMTEST_THREAD_WORKER_AC);
      break;

    case LSMTEST_MODE_BACKGROUND_BOTH:
      pDb->nWorker = 2;
      test_lsm_config_str(0, pDb->db, 0, "autowork=0", 0);
      rc = mt_start_worker(pDb, 0, zFilename, zCfg, LSMTEST_THREAD_WORKER);
      if( rc==0 ){
        rc = mt_start_worker(pDb, 1, zFilename, zCfg, LSMTEST_THREAD_CKPT);
      }
      break;
  }

  return rc;
}


int test_lsm_mt2(
  const char *zSpec, 
  const char *zFilename, 
  int bClear, 
  TestDb **ppDb
){
  const char *zCfg = "mt_mode=2";
  return testLsmOpen(zCfg, zFilename, bClear, ppDb);
}

int test_lsm_mt3(
  const char *zSpec, 
  const char *zFilename, 
  int bClear, 
  TestDb **ppDb
){
  const char *zCfg = "mt_mode=4";
  return testLsmOpen(zCfg, zFilename, bClear, ppDb);
}

#else
static void mt_shutdown(LsmDb *pDb) { 
  unused_parameter(pDb); 
}
int test_lsm_mt(const char *zFilename, int bClear, TestDb **ppDb){
  unused_parameter(zFilename);
  unused_parameter(bClear);
  unused_parameter(ppDb);
  testPrintError("threads unavailable - recompile with LSM_MUTEX_PTHREADS\n");
  return 1;
}
#endif

Added ext/lsm1/lsm-test/lsmtest_tdb4.c.













































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982

/*
** This file contains the TestDb bt wrapper.
*/

#include "lsmtest_tdb.h"
#include "lsmtest.h"
#include <unistd.h>
#include "bt.h"

#include <pthread.h>

typedef struct BtDb BtDb;
typedef struct BtFile BtFile;

/* Background checkpointer interface (see implementations below). */
typedef struct bt_ckpter bt_ckpter;
static int bgc_attach(BtDb *pDb, const char*);
static int bgc_detach(BtDb *pDb);

/*
** Each database or log file opened by a database handle is wrapped by
** an object of the following type.
*/
struct BtFile {
  BtDb *pBt;                      /* Database handle that opened this file */
  bt_env *pVfs;                   /* Underlying VFS */
  bt_file *pFile;                 /* File handle belonging to underlying VFS */
  int nSectorSize;                /* Size of sectors in bytes */
  int nSector;                    /* Allocated size of nSector array */
  u8 **apSector;                  /* Original sector data */
};

/*
** nCrashSync:
**   If this value is non-zero, then a "crash-test" is running. If
**   nCrashSync==1, then the crash is simulated during the very next 
**   call to the xSync() VFS method (on either the db or log file).
**   If nCrashSync==2, the following call to xSync(), and so on.
**
** bCrash:
**   After a crash is simulated, this variable is set. Any subsequent
**   attempts to write to a file or modify the file system in any way 
**   fail once this is set. All the caller can do is close the connection.
**
** bFastInsert:
**   If this variable is set to true, then a BT_CONTROL_FAST_INSERT_OP
**   control is issued before each callto BtReplace() or BtCsrOpen().
*/
struct BtDb {
  TestDb base;                    /* Base class */
  bt_db *pBt;                     /* bt database handle */
  sqlite4_env *pEnv;              /* SQLite environment (for malloc/free) */
  bt_env *pVfs;                   /* Underlying VFS */
  int bFastInsert;                /* True to use fast-insert */

  /* Space for bt_fetch() results */
  u8 *aBuffer;                    /* Space to store results */
  int nBuffer;                    /* Allocated size of aBuffer[] in bytes */
  int nRef;

  /* Background checkpointer used by mt connections */
  bt_ckpter *pCkpter;

  /* Stuff used for crash test simulation */
  BtFile *apFile[2];              /* Database and log files used by pBt */
  bt_env env;                     /* Private VFS for this object */
  int nCrashSync;                 /* Number of syncs until crash (see above) */
  int bCrash;                     /* True once a crash has been simulated */
};

static int btVfsFullpath(
  sqlite4_env *pEnv, 
  bt_env *pVfs, 
  const char *z, 
  char **pzOut
){
  BtDb *pBt = (BtDb*)pVfs->pVfsCtx;
  if( pBt->bCrash ) return SQLITE4_IOERR;
  return pBt->pVfs->xFullpath(pEnv, pBt->pVfs, z, pzOut);
}

static int btVfsOpen(
  sqlite4_env *pEnv, 
  bt_env *pVfs, 
  const char *zFile, 
  int flags, bt_file **ppFile
){
  BtFile *p;
  BtDb *pBt = (BtDb*)pVfs->pVfsCtx;
  int rc;

  if( pBt->bCrash ) return SQLITE4_IOERR;

  p = (BtFile*)testMalloc(sizeof(BtFile));
  if( !p ) return SQLITE4_NOMEM;
  if( flags & BT_OPEN_DATABASE ){
    pBt->apFile[0] = p;
  }else if( flags & BT_OPEN_LOG ){
    pBt->apFile[1] = p;
  }
  if( (flags & BT_OPEN_SHARED)==0 ){
    p->pBt = pBt; 
  }
  p->pVfs = pBt->pVfs; 

  rc = pBt->pVfs->xOpen(pEnv, pVfs, zFile, flags, &p->pFile);
  if( rc!=SQLITE4_OK ){
    testFree(p);
    p = 0;
  }else{
    pBt->nRef++;
  }

  *ppFile = (bt_file*)p;
  return rc;
}

static int btVfsSize(bt_file *pFile, sqlite4_int64 *piRes){
  BtFile *p = (BtFile*)pFile;
  if( p->pBt && p->pBt->bCrash ) return SQLITE4_IOERR;
  return p->pVfs->xSize(p->pFile, piRes);
}

static int btVfsRead(bt_file *pFile, sqlite4_int64 iOff, void *pBuf, int nBuf){
  BtFile *p = (BtFile*)pFile;
  if( p->pBt && p->pBt->bCrash ) return SQLITE4_IOERR;
  return p->pVfs->xRead(p->pFile, iOff, pBuf, nBuf);
}

static int btFlushSectors(BtFile *p, int iFile){
  sqlite4_int64 iSz;
  int rc;
  int i;
  u8 *aTmp = 0;

  rc = p->pBt->pVfs->xSize(p->pFile, &iSz);
  for(i=0; rc==SQLITE4_OK && i<p->nSector; i++){
    if( p->pBt->bCrash && p->apSector[i] ){

      /* The system is simulating a crash. There are three choices for
      ** this sector:
      **
      **   1) Leave it as it is (simulating a successful write),
      **   2) Restore the original data (simulating a lost write),
      **   3) Populate the disk sector with garbage data.
      */
      sqlite4_int64 iSOff = p->nSectorSize*i;
      int nWrite = MIN(p->nSectorSize, iSz - iSOff);

      if( nWrite ){
        u8 *aWrite = 0;
        int iOpt = (testPrngValue(i) % 3) + 1;
        if( iOpt==1 ){
          aWrite = p->apSector[i];
        }else if( iOpt==3 ){
          if( aTmp==0 ) aTmp = testMalloc(p->nSectorSize);
          aWrite = aTmp;
          testPrngArray(i*13, (u32*)aWrite, nWrite/sizeof(u32));
        }

#if 0
fprintf(stderr, "handle sector %d of %s with %s\n", i, 
    iFile==0 ? "db" : "log",
    iOpt==1 ? "rollback" : iOpt==2 ? "write" : "omit"
);
fflush(stderr);
#endif

        if( aWrite ){
          rc = p->pBt->pVfs->xWrite(p->pFile, iSOff, aWrite, nWrite);
        }
      }
    }
    testFree(p->apSector[i]);
    p->apSector[i] = 0;
  }

  testFree(aTmp);
  return rc;
}

static int btSaveSectors(BtFile *p, sqlite4_int64 iOff, int nBuf){
  int rc;
  sqlite4_int64 iSz;              /* Size of file on disk */
  int iFirst;                     /* First sector affected */
  int iSector;                    /* Current sector */
  int iLast;                      /* Last sector affected */

  if( p->nSectorSize==0 ){
    p->nSectorSize = p->pBt->pVfs->xSectorSize(p->pFile);
    if( p->nSectorSize<512 ) p->nSectorSize = 512;
  }
  iLast = (iOff+nBuf-1) / p->nSectorSize;
  iFirst = iOff / p->nSectorSize;

  rc = p->pBt->pVfs->xSize(p->pFile, &iSz);
  for(iSector=iFirst; rc==SQLITE4_OK && iSector<=iLast; iSector++){
    int nRead;
    sqlite4_int64 iSOff = iSector * p->nSectorSize;
    u8 *aBuf = testMalloc(p->nSectorSize);
    nRead = MIN(p->nSectorSize, (iSz - iSOff));
    if( nRead>0 ){
      rc = p->pBt->pVfs->xRead(p->pFile, iSOff, aBuf, nRead);
    }

    while( rc==SQLITE4_OK && iSector>=p->nSector ){
      int nNew = p->nSector + 32;
      u8 **apNew = (u8**)testMalloc(nNew * sizeof(u8*));
      memcpy(apNew, p->apSector, p->nSector*sizeof(u8*));
      testFree(p->apSector);
      p->apSector = apNew;
      p->nSector = nNew;
    }

    p->apSector[iSector] = aBuf;
  }

  return rc;
}

static int btVfsWrite(bt_file *pFile, sqlite4_int64 iOff, void *pBuf, int nBuf){
  BtFile *p = (BtFile*)pFile;
  if( p->pBt && p->pBt->bCrash ) return SQLITE4_IOERR;
  if( p->pBt && p->pBt->nCrashSync ){
    btSaveSectors(p, iOff, nBuf);
  }
  return p->pVfs->xWrite(p->pFile, iOff, pBuf, nBuf);
}

static int btVfsTruncate(bt_file *pFile, sqlite4_int64 iOff){
  BtFile *p = (BtFile*)pFile;
  if( p->pBt && p->pBt->bCrash ) return SQLITE4_IOERR;
  return p->pVfs->xTruncate(p->pFile, iOff);
}

static int btVfsSync(bt_file *pFile){
  int rc = SQLITE4_OK;
  BtFile *p = (BtFile*)pFile;
  BtDb *pBt = p->pBt;

  if( pBt ){
    if( pBt->bCrash ) return SQLITE4_IOERR;
    if( pBt->nCrashSync ){
      pBt->nCrashSync--;
      pBt->bCrash = (pBt->nCrashSync==0);
      if( pBt->bCrash ){
        btFlushSectors(pBt->apFile[0], 0);
        btFlushSectors(pBt->apFile[1], 1);
        rc = SQLITE4_IOERR;
      }else{
        btFlushSectors(p, 0);
      }
    }
  }

  if( rc==SQLITE4_OK ){
    rc = p->pVfs->xSync(p->pFile);
  }
  return rc;
}

static int btVfsSectorSize(bt_file *pFile){
  BtFile *p = (BtFile*)pFile;
  return p->pVfs->xSectorSize(p->pFile);
}

static void btDeref(BtDb *p){
  p->nRef--;
  assert( p->nRef>=0 );
  if( p->nRef<=0 ) testFree(p);
}

static int btVfsClose(bt_file *pFile){
  BtFile *p = (BtFile*)pFile;
  BtDb *pBt = p->pBt;
  int rc;
  if( pBt ){
    btFlushSectors(p, 0);
    if( p==pBt->apFile[0] ) pBt->apFile[0] = 0;
    if( p==pBt->apFile[1] ) pBt->apFile[1] = 0;
  }
  testFree(p->apSector);
  rc = p->pVfs->xClose(p->pFile);
#if 0
  btDeref(p->pBt);
#endif
  testFree(p);
  return rc;
}

static int btVfsUnlink(sqlite4_env *pEnv, bt_env *pVfs, const char *zFile){
  BtDb *pBt = (BtDb*)pVfs->pVfsCtx;
  if( pBt->bCrash ) return SQLITE4_IOERR;
  return pBt->pVfs->xUnlink(pEnv, pBt->pVfs, zFile);
}

static int btVfsLock(bt_file *pFile, int iLock, int eType){
  BtFile *p = (BtFile*)pFile;
  if( p->pBt && p->pBt->bCrash ) return SQLITE4_IOERR;
  return p->pVfs->xLock(p->pFile, iLock, eType);
}

static int btVfsTestLock(bt_file *pFile, int iLock, int nLock, int eType){
  BtFile *p = (BtFile*)pFile;
  if( p->pBt && p->pBt->bCrash ) return SQLITE4_IOERR;
  return p->pVfs->xTestLock(p->pFile, iLock, nLock, eType);
}

static int btVfsShmMap(bt_file *pFile, int iChunk, int sz, void **ppOut){
  BtFile *p = (BtFile*)pFile;
  if( p->pBt && p->pBt->bCrash ) return SQLITE4_IOERR;
  return p->pVfs->xShmMap(p->pFile, iChunk, sz, ppOut);
}

static void btVfsShmBarrier(bt_file *pFile){
  BtFile *p = (BtFile*)pFile;
  return p->pVfs->xShmBarrier(p->pFile);
}

static int btVfsShmUnmap(bt_file *pFile, int bDelete){
  BtFile *p = (BtFile*)pFile;
  if( p->pBt && p->pBt->bCrash ) return SQLITE4_IOERR;
  return p->pVfs->xShmUnmap(p->pFile, bDelete);
}

static int bt_close(TestDb *pTestDb){
  BtDb *p = (BtDb*)pTestDb;
  int rc = sqlite4BtClose(p->pBt);
  free(p->aBuffer);
  if( p->apFile[0] ) p->apFile[0]->pBt = 0;
  if( p->apFile[1] ) p->apFile[1]->pBt = 0;
  bgc_detach(p);
  testFree(p);
  return rc;
}

static int btMinTransaction(BtDb *p, int iMin, int *piLevel){
  int iLevel;
  int rc = SQLITE4_OK;

  iLevel = sqlite4BtTransactionLevel(p->pBt);
  if( iLevel<iMin ){ 
    rc = sqlite4BtBegin(p->pBt, iMin); 
    *piLevel = iLevel;
  }else{
    *piLevel = -1;
  }

  return rc;
}
static int btRestoreTransaction(BtDb *p, int iLevel, int rcin){
  int rc = rcin;
  if( iLevel>=0 ){
    if( rc==SQLITE4_OK ){
      rc = sqlite4BtCommit(p->pBt, iLevel);
    }else{
      sqlite4BtRollback(p->pBt, iLevel);
    }
    assert( iLevel==sqlite4BtTransactionLevel(p->pBt) );
  }
  return rc;
}

static int bt_write(TestDb *pTestDb, void *pK, int nK, void *pV, int nV){
  BtDb *p = (BtDb*)pTestDb;
  int iLevel;
  int rc;

  rc = btMinTransaction(p, 2, &iLevel);
  if( rc==SQLITE4_OK ){
    if( p->bFastInsert ) sqlite4BtControl(p->pBt, BT_CONTROL_FAST_INSERT_OP, 0);
    rc = sqlite4BtReplace(p->pBt, pK, nK, pV, nV);
    rc = btRestoreTransaction(p, iLevel, rc);
  }
  return rc;
}

static int bt_delete(TestDb *pTestDb, void *pK, int nK){
  return bt_write(pTestDb, pK, nK, 0, -1);
}

static int bt_delete_range(
  TestDb *pTestDb, 
  void *pKey1, int nKey1,
  void *pKey2, int nKey2
){
  BtDb *p = (BtDb*)pTestDb;
  bt_cursor *pCsr = 0;
  int rc = SQLITE4_OK;
  int iLevel;

  rc = btMinTransaction(p, 2, &iLevel);
  if( rc==SQLITE4_OK ){
    if( p->bFastInsert ) sqlite4BtControl(p->pBt, BT_CONTROL_FAST_INSERT_OP, 0);
    rc = sqlite4BtCsrOpen(p->pBt, 0, &pCsr);
  }
  while( rc==SQLITE4_OK ){
    const void *pK;
    int n;
    int nCmp;
    int res;

    rc = sqlite4BtCsrSeek(pCsr, pKey1, nKey1, BT_SEEK_GE);
    if( rc==SQLITE4_INEXACT ) rc = SQLITE4_OK;
    if( rc!=SQLITE4_OK ) break;

    rc = sqlite4BtCsrKey(pCsr, &pK, &n);
    if( rc!=SQLITE4_OK ) break;

    nCmp = MIN(n, nKey1);
    res = memcmp(pKey1, pK, nCmp);
    assert( res<0 || (res==0 && nKey1<=n) );
    if( res==0 && nKey1==n ){
      rc = sqlite4BtCsrNext(pCsr);
      if( rc!=SQLITE4_OK ) break;
      rc = sqlite4BtCsrKey(pCsr, &pK, &n);
      if( rc!=SQLITE4_OK ) break;
    }

    nCmp = MIN(n, nKey2);
    res = memcmp(pKey2, pK, nCmp);
    if( res<0 || (res==0 && nKey2<=n) ) break;
    
    rc = sqlite4BtDelete(pCsr);
  }
  if( rc==SQLITE4_NOTFOUND ) rc = SQLITE4_OK;

  sqlite4BtCsrClose(pCsr);

  rc = btRestoreTransaction(p, iLevel, rc);
  return rc;
}

static int bt_fetch(
  TestDb *pTestDb, 
  void *pK, int nK, 
  void **ppVal, int *pnVal
){
  BtDb *p = (BtDb*)pTestDb;
  bt_cursor *pCsr = 0;
  int iLevel;
  int rc = SQLITE4_OK;

  iLevel = sqlite4BtTransactionLevel(p->pBt);
  if( iLevel==0 ){ 
    rc = sqlite4BtBegin(p->pBt, 1); 
    if( rc!=SQLITE4_OK ) return rc;
  }

  if( p->bFastInsert ) sqlite4BtControl(p->pBt, BT_CONTROL_FAST_INSERT_OP, 0);
  rc = sqlite4BtCsrOpen(p->pBt, 0, &pCsr);
  if( rc==SQLITE4_OK ){
    rc = sqlite4BtCsrSeek(pCsr, pK, nK, BT_SEEK_EQ);
    if( rc==SQLITE4_OK ){
      const void *pV = 0;
      int nV = 0;
      rc = sqlite4BtCsrData(pCsr, 0, -1, &pV, &nV);
      if( rc==SQLITE4_OK ){
        if( nV>p->nBuffer ){
          free(p->aBuffer);
          p->aBuffer = (u8*)malloc(nV*2);
          p->nBuffer = nV*2;
        }
        memcpy(p->aBuffer, pV, nV);
        *pnVal = nV;
        *ppVal = (void*)(p->aBuffer);
      }

    }else if( rc==SQLITE4_INEXACT || rc==SQLITE4_NOTFOUND ){
      *ppVal = 0;
      *pnVal = -1;
      rc = SQLITE4_OK;
    }
    sqlite4BtCsrClose(pCsr);
  }

  if( iLevel==0 ) sqlite4BtCommit(p->pBt, 0); 
  return rc;
}

static int bt_scan(
  TestDb *pTestDb,
  void *pCtx,
  int bReverse,
  void *pFirst, int nFirst,
  void *pLast, int nLast,
  void (*xCallback)(void *, void *, int , void *, int)
){
  BtDb *p = (BtDb*)pTestDb;
  bt_cursor *pCsr = 0;
  int rc;
  int iLevel;

  rc = btMinTransaction(p, 1, &iLevel);

  if( rc==SQLITE4_OK ){
    if( p->bFastInsert ) sqlite4BtControl(p->pBt, BT_CONTROL_FAST_INSERT_OP, 0);
    rc = sqlite4BtCsrOpen(p->pBt, 0, &pCsr);
  }
  if( rc==SQLITE4_OK ){
    if( bReverse ){
      if( pLast ){
        rc = sqlite4BtCsrSeek(pCsr, pLast, nLast, BT_SEEK_LE);
      }else{
        rc = sqlite4BtCsrLast(pCsr);
      }
    }else{
      rc = sqlite4BtCsrSeek(pCsr, pFirst, nFirst, BT_SEEK_GE);
    }
    if( rc==SQLITE4_INEXACT ) rc = SQLITE4_OK;

    while( rc==SQLITE4_OK ){
      const void *pK = 0; int nK = 0;
      const void *pV = 0; int nV = 0;

      rc = sqlite4BtCsrKey(pCsr, &pK, &nK);
      if( rc==SQLITE4_OK ){
        rc = sqlite4BtCsrData(pCsr, 0, -1, &pV, &nV);
      }

      if( rc!=SQLITE4_OK ) break;
      if( bReverse ){
        if( pFirst ){
          int res;
          int nCmp = MIN(nK, nFirst);
          res = memcmp(pFirst, pK, nCmp);
          if( res>0 || (res==0 && nK<nFirst) ) break;
        }
      }else{
        if( pLast ){
          int res;
          int nCmp = MIN(nK, nLast);
          res = memcmp(pLast, pK, nCmp);
          if( res<0 || (res==0 && nK>nLast) ) break;
        }
      }

      xCallback(pCtx, (void*)pK, nK, (void*)pV, nV);
      if( bReverse ){
        rc = sqlite4BtCsrPrev(pCsr);
      }else{
        rc = sqlite4BtCsrNext(pCsr);
      }
    }
    if( rc==SQLITE4_NOTFOUND ) rc = SQLITE4_OK;

    sqlite4BtCsrClose(pCsr);
  }

  rc = btRestoreTransaction(p, iLevel, rc);
  return rc;
}

static int bt_begin(TestDb *pTestDb, int iLvl){
  BtDb *p = (BtDb*)pTestDb;
  int rc = sqlite4BtBegin(p->pBt, iLvl);
  return rc;
}

static int bt_commit(TestDb *pTestDb, int iLvl){
  BtDb *p = (BtDb*)pTestDb;
  int rc = sqlite4BtCommit(p->pBt, iLvl);
  return rc;
}

static int bt_rollback(TestDb *pTestDb, int iLvl){
  BtDb *p = (BtDb*)pTestDb;
  int rc = sqlite4BtRollback(p->pBt, iLvl);
  return rc;
}

static int testParseOption(
  const char **pzIn,              /* IN/OUT: pointer to next option */
  const char **pzOpt,             /* OUT: nul-terminated option name */
  const char **pzArg,             /* OUT: nul-terminated option argument */
  char *pSpace                    /* Temporary space for output params */
){
  const char *p = *pzIn;
  const char *pStart;
  int n;

  char *pOut = pSpace;

  while( *p==' ' ) p++;
  pStart = p;
  while( *p && *p!='=' ) p++;
  if( *p==0 ) return 1;

  n = (p - pStart);
  memcpy(pOut, pStart, n);
  *pzOpt = pOut;
  pOut += n;
  *pOut++ = '\0';

  p++;
  pStart = p;
  while( *p && *p!=' ' ) p++;
  n = (p - pStart);

  memcpy(pOut, pStart, n);
  *pzArg = pOut;
  pOut += n;
  *pOut++ = '\0';

  *pzIn = p;
  return 0;
}

static int testParseInt(const char *z, int *piVal){
  int i = 0;
  const char *p = z;

  while( *p>='0' && *p<='9' ){
    i = i*10 + (*p - '0');
    p++;
  }
  if( *p=='K' || *p=='k' ){
    i = i * 1024;
    p++;
  }else if( *p=='M' || *p=='m' ){
    i = i * 1024 * 1024;
    p++;
  }

  if( *p ) return SQLITE4_ERROR;
  *piVal = i;
  return SQLITE4_OK;
}

static int testBtConfigure(BtDb *pDb, const char *zCfg, int *pbMt){
  int rc = SQLITE4_OK;

  if( zCfg ){
    struct CfgParam {
      const char *zParam;
      int eParam;
    } aParam[] = {
      { "safety",         BT_CONTROL_SAFETY },
      { "autockpt",       BT_CONTROL_AUTOCKPT },
      { "multiproc",      BT_CONTROL_MULTIPROC },
      { "blksz",          BT_CONTROL_BLKSZ },
      { "pagesz",         BT_CONTROL_PAGESZ },
      { "mt",             -1 },
      { "fastinsert",     -2 },
      { 0, 0 }
    };
    const char *z = zCfg;
    int n = strlen(z);
    char *aSpace;
    const char *zOpt;
    const char *zArg;

    aSpace = (char*)testMalloc(n+2);
    while( rc==SQLITE4_OK && 0==testParseOption(&z, &zOpt, &zArg, aSpace) ){
      int i;
      int iVal;
      rc = testArgSelect(aParam, "param", zOpt, &i);
      if( rc!=SQLITE4_OK ) break;

      rc = testParseInt(zArg, &iVal);
      if( rc!=SQLITE4_OK ) break;

      switch( aParam[i].eParam ){
        case -1:
          *pbMt = iVal;
          break;
        case -2:
          pDb->bFastInsert = 1;
          break;
        default:
          rc = sqlite4BtControl(pDb->pBt, aParam[i].eParam, (void*)&iVal);
          break;
      }
    }
    testFree(aSpace);
  }

  return rc;
}


int test_bt_open(
  const char *zSpec, 
  const char *zFilename, 
  int bClear, 
  TestDb **ppDb
){

  static const DatabaseMethods SqlMethods = {
    bt_close,
    bt_write,
    bt_delete,
    bt_delete_range,
    bt_fetch,
    bt_scan,
    bt_begin,
    bt_commit,
    bt_rollback
  };
  BtDb *p = 0;
  bt_db *pBt = 0;
  int rc;
  sqlite4_env *pEnv = sqlite4_env_default();

  if( bClear && zFilename && zFilename[0] ){
    char *zLog = sqlite3_mprintf("%s-wal", zFilename);
    unlink(zFilename);
    unlink(zLog);
    sqlite3_free(zLog);
  }
  
  rc = sqlite4BtNew(pEnv, 0, &pBt);
  if( rc==SQLITE4_OK ){
    int mt = 0;                   /* True for multi-threaded connection */

    p = (BtDb*)testMalloc(sizeof(BtDb));
    p->base.pMethods = &SqlMethods;
    p->pBt = pBt;
    p->pEnv = pEnv;
    p->nRef = 1;

    p->env.pVfsCtx = (void*)p;
    p->env.xFullpath = btVfsFullpath;
    p->env.xOpen = btVfsOpen;
    p->env.xSize = btVfsSize;
    p->env.xRead = btVfsRead;
    p->env.xWrite = btVfsWrite;
    p->env.xTruncate = btVfsTruncate;
    p->env.xSync = btVfsSync;
    p->env.xSectorSize = btVfsSectorSize;
    p->env.xClose = btVfsClose;
    p->env.xUnlink = btVfsUnlink;
    p->env.xLock = btVfsLock;
    p->env.xTestLock = btVfsTestLock;
    p->env.xShmMap = btVfsShmMap;
    p->env.xShmBarrier = btVfsShmBarrier;
    p->env.xShmUnmap = btVfsShmUnmap;

    sqlite4BtControl(pBt, BT_CONTROL_GETVFS, (void*)&p->pVfs);
    sqlite4BtControl(pBt, BT_CONTROL_SETVFS, (void*)&p->env);

    rc = testBtConfigure(p, zSpec, &mt);
    if( rc==SQLITE4_OK ){
      rc = sqlite4BtOpen(pBt, zFilename);
    }

    if( rc==SQLITE4_OK && mt ){
      int nAuto = 0;
      rc = bgc_attach(p, zSpec);
      sqlite4BtControl(pBt, BT_CONTROL_AUTOCKPT, (void*)&nAuto);
    }
  }

  if( rc!=SQLITE4_OK && p ){
    bt_close(&p->base);
  }

  *ppDb = &p->base;
  return rc;
}

int test_fbt_open(
  const char *zSpec, 
  const char *zFilename, 
  int bClear, 
  TestDb **ppDb
){
  return test_bt_open("fast=1", zFilename, bClear, ppDb);
}

int test_fbts_open(
  const char *zSpec, 
  const char *zFilename, 
  int bClear, 
  TestDb **ppDb
){
  return test_bt_open("fast=1 blksz=32K pagesz=512", zFilename, bClear, ppDb);
}


void tdb_bt_prepare_sync_crash(TestDb *pTestDb, int iSync){
  BtDb *p = (BtDb*)pTestDb;
  assert( pTestDb->pMethods->xClose==bt_close );
  assert( p->bCrash==0 );
  p->nCrashSync = iSync;
}

bt_db *tdb_bt(TestDb *pDb){
  if( pDb->pMethods->xClose==bt_close ){
    return ((BtDb *)pDb)->pBt;
  }
  return 0;
}

/*************************************************************************
** Beginning of code for background checkpointer.
*/

struct bt_ckpter {
  sqlite4_buffer file;            /* File name */
  sqlite4_buffer spec;            /* Options */
  int nLogsize;                   /* Minimum log size to checkpoint */
  int nRef;                       /* Number of clients */

  int bDoWork;                    /* Set by client threads */
  pthread_t ckpter_thread;        /* Checkpointer thread */
  pthread_cond_t ckpter_cond;     /* Condition var the ckpter waits on */
  pthread_mutex_t ckpter_mutex;   /* Mutex used with ckpter_cond */

  bt_ckpter *pNext;               /* Next object in list at gBgc.pCkpter */
};

static struct GlobalBackgroundCheckpointer {
  bt_ckpter *pCkpter;             /* Linked list of checkpointers */
} gBgc;

static void *bgc_main(void *pArg){
  BtDb *pDb = 0;
  int rc;
  int mt;
  bt_ckpter *pCkpter = (bt_ckpter*)pArg;

  rc = test_bt_open("", (char*)pCkpter->file.p, 0, (TestDb**)&pDb);
  assert( rc==SQLITE4_OK );
  rc = testBtConfigure(pDb, (char*)pCkpter->spec.p, &mt);

  while( pCkpter->nRef>0 ){
    bt_db *db = pDb->pBt;
    int nLog = 0;

    sqlite4BtBegin(db, 1);
    sqlite4BtCommit(db, 0);
    sqlite4BtControl(db, BT_CONTROL_LOGSIZE, (void*)&nLog);

    if( nLog>=pCkpter->nLogsize ){
      int rc;
      bt_checkpoint ckpt;
      memset(&ckpt, 0, sizeof(bt_checkpoint));
      ckpt.nFrameBuffer = nLog/2;
      rc = sqlite4BtControl(db, BT_CONTROL_CHECKPOINT, (void*)&ckpt);
      assert( rc==SQLITE4_OK );
      sqlite4BtControl(db, BT_CONTROL_LOGSIZE, (void*)&nLog);
    }

    /* The thread will wake up when it is signaled either because another
    ** thread has created some work for this one or because the connection
    ** is being closed.  */
    pthread_mutex_lock(&pCkpter->ckpter_mutex);
    if( pCkpter->bDoWork==0 ){
      pthread_cond_wait(&pCkpter->ckpter_cond, &pCkpter->ckpter_mutex);
    }
    pCkpter->bDoWork = 0;
    pthread_mutex_unlock(&pCkpter->ckpter_mutex);
  }

  if( pDb ) bt_close((TestDb*)pDb);
  return 0;
}

static void bgc_logsize_cb(void *pCtx, int nLogsize){
  bt_ckpter *p = (bt_ckpter*)pCtx;
  if( nLogsize>=p->nLogsize ){
    pthread_mutex_lock(&p->ckpter_mutex);
    p->bDoWork = 1;
    pthread_cond_signal(&p->ckpter_cond);
    pthread_mutex_unlock(&p->ckpter_mutex);
  }
}

static int bgc_attach(BtDb *pDb, const char *zSpec){
  int rc;
  int n;
  bt_info info;
  bt_ckpter *pCkpter;

  /* Figure out the full path to the database opened by handle pDb. */
  info.eType = BT_INFO_FILENAME;
  info.pgno = 0;
  sqlite4_buffer_init(&info.output, 0);
  rc = sqlite4BtControl(pDb->pBt, BT_CONTROL_INFO, (void*)&info);
  if( rc!=SQLITE4_OK ) return rc;

  sqlite4_mutex_enter(sqlite4_mutex_alloc(pDb->pEnv, SQLITE4_MUTEX_STATIC_KV));

  /* Search for an existing bt_ckpter object. */
  n = info.output.n;
  for(pCkpter=gBgc.pCkpter; pCkpter; pCkpter=pCkpter->pNext){
    if( n==pCkpter->file.n && 0==memcmp(info.output.p, pCkpter->file.p, n) ){
      break;
    }
  }

  /* Failed to find a suitable checkpointer. Create a new one. */
  if( pCkpter==0 ){
    bt_logsizecb cb;

    pCkpter = testMalloc(sizeof(bt_ckpter));
    memcpy(&pCkpter->file, &info.output, sizeof(sqlite4_buffer));
    info.output.p = 0;
    pCkpter->pNext = gBgc.pCkpter;
    pCkpter->nLogsize = 1000;
    gBgc.pCkpter = pCkpter;
    pCkpter->nRef = 1;

    sqlite4_buffer_init(&pCkpter->spec, 0);
    rc = sqlite4_buffer_set(&pCkpter->spec, zSpec, strlen(zSpec)+1);
    assert( rc==SQLITE4_OK );

    /* Kick off the checkpointer thread. */
    if( rc==0 ) rc = pthread_cond_init(&pCkpter->ckpter_cond, 0);
    if( rc==0 ) rc = pthread_mutex_init(&pCkpter->ckpter_mutex, 0);
    if( rc==0 ){
      rc = pthread_create(&pCkpter->ckpter_thread, 0, bgc_main, (void*)pCkpter);
    }
    assert( rc==0 ); /* todo: Fix this */

    /* Set up the logsize callback for the client thread */
    cb.pCtx = (void*)pCkpter;
    cb.xLogsize = bgc_logsize_cb;
    sqlite4BtControl(pDb->pBt, BT_CONTROL_LOGSIZECB, (void*)&cb);
  }else{
    pCkpter->nRef++;
  }

  /* Assuming a checkpointer was encountered or effected, attach the 
  ** connection to it.  */
  if( pCkpter ){
    pDb->pCkpter = pCkpter;
  }

  sqlite4_mutex_leave(sqlite4_mutex_alloc(pDb->pEnv, SQLITE4_MUTEX_STATIC_KV));
  sqlite4_buffer_clear(&info.output);
  return rc;
}

static int bgc_detach(BtDb *pDb){
  int rc = SQLITE4_OK;
  bt_ckpter *pCkpter = pDb->pCkpter;
  if( pCkpter ){
    int bShutdown = 0;            /* True if this is the last reference */

    sqlite4_mutex_enter(sqlite4_mutex_alloc(pDb->pEnv,SQLITE4_MUTEX_STATIC_KV));
    pCkpter->nRef--;
    if( pCkpter->nRef==0 ){
      bt_ckpter **pp;

      *pp = pCkpter->pNext;
      for(pp=&gBgc.pCkpter; *pp!=pCkpter; pp=&((*pp)->pNext));
      bShutdown = 1;
    }
    sqlite4_mutex_leave(sqlite4_mutex_alloc(pDb->pEnv,SQLITE4_MUTEX_STATIC_KV));

    if( bShutdown ){
      void *pDummy;

      /* Signal the checkpointer thread. */
      pthread_mutex_lock(&pCkpter->ckpter_mutex);
      pCkpter->bDoWork = 1;
      pthread_cond_signal(&pCkpter->ckpter_cond);
      pthread_mutex_unlock(&pCkpter->ckpter_mutex);

      /* Join the checkpointer thread. */
      pthread_join(pCkpter->ckpter_thread, &pDummy);
      pthread_cond_destroy(&pCkpter->ckpter_cond);
      pthread_mutex_destroy(&pCkpter->ckpter_mutex);

      sqlite4_buffer_clear(&pCkpter->file);
      sqlite4_buffer_clear(&pCkpter->spec);
      testFree(pCkpter);
    }

    pDb->pCkpter = 0;
  }
  return rc;
}

/*
** End of background checkpointer.
*************************************************************************/


Added ext/lsm1/lsm-test/lsmtest_util.c.































































































































































































































































































































































































































































>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223

#include "lsmtest.h"
#include <stdarg.h>
#include <stdio.h>
#include <string.h>
#ifndef _WIN32
# include <sys/time.h>
#endif

/*
** Global variables used within this module.
*/
static struct TestutilGlobal {
  char **argv;
  int argc;
} g = {0, 0};

static struct TestutilRnd {
  unsigned int aRand1[2048];          /* Bits 0..10 */
  unsigned int aRand2[2048];          /* Bits 11..21 */
  unsigned int aRand3[1024];          /* Bits 22..31 */
} r;

/*************************************************************************
** The following block is a copy of the implementation of SQLite function
** sqlite3_randomness. This version has two important differences:
**
**   1. It always uses the same seed. So the sequence of random data output
**      is the same for every run of the program.
**
**   2. It is not threadsafe.
*/
static struct sqlite3PrngType {
  unsigned char i, j;             /* State variables */
  unsigned char s[256];           /* State variables */
} sqlite3Prng = {
    0xAF, 0x28,
  {
    0x71, 0xF5, 0xB4, 0x6E, 0x80, 0xAB, 0x1D, 0xB8, 
    0xFB, 0xB7, 0x49, 0xBF, 0xFF, 0x72, 0x2D, 0x14, 
    0x79, 0x09, 0xE3, 0x78, 0x76, 0xB0, 0x2C, 0x0A, 
    0x8E, 0x23, 0xEE, 0xDF, 0xE0, 0x9A, 0x2F, 0x67, 
    0xE1, 0xBE, 0x0E, 0xA7, 0x08, 0x97, 0xEB, 0x77, 
    0x78, 0xBA, 0x9D, 0xCA, 0x49, 0x4C, 0x60, 0x9A, 
    0xF6, 0xBD, 0xDA, 0x7F, 0xBC, 0x48, 0x58, 0x52, 
    0xE5, 0xCD, 0x83, 0x72, 0x23, 0x52, 0xFF, 0x6D, 
    0xEF, 0x0F, 0x82, 0x29, 0xA0, 0x83, 0x3F, 0x7D, 
    0xA4, 0x88, 0x31, 0xE7, 0x88, 0x92, 0x3B, 0x9B, 
    0x3B, 0x2C, 0xC2, 0x4C, 0x71, 0xA2, 0xB0, 0xEA, 
    0x36, 0xD0, 0x00, 0xF1, 0xD3, 0x39, 0x17, 0x5D, 
    0x2A, 0x7A, 0xE4, 0xAD, 0xE1, 0x64, 0xCE, 0x0F, 
    0x9C, 0xD9, 0xF5, 0xED, 0xB0, 0x22, 0x5E, 0x62, 
    0x97, 0x02, 0xA3, 0x8C, 0x67, 0x80, 0xFC, 0x88, 
    0x14, 0x0B, 0x15, 0x10, 0x0F, 0xC7, 0x40, 0xD4, 
    0xF1, 0xF9, 0x0E, 0x1A, 0xCE, 0xB9, 0x1E, 0xA1, 
    0x72, 0x8E, 0xD7, 0x78, 0x39, 0xCD, 0xF4, 0x5D, 
    0x2A, 0x59, 0x26, 0x34, 0xF2, 0x73, 0x0B, 0xA0, 
    0x02, 0x51, 0x2C, 0x03, 0xA3, 0xA7, 0x43, 0x13, 
    0xE8, 0x98, 0x2B, 0xD2, 0x53, 0xF8, 0xEE, 0x91, 
    0x7D, 0xE7, 0xE3, 0xDA, 0xD5, 0xBB, 0xC0, 0x92, 
    0x9D, 0x98, 0x01, 0x2C, 0xF9, 0xB9, 0xA0, 0xEB, 
    0xCF, 0x32, 0xFA, 0x01, 0x49, 0xA5, 0x1D, 0x9A, 
    0x76, 0x86, 0x3F, 0x40, 0xD4, 0x89, 0x8F, 0x9C, 
    0xE2, 0xE3, 0x11, 0x31, 0x37, 0xB2, 0x49, 0x28, 
    0x35, 0xC0, 0x99, 0xB6, 0xD0, 0xBC, 0x66, 0x35, 
    0xF7, 0x83, 0x5B, 0xD7, 0x37, 0x1A, 0x2B, 0x18, 
    0xA6, 0xFF, 0x8D, 0x7C, 0x81, 0xA8, 0xFC, 0x9E, 
    0xC4, 0xEC, 0x80, 0xD0, 0x98, 0xA7, 0x76, 0xCC, 
    0x9C, 0x2F, 0x7B, 0xFF, 0x8E, 0x0E, 0xBB, 0x90, 
    0xAE, 0x13, 0x06, 0xF5, 0x1C, 0x4E, 0x52, 0xF7
  }
};

/* Generate and return single random byte */
static unsigned char randomByte(void){
  unsigned char t;
  sqlite3Prng.i++;
  t = sqlite3Prng.s[sqlite3Prng.i];
  sqlite3Prng.j += t;
  sqlite3Prng.s[sqlite3Prng.i] = sqlite3Prng.s[sqlite3Prng.j];
  sqlite3Prng.s[sqlite3Prng.j] = t;
  t += sqlite3Prng.s[sqlite3Prng.i];
  return sqlite3Prng.s[t];
}

/*
** Return N random bytes.
*/
static void randomBlob(int nBuf, unsigned char *zBuf){
  int i;
  for(i=0; i<nBuf; i++){
    zBuf[i] = randomByte();
  }
}
/*
** End of code copied from SQLite.
*************************************************************************/


int testPrngInit(void){
  randomBlob(sizeof(r.aRand1), (unsigned char *)r.aRand1);
  randomBlob(sizeof(r.aRand2), (unsigned char *)r.aRand2);
  randomBlob(sizeof(r.aRand3), (unsigned char *)r.aRand3);
  return 0;
}

unsigned int testPrngValue(unsigned int iVal){
  return
    r.aRand1[iVal & 0x000007FF] ^
    r.aRand2[(iVal>>11) & 0x000007FF] ^
    r.aRand3[(iVal>>22) & 0x000003FF]
  ;
}

void testPrngArray(unsigned int iVal, unsigned int *aOut, int nOut){
  int i;
  for(i=0; i<nOut; i++){
    aOut[i] = testPrngValue(iVal+i);
  }
}

void testPrngString(unsigned int iVal, char *aOut, int nOut){
  int i;
  for(i=0; i<(nOut-1); i++){
    aOut[i] = 'a' + (testPrngValue(iVal+i) % 26);
  }
  aOut[i] = '\0';
}

void testErrorInit(int argc, char **argv){
  g.argc = argc;
  g.argv = argv;
}

void testPrintError(const char *zFormat, ...){
  va_list ap;
  va_start(ap, zFormat);
  vfprintf(stderr, zFormat, ap);
  va_end(ap);
}

void testPrintFUsage(const char *zFormat, ...){
  va_list ap;
  va_start(ap, zFormat);
  fprintf(stderr, "Usage: %s %s ", g.argv[0], g.argv[1]);
  vfprintf(stderr, zFormat, ap);
  fprintf(stderr, "\n");
  va_end(ap);
}

void testPrintUsage(const char *zArgs){
  testPrintError("Usage: %s %s %s\n", g.argv[0], g.argv[1], zArgs);
}


static void argError(void *aData, const char *zType, int sz, const char *zArg){
  struct Entry { const char *zName; };
  struct Entry *pEntry;
  const char *zPrev = 0;

  testPrintError("unrecognized %s \"%s\": must be ", zType, zArg);
  for(pEntry=(struct Entry *)aData; 
      pEntry->zName; 
      pEntry=(struct Entry *)&((unsigned char *)pEntry)[sz]
  ){
    if( zPrev ){ testPrintError("%s, ", zPrev); }
    zPrev = pEntry->zName;
  }
  testPrintError("or %s\n", zPrev);
}

int testArgSelectX(
  void *aData, 
  const char *zType, 
  int sz, 
  const char *zArg, 
  int *piOut
){
  struct Entry { const char *zName; };
  struct Entry *pEntry;
  int nArg = strlen(zArg);

  int i = 0;
  int iOut = -1;
  int nOut = 0;

  for(pEntry=(struct Entry *)aData; 
      pEntry->zName; 
      pEntry=(struct Entry *)&((unsigned char *)pEntry)[sz]
  ){
    int nName = strlen(pEntry->zName);
    if( nArg<=nName && memcmp(pEntry->zName, zArg, nArg)==0 ){
      iOut = i;
      if( nName==nArg ){
        nOut = 1;
        break;
      }
      nOut++;
    }
    i++;
  }

  if( nOut!=1 ){
    argError(aData, zType, sz, zArg);
  }else{
    *piOut = iOut;
  }
  return (nOut!=1);
}

struct timeval zero_time;

void testTimeInit(void){
  gettimeofday(&zero_time, 0);
}

int testTimeGet(void){
  struct timeval now;
  gettimeofday(&now, 0);
  return
    (((int)now.tv_sec - (int)zero_time.tv_sec)*1000) +
    (((int)now.tv_usec - (int)zero_time.tv_usec)/1000);
}

Added ext/lsm1/lsm-test/lsmtest_win32.c.





























































>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30

#include "lsmtest.h"

#ifdef _WIN32

#define TICKS_PER_SECOND      (10000000)
#define TICKS_PER_MICROSECOND (10)
#define TICKS_UNIX_EPOCH      (116444736000000000LL)

int win32GetTimeOfDay(
  struct timeval *tp,
  void *tzp
){
  FILETIME fileTime;
  ULARGE_INTEGER largeInteger;
  ULONGLONG ticks;

  unused_parameter(tzp);
  memset(&fileTime, 0, sizeof(FILETIME));
  GetSystemTimeAsFileTime(&fileTime);
  memset(&largeInteger, 0, sizeof(ULARGE_INTEGER));
  largeInteger.LowPart = fileTime.dwLowDateTime;
  largeInteger.HighPart = fileTime.dwHighDateTime;
  ticks = largeInteger.QuadPart - TICKS_UNIX_EPOCH;
  tp->tv_sec = (long)(ticks / TICKS_PER_SECOND);
  ticks -= ((ULONGLONG)tp->tv_sec * TICKS_PER_SECOND);
  tp->tv_usec = (long)(ticks / TICKS_PER_MICROSECOND);
  return 0;
}
#endif

Added ext/lsm1/lsm.h.

























































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
/*
** 2011-08-10
**
** The author disclaims copyright to this source code.  In place of
** a legal notice, here is a blessing:
**
**    May you do good and not evil.
**    May you find forgiveness for yourself and forgive others.
**    May you share freely, never taking more than you give.
**
*************************************************************************
**
** This file defines the LSM API.
*/
#ifndef _LSM_H
#define _LSM_H
#include <stddef.h>
#ifdef __cplusplus
extern "C" {
#endif

/*
** Opaque handle types.
*/
typedef struct lsm_compress lsm_compress;   /* Compression library functions */
typedef struct lsm_compress_factory lsm_compress_factory;
typedef struct lsm_cursor lsm_cursor;       /* Database cursor handle */
typedef struct lsm_db lsm_db;               /* Database connection handle */
typedef struct lsm_env lsm_env;             /* Runtime environment */
typedef struct lsm_file lsm_file;           /* OS file handle */
typedef struct lsm_mutex lsm_mutex;         /* Mutex handle */

/* 64-bit integer type used for file offsets. */
typedef long long int lsm_i64;              /* 64-bit signed integer type */

/* Candidate values for the 3rd argument to lsm_env.xLock() */
#define LSM_LOCK_UNLOCK 0
#define LSM_LOCK_SHARED 1
#define LSM_LOCK_EXCL   2

/* Flags for lsm_env.xOpen() */
#define LSM_OPEN_READONLY 0x0001

/*
** CAPI: Database Runtime Environment
**
** Run-time environment used by LSM
*/
struct lsm_env {
  int nByte;                 /* Size of this structure in bytes */
  int iVersion;              /* Version number of this structure (1) */
  /****** file i/o ***********************************************/
  void *pVfsCtx;
  int (*xFullpath)(lsm_env*, const char *, char *, int *);
  int (*xOpen)(lsm_env*, const char *, int flags, lsm_file **);
  int (*xRead)(lsm_file *, lsm_i64, void *, int);
  int (*xWrite)(lsm_file *, lsm_i64, void *, int);
  int (*xTruncate)(lsm_file *, lsm_i64);
  int (*xSync)(lsm_file *);
  int (*xSectorSize)(lsm_file *);
  int (*xRemap)(lsm_file *, lsm_i64, void **, lsm_i64*);
  int (*xFileid)(lsm_file *, void *pBuf, int *pnBuf);
  int (*xClose)(lsm_file *);
  int (*xUnlink)(lsm_env*, const char *);
  int (*xLock)(lsm_file*, int, int);
  int (*xTestLock)(lsm_file*, int, int, int);
  int (*xShmMap)(lsm_file*, int, int, void **);
  void (*xShmBarrier)(void);
  int (*xShmUnmap)(lsm_file*, int);
  /****** memory allocation ****************************************/
  void *pMemCtx;
  void *(*xMalloc)(lsm_env*, size_t);            /* malloc(3) function */
  void *(*xRealloc)(lsm_env*, void *, size_t);   /* realloc(3) function */
  void (*xFree)(lsm_env*, void *);               /* free(3) function */
  size_t (*xSize)(lsm_env*, void *);             /* xSize function */
  /****** mutexes ****************************************************/
  void *pMutexCtx;
  int (*xMutexStatic)(lsm_env*,int,lsm_mutex**); /* Obtain a static mutex */
  int (*xMutexNew)(lsm_env*, lsm_mutex**);       /* Get a new dynamic mutex */
  void (*xMutexDel)(lsm_mutex *);           /* Delete an allocated mutex */
  void (*xMutexEnter)(lsm_mutex *);         /* Grab a mutex */
  int (*xMutexTry)(lsm_mutex *);            /* Attempt to obtain a mutex */
  void (*xMutexLeave)(lsm_mutex *);         /* Leave a mutex */
  int (*xMutexHeld)(lsm_mutex *);           /* Return true if mutex is held */
  int (*xMutexNotHeld)(lsm_mutex *);        /* Return true if mutex not held */
  /****** other ****************************************************/
  int (*xSleep)(lsm_env*, int microseconds);

  /* New fields may be added in future releases, in which case the
  ** iVersion value will increase. */
};

/* 
** Values that may be passed as the second argument to xMutexStatic. 
*/
#define LSM_MUTEX_GLOBAL 1
#define LSM_MUTEX_HEAP   2

/*
** CAPI: LSM Error Codes
*/
#define LSM_OK         0
#define LSM_ERROR      1
#define LSM_BUSY       5
#define LSM_NOMEM      7
#define LSM_READONLY   8
#define LSM_IOERR     10
#define LSM_CORRUPT   11
#define LSM_FULL      13
#define LSM_CANTOPEN  14
#define LSM_PROTOCOL  15
#define LSM_MISUSE    21

#define LSM_MISMATCH  50


#define LSM_IOERR_NOENT (LSM_IOERR | (1<<8))

/* 
** CAPI: Creating and Destroying Database Connection Handles
**
** Open and close a database connection handle.
*/
int lsm_new(lsm_env*, lsm_db **ppDb);
int lsm_close(lsm_db *pDb);

/* 
** CAPI: Connecting to a Database
*/
int lsm_open(lsm_db *pDb, const char *zFilename);

/*
** CAPI: Obtaining pointers to database environments
**
** Return a pointer to the environment used by the database connection 
** passed as the first argument. Assuming the argument is valid, this 
** function always returns a valid environment pointer - it cannot fail.
*/
lsm_env *lsm_get_env(lsm_db *pDb);

/*
** The lsm_default_env() function returns a pointer to the default LSM
** environment for the current platform.
*/
lsm_env *lsm_default_env(void);


/*
** CAPI: Configuring a database connection.
**
** The lsm_config() function is used to configure a database connection.
*/
int lsm_config(lsm_db *, int, ...);

/*
** The following values may be passed as the second argument to lsm_config().
**
** LSM_CONFIG_AUTOFLUSH:
**   A read/write integer parameter. 
**
**   This value determines the amount of data allowed to accumulate in a
**   live in-memory tree before it is marked as old. After committing a
**   transaction, a connection checks if the size of the live in-memory tree,
**   including data structure overhead, is greater than the value of this
**   option in KB. If it is, and there is not already an old in-memory tree,
**   the live in-memory tree is marked as old.
**
**   The maximum allowable value is 1048576 (1GB). There is no minimum 
**   value. If this parameter is set to zero, then an attempt is made to
**   mark the live in-memory tree as old after each transaction is committed.
**
**   The default value is 1024 (1MB).
**
** LSM_CONFIG_PAGE_SIZE:
**   A read/write integer parameter. This parameter may only be set before
**   lsm_open() has been called.
**
** LSM_CONFIG_BLOCK_SIZE:
**   A read/write integer parameter. 
**
**   This parameter may only be set before lsm_open() has been called. It
**   must be set to a power of two between 64 and 65536, inclusive (block 
**   sizes between 64KB and 64MB).
**
**   If the connection creates a new database, the block size of the new
**   database is set to the value of this option in KB. After lsm_open()
**   has been called, querying this parameter returns the actual block
**   size of the opened database.
**
**   The default value is 1024 (1MB blocks).
**
** LSM_CONFIG_SAFETY:
**   A read/write integer parameter. Valid values are 0, 1 (the default) 
**   and 2. This parameter determines how robust the database is in the
**   face of a system crash (e.g. a power failure or operating system 
**   crash). As follows:
**
**     0 (off):    No robustness. A system crash may corrupt the database.
**
**     1 (normal): Some robustness. A system crash may not corrupt the
**                 database file, but recently committed transactions may
**                 be lost following recovery.
**
**     2 (full):   Full robustness. A system crash may not corrupt the
**                 database file. Following recovery the database file
**                 contains all successfully committed transactions.
**
** LSM_CONFIG_AUTOWORK:
**   A read/write integer parameter.
**
** LSM_CONFIG_AUTOCHECKPOINT:
**   A read/write integer parameter.
**
**   If this option is set to non-zero value N, then a checkpoint is
**   automatically attempted after each N KB of data have been written to 
**   the database file.
**
**   The amount of uncheckpointed data already written to the database file
**   is a global parameter. After performing database work (writing to the
**   database file), the process checks if the total amount of uncheckpointed 
**   data exceeds the value of this paramter. If so, a checkpoint is performed.
**   This means that this option may cause the connection to perform a 
**   checkpoint even if the current connection has itself written very little
**   data into the database file.
**
**   The default value is 2048 (checkpoint every 2MB).
**
** LSM_CONFIG_MMAP:
**   A read/write integer parameter. If this value is set to 0, then the 
**   database file is accessed using ordinary read/write IO functions. Or,
**   if it is set to 1, then the database file is memory mapped and accessed
**   that way. If this parameter is set to any value N greater than 1, then
**   up to the first N KB of the file are memory mapped, and any remainder
**   accessed using read/write IO.
**
**   The default value is 1 on 64-bit platforms and 32768 on 32-bit platforms.
**   
**
** LSM_CONFIG_USE_LOG:
**   A read/write boolean parameter. True (the default) to use the log
**   file normally. False otherwise.
**
** LSM_CONFIG_AUTOMERGE:
**   A read/write integer parameter. The minimum number of segments to
**   merge together at a time. Default value 4.
**
** LSM_CONFIG_MAX_FREELIST:
**   A read/write integer parameter. The maximum number of free-list 
**   entries that are stored in a database checkpoint (the others are
**   stored elsewhere in the database).
**
**   There is no reason for an application to configure or query this
**   parameter. It is only present because configuring a small value
**   makes certain parts of the lsm code easier to test.
**
** LSM_CONFIG_MULTIPLE_PROCESSES:
**   A read/write boolean parameter. This parameter may only be set before
**   lsm_open() has been called. If true, the library uses shared-memory
**   and posix advisory locks to co-ordinate access by clients from within
**   multiple processes. Otherwise, if false, all database clients must be 
**   located in the same process. The default value is true.
**
** LSM_CONFIG_SET_COMPRESSION:
**   Set the compression methods used to compress and decompress database
**   content. The argument to this option should be a pointer to a structure
**   of type lsm_compress. The lsm_config() method takes a copy of the 
**   structures contents.
**
**   This option may only be used before lsm_open() is called. Invoking it
**   after lsm_open() has been called results in an LSM_MISUSE error.
**
** LSM_CONFIG_GET_COMPRESSION:
**   Query the compression methods used to compress and decompress database
**   content.
**
** LSM_CONFIG_SET_COMPRESSION_FACTORY:
**   Configure a factory method to be invoked in case of an LSM_MISMATCH
**   error.
**
** LSM_CONFIG_READONLY:
**   A read/write boolean parameter. This parameter may only be set before
**   lsm_open() is called.
*/
#define LSM_CONFIG_AUTOFLUSH                1
#define LSM_CONFIG_PAGE_SIZE                2
#define LSM_CONFIG_SAFETY                   3
#define LSM_CONFIG_BLOCK_SIZE               4
#define LSM_CONFIG_AUTOWORK                 5
#define LSM_CONFIG_MMAP                     7
#define LSM_CONFIG_USE_LOG                  8
#define LSM_CONFIG_AUTOMERGE                9
#define LSM_CONFIG_MAX_FREELIST            10
#define LSM_CONFIG_MULTIPLE_PROCESSES      11
#define LSM_CONFIG_AUTOCHECKPOINT          12
#define LSM_CONFIG_SET_COMPRESSION         13
#define LSM_CONFIG_GET_COMPRESSION         14
#define LSM_CONFIG_SET_COMPRESSION_FACTORY 15
#define LSM_CONFIG_READONLY                16

#define LSM_SAFETY_OFF    0
#define LSM_SAFETY_NORMAL 1
#define LSM_SAFETY_FULL   2

/*
** CAPI: Compression and/or Encryption Hooks
*/
struct lsm_compress {
  void *pCtx;
  unsigned int iId;
  int (*xBound)(void *, int nSrc);
  int (*xCompress)(void *, char *, int *, const char *, int);
  int (*xUncompress)(void *, char *, int *, const char *, int);
  void (*xFree)(void *pCtx);
};

struct lsm_compress_factory {
  void *pCtx;
  int (*xFactory)(void *, lsm_db *, unsigned int);
  void (*xFree)(void *pCtx);
};

#define LSM_COMPRESSION_EMPTY 0
#define LSM_COMPRESSION_NONE  1

/*
** CAPI: Allocating and Freeing Memory
**
** Invoke the memory allocation functions that belong to environment
** pEnv. Or the system defaults if no memory allocation functions have 
** been registered.
*/
void *lsm_malloc(lsm_env*, size_t);
void *lsm_realloc(lsm_env*, void *, size_t);
void lsm_free(lsm_env*, void *);

/*
** CAPI: Querying a Connection For Operational Data
**
** Query a database connection for operational statistics or data.
*/
int lsm_info(lsm_db *, int, ...);

int lsm_get_user_version(lsm_db *, unsigned int *);
int lsm_set_user_version(lsm_db *, unsigned int);

/*
** The following values may be passed as the second argument to lsm_info().
**
** LSM_INFO_NWRITE:
**   The third parameter should be of type (int *). The location pointed
**   to by the third parameter is set to the number of 4KB pages written to
**   the database file during the lifetime of this connection. 
**
** LSM_INFO_NREAD:
**   The third parameter should be of type (int *). The location pointed
**   to by the third parameter is set to the number of 4KB pages read from
**   the database file during the lifetime of this connection.
**
** LSM_INFO_DB_STRUCTURE:
**   The third argument should be of type (char **). The location pointed
**   to is populated with a pointer to a nul-terminated string containing
**   the string representation of a Tcl data-structure reflecting the 
**   current structure of the database file. Specifically, the current state
**   of the worker snapshot. The returned string should be eventually freed 
**   by the caller using lsm_free().
**
**   The returned list contains one element for each level in the database,
**   in order from most to least recent. Each element contains a 
**   single element for each segment comprising the corresponding level,
**   starting with the lhs segment, then each of the rhs segments (if any)
**   in order from most to least recent.
**
**   Each segment element is itself a list of 4 integer values, as follows:
**
**   <ol><li> First page of segment
**       <li> Last page of segment
**       <li> Root page of segment (if applicable)
**       <li> Total number of pages in segment
**   </ol>
**
** LSM_INFO_ARRAY_STRUCTURE:
**   There should be two arguments passed following this option (i.e. a 
**   total of four arguments passed to lsm_info()). The first argument 
**   should be the page number of the first page in a database array 
**   (perhaps obtained from an earlier INFO_DB_STRUCTURE call). The second 
**   trailing argument should be of type (char **). The location pointed 
**   to is populated with a pointer to a nul-terminated string that must 
**   be eventually freed using lsm_free() by the caller.
**
**   The output string contains the text representation of a Tcl list of
**   integers. Each pair of integers represent a range of pages used by
**   the identified array. For example, if the array occupies database
**   pages 993 to 1024, then pages 2048 to 2777, then the returned string
**   will be "993 1024 2048 2777".
**
**   If the specified integer argument does not correspond to the first
**   page of any database array, LSM_ERROR is returned and the output
**   pointer is set to a NULL value.
**
** LSM_INFO_LOG_STRUCTURE:
**   The third argument should be of type (char **). The location pointed
**   to is populated with a pointer to a nul-terminated string containing
**   the string representation of a Tcl data-structure. The returned 
**   string should be eventually freed by the caller using lsm_free().
**
**   The Tcl structure returned is a list of six integers that describe
**   the current structure of the log file.
**
** LSM_INFO_ARRAY_PAGES:
**
** LSM_INFO_PAGE_ASCII_DUMP:
**   As with LSM_INFO_ARRAY_STRUCTURE, there should be two arguments passed
**   with calls that specify this option - an integer page number and a
**   (char **) used to return a nul-terminated string that must be later
**   freed using lsm_free(). In this case the output string is populated
**   with a human-readable description of the page content.
**
**   If the page cannot be decoded, it is not an error. In this case the
**   human-readable output message will report the systems failure to 
**   interpret the page data.
**
** LSM_INFO_PAGE_HEX_DUMP:
**   This argument is similar to PAGE_ASCII_DUMP, except that keys and
**   values are represented using hexadecimal notation instead of ascii.
**
** LSM_INFO_FREELIST:
**   The third argument should be of type (char **). The location pointed
**   to is populated with a pointer to a nul-terminated string containing
**   the string representation of a Tcl data-structure. The returned 
**   string should be eventually freed by the caller using lsm_free().
**
**   The Tcl structure returned is a list containing one element for each
**   free block in the database. The element itself consists of two 
**   integers - the block number and the id of the snapshot that freed it.
**
** LSM_INFO_CHECKPOINT_SIZE:
**   The third argument should be of type (int *). The location pointed to
**   by this argument is populated with the number of KB written to the
**   database file since the most recent checkpoint.
**
** LSM_INFO_TREE_SIZE:
**   If this value is passed as the second argument to an lsm_info() call, it
**   should be followed by two arguments of type (int *) (for a total of four
**   arguments).
**
**   At any time, there are either one or two tree structures held in shared
**   memory that new database clients will access (there may also be additional
**   tree structures being used by older clients - this API does not provide
**   information on them). One tree structure - the current tree - is used to
**   accumulate new data written to the database. The other tree structure -
**   the old tree - is a read-only tree holding older data and may be flushed 
**   to disk at any time.
** 
**   Assuming no error occurs, the location pointed to by the first of the two
**   (int *) arguments is set to the size of the old in-memory tree in KB.
**   The second is set to the size of the current, or live in-memory tree.
**
** LSM_INFO_COMPRESSION_ID:
**   This value should be followed by a single argument of type 
**   (unsigned int *). If successful, the location pointed to is populated 
**   with the database compression id before returning.
*/
#define LSM_INFO_NWRITE           1
#define LSM_INFO_NREAD            2
#define LSM_INFO_DB_STRUCTURE     3
#define LSM_INFO_LOG_STRUCTURE    4
#define LSM_INFO_ARRAY_STRUCTURE  5
#define LSM_INFO_PAGE_ASCII_DUMP  6
#define LSM_INFO_PAGE_HEX_DUMP    7
#define LSM_INFO_FREELIST         8
#define LSM_INFO_ARRAY_PAGES      9
#define LSM_INFO_CHECKPOINT_SIZE 10
#define LSM_INFO_TREE_SIZE       11
#define LSM_INFO_FREELIST_SIZE   12
#define LSM_INFO_COMPRESSION_ID  13


/* 
** CAPI: Opening and Closing Write Transactions
**
** These functions are used to open and close transactions and nested 
** sub-transactions.
**
** The lsm_begin() function is used to open transactions and sub-transactions. 
** A successful call to lsm_begin() ensures that there are at least iLevel 
** nested transactions open. To open a top-level transaction, pass iLevel=1. 
** To open a sub-transaction within the top-level transaction, iLevel=2. 
** Passing iLevel=0 is a no-op.
**
** lsm_commit() is used to commit transactions and sub-transactions. A
** successful call to lsm_commit() ensures that there are at most iLevel 
** nested transactions open. To commit a top-level transaction, pass iLevel=0. 
** To commit all sub-transactions inside the main transaction, pass iLevel=1.
**
** Function lsm_rollback() is used to roll back transactions and
** sub-transactions. A successful call to lsm_rollback() restores the database 
** to the state it was in when the iLevel'th nested sub-transaction (if any) 
** was first opened. And then closes transactions to ensure that there are 
** at most iLevel nested transactions open. Passing iLevel=0 rolls back and 
** closes the top-level transaction. iLevel=1 also rolls back the top-level 
** transaction, but leaves it open. iLevel=2 rolls back the sub-transaction 
** nested directly inside the top-level transaction (and leaves it open).
*/
int lsm_begin(lsm_db *pDb, int iLevel);
int lsm_commit(lsm_db *pDb, int iLevel);
int lsm_rollback(lsm_db *pDb, int iLevel);

/* 
** CAPI: Writing to a Database
**
** Write a new value into the database. If a value with a duplicate key 
** already exists it is replaced.
*/
int lsm_insert(lsm_db*, const void *pKey, int nKey, const void *pVal, int nVal);

/*
** Delete a value from the database. No error is returned if the specified
** key value does not exist in the database.
*/
int lsm_delete(lsm_db *, const void *pKey, int nKey);

/*
** Delete all database entries with keys that are greater than (pKey1/nKey1) 
** and smaller than (pKey2/nKey2). Note that keys (pKey1/nKey1) and
** (pKey2/nKey2) themselves, if they exist in the database, are not deleted.
**
** Return LSM_OK if successful, or an LSM error code otherwise.
*/
int lsm_delete_range(lsm_db *, 
    const void *pKey1, int nKey1, const void *pKey2, int nKey2
);

/*
** CAPI: Explicit Database Work and Checkpointing
**
** This function is called by a thread to work on the database structure.
*/
int lsm_work(lsm_db *pDb, int nMerge, int nKB, int *pnWrite);

int lsm_flush(lsm_db *pDb);

/*
** Attempt to checkpoint the current database snapshot. Return an LSM
** error code if an error occurs or LSM_OK otherwise.
**
** If the current snapshot has already been checkpointed, calling this 
** function is a no-op. In this case if pnKB is not NULL, *pnKB is
** set to 0. Or, if the current snapshot is successfully checkpointed
** by this function and pbKB is not NULL, *pnKB is set to the number
** of bytes written to the database file since the previous checkpoint
** (the same measure as returned by the LSM_INFO_CHECKPOINT_SIZE query).
*/
int lsm_checkpoint(lsm_db *pDb, int *pnKB);

/*
** CAPI: Opening and Closing Database Cursors
**
** Open and close a database cursor.
*/
int lsm_csr_open(lsm_db *pDb, lsm_cursor **ppCsr);
int lsm_csr_close(lsm_cursor *pCsr);

/* 
** CAPI: Positioning Database Cursors
**
** If the fourth parameter is LSM_SEEK_EQ, LSM_SEEK_GE or LSM_SEEK_LE,
** this function searches the database for an entry with key (pKey/nKey). 
** If an error occurs, an LSM error code is returned. Otherwise, LSM_OK.
**
** If no error occurs and the requested key is present in the database, the
** cursor is left pointing to the entry with the specified key. Or, if the 
** specified key is not present in the database the state of the cursor 
** depends on the value passed as the final parameter, as follows:
**
** LSM_SEEK_EQ:
**   The cursor is left at EOF (invalidated). A call to lsm_csr_valid()
**   returns non-zero.
**
** LSM_SEEK_LE:
**   The cursor is left pointing to the largest key in the database that
**   is smaller than (pKey/nKey). If the database contains no keys smaller
**   than (pKey/nKey), the cursor is left at EOF.
**
** LSM_SEEK_GE:
**   The cursor is left pointing to the smallest key in the database that
**   is larger than (pKey/nKey). If the database contains no keys larger
**   than (pKey/nKey), the cursor is left at EOF.
**
** If the fourth parameter is LSM_SEEK_LEFAST, this function searches the
** database in a similar manner to LSM_SEEK_LE, with two differences:
**
** <ol><li>Even if a key can be found (the cursor is not left at EOF), the
** lsm_csr_value() function may not be used (attempts to do so return
** LSM_MISUSE).
**
** <li>The key that the cursor is left pointing to may be one that has 
** been recently deleted from the database. In this case it is
** guaranteed that the returned key is larger than any key currently 
** in the database that is less than or equal to (pKey/nKey).
** </ol>
**
** LSM_SEEK_LEFAST requests are intended to be used to allocate database
** keys.
*/
int lsm_csr_seek(lsm_cursor *pCsr, const void *pKey, int nKey, int eSeek);

int lsm_csr_first(lsm_cursor *pCsr);
int lsm_csr_last(lsm_cursor *pCsr);

/*
** Advance the specified cursor to the next or previous key in the database.
** Return LSM_OK if successful, or an LSM error code otherwise.
**
** Functions lsm_csr_seek(), lsm_csr_first() and lsm_csr_last() are "seek"
** functions. Whether or not lsm_csr_next and lsm_csr_prev may be called
** successfully also depends on the most recent seek function called on
** the cursor. Specifically:
**
** <ul>
** <li> At least one seek function must have been called on the cursor.
** <li> To call lsm_csr_next(), the most recent call to a seek function must
** have been either lsm_csr_first() or a call to lsm_csr_seek() specifying
** LSM_SEEK_GE.
** <li> To call lsm_csr_prev(), the most recent call to a seek function must
** have been either lsm_csr_last() or a call to lsm_csr_seek() specifying
** LSM_SEEK_LE.
** </ul>
**
** Otherwise, if the above conditions are not met when lsm_csr_next or 
** lsm_csr_prev is called, LSM_MISUSE is returned and the cursor position
** remains unchanged.
*/
int lsm_csr_next(lsm_cursor *pCsr);
int lsm_csr_prev(lsm_cursor *pCsr);

/*
** Values that may be passed as the fourth argument to lsm_csr_seek().
*/
#define LSM_SEEK_LEFAST   -2
#define LSM_SEEK_LE       -1
#define LSM_SEEK_EQ        0
#define LSM_SEEK_GE        1

/* 
** CAPI: Extracting Data From Database Cursors
**
** Retrieve data from a database cursor.
*/
int lsm_csr_valid(lsm_cursor *pCsr);
int lsm_csr_key(lsm_cursor *pCsr, const void **ppKey, int *pnKey);
int lsm_csr_value(lsm_cursor *pCsr, const void **ppVal, int *pnVal);

/*
** If no error occurs, this function compares the database key passed via
** the pKey/nKey arguments with the key that the cursor passed as the first
** argument currently points to. If the cursors key is less than, equal to
** or greater than pKey/nKey, *piRes is set to less than, equal to or greater
** than zero before returning. LSM_OK is returned in this case.
**
** Or, if an error occurs, an LSM error code is returned and the final 
** value of *piRes is undefined. If the cursor does not point to a valid
** key when this function is called, LSM_MISUSE is returned.
*/
int lsm_csr_cmp(lsm_cursor *pCsr, const void *pKey, int nKey, int *piRes);

/*
** CAPI: Change these!!
**
** Configure a callback to which debugging and other messages should 
** be directed. Only useful for debugging lsm.
*/
void lsm_config_log(lsm_db *, void (*)(void *, int, const char *), void *);

/*
** Configure a callback that is invoked if the database connection ever
** writes to the database file.
*/
void lsm_config_work_hook(lsm_db *, void (*)(lsm_db *, void *), void *);

/* ENDOFAPI */
#ifdef __cplusplus
}  /* End of the 'extern "C"' block */
#endif
#endif /* ifndef _LSM_H */

Added ext/lsm1/lsmInt.h.









































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
/*
** 2011-08-18
**
** The author disclaims copyright to this source code.  In place of
** a legal notice, here is a blessing:
**
**    May you do good and not evil.
**    May you find forgiveness for yourself and forgive others.
**    May you share freely, never taking more than you give.
**
*************************************************************************
** Internal structure definitions for the LSM module.
*/
#ifndef _LSM_INT_H
#define _LSM_INT_H

#include "lsm.h"
#include <assert.h>
#include <string.h>

#include <stdarg.h>
#include <stdlib.h>
#include <stdio.h>
#include <ctype.h>

#ifdef _WIN32
# ifdef _MSC_VER
#  define snprintf _snprintf
# endif
#else
# include <unistd.h>
#endif

#ifdef NDEBUG
# ifdef LSM_DEBUG_EXPENSIVE
#  undef LSM_DEBUG_EXPENSIVE
# endif
# ifdef LSM_DEBUG
#  undef LSM_DEBUG
# endif
#else
# ifndef LSM_DEBUG
#  define LSM_DEBUG
# endif
#endif

/*
** Default values for various data structure parameters. These may be
** overridden by calls to lsm_config().
*/
#define LSM_DFLT_PAGE_SIZE          (4 * 1024)
#define LSM_DFLT_BLOCK_SIZE         (1 * 1024 * 1024)
#define LSM_DFLT_AUTOFLUSH          (1 * 1024 * 1024)
#define LSM_DFLT_AUTOCHECKPOINT     (i64)(2 * 1024 * 1024)
#define LSM_DFLT_AUTOWORK           1
#define LSM_DFLT_LOG_SIZE           (128*1024)
#define LSM_DFLT_AUTOMERGE          4
#define LSM_DFLT_SAFETY             LSM_SAFETY_NORMAL
#define LSM_DFLT_MMAP               (LSM_IS_64_BIT ? 1 : 32768)
#define LSM_DFLT_MULTIPLE_PROCESSES 1
#define LSM_DFLT_USE_LOG            1

/* Initial values for log file checksums. These are only used if the 
** database file does not contain a valid checkpoint.  */
#define LSM_CKSUM0_INIT 42
#define LSM_CKSUM1_INIT 42

#define LSM_META_PAGE_SIZE 4096

/* "mmap" mode is currently only used in environments with 64-bit address 
** spaces. The following macro is used to test for this.  */
#define LSM_IS_64_BIT (sizeof(void*)==8)

#define LSM_AUTOWORK_QUANT 32

typedef struct Database Database;
typedef struct DbLog DbLog;
typedef struct FileSystem FileSystem;
typedef struct Freelist Freelist;
typedef struct FreelistEntry FreelistEntry;
typedef struct Level Level;
typedef struct LogMark LogMark;
typedef struct LogRegion LogRegion;
typedef struct LogWriter LogWriter;
typedef struct LsmString LsmString;
typedef struct Mempool Mempool;
typedef struct Merge Merge;
typedef struct MergeInput MergeInput;
typedef struct MetaPage MetaPage;
typedef struct MultiCursor MultiCursor;
typedef struct Page Page;
typedef struct Redirect Redirect;
typedef struct Segment Segment;
typedef struct SegmentMerger SegmentMerger;
typedef struct ShmChunk ShmChunk;
typedef struct ShmHeader ShmHeader;
typedef struct ShmReader ShmReader;
typedef struct Snapshot Snapshot;
typedef struct TransMark TransMark;
typedef struct Tree Tree;
typedef struct TreeCursor TreeCursor;
typedef struct TreeHeader TreeHeader;
typedef struct TreeMark TreeMark;
typedef struct TreeRoot TreeRoot;

#ifndef _SQLITEINT_H_
typedef unsigned char u8;
typedef unsigned short int u16;
typedef unsigned int u32;
typedef lsm_i64 i64;
typedef unsigned long long int u64;
#endif

/* A page number is a 64-bit integer. */
typedef i64 Pgno;

#ifdef LSM_DEBUG
int lsmErrorBkpt(int);
#else
# define lsmErrorBkpt(x) (x)
#endif

#define LSM_PROTOCOL_BKPT lsmErrorBkpt(LSM_PROTOCOL)
#define LSM_IOERR_BKPT    lsmErrorBkpt(LSM_IOERR)
#define LSM_NOMEM_BKPT    lsmErrorBkpt(LSM_NOMEM)
#define LSM_CORRUPT_BKPT  lsmErrorBkpt(LSM_CORRUPT)
#define LSM_MISUSE_BKPT   lsmErrorBkpt(LSM_MISUSE)

#define unused_parameter(x) (void)(x)
#define array_size(x) (sizeof(x)/sizeof(x[0]))


/* The size of each shared-memory chunk */
#define LSM_SHM_CHUNK_SIZE (32*1024)

/* The number of bytes reserved at the start of each shm chunk for MM. */
#define LSM_SHM_CHUNK_HDR  (sizeof(ShmChunk))

/* The number of available read locks. */
#define LSM_LOCK_NREADER   6

/* The number of available read-write client locks. */
#define LSM_LOCK_NRWCLIENT   16

/* Lock definitions. 
*/
#define LSM_LOCK_DMS1         1   /* Serialize connect/disconnect ops */
#define LSM_LOCK_DMS2         2   /* Read-write connections */
#define LSM_LOCK_DMS3         3   /* Read-only connections */
#define LSM_LOCK_WRITER       4
#define LSM_LOCK_WORKER       5
#define LSM_LOCK_CHECKPOINTER 6
#define LSM_LOCK_ROTRANS      7
#define LSM_LOCK_READER(i)    ((i) + LSM_LOCK_ROTRANS + 1)
#define LSM_LOCK_RWCLIENT(i)  ((i) + LSM_LOCK_READER(LSM_LOCK_NREADER))

/*
** Hard limit on the number of free-list entries that may be stored in 
** a checkpoint (the remainder are stored as a system record in the LSM).
** See also LSM_CONFIG_MAX_FREELIST.
*/
#define LSM_MAX_FREELIST_ENTRIES 24

#define LSM_MAX_BLOCK_REDIRECTS 16

#define LSM_ATTEMPTS_BEFORE_PROTOCOL 10000


/*
** Each entry stored in the LSM (or in-memory tree structure) has an
** associated mask of the following flags.
*/
#define LSM_START_DELETE 0x01     /* Start of open-ended delete range */
#define LSM_END_DELETE   0x02     /* End of open-ended delete range */
#define LSM_POINT_DELETE 0x04     /* Delete this key */
#define LSM_INSERT       0x08     /* Insert this key and value */
#define LSM_SEPARATOR    0x10     /* True if entry is separator key only */
#define LSM_SYSTEMKEY    0x20     /* True if entry is a system key (FREELIST) */

#define LSM_CONTIGUOUS   0x40     /* Used in lsm_tree.c */

/*
** A string that can grow by appending.
*/
struct LsmString {
  lsm_env *pEnv;              /* Run-time environment */
  int n;                      /* Size of string.  -1 indicates error */
  int nAlloc;                 /* Space allocated for z[] */
  char *z;                    /* The string content */
};

typedef struct LsmFile LsmFile;
struct LsmFile {
  lsm_file *pFile;
  LsmFile *pNext;
};

/*
** An instance of the following type is used to store an ordered list of
** u32 values. 
**
** Note: This is a place-holder implementation. It should be replaced by
** a version that avoids making a single large allocation when the array
** contains a large number of values. For this reason, the internals of 
** this object should only manipulated by the intArrayXXX() functions in 
** lsm_tree.c.
*/
typedef struct IntArray IntArray;
struct IntArray {
  int nAlloc;
  int nArray;
  u32 *aArray;
};

struct Redirect {
  int n;                          /* Number of redirects */
  struct RedirectEntry {
    int iFrom;
    int iTo;
  } *a;
};

/*
** An instance of this structure represents a point in the history of the
** tree structure to roll back to. Refer to comments in lsm_tree.c for 
** details.
*/
struct TreeMark {
  u32 iRoot;                      /* Offset of root node in shm file */
  u32 nHeight;                    /* Current height of tree structure */
  u32 iWrite;                     /* Write offset in shm file */
  u32 nChunk;                     /* Number of chunks in shared-memory file */
  u32 iFirst;                     /* First chunk in linked list */
  u32 iNextShmid;                 /* Next id to allocate */
  int iRollback;                  /* Index in lsm->rollback to revert to */
};

/*
** An instance of this structure represents a point in the database log.
*/
struct LogMark {
  i64 iOff;                       /* Offset into log (see lsm_log.c) */
  int nBuf;                       /* Size of in-memory buffer here */
  u8 aBuf[8];                     /* Bytes of content in aBuf[] */
  u32 cksum0;                     /* Checksum 0 at offset (iOff-nBuf) */
  u32 cksum1;                     /* Checksum 1 at offset (iOff-nBuf) */
};

struct TransMark {
  TreeMark tree;
  LogMark log;
};

/*
** A structure that defines the start and end offsets of a region in the
** log file. The size of the region in bytes is (iEnd - iStart), so if
** iEnd==iStart the region is zero bytes in size.
*/
struct LogRegion {
  i64 iStart;                     /* Start of region in log file */
  i64 iEnd;                       /* End of region in log file */
};

struct DbLog {
  u32 cksum0;                     /* Checksum 0 at offset iOff */
  u32 cksum1;                     /* Checksum 1 at offset iOff */
  i64 iSnapshotId;                /* Log space has been reclaimed to this ss */
  LogRegion aRegion[3];           /* Log file regions (see docs in lsm_log.c) */
};

struct TreeRoot {
  u32 iRoot;
  u32 nHeight;
  u32 nByte;                      /* Total size of this tree in bytes */
  u32 iTransId;
};

/*
** Tree header structure. 
*/
struct TreeHeader {
  u32 iUsedShmid;                 /* Id of first shm chunk used by this tree */
  u32 iNextShmid;                 /* Shm-id of next chunk allocated */
  u32 iFirst;                     /* Chunk number of smallest shm-id */
  u32 nChunk;                     /* Number of chunks in shared-memory file */
  TreeRoot root;                  /* Root and height of current tree */
  u32 iWrite;                     /* Write offset in shm file */
  TreeRoot oldroot;               /* Root and height of the previous tree */
  u32 iOldShmid;                  /* Last shm-id used by previous tree */
  u32 iUsrVersion;                /* get/set_user_version() value */
  i64 iOldLog;                    /* Log offset associated with old tree */
  u32 oldcksum0;
  u32 oldcksum1;
  DbLog log;                      /* Current layout of log file */ 
  u32 aCksum[2];                  /* Checksums 1 and 2. */
};

/*
** Database handle structure.
**
** mLock:
**   A bitmask representing the locks currently held by the connection.
**   An LSM database supports N distinct locks, where N is some number less
**   than or equal to 32. Locks are numbered starting from 1 (see the 
**   definitions for LSM_LOCK_WRITER and co.).
**
**   The least significant 32-bits in mLock represent EXCLUSIVE locks. The
**   most significant are SHARED locks. So, if a connection holds a SHARED
**   lock on lock region iLock, then the following is true:
**
**       (mLock & ((iLock+32-1) << 1))
**
**   Or for an EXCLUSIVE lock:
**
**       (mLock & ((iLock-1) << 1))
** 
** pCsr:
**   Points to the head of a linked list that contains all currently open
**   cursors. Once this list becomes empty, the user has no outstanding
**   cursors and the database handle can be successfully closed.
**
** pCsrCache:
**   This list contains cursor objects that have been closed using
**   lsm_csr_close(). Each time a cursor is closed, it is shifted from 
**   the pCsr list to this list. When a new cursor is opened, this list
**   is inspected to see if there exists a cursor object that can be
**   reused. This is an optimization only.
*/
struct lsm_db {

  /* Database handle configuration */
  lsm_env *pEnv;                            /* runtime environment */
  int (*xCmp)(void *, int, void *, int);    /* Compare function */

  /* Values configured by calls to lsm_config */
  int eSafety;                    /* LSM_SAFETY_OFF, NORMAL or FULL */
  int bAutowork;                  /* Configured by LSM_CONFIG_AUTOWORK */
  int nTreeLimit;                 /* Configured by LSM_CONFIG_AUTOFLUSH */
  int nMerge;                     /* Configured by LSM_CONFIG_AUTOMERGE */
  int bUseLog;                    /* Configured by LSM_CONFIG_USE_LOG */
  int nDfltPgsz;                  /* Configured by LSM_CONFIG_PAGE_SIZE */
  int nDfltBlksz;                 /* Configured by LSM_CONFIG_BLOCK_SIZE */
  int nMaxFreelist;               /* Configured by LSM_CONFIG_MAX_FREELIST */
  int iMmap;                      /* Configured by LSM_CONFIG_MMAP */
  i64 nAutockpt;                  /* Configured by LSM_CONFIG_AUTOCHECKPOINT */
  int bMultiProc;                 /* Configured by L_C_MULTIPLE_PROCESSES */
  int bReadonly;                  /* Configured by LSM_CONFIG_READONLY */
  lsm_compress compress;          /* Compression callbacks */
  lsm_compress_factory factory;   /* Compression callback factory */

  /* Sub-system handles */
  FileSystem *pFS;                /* On-disk portion of database */
  Database *pDatabase;            /* Database shared data */

  int iRwclient;                  /* Read-write client lock held (-1 == none) */

  /* Client transaction context */
  Snapshot *pClient;              /* Client snapshot */
  int iReader;                    /* Read lock held (-1 == unlocked) */
  int bRoTrans;                   /* True if a read-only db trans is open */
  MultiCursor *pCsr;              /* List of all open cursors */
  LogWriter *pLogWriter;          /* Context for writing to the log file */
  int nTransOpen;                 /* Number of opened write transactions */
  int nTransAlloc;                /* Allocated size of aTrans[] array */
  TransMark *aTrans;              /* Array of marks for transaction rollback */
  IntArray rollback;              /* List of tree-nodes to roll back */
  int bDiscardOld;                /* True if lsmTreeDiscardOld() was called */

  MultiCursor *pCsrCache;         /* List of all closed cursors */

  /* Worker context */
  Snapshot *pWorker;              /* Worker snapshot (or NULL) */
  Freelist *pFreelist;            /* See sortedNewToplevel() */
  int bUseFreelist;               /* True to use pFreelist */
  int bIncrMerge;                 /* True if currently doing a merge */

  int bInFactory;                 /* True if within factory.xFactory() */

  /* Debugging message callback */
  void (*xLog)(void *, int, const char *);
  void *pLogCtx;

  /* Work done notification callback */
  void (*xWork)(lsm_db *, void *);
  void *pWorkCtx;

  u64 mLock;                      /* Mask of current locks. See lsmShmLock(). */
  lsm_db *pNext;                  /* Next connection to same database */

  int nShm;                       /* Size of apShm[] array */
  void **apShm;                   /* Shared memory chunks */
  ShmHeader *pShmhdr;             /* Live shared-memory header */
  TreeHeader treehdr;             /* Local copy of tree-header */
  u32 aSnapshot[LSM_META_PAGE_SIZE / sizeof(u32)];
};

struct Segment {
  Pgno iFirst;                     /* First page of this run */
  Pgno iLastPg;                    /* Last page of this run */
  Pgno iRoot;                      /* Root page number (if any) */
  int nSize;                       /* Size of this run in pages */

  Redirect *pRedirect;             /* Block redirects (or NULL) */
};

/*
** iSplitTopic/pSplitKey/nSplitKey:
**   If nRight>0, this buffer contains a copy of the largest key that has
**   already been written to the left-hand-side of the level.
*/
struct Level {
  Segment lhs;                    /* Left-hand (main) segment */
  int nRight;                     /* Size of apRight[] array */
  Segment *aRhs;                  /* Old segments being merged into this */
  int iSplitTopic;                /* Split key topic (if nRight>0) */
  void *pSplitKey;                /* Pointer to split-key (if nRight>0) */
  int nSplitKey;                  /* Number of bytes in split-key */

  u16 iAge;                       /* Number of times data has been written */
  u16 flags;                      /* Mask of LEVEL_XXX bits */
  Merge *pMerge;                  /* Merge operation currently underway */
  Level *pNext;                   /* Next level in tree */
};

/*
** The Level.flags field is set to a combination of the following bits.
**
** LEVEL_FREELIST_ONLY:
**   Set if the level consists entirely of free-list entries. 
**
** LEVEL_INCOMPLETE:
**   This is set while a new toplevel level is being constructed. It is
**   never set for any level other than a new toplevel.
*/
#define LEVEL_FREELIST_ONLY      0x0001
#define LEVEL_INCOMPLETE         0x0002


/*
** A structure describing an ongoing merge. There is an instance of this
** structure for every Level currently undergoing a merge in the worker
** snapshot.
**
** It is assumed that code that uses an instance of this structure has
** access to the associated Level struct.
**
** iOutputOff:
**   The byte offset to write to next within the last page of the 
**   output segment.
*/
struct MergeInput {
  Pgno iPg;                       /* Page on which next input is stored */
  int iCell;                      /* Cell containing next input to merge */
};
struct Merge {
  int nInput;                     /* Number of input runs being merged */
  MergeInput *aInput;             /* Array nInput entries in size */
  MergeInput splitkey;            /* Location in file of current splitkey */
  int nSkip;                      /* Number of separators entries to skip */
  int iOutputOff;                 /* Write offset on output page */
  Pgno iCurrentPtr;               /* Current pointer value */
};

/* 
** The first argument to this macro is a pointer to a Segment structure.
** Returns true if the structure instance indicates that the separators
** array is valid.
*/
#define segmentHasSeparators(pSegment) ((pSegment)->sep.iFirst>0)

/*
** The values that accompany the lock held by a database reader.
*/
struct ShmReader {
  u32 iTreeId;
  i64 iLsmId;
};

/*
** An instance of this structure is stored in the first shared-memory
** page. The shared-memory header.
**
** bWriter:
**   Immediately after opening a write transaction taking the WRITER lock, 
**   each writer client sets this flag. It is cleared right before the 
**   WRITER lock is relinquished. If a subsequent writer finds that this
**   flag is already set when a write transaction is opened, this indicates
**   that a previous writer failed mid-transaction.
**
** iMetaPage:
**   If the database file does not contain a valid, synced, checkpoint, this
**   value is set to 0. Otherwise, it is set to the meta-page number that
**   contains the most recently written checkpoint (either 1 or 2).
**
** hdr1, hdr2:
**   The two copies of the in-memory tree header. Two copies are required
**   in case a writer fails while updating one of them.
*/
struct ShmHeader {
  u32 aSnap1[LSM_META_PAGE_SIZE / 4];
  u32 aSnap2[LSM_META_PAGE_SIZE / 4];
  u32 bWriter;
  u32 iMetaPage;
  TreeHeader hdr1;
  TreeHeader hdr2;
  ShmReader aReader[LSM_LOCK_NREADER];
};

/*
** An instance of this structure is stored at the start of each shared-memory
** chunk except the first (which is the header chunk - see above).
*/
struct ShmChunk {
  u32 iShmid;
  u32 iNext;
};

/*
** Maximum number of shared-memory chunks allowed in the *-shm file. Since
** each shared-memory chunk is 32KB in size, this is a theoretical limit only.
*/
#define LSM_MAX_SHMCHUNKS  (1<<30)

/* Return true if shm-sequence "a" is larger than or equal to "b" */
#define shm_sequence_ge(a, b) (((u32)a-(u32)b) < LSM_MAX_SHMCHUNKS)

#define LSM_APPLIST_SZ 4

/*
** An instance of the following structure stores the in-memory part of
** the current free block list. This structure is to the free block list
** as the in-memory tree is to the users database content. The contents 
** of the free block list is found by merging the in-memory components 
** with those stored in the LSM, just as the contents of the database is
** found by merging the in-memory tree with the user data entries in the
** LSM.
**
** Each FreelistEntry structure in the array represents either an insert
** or delete operation on the free-list. For deletes, the FreelistEntry.iId
** field is set to -1. For inserts, it is set to zero or greater. 
**
** The array of FreelistEntry structures is always sorted in order of
** block number (ascending).
**
** When the in-memory free block list is written into the LSM, each insert
** operation is written separately. The entry key is the bitwise inverse
** of the block number as a 32-bit big-endian integer. This is done so that
** the entries in the LSM are sorted in descending order of block id. 
** The associated value is the snapshot id, formated as a varint.
*/
struct Freelist {
  FreelistEntry *aEntry;          /* Free list entries */
  int nEntry;                     /* Number of valid slots in aEntry[] */
  int nAlloc;                     /* Allocated size of aEntry[] */
};
struct FreelistEntry {
  u32 iBlk;                       /* Block number */
  i64 iId;                        /* Largest snapshot id to use this block */
};

/*
** A snapshot of a database. A snapshot contains all the information required
** to read or write a database file on disk. See the description of struct
** Database below for futher details.
*/
struct Snapshot {
  Database *pDatabase;            /* Database this snapshot belongs to */
  u32 iCmpId;                     /* Id of compression scheme */
  Level *pLevel;                  /* Pointer to level 0 of snapshot (or NULL) */
  i64 iId;                        /* Snapshot id */
  i64 iLogOff;                    /* Log file offset */
  Redirect redirect;              /* Block redirection array */

  /* Used by worker snapshots only */
  int nBlock;                     /* Number of blocks in database file */
  Pgno aiAppend[LSM_APPLIST_SZ];  /* Append point list */
  Freelist freelist;              /* Free block list */
  u32 nWrite;                     /* Total number of pages written to disk */
};
#define LSM_INITIAL_SNAPSHOT_ID 11

/*
** Functions from file "lsm_ckpt.c".
*/
int lsmCheckpointWrite(lsm_db *, int, u32 *);
int lsmCheckpointLevels(lsm_db *, int, void **, int *);
int lsmCheckpointLoadLevels(lsm_db *pDb, void *pVal, int nVal);

int lsmCheckpointRecover(lsm_db *);
int lsmCheckpointDeserialize(lsm_db *, int, u32 *, Snapshot **);

int lsmCheckpointLoadWorker(lsm_db *pDb);
int lsmCheckpointStore(lsm_db *pDb, int);

int lsmCheckpointLoad(lsm_db *pDb, int *);
int lsmCheckpointLoadOk(lsm_db *pDb, int);
int lsmCheckpointClientCacheOk(lsm_db *);

u32 lsmCheckpointNBlock(u32 *);
i64 lsmCheckpointId(u32 *, int);
u32 lsmCheckpointNWrite(u32 *, int);
i64 lsmCheckpointLogOffset(u32 *);
int lsmCheckpointPgsz(u32 *);
int lsmCheckpointBlksz(u32 *);
void lsmCheckpointLogoffset(u32 *aCkpt, DbLog *pLog);
void lsmCheckpointZeroLogoffset(lsm_db *);

int lsmCheckpointSaveWorker(lsm_db *pDb, int);
int lsmDatabaseFull(lsm_db *pDb);
int lsmCheckpointSynced(lsm_db *pDb, i64 *piId, i64 *piLog, u32 *pnWrite);

int lsmCheckpointSize(lsm_db *db, int *pnByte);

int lsmInfoCompressionId(lsm_db *db, u32 *piCmpId);

/* 
** Functions from file "lsm_tree.c".
*/
int lsmTreeNew(lsm_env *, int (*)(void *, int, void *, int), Tree **ppTree);
void lsmTreeRelease(lsm_env *, Tree *);
int lsmTreeInit(lsm_db *);
int lsmTreeRepair(lsm_db *);

void lsmTreeMakeOld(lsm_db *pDb);
void lsmTreeDiscardOld(lsm_db *pDb);
int lsmTreeHasOld(lsm_db *pDb);

int lsmTreeSize(lsm_db *);
int lsmTreeEndTransaction(lsm_db *pDb, int bCommit);
int lsmTreeLoadHeader(lsm_db *pDb, int *);
int lsmTreeLoadHeaderOk(lsm_db *, int);

int lsmTreeInsert(lsm_db *pDb, void *pKey, int nKey, void *pVal, int nVal);
int lsmTreeDelete(lsm_db *db, void *pKey1, int nKey1, void *pKey2, int nKey2);
void lsmTreeRollback(lsm_db *pDb, TreeMark *pMark);
void lsmTreeMark(lsm_db *pDb, TreeMark *pMark);

int lsmTreeCursorNew(lsm_db *pDb, int, TreeCursor **);
void lsmTreeCursorDestroy(TreeCursor *);

int lsmTreeCursorSeek(TreeCursor *pCsr, void *pKey, int nKey, int *pRes);
int lsmTreeCursorNext(TreeCursor *pCsr);
int lsmTreeCursorPrev(TreeCursor *pCsr);
int lsmTreeCursorEnd(TreeCursor *pCsr, int bLast);
void lsmTreeCursorReset(TreeCursor *pCsr);
int lsmTreeCursorKey(TreeCursor *pCsr, int *pFlags, void **ppKey, int *pnKey);
int lsmTreeCursorFlags(TreeCursor *pCsr);
int lsmTreeCursorValue(TreeCursor *pCsr, void **ppVal, int *pnVal);
int lsmTreeCursorValid(TreeCursor *pCsr);
int lsmTreeCursorSave(TreeCursor *pCsr);

void lsmFlagsToString(int flags, char *zFlags);

/* 
** Functions from file "mem.c".
*/
void *lsmMalloc(lsm_env*, size_t);
void lsmFree(lsm_env*, void *);
void *lsmRealloc(lsm_env*, void *, size_t);
void *lsmReallocOrFree(lsm_env*, void *, size_t);
void *lsmReallocOrFreeRc(lsm_env *, void *, size_t, int *);

void *lsmMallocZeroRc(lsm_env*, size_t, int *);
void *lsmMallocRc(lsm_env*, size_t, int *);

void *lsmMallocZero(lsm_env *pEnv, size_t);
char *lsmMallocStrdup(lsm_env *pEnv, const char *);

/* 
** Functions from file "lsm_mutex.c".
*/
int lsmMutexStatic(lsm_env*, int, lsm_mutex **);
int lsmMutexNew(lsm_env*, lsm_mutex **);
void lsmMutexDel(lsm_env*, lsm_mutex *);
void lsmMutexEnter(lsm_env*, lsm_mutex *);
int lsmMutexTry(lsm_env*, lsm_mutex *);
void lsmMutexLeave(lsm_env*, lsm_mutex *);

#ifndef NDEBUG
int lsmMutexHeld(lsm_env *, lsm_mutex *);
int lsmMutexNotHeld(lsm_env *, lsm_mutex *);
#endif

/**************************************************************************
** Start of functions from "lsm_file.c".
*/
int lsmFsOpen(lsm_db *, const char *, int);
int lsmFsOpenLog(lsm_db *, int *);
void lsmFsCloseLog(lsm_db *);
void lsmFsClose(FileSystem *);

int lsmFsConfigure(lsm_db *db);

int lsmFsBlockSize(FileSystem *);
void lsmFsSetBlockSize(FileSystem *, int);
int lsmFsMoveBlock(FileSystem *pFS, Segment *pSeg, int iTo, int iFrom);

int lsmFsPageSize(FileSystem *);
void lsmFsSetPageSize(FileSystem *, int);

int lsmFsFileid(lsm_db *pDb, void **ppId, int *pnId);

/* Creating, populating, gobbling and deleting sorted runs. */
void lsmFsGobble(lsm_db *, Segment *, Pgno *, int);
int lsmFsSortedDelete(FileSystem *, Snapshot *, int, Segment *);
int lsmFsSortedFinish(FileSystem *, Segment *);
int lsmFsSortedAppend(FileSystem *, Snapshot *, Level *, int, Page **);
int lsmFsSortedPadding(FileSystem *, Snapshot *, Segment *);

/* Functions to retrieve the lsm_env pointer from a FileSystem or Page object */
lsm_env *lsmFsEnv(FileSystem *);
lsm_env *lsmPageEnv(Page *);
FileSystem *lsmPageFS(Page *);

int lsmFsSectorSize(FileSystem *);

void lsmSortedSplitkey(lsm_db *, Level *, int *);

/* Reading sorted run content. */
int lsmFsDbPageLast(FileSystem *pFS, Segment *pSeg, Page **ppPg);
int lsmFsDbPageGet(FileSystem *, Segment *, Pgno, Page **);
int lsmFsDbPageNext(Segment *, Page *, int eDir, Page **);

u8 *lsmFsPageData(Page *, int *);
int lsmFsPageRelease(Page *);
int lsmFsPagePersist(Page *);
void lsmFsPageRef(Page *);
Pgno lsmFsPageNumber(Page *);

int lsmFsNRead(FileSystem *);
int lsmFsNWrite(FileSystem *);

int lsmFsMetaPageGet(FileSystem *, int, int, MetaPage **);
int lsmFsMetaPageRelease(MetaPage *);
u8 *lsmFsMetaPageData(MetaPage *, int *);

#ifdef LSM_DEBUG
int lsmFsDbPageIsLast(Segment *pSeg, Page *pPg);
int lsmFsIntegrityCheck(lsm_db *);
#endif

Pgno lsmFsRedirectPage(FileSystem *, Redirect *, Pgno);

int lsmFsPageWritable(Page *);

/* Functions to read, write and sync the log file. */
int lsmFsWriteLog(FileSystem *pFS, i64 iOff, LsmString *pStr);
int lsmFsSyncLog(FileSystem *pFS);
int lsmFsReadLog(FileSystem *pFS, i64 iOff, int nRead, LsmString *pStr);
int lsmFsTruncateLog(FileSystem *pFS, i64 nByte);
int lsmFsTruncateDb(FileSystem *pFS, i64 nByte);
int lsmFsCloseAndDeleteLog(FileSystem *pFS);

LsmFile *lsmFsDeferClose(FileSystem *pFS);

/* And to sync the db file */
int lsmFsSyncDb(FileSystem *, int);

void lsmFsFlushWaiting(FileSystem *, int *);

/* Used by lsm_info(ARRAY_STRUCTURE) and lsm_config(MMAP) */
int lsmInfoArrayStructure(lsm_db *pDb, int bBlock, Pgno iFirst, char **pzOut);
int lsmInfoArrayPages(lsm_db *pDb, Pgno iFirst, char **pzOut);
int lsmConfigMmap(lsm_db *pDb, int *piParam);

int lsmEnvOpen(lsm_env *, const char *, int, lsm_file **);
int lsmEnvClose(lsm_env *pEnv, lsm_file *pFile);
int lsmEnvLock(lsm_env *pEnv, lsm_file *pFile, int iLock, int eLock);
int lsmEnvTestLock(lsm_env *pEnv, lsm_file *pFile, int iLock, int nLock, int);

int lsmEnvShmMap(lsm_env *, lsm_file *, int, int, void **); 
void lsmEnvShmBarrier(lsm_env *);
void lsmEnvShmUnmap(lsm_env *, lsm_file *, int);

void lsmEnvSleep(lsm_env *, int);

int lsmFsReadSyncedId(lsm_db *db, int, i64 *piVal);

int lsmFsSegmentContainsPg(FileSystem *pFS, Segment *, Pgno, int *);

void lsmFsPurgeCache(FileSystem *);

/*
** End of functions from "lsm_file.c".
**************************************************************************/

/* 
** Functions from file "lsm_sorted.c".
*/
int lsmInfoPageDump(lsm_db *, Pgno, int, char **);
void lsmSortedCleanup(lsm_db *);
int lsmSortedAutoWork(lsm_db *, int nUnit);

int lsmSortedWalkFreelist(lsm_db *, int, int (*)(void *, int, i64), void *);

int lsmSaveWorker(lsm_db *, int);

int lsmFlushTreeToDisk(lsm_db *pDb);

void lsmSortedRemap(lsm_db *pDb);

void lsmSortedFreeLevel(lsm_env *pEnv, Level *);

int lsmSortedAdvanceAll(lsm_db *pDb);

int lsmSortedLoadMerge(lsm_db *, Level *, u32 *, int *);
int lsmSortedLoadFreelist(lsm_db *pDb, void **, int *);

void *lsmSortedSplitKey(Level *pLevel, int *pnByte);

void lsmSortedSaveTreeCursors(lsm_db *);

int lsmMCursorNew(lsm_db *, MultiCursor **);
void lsmMCursorClose(MultiCursor *, int);
int lsmMCursorSeek(MultiCursor *, int, void *, int , int);
int lsmMCursorFirst(MultiCursor *);
int lsmMCursorPrev(MultiCursor *);
int lsmMCursorLast(MultiCursor *);
int lsmMCursorValid(MultiCursor *);
int lsmMCursorNext(MultiCursor *);
int lsmMCursorKey(MultiCursor *, void **, int *);
int lsmMCursorValue(MultiCursor *, void **, int *);
int lsmMCursorType(MultiCursor *, int *);
lsm_db *lsmMCursorDb(MultiCursor *);
void lsmMCursorFreeCache(lsm_db *);

int lsmSaveCursors(lsm_db *pDb);
int lsmRestoreCursors(lsm_db *pDb);

void lsmSortedDumpStructure(lsm_db *pDb, Snapshot *, int, int, const char *);
void lsmFsDumpBlocklists(lsm_db *);

void lsmSortedExpandBtreePage(Page *pPg, int nOrig);

void lsmPutU32(u8 *, u32);
u32 lsmGetU32(u8 *);
u64 lsmGetU64(u8 *);

/*
** Functions from "lsm_varint.c".
*/
int lsmVarintPut32(u8 *, int);
int lsmVarintGet32(u8 *, int *);
int lsmVarintPut64(u8 *aData, i64 iVal);
int lsmVarintGet64(const u8 *aData, i64 *piVal);

int lsmVarintLen32(int);
int lsmVarintSize(u8 c);

/* 
** Functions from file "main.c".
*/
void lsmLogMessage(lsm_db *, int, const char *, ...);
int lsmInfoFreelist(lsm_db *pDb, char **pzOut);

/*
** Functions from file "lsm_log.c".
*/
int lsmLogBegin(lsm_db *pDb);
int lsmLogWrite(lsm_db *, void *, int, void *, int);
int lsmLogCommit(lsm_db *);
void lsmLogEnd(lsm_db *pDb, int bCommit);
void lsmLogTell(lsm_db *, LogMark *);
void lsmLogSeek(lsm_db *, LogMark *);
void lsmLogClose(lsm_db *);

int lsmLogRecover(lsm_db *);
int lsmInfoLogStructure(lsm_db *pDb, char **pzVal);


/**************************************************************************
** Functions from file "lsm_shared.c".
*/

int lsmDbDatabaseConnect(lsm_db*, const char *);
void lsmDbDatabaseRelease(lsm_db *);

int lsmBeginReadTrans(lsm_db *);
int lsmBeginWriteTrans(lsm_db *);
int lsmBeginFlush(lsm_db *);

int lsmDetectRoTrans(lsm_db *db, int *);
int lsmBeginRoTrans(lsm_db *db);

int lsmBeginWork(lsm_db *);
void lsmFinishWork(lsm_db *, int, int *);

int lsmFinishRecovery(lsm_db *);
void lsmFinishReadTrans(lsm_db *);
int lsmFinishWriteTrans(lsm_db *, int);
int lsmFinishFlush(lsm_db *, int);

int lsmSnapshotSetFreelist(lsm_db *, int *, int);

Snapshot *lsmDbSnapshotClient(lsm_db *);
Snapshot *lsmDbSnapshotWorker(lsm_db *);

void lsmSnapshotSetCkptid(Snapshot *, i64);

Level *lsmDbSnapshotLevel(Snapshot *);
void lsmDbSnapshotSetLevel(Snapshot *, Level *);

void lsmDbRecoveryComplete(lsm_db *, int);

int lsmBlockAllocate(lsm_db *, int, int *);
int lsmBlockFree(lsm_db *, int);
int lsmBlockRefree(lsm_db *, int);

void lsmFreelistDeltaBegin(lsm_db *);
void lsmFreelistDeltaEnd(lsm_db *);
int lsmFreelistDelta(lsm_db *pDb);

DbLog *lsmDatabaseLog(lsm_db *pDb);

#ifdef LSM_DEBUG
  int lsmHoldingClientMutex(lsm_db *pDb);
  int lsmShmAssertLock(lsm_db *db, int iLock, int eOp);
  int lsmShmAssertWorker(lsm_db *db);
#endif

void lsmFreeSnapshot(lsm_env *, Snapshot *);


/* Candidate values for the 3rd argument to lsmShmLock() */
#define LSM_LOCK_UNLOCK 0
#define LSM_LOCK_SHARED 1
#define LSM_LOCK_EXCL   2

int lsmShmCacheChunks(lsm_db *db, int nChunk);
int lsmShmLock(lsm_db *db, int iLock, int eOp, int bBlock);
int lsmShmTestLock(lsm_db *db, int iLock, int nLock, int eOp);
void lsmShmBarrier(lsm_db *db);

#ifdef LSM_DEBUG
void lsmShmHasLock(lsm_db *db, int iLock, int eOp);
#else
# define lsmShmHasLock(x,y,z)
#endif

int lsmReadlock(lsm_db *, i64 iLsm, u32 iShmMin, u32 iShmMax);

int lsmLsmInUse(lsm_db *db, i64 iLsmId, int *pbInUse);
int lsmTreeInUse(lsm_db *db, u32 iLsmId, int *pbInUse);
int lsmFreelistAppend(lsm_env *pEnv, Freelist *p, int iBlk, i64 iId);

int lsmDbMultiProc(lsm_db *);
void lsmDbDeferredClose(lsm_db *, lsm_file *, LsmFile *);
LsmFile *lsmDbRecycleFd(lsm_db *);

int lsmWalkFreelist(lsm_db *, int, int (*)(void *, int, i64), void *);

int lsmCheckCompressionId(lsm_db *, u32);


/**************************************************************************
** functions in lsm_str.c
*/
void lsmStringInit(LsmString*, lsm_env *pEnv);
int lsmStringExtend(LsmString*, int);
int lsmStringAppend(LsmString*, const char *, int);
void lsmStringVAppendf(LsmString*, const char *zFormat, va_list, va_list);
void lsmStringAppendf(LsmString*, const char *zFormat, ...);
void lsmStringClear(LsmString*);
char *lsmMallocPrintf(lsm_env*, const char*, ...);
int lsmStringBinAppend(LsmString *pStr, const u8 *a, int n);

int lsmStrlen(const char *zName);



/* 
** Round up a number to the next larger multiple of 8.  This is used
** to force 8-byte alignment on 64-bit architectures.
*/
#define ROUND8(x)     (((x)+7)&~7)

#define LSM_MIN(x,y) ((x)>(y) ? (y) : (x))
#define LSM_MAX(x,y) ((x)>(y) ? (x) : (y))

#endif

Added ext/lsm1/lsm_ckpt.c.











































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
/*
** 2011-09-11
**
** The author disclaims copyright to this source code.  In place of
** a legal notice, here is a blessing:
**
**    May you do good and not evil.
**    May you find forgiveness for yourself and forgive others.
**    May you share freely, never taking more than you give.
**
*************************************************************************
**
** This file contains code to read and write checkpoints.
**
** A checkpoint represents the database layout at a single point in time.
** It includes a log offset. When an existing database is opened, the
** current state is determined by reading the newest checkpoint and updating
** it with all committed transactions from the log that follow the specified
** offset.
*/
#include "lsmInt.h"

/*
** CHECKPOINT BLOB FORMAT:
**
** A checkpoint blob is a series of unsigned 32-bit integers stored in
** big-endian byte order. As follows:
**
**   Checkpoint header (see the CKPT_HDR_XXX #defines):
**
**     1. The checkpoint id MSW.
**     2. The checkpoint id LSW.
**     3. The number of integer values in the entire checkpoint, including 
**        the two checksum values.
**     4. The compression scheme id.
**     5. The total number of blocks in the database.
**     6. The block size.
**     7. The number of levels.
**     8. The nominal database page size.
**     9. The number of pages (in total) written to the database file.
**
**   Log pointer:
**
**     1. The log offset MSW.
**     2. The log offset LSW.
**     3. Log checksum 0.
**     4. Log checksum 1.
**
**     Note that the "log offset" is not the literal byte offset. Instead,
**     it is the byte offset multiplied by 2, with least significant bit
**     toggled each time the log pointer value is changed. This is to make
**     sure that this field changes each time the log pointer is updated,
**     even if the log file itself is disabled. See lsmTreeMakeOld().
**
**     See ckptExportLog() and ckptImportLog().
**
**   Append points:
**
**     8 integers (4 * 64-bit page numbers). See ckptExportAppendlist().
**
**   For each level in the database, a level record. Formatted as follows:
**
**     0. Age of the level (least significant 16-bits). And flags mask (most
**        significant 16-bits).
**     1. The number of right-hand segments (nRight, possibly 0),
**     2. Segment record for left-hand segment (8 integers defined below),
**     3. Segment record for each right-hand segment (8 integers defined below),
**     4. If nRight>0, The number of segments involved in the merge
**     5. if nRight>0, Current nSkip value (see Merge structure defn.),
**     6. For each segment in the merge:
**        5a. Page number of next cell to read during merge (this field
**            is 64-bits - 2 integers)
**        5b. Cell number of next cell to read during merge
**     7. Page containing current split-key (64-bits - 2 integers).
**     8. Cell within page containing current split-key.
**     9. Current pointer value (64-bits - 2 integers).
**
**   The block redirect array:
**
**     1. Number of redirections (maximum LSM_MAX_BLOCK_REDIRECTS).
**     2. For each redirection:
**        a. "from" block number
**        b. "to" block number
**
**   The in-memory freelist entries. Each entry is either an insert or a
**   delete. The in-memory freelist is to the free-block-list as the
**   in-memory tree is to the users database content.
**
**     1. Number of free-list entries stored in checkpoint header.
**     2. Number of free blocks (in total).
**     3. Total number of blocks freed during database lifetime.
**     4. For each entry:
**        2a. Block number of free block.
**        2b. A 64-bit integer (MSW followed by LSW). -1 for a delete entry,
**            or the associated checkpoint id for an insert.
**
**   The checksum:
**
**     1. Checksum value 1.
**     2. Checksum value 2.
**
** In the above, a segment record consists of the following four 64-bit 
** fields (converted to 2 * u32 by storing the MSW followed by LSW):
**
**     1. First page of array,
**     2. Last page of array,
**     3. Root page of array (or 0),
**     4. Size of array in pages.
*/

/*
** LARGE NUMBERS OF LEVEL RECORDS:
**
** A limit on the number of rhs segments that may be present in the database
** file. Defining this limit ensures that all level records fit within
** the 4096 byte limit for checkpoint blobs.
**
** The number of right-hand-side segments in a database is counted as 
** follows:
**
**   * For each level in the database not undergoing a merge, add 1.
**
**   * For each level in the database that is undergoing a merge, add 
**     the number of segments on the rhs of the level.
**
** A level record not undergoing a merge is 10 integers. A level record 
** with nRhs rhs segments and (nRhs+1) input segments (i.e. including the 
** separators from the next level) is (11*nRhs+20) integers. The maximum
** per right-hand-side level is therefore 21 integers. So the maximum
** size of all level records in a checkpoint is 21*40=820 integers.
**
** TODO: Before pointer values were changed from 32 to 64 bits, the above
** used to come to 420 bytes - leaving significant space for a free-list
** prefix. No more. To fix this, reduce the size of the level records in
** a db snapshot, and improve management of the free-list tail in 
** lsm_sorted.c. 
*/
#define LSM_MAX_RHS_SEGMENTS 40

/*
** LARGE NUMBERS OF FREELIST ENTRIES:
**
** There is also a limit (LSM_MAX_FREELIST_ENTRIES - defined in lsmInt.h)
** on the number of free-list entries stored in a checkpoint. Since each 
** free-list entry consists of 3 integers, the maximum free-list size is 
** 3*100=300 integers. Combined with the limit on rhs segments defined
** above, this ensures that a checkpoint always fits within a 4096 byte
** meta page.
**
** If the database contains more than 100 free blocks, the "overflow" flag
** in the checkpoint header is set and the remainder are stored in the
** system FREELIST entry in the LSM (along with user data). The value
** accompanying the FREELIST key in the LSM is, like a checkpoint, an array
** of 32-bit big-endian integers. As follows:
**
**     For each entry:
**       a. Block number of free block.
**       b. MSW of associated checkpoint id.
**       c. LSW of associated checkpoint id.
**
** The number of entries is not required - it is implied by the size of the
** value blob containing the integer array.
**
** Note that the limit defined by LSM_MAX_FREELIST_ENTRIES is a hard limit.
** The actual value used may be configured using LSM_CONFIG_MAX_FREELIST.
*/

/*
** The argument to this macro must be of type u32. On a little-endian
** architecture, it returns the u32 value that results from interpreting
** the 4 bytes as a big-endian value. On a big-endian architecture, it
** returns the value that would be produced by intepreting the 4 bytes
** of the input value as a little-endian integer.
*/
#define BYTESWAP32(x) ( \
   (((x)&0x000000FF)<<24) + (((x)&0x0000FF00)<<8)  \
 + (((x)&0x00FF0000)>>8)  + (((x)&0xFF000000)>>24) \
)

static const int one = 1;
#define LSM_LITTLE_ENDIAN (*(u8 *)(&one))

/* Sizes, in integers, of various parts of the checkpoint. */
#define CKPT_HDR_SIZE         9
#define CKPT_LOGPTR_SIZE      4
#define CKPT_APPENDLIST_SIZE  (LSM_APPLIST_SZ * 2)

/* A #define to describe each integer in the checkpoint header. */
#define CKPT_HDR_ID_MSW   0
#define CKPT_HDR_ID_LSW   1
#define CKPT_HDR_NCKPT    2
#define CKPT_HDR_CMPID    3
#define CKPT_HDR_NBLOCK   4
#define CKPT_HDR_BLKSZ    5
#define CKPT_HDR_NLEVEL   6
#define CKPT_HDR_PGSZ     7
#define CKPT_HDR_NWRITE   8

#define CKPT_HDR_LO_MSW     9
#define CKPT_HDR_LO_LSW    10
#define CKPT_HDR_LO_CKSUM1 11
#define CKPT_HDR_LO_CKSUM2 12

typedef struct CkptBuffer CkptBuffer;

/*
** Dynamic buffer used to accumulate data for a checkpoint.
*/
struct CkptBuffer {
  lsm_env *pEnv;
  int nAlloc;
  u32 *aCkpt;
};

/*
** Calculate the checksum of the checkpoint specified by arguments aCkpt and
** nCkpt. Store the checksum in *piCksum1 and *piCksum2 before returning.
**
** The value of the nCkpt parameter includes the two checksum values at
** the end of the checkpoint. They are not used as inputs to the checksum 
** calculation. The checksum is based on the array of (nCkpt-2) integers
** at aCkpt[].
*/
static void ckptChecksum(u32 *aCkpt, u32 nCkpt, u32 *piCksum1, u32 *piCksum2){
  int i;
  u32 cksum1 = 1;
  u32 cksum2 = 2;

  if( nCkpt % 2 ){
    cksum1 += aCkpt[nCkpt-3] & 0x0000FFFF;
    cksum2 += aCkpt[nCkpt-3] & 0xFFFF0000;
  }

  for(i=0; (i+3)<nCkpt; i+=2){
    cksum1 += cksum2 + aCkpt[i];
    cksum2 += cksum1 + aCkpt[i+1];
  }

  *piCksum1 = cksum1;
  *piCksum2 = cksum2;
}

/*
** Set integer iIdx of the checkpoint accumulating in buffer *p to iVal.
*/
static void ckptSetValue(CkptBuffer *p, int iIdx, u32 iVal, int *pRc){
  if( *pRc ) return;
  if( iIdx>=p->nAlloc ){
    int nNew = LSM_MAX(8, iIdx*2);
    p->aCkpt = (u32 *)lsmReallocOrFree(p->pEnv, p->aCkpt, nNew*sizeof(u32));
    if( !p->aCkpt ){
      *pRc = LSM_NOMEM_BKPT;
      return;
    }
    p->nAlloc = nNew;
  }
  p->aCkpt[iIdx] = iVal;
}

/*
** Argument aInt points to an array nInt elements in size. Switch the 
** endian-ness of each element of the array.
*/
static void ckptChangeEndianness(u32 *aInt, int nInt){
  if( LSM_LITTLE_ENDIAN ){
    int i;
    for(i=0; i<nInt; i++) aInt[i] = BYTESWAP32(aInt[i]);
  }
}

/*
** Object *p contains a checkpoint in native byte-order. The checkpoint is
** nCkpt integers in size, not including any checksum. This function sets
** the two checksum elements of the checkpoint accordingly.
*/
static void ckptAddChecksum(CkptBuffer *p, int nCkpt, int *pRc){
  if( *pRc==LSM_OK ){
    u32 aCksum[2] = {0, 0};
    ckptChecksum(p->aCkpt, nCkpt+2, &aCksum[0], &aCksum[1]);
    ckptSetValue(p, nCkpt, aCksum[0], pRc);
    ckptSetValue(p, nCkpt+1, aCksum[1], pRc);
  }
}

static void ckptAppend64(CkptBuffer *p, int *piOut, i64 iVal, int *pRc){
  int iOut = *piOut;
  ckptSetValue(p, iOut++, (iVal >> 32) & 0xFFFFFFFF, pRc);
  ckptSetValue(p, iOut++, (iVal & 0xFFFFFFFF), pRc);
  *piOut = iOut;
}

static i64 ckptRead64(u32 *a){
  return (((i64)a[0]) << 32) + (i64)a[1];
}

static i64 ckptGobble64(u32 *a, int *piIn){
  int iIn = *piIn;
  *piIn += 2;
  return ckptRead64(&a[iIn]);
}


/*
** Append a 6-value segment record corresponding to pSeg to the checkpoint 
** buffer passed as the third argument.
*/
static void ckptExportSegment(
  Segment *pSeg, 
  CkptBuffer *p, 
  int *piOut, 
  int *pRc
){
  ckptAppend64(p, piOut, pSeg->iFirst, pRc);
  ckptAppend64(p, piOut, pSeg->iLastPg, pRc);
  ckptAppend64(p, piOut, pSeg->iRoot, pRc);
  ckptAppend64(p, piOut, pSeg->nSize, pRc);
}

static void ckptExportLevel(
  Level *pLevel,                  /* Level object to serialize */
  CkptBuffer *p,                  /* Append new level record to this ckpt */
  int *piOut,                     /* IN/OUT: Size of checkpoint so far */
  int *pRc                        /* IN/OUT: Error code */
){
  int iOut = *piOut;
  Merge *pMerge;

  pMerge = pLevel->pMerge;
  ckptSetValue(p, iOut++, (u32)pLevel->iAge + (u32)(pLevel->flags<<16), pRc);
  ckptSetValue(p, iOut++, pLevel->nRight, pRc);
  ckptExportSegment(&pLevel->lhs, p, &iOut, pRc);

  assert( (pLevel->nRight>0)==(pMerge!=0) );
  if( pMerge ){
    int i;
    for(i=0; i<pLevel->nRight; i++){
      ckptExportSegment(&pLevel->aRhs[i], p, &iOut, pRc);
    }
    assert( pMerge->nInput==pLevel->nRight 
         || pMerge->nInput==pLevel->nRight+1 
    );
    ckptSetValue(p, iOut++, pMerge->nInput, pRc);
    ckptSetValue(p, iOut++, pMerge->nSkip, pRc);
    for(i=0; i<pMerge->nInput; i++){
      ckptAppend64(p, &iOut, pMerge->aInput[i].iPg, pRc);
      ckptSetValue(p, iOut++, pMerge->aInput[i].iCell, pRc);
    }
    ckptAppend64(p, &iOut, pMerge->splitkey.iPg, pRc);
    ckptSetValue(p, iOut++, pMerge->splitkey.iCell, pRc);
    ckptAppend64(p, &iOut, pMerge->iCurrentPtr, pRc);
  }

  *piOut = iOut;
}

/*
** Populate the log offset fields of the checkpoint buffer. 4 values.
*/
static void ckptExportLog(
  lsm_db *pDb, 
  int bFlush,
  CkptBuffer *p, 
  int *piOut, 
  int *pRc
){
  int iOut = *piOut;

  assert( iOut==CKPT_HDR_LO_MSW );

  if( bFlush ){
    i64 iOff = pDb->treehdr.iOldLog;
    ckptAppend64(p, &iOut, iOff, pRc);
    ckptSetValue(p, iOut++, pDb->treehdr.oldcksum0, pRc);
    ckptSetValue(p, iOut++, pDb->treehdr.oldcksum1, pRc);
  }else{
    for(; iOut<=CKPT_HDR_LO_CKSUM2; iOut++){
      ckptSetValue(p, iOut, pDb->pShmhdr->aSnap2[iOut], pRc);
    }
  }

  assert( *pRc || iOut==CKPT_HDR_LO_CKSUM2+1 );
  *piOut = iOut;
}

static void ckptExportAppendlist(
  lsm_db *db,                     /* Database connection */
  CkptBuffer *p,                  /* Checkpoint buffer to write to */
  int *piOut,                     /* IN/OUT: Offset within checkpoint buffer */
  int *pRc                        /* IN/OUT: Error code */
){
  int i;
  Pgno *aiAppend = db->pWorker->aiAppend;

  for(i=0; i<LSM_APPLIST_SZ; i++){
    ckptAppend64(p, piOut, aiAppend[i], pRc);
  }
};

static int ckptExportSnapshot( 
  lsm_db *pDb,                    /* Connection handle */
  int bLog,                       /* True to update log-offset fields */
  i64 iId,                        /* Checkpoint id */
  int bCksum,                     /* If true, include checksums */
  void **ppCkpt,                  /* OUT: Buffer containing checkpoint */
  int *pnCkpt                     /* OUT: Size of checkpoint in bytes */
){
  int rc = LSM_OK;                /* Return Code */
  FileSystem *pFS = pDb->pFS;     /* File system object */
  Snapshot *pSnap = pDb->pWorker; /* Worker snapshot */
  int nLevel = 0;                 /* Number of levels in checkpoint */
  int iLevel;                     /* Used to count out nLevel levels */
  int iOut = 0;                   /* Current offset in aCkpt[] */
  Level *pLevel;                  /* Level iterator */
  int i;                          /* Iterator used while serializing freelist */
  CkptBuffer ckpt;

  /* Initialize the output buffer */
  memset(&ckpt, 0, sizeof(CkptBuffer));
  ckpt.pEnv = pDb->pEnv;
  iOut = CKPT_HDR_SIZE;

  /* Write the log offset into the checkpoint. */
  ckptExportLog(pDb, bLog, &ckpt, &iOut, &rc);

  /* Write the append-point list */
  ckptExportAppendlist(pDb, &ckpt, &iOut, &rc);

  /* Figure out how many levels will be written to the checkpoint. */
  for(pLevel=lsmDbSnapshotLevel(pSnap); pLevel; pLevel=pLevel->pNext) nLevel++;

  /* Serialize nLevel levels. */
  iLevel = 0;
  for(pLevel=lsmDbSnapshotLevel(pSnap); iLevel<nLevel; pLevel=pLevel->pNext){
    ckptExportLevel(pLevel, &ckpt, &iOut, &rc);
    iLevel++;
  }

  /* Write the block-redirect list */
  ckptSetValue(&ckpt, iOut++, pSnap->redirect.n, &rc);
  for(i=0; i<pSnap->redirect.n; i++){
    ckptSetValue(&ckpt, iOut++, pSnap->redirect.a[i].iFrom, &rc);
    ckptSetValue(&ckpt, iOut++, pSnap->redirect.a[i].iTo, &rc);
  }

  /* Write the freelist */
  assert( pSnap->freelist.nEntry<=pDb->nMaxFreelist );
  if( rc==LSM_OK ){
    int nFree = pSnap->freelist.nEntry;
    ckptSetValue(&ckpt, iOut++, nFree, &rc);
    for(i=0; i<nFree; i++){
      FreelistEntry *p = &pSnap->freelist.aEntry[i];
      ckptSetValue(&ckpt, iOut++, p->iBlk, &rc);
      ckptSetValue(&ckpt, iOut++, (p->iId >> 32) & 0xFFFFFFFF, &rc);
      ckptSetValue(&ckpt, iOut++, p->iId & 0xFFFFFFFF, &rc);
    }
  }

  /* Write the checkpoint header */
  assert( iId>=0 );
  assert( pSnap->iCmpId==pDb->compress.iId
       || pSnap->iCmpId==LSM_COMPRESSION_EMPTY 
  );
  ckptSetValue(&ckpt, CKPT_HDR_ID_MSW, (u32)(iId>>32), &rc);
  ckptSetValue(&ckpt, CKPT_HDR_ID_LSW, (u32)(iId&0xFFFFFFFF), &rc);
  ckptSetValue(&ckpt, CKPT_HDR_NCKPT, iOut+2, &rc);
  ckptSetValue(&ckpt, CKPT_HDR_CMPID, pDb->compress.iId, &rc);
  ckptSetValue(&ckpt, CKPT_HDR_NBLOCK, pSnap->nBlock, &rc);
  ckptSetValue(&ckpt, CKPT_HDR_BLKSZ, lsmFsBlockSize(pFS), &rc);
  ckptSetValue(&ckpt, CKPT_HDR_NLEVEL, nLevel, &rc);
  ckptSetValue(&ckpt, CKPT_HDR_PGSZ, lsmFsPageSize(pFS), &rc);
  ckptSetValue(&ckpt, CKPT_HDR_NWRITE, pSnap->nWrite, &rc);

  if( bCksum ){
    ckptAddChecksum(&ckpt, iOut, &rc);
  }else{
    ckptSetValue(&ckpt, iOut, 0, &rc);
    ckptSetValue(&ckpt, iOut+1, 0, &rc);
  }
  iOut += 2;
  assert( iOut<=1024 );

#ifdef LSM_LOG_FREELIST
  lsmLogMessage(pDb, rc, 
      "ckptExportSnapshot(): id=%lld freelist: %d", iId, pSnap->freelist.nEntry
  );
  for(i=0; i<pSnap->freelist.nEntry; i++){
  lsmLogMessage(pDb, rc, 
      "ckptExportSnapshot(): iBlk=%d id=%lld", 
      pSnap->freelist.aEntry[i].iBlk,
      pSnap->freelist.aEntry[i].iId
  );
  }
#endif

  *ppCkpt = (void *)ckpt.aCkpt;
  if( pnCkpt ) *pnCkpt = sizeof(u32)*iOut;
  return rc;
}


/*
** Helper function for ckptImport().
*/
static void ckptNewSegment(
  u32 *aIn,
  int *piIn,
  Segment *pSegment               /* Populate this structure */
){
  assert( pSegment->iFirst==0 && pSegment->iLastPg==0 );
  assert( pSegment->nSize==0 && pSegment->iRoot==0 );
  pSegment->iFirst = ckptGobble64(aIn, piIn);
  pSegment->iLastPg = ckptGobble64(aIn, piIn);
  pSegment->iRoot = ckptGobble64(aIn, piIn);
  pSegment->nSize = ckptGobble64(aIn, piIn);
  assert( pSegment->iFirst );
}

static int ckptSetupMerge(lsm_db *pDb, u32 *aInt, int *piIn, Level *pLevel){
  Merge *pMerge;                  /* Allocated Merge object */
  int nInput;                     /* Number of input segments in merge */
  int iIn = *piIn;                /* Next value to read from aInt[] */
  int i;                          /* Iterator variable */
  int nByte;                      /* Number of bytes to allocate */

  /* Allocate the Merge object. If malloc() fails, return LSM_NOMEM. */
  nInput = (int)aInt[iIn++];
  nByte = sizeof(Merge) + sizeof(MergeInput) * nInput;
  pMerge = (Merge *)lsmMallocZero(pDb->pEnv, nByte);
  if( !pMerge ) return LSM_NOMEM_BKPT;
  pLevel->pMerge = pMerge;

  /* Populate the Merge object. */
  pMerge->aInput = (MergeInput *)&pMerge[1];
  pMerge->nInput = nInput;
  pMerge->iOutputOff = -1;
  pMerge->nSkip = (int)aInt[iIn++];
  for(i=0; i<nInput; i++){
    pMerge->aInput[i].iPg = ckptGobble64(aInt, &iIn);
    pMerge->aInput[i].iCell = (int)aInt[iIn++];
  }
  pMerge->splitkey.iPg = ckptGobble64(aInt, &iIn);
  pMerge->splitkey.iCell = (int)aInt[iIn++];
  pMerge->iCurrentPtr = ckptGobble64(aInt, &iIn);

  /* Set *piIn and return LSM_OK. */
  *piIn = iIn;
  return LSM_OK;
}


static int ckptLoadLevels(
  lsm_db *pDb,
  u32 *aIn, 
  int *piIn, 
  int nLevel,
  Level **ppLevel
){
  int i;
  int rc = LSM_OK;
  Level *pRet = 0;
  Level **ppNext;
  int iIn = *piIn;

  ppNext = &pRet;
  for(i=0; rc==LSM_OK && i<nLevel; i++){
    int iRight;
    Level *pLevel;

    /* Allocate space for the Level structure and Level.apRight[] array */
    pLevel = (Level *)lsmMallocZeroRc(pDb->pEnv, sizeof(Level), &rc);
    if( rc==LSM_OK ){
      pLevel->iAge = (u16)(aIn[iIn] & 0x0000FFFF);
      pLevel->flags = (u16)((aIn[iIn]>>16) & 0x0000FFFF);
      iIn++;
      pLevel->nRight = aIn[iIn++];
      if( pLevel->nRight ){
        int nByte = sizeof(Segment) * pLevel->nRight;
        pLevel->aRhs = (Segment *)lsmMallocZeroRc(pDb->pEnv, nByte, &rc);
      }
      if( rc==LSM_OK ){
        *ppNext = pLevel;
        ppNext = &pLevel->pNext;

        /* Allocate the main segment */
        ckptNewSegment(aIn, &iIn, &pLevel->lhs);

        /* Allocate each of the right-hand segments, if any */
        for(iRight=0; iRight<pLevel->nRight; iRight++){
          ckptNewSegment(aIn, &iIn, &pLevel->aRhs[iRight]);
        }

        /* Set up the Merge object, if required */
        if( pLevel->nRight>0 ){
          rc = ckptSetupMerge(pDb, aIn, &iIn, pLevel);
        }
      }
    }
  }

  if( rc!=LSM_OK ){
    /* An OOM must have occurred. Free any level structures allocated and
    ** return the error to the caller. */
    lsmSortedFreeLevel(pDb->pEnv, pRet);
    pRet = 0;
  }

  *ppLevel = pRet;
  *piIn = iIn;
  return rc;
}


int lsmCheckpointLoadLevels(lsm_db *pDb, void *pVal, int nVal){
  int rc = LSM_OK;
  if( nVal>0 ){
    u32 *aIn;

    aIn = lsmMallocRc(pDb->pEnv, nVal, &rc);
    if( aIn ){
      Level *pLevel = 0;
      Level *pParent;

      int nIn;
      int nLevel;
      int iIn = 1;
      memcpy(aIn, pVal, nVal);
      nIn = nVal / sizeof(u32);

      ckptChangeEndianness(aIn, nIn);
      nLevel = aIn[0];
      rc = ckptLoadLevels(pDb, aIn, &iIn, nLevel, &pLevel);
      lsmFree(pDb->pEnv, aIn);
      assert( rc==LSM_OK || pLevel==0 );
      if( rc==LSM_OK ){
        pParent = lsmDbSnapshotLevel(pDb->pWorker);
        assert( pParent );
        while( pParent->pNext ) pParent = pParent->pNext;
        pParent->pNext = pLevel;
      }
    }
  }

  return rc;
}

/*
** Return the data for the LEVELS record.
**
** The size of the checkpoint that can be stored in the database header
** must not exceed 1024 32-bit integers. Normally, it does not. However,
** if it does, part of the checkpoint must be stored in the LSM. This
** routine returns that part.
*/
int lsmCheckpointLevels(
  lsm_db *pDb,                    /* Database handle */
  int nLevel,                     /* Number of levels to write to blob */
  void **paVal,                   /* OUT: Pointer to LEVELS blob */
  int *pnVal                      /* OUT: Size of LEVELS blob in bytes */
){
  Level *p;                       /* Used to iterate through levels */
  int nAll= 0;
  int rc;
  int i;
  int iOut;
  CkptBuffer ckpt;
  assert( nLevel>0 );

  for(p=lsmDbSnapshotLevel(pDb->pWorker); p; p=p->pNext) nAll++;

  assert( nAll>nLevel );
  nAll -= nLevel;
  for(p=lsmDbSnapshotLevel(pDb->pWorker); p && nAll>0; p=p->pNext) nAll--;

  memset(&ckpt, 0, sizeof(CkptBuffer));
  ckpt.pEnv = pDb->pEnv;

  ckptSetValue(&ckpt, 0, nLevel, &rc);
  iOut = 1;
  for(i=0; rc==LSM_OK && i<nLevel; i++){
    ckptExportLevel(p, &ckpt, &iOut, &rc);
    p = p->pNext;
  }
  assert( rc!=LSM_OK || p==0 );

  if( rc==LSM_OK ){
    ckptChangeEndianness(ckpt.aCkpt, iOut);
    *paVal = (void *)ckpt.aCkpt;
    *pnVal = iOut * sizeof(u32);
  }else{
    *pnVal = 0;
    *paVal = 0;
  }

  return rc;
}

/*
** Read the checkpoint id from meta-page pPg.
*/
static i64 ckptLoadId(MetaPage *pPg){
  i64 ret = 0;
  if( pPg ){
    int nData;
    u8 *aData = lsmFsMetaPageData(pPg, &nData);
    ret = (((i64)lsmGetU32(&aData[CKPT_HDR_ID_MSW*4])) << 32) + 
          ((i64)lsmGetU32(&aData[CKPT_HDR_ID_LSW*4]));
  }
  return ret;
}

/*
** Return true if the buffer passed as an argument contains a valid
** checkpoint.
*/
static int ckptChecksumOk(u32 *aCkpt){
  u32 nCkpt = aCkpt[CKPT_HDR_NCKPT];
  u32 cksum1;
  u32 cksum2;

  if( nCkpt<CKPT_HDR_NCKPT || nCkpt>(LSM_META_PAGE_SIZE)/sizeof(u32) ) return 0;
  ckptChecksum(aCkpt, nCkpt, &cksum1, &cksum2);
  return (cksum1==aCkpt[nCkpt-2] && cksum2==aCkpt[nCkpt-1]);
}

/*
** Attempt to load a checkpoint from meta page iMeta.
**
** This function is a no-op if *pRc is set to any value other than LSM_OK
** when it is called. If an error occurs, *pRc is set to an LSM error code
** before returning.
**
** If no error occurs and the checkpoint is successfully loaded, copy it to
** ShmHeader.aSnap1[] and ShmHeader.aSnap2[], and set ShmHeader.iMetaPage 
** to indicate its origin. In this case return 1. Or, if the checkpoint 
** cannot be loaded (because the checksum does not compute), return 0.
*/
static int ckptTryLoad(lsm_db *pDb, MetaPage *pPg, u32 iMeta, int *pRc){
  int bLoaded = 0;                /* Return value */
  if( *pRc==LSM_OK ){
    int rc = LSM_OK;              /* Error code */
    u32 *aCkpt = 0;               /* Pointer to buffer containing checkpoint */
    u32 nCkpt;                    /* Number of elements in aCkpt[] */
    int nData;                    /* Bytes of data in aData[] */
    u8 *aData;                    /* Meta page data */
   
    aData = lsmFsMetaPageData(pPg, &nData);
    nCkpt = (u32)lsmGetU32(&aData[CKPT_HDR_NCKPT*sizeof(u32)]);
    if( nCkpt<=nData/sizeof(u32) && nCkpt>CKPT_HDR_NCKPT ){
      aCkpt = (u32 *)lsmMallocRc(pDb->pEnv, nCkpt*sizeof(u32), &rc);
    }
    if( aCkpt ){
      memcpy(aCkpt, aData, nCkpt*sizeof(u32));
      ckptChangeEndianness(aCkpt, nCkpt);
      if( ckptChecksumOk(aCkpt) ){
        ShmHeader *pShm = pDb->pShmhdr;
        memcpy(pShm->aSnap1, aCkpt, nCkpt*sizeof(u32));
        memcpy(pShm->aSnap2, aCkpt, nCkpt*sizeof(u32));
        memcpy(pDb->aSnapshot, aCkpt, nCkpt*sizeof(u32));
        pShm->iMetaPage = iMeta;
        bLoaded = 1;
      }
    }

    lsmFree(pDb->pEnv, aCkpt);
    *pRc = rc;
  }
  return bLoaded;
}

/*
** Initialize the shared-memory header with an empty snapshot. This function
** is called when no valid snapshot can be found in the database header.
*/
static void ckptLoadEmpty(lsm_db *pDb){
  u32 aCkpt[] = {
    0,                       /* CKPT_HDR_ID_MSW */
    10,                      /* CKPT_HDR_ID_LSW */
    0,                       /* CKPT_HDR_NCKPT */
    LSM_COMPRESSION_EMPTY,   /* CKPT_HDR_CMPID */
    0,                       /* CKPT_HDR_NBLOCK */
    0,                       /* CKPT_HDR_BLKSZ */
    0,                       /* CKPT_HDR_NLEVEL */
    0,                       /* CKPT_HDR_PGSZ */
    0,                       /* CKPT_HDR_NWRITE */
    0, 0, 1234, 5678,        /* The log pointer and initial checksum */
    0,0,0,0, 0,0,0,0,        /* The append list */
    0,                       /* The redirected block list */
    0,                       /* The free block list */
    0, 0                     /* Space for checksum values */
  };
  u32 nCkpt = array_size(aCkpt);
  ShmHeader *pShm = pDb->pShmhdr;

  aCkpt[CKPT_HDR_NCKPT] = nCkpt;
  aCkpt[CKPT_HDR_BLKSZ] = pDb->nDfltBlksz;
  aCkpt[CKPT_HDR_PGSZ] = pDb->nDfltPgsz;
  ckptChecksum(aCkpt, array_size(aCkpt), &aCkpt[nCkpt-2], &aCkpt[nCkpt-1]);

  memcpy(pShm->aSnap1, aCkpt, nCkpt*sizeof(u32));
  memcpy(pShm->aSnap2, aCkpt, nCkpt*sizeof(u32));
  memcpy(pDb->aSnapshot, aCkpt, nCkpt*sizeof(u32));
}

/*
** This function is called as part of database recovery to initialize the
** ShmHeader.aSnap1[] and ShmHeader.aSnap2[] snapshots.
*/
int lsmCheckpointRecover(lsm_db *pDb){
  int rc = LSM_OK;                /* Return Code */
  i64 iId1;                       /* Id of checkpoint on meta-page 1 */
  i64 iId2;                       /* Id of checkpoint on meta-page 2 */
  int bLoaded = 0;                /* True once checkpoint has been loaded */
  int cmp;                        /* True if (iId2>iId1) */
  MetaPage *apPg[2] = {0, 0};     /* Meta-pages 1 and 2 */

  rc = lsmFsMetaPageGet(pDb->pFS, 0, 1, &apPg[0]);
  if( rc==LSM_OK ) rc = lsmFsMetaPageGet(pDb->pFS, 0, 2, &apPg[1]);

  iId1 = ckptLoadId(apPg[0]);
  iId2 = ckptLoadId(apPg[1]);
  cmp = (iId2 > iId1);
  bLoaded = ckptTryLoad(pDb, apPg[cmp?1:0], (cmp?2:1), &rc);
  if( bLoaded==0 ){
    bLoaded = ckptTryLoad(pDb, apPg[cmp?0:1], (cmp?1:2), &rc);
  }

  /* The database does not contain a valid checkpoint. Initialize the shared
  ** memory header with an empty checkpoint.  */
  if( bLoaded==0 ){
    ckptLoadEmpty(pDb);
  }

  lsmFsMetaPageRelease(apPg[0]);
  lsmFsMetaPageRelease(apPg[1]);

  return rc;
}

/* 
** Store the snapshot in pDb->aSnapshot[] in meta-page iMeta.
*/
int lsmCheckpointStore(lsm_db *pDb, int iMeta){
  MetaPage *pPg = 0;
  int rc;

  assert( iMeta==1 || iMeta==2 );
  rc = lsmFsMetaPageGet(pDb->pFS, 1, iMeta, &pPg);
  if( rc==LSM_OK ){
    u8 *aData;
    int nData;
    int nCkpt;

    nCkpt = (int)pDb->aSnapshot[CKPT_HDR_NCKPT];
    aData = lsmFsMetaPageData(pPg, &nData);
    memcpy(aData, pDb->aSnapshot, nCkpt*sizeof(u32));
    ckptChangeEndianness((u32 *)aData, nCkpt);
    rc = lsmFsMetaPageRelease(pPg);
  }
      
  return rc;
}

/*
** Copy the current client snapshot from shared-memory to pDb->aSnapshot[].
*/
int lsmCheckpointLoad(lsm_db *pDb, int *piRead){
  int nRem = LSM_ATTEMPTS_BEFORE_PROTOCOL;
  ShmHeader *pShm = pDb->pShmhdr;
  while( (nRem--)>0 ){
    int nInt;

    nInt = pShm->aSnap1[CKPT_HDR_NCKPT];
    if( nInt<=(LSM_META_PAGE_SIZE / sizeof(u32)) ){
      memcpy(pDb->aSnapshot, pShm->aSnap1, nInt*sizeof(u32));
      if( ckptChecksumOk(pDb->aSnapshot) ){
        if( piRead ) *piRead = 1;
        return LSM_OK;
      }
    }

    nInt = pShm->aSnap2[CKPT_HDR_NCKPT];
    if( nInt<=(LSM_META_PAGE_SIZE / sizeof(u32)) ){
      memcpy(pDb->aSnapshot, pShm->aSnap2, nInt*sizeof(u32));
      if( ckptChecksumOk(pDb->aSnapshot) ){
        if( piRead ) *piRead = 2;
        return LSM_OK;
      }
    }

    lsmShmBarrier(pDb);
  }
  return LSM_PROTOCOL_BKPT;
}

int lsmInfoCompressionId(lsm_db *db, u32 *piCmpId){
  int rc;

  assert( db->pClient==0 && db->pWorker==0 );
  rc = lsmCheckpointLoad(db, 0);
  if( rc==LSM_OK ){
    *piCmpId = db->aSnapshot[CKPT_HDR_CMPID];
  }

  return rc;
}

int lsmCheckpointLoadOk(lsm_db *pDb, int iSnap){
  u32 *aShm;
  assert( iSnap==1 || iSnap==2 );
  aShm = (iSnap==1) ? pDb->pShmhdr->aSnap1 : pDb->pShmhdr->aSnap2;
  return (lsmCheckpointId(pDb->aSnapshot, 0)==lsmCheckpointId(aShm, 0) );
}

int lsmCheckpointClientCacheOk(lsm_db *pDb){
  return ( pDb->pClient 
        && pDb->pClient->iId==lsmCheckpointId(pDb->aSnapshot, 0)
        && pDb->pClient->iId==lsmCheckpointId(pDb->pShmhdr->aSnap1, 0)
        && pDb->pClient->iId==lsmCheckpointId(pDb->pShmhdr->aSnap2, 0)
  );
}

int lsmCheckpointLoadWorker(lsm_db *pDb){
  int rc;
  ShmHeader *pShm = pDb->pShmhdr;
  int nInt1;
  int nInt2;

  /* Must be holding the WORKER lock to do this. Or DMS2. */
  assert( 
      lsmShmAssertLock(pDb, LSM_LOCK_WORKER, LSM_LOCK_EXCL) 
   || lsmShmAssertLock(pDb, LSM_LOCK_DMS1, LSM_LOCK_EXCL) 
  );

  /* Check that the two snapshots match. If not, repair them. */
  nInt1 = pShm->aSnap1[CKPT_HDR_NCKPT];
  nInt2 = pShm->aSnap2[CKPT_HDR_NCKPT];
  if( nInt1!=nInt2 || memcmp(pShm->aSnap1, pShm->aSnap2, nInt2*sizeof(u32)) ){
    if( ckptChecksumOk(pShm->aSnap1) ){
      memcpy(pShm->aSnap2, pShm->aSnap1, sizeof(u32)*nInt1);
    }else if( ckptChecksumOk(pShm->aSnap2) ){
      memcpy(pShm->aSnap1, pShm->aSnap2, sizeof(u32)*nInt2);
    }else{
      return LSM_PROTOCOL_BKPT;
    }
  }

  rc = lsmCheckpointDeserialize(pDb, 1, pShm->aSnap1, &pDb->pWorker);
  if( pDb->pWorker ) pDb->pWorker->pDatabase = pDb->pDatabase;

  if( rc==LSM_OK ){
    rc = lsmCheckCompressionId(pDb, pDb->pWorker->iCmpId);
  }

#if 0
  assert( rc!=LSM_OK || lsmFsIntegrityCheck(pDb) );
#endif
  return rc;
}

int lsmCheckpointDeserialize(
  lsm_db *pDb, 
  int bInclFreelist,              /* If true, deserialize free-list */
  u32 *aCkpt, 
  Snapshot **ppSnap
){
  int rc = LSM_OK;
  Snapshot *pNew;

  pNew = (Snapshot *)lsmMallocZeroRc(pDb->pEnv, sizeof(Snapshot), &rc);
  if( rc==LSM_OK ){
    Level *pLvl;
    int nFree;
    int i;
    int nLevel = (int)aCkpt[CKPT_HDR_NLEVEL];
    int iIn = CKPT_HDR_SIZE + CKPT_APPENDLIST_SIZE + CKPT_LOGPTR_SIZE;

    pNew->iId = lsmCheckpointId(aCkpt, 0);
    pNew->nBlock = aCkpt[CKPT_HDR_NBLOCK];
    pNew->nWrite = aCkpt[CKPT_HDR_NWRITE];
    rc = ckptLoadLevels(pDb, aCkpt, &iIn, nLevel, &pNew->pLevel);
    pNew->iLogOff = lsmCheckpointLogOffset(aCkpt);
    pNew->iCmpId = aCkpt[CKPT_HDR_CMPID];

    /* Make a copy of the append-list */
    for(i=0; i<LSM_APPLIST_SZ; i++){
      u32 *a = &aCkpt[CKPT_HDR_SIZE + CKPT_LOGPTR_SIZE + i*2];
      pNew->aiAppend[i] = ckptRead64(a);
    }

    /* Read the block-redirect list */
    pNew->redirect.n = aCkpt[iIn++];
    if( pNew->redirect.n ){
      pNew->redirect.a = lsmMallocZeroRc(pDb->pEnv, 
          (sizeof(struct RedirectEntry) * LSM_MAX_BLOCK_REDIRECTS), &rc
      );
      if( rc==LSM_OK ){
        for(i=0; i<pNew->redirect.n; i++){
          pNew->redirect.a[i].iFrom = aCkpt[iIn++];
          pNew->redirect.a[i].iTo = aCkpt[iIn++];
        }
      }
      for(pLvl=pNew->pLevel; pLvl->pNext; pLvl=pLvl->pNext);
      if( pLvl->nRight ){
        pLvl->aRhs[pLvl->nRight-1].pRedirect = &pNew->redirect;
      }else{
        pLvl->lhs.pRedirect = &pNew->redirect;
      }
    }

    /* Copy the free-list */
    if( rc==LSM_OK && bInclFreelist ){
      nFree = aCkpt[iIn++];
      if( nFree ){
        pNew->freelist.aEntry = (FreelistEntry *)lsmMallocZeroRc(
            pDb->pEnv, sizeof(FreelistEntry)*nFree, &rc
        );
        if( rc==LSM_OK ){
          int i;
          for(i=0; i<nFree; i++){
            FreelistEntry *p = &pNew->freelist.aEntry[i];
            p->iBlk = aCkpt[iIn++];
            p->iId = ((i64)(aCkpt[iIn])<<32) + aCkpt[iIn+1];
            iIn += 2;
          }
          pNew->freelist.nEntry = pNew->freelist.nAlloc = nFree;
        }
      }
    }
  }

  if( rc!=LSM_OK ){
    lsmFreeSnapshot(pDb->pEnv, pNew);
    pNew = 0;
  }

  *ppSnap = pNew;
  return rc;
}

/*
** Connection pDb must be the worker connection in order to call this
** function. It returns true if the database already contains the maximum
** number of levels or false otherwise.
**
** This is used when flushing the in-memory tree to disk. If the database
** is already full, then the caller should invoke lsm_work() or similar
** until it is not full before creating a new level by flushing the in-memory
** tree to disk. Limiting the number of levels in the database ensures that
** the records describing them always fit within the checkpoint blob.
*/
int lsmDatabaseFull(lsm_db *pDb){
  Level *p;
  int nRhs = 0;

  assert( lsmShmAssertLock(pDb, LSM_LOCK_WORKER, LSM_LOCK_EXCL) );
  assert( pDb->pWorker );

  for(p=pDb->pWorker->pLevel; p; p=p->pNext){
    nRhs += (p->nRight ? p->nRight : 1);
  }

  return (nRhs >= LSM_MAX_RHS_SEGMENTS);
}

/*
** The connection passed as the only argument is currently the worker
** connection. Some work has been performed on the database by the connection,
** but no new snapshot has been written into shared memory.
**
** This function updates the shared-memory worker and client snapshots with
** the new snapshot produced by the work performed by pDb.
**
** If successful, LSM_OK is returned. Otherwise, if an error occurs, an LSM
** error code is returned.
*/
int lsmCheckpointSaveWorker(lsm_db *pDb, int bFlush){
  Snapshot *pSnap = pDb->pWorker;
  ShmHeader *pShm = pDb->pShmhdr;
  void *p = 0;
  int n = 0;
  int rc;

  pSnap->iId++;
  rc = ckptExportSnapshot(pDb, bFlush, pSnap->iId, 1, &p, &n);
  if( rc!=LSM_OK ) return rc;
  assert( ckptChecksumOk((u32 *)p) );

  assert( n<=LSM_META_PAGE_SIZE );
  memcpy(pShm->aSnap2, p, n);
  lsmShmBarrier(pDb);
  memcpy(pShm->aSnap1, p, n);
  lsmFree(pDb->pEnv, p);

  assert( lsmFsIntegrityCheck(pDb) );
  return LSM_OK;
}

/*
** This function is used to determine the snapshot-id of the most recently
** checkpointed snapshot. Variable ShmHeader.iMetaPage indicates which of
** the two meta-pages said snapshot resides on (if any). 
**
** If successful, this function loads the snapshot from the meta-page, 
** verifies its checksum and sets *piId to the snapshot-id before returning
** LSM_OK. Or, if the checksum attempt fails, *piId is set to zero and
** LSM_OK returned. If an error occurs, an LSM error code is returned and
** the final value of *piId is undefined.
*/
int lsmCheckpointSynced(lsm_db *pDb, i64 *piId, i64 *piLog, u32 *pnWrite){
  int rc = LSM_OK;
  MetaPage *pPg;
  u32 iMeta;

  iMeta = pDb->pShmhdr->iMetaPage;
  if( iMeta==1 || iMeta==2 ){
    rc = lsmFsMetaPageGet(pDb->pFS, 0, iMeta, &pPg);
    if( rc==LSM_OK ){
      int nCkpt;
      int nData;
      u8 *aData; 

      aData = lsmFsMetaPageData(pPg, &nData);
      assert( nData==LSM_META_PAGE_SIZE );
      nCkpt = lsmGetU32(&aData[CKPT_HDR_NCKPT*sizeof(u32)]);
      if( nCkpt<(LSM_META_PAGE_SIZE/sizeof(u32)) ){
        u32 *aCopy = lsmMallocRc(pDb->pEnv, sizeof(u32) * nCkpt, &rc);
        if( aCopy ){
          memcpy(aCopy, aData, nCkpt*sizeof(u32));
          ckptChangeEndianness(aCopy, nCkpt);
          if( ckptChecksumOk(aCopy) ){
            if( piId ) *piId = lsmCheckpointId(aCopy, 0);
            if( piLog ) *piLog = (lsmCheckpointLogOffset(aCopy) >> 1);
            if( pnWrite ) *pnWrite = aCopy[CKPT_HDR_NWRITE];
          }
          lsmFree(pDb->pEnv, aCopy);
        }
      }
      lsmFsMetaPageRelease(pPg);
    }
  }

  if( (iMeta!=1 && iMeta!=2) || rc!=LSM_OK || pDb->pShmhdr->iMetaPage!=iMeta ){
    if( piId ) *piId = 0;
    if( piLog ) *piLog = 0;
    if( pnWrite ) *pnWrite = 0;
  }
  return rc;
}

/*
** Return the checkpoint-id of the checkpoint array passed as the first
** argument to this function. If the second argument is true, then assume
** that the checkpoint is made up of 32-bit big-endian integers. If it
** is false, assume that the integers are in machine byte order.
*/
i64 lsmCheckpointId(u32 *aCkpt, int bDisk){
  i64 iId;
  if( bDisk ){
    u8 *aData = (u8 *)aCkpt;
    iId = (((i64)lsmGetU32(&aData[CKPT_HDR_ID_MSW*4])) << 32);
    iId += ((i64)lsmGetU32(&aData[CKPT_HDR_ID_LSW*4]));
  }else{
    iId = ((i64)aCkpt[CKPT_HDR_ID_MSW] << 32) + (i64)aCkpt[CKPT_HDR_ID_LSW];
  }
  return iId;
}

u32 lsmCheckpointNBlock(u32 *aCkpt){
  return aCkpt[CKPT_HDR_NBLOCK];
}

u32 lsmCheckpointNWrite(u32 *aCkpt, int bDisk){
  if( bDisk ){
    return lsmGetU32((u8 *)&aCkpt[CKPT_HDR_NWRITE]);
  }else{
    return aCkpt[CKPT_HDR_NWRITE];
  }
}

i64 lsmCheckpointLogOffset(u32 *aCkpt){
  return ((i64)aCkpt[CKPT_HDR_LO_MSW] << 32) + (i64)aCkpt[CKPT_HDR_LO_LSW];
}

int lsmCheckpointPgsz(u32 *aCkpt){ return (int)aCkpt[CKPT_HDR_PGSZ]; }

int lsmCheckpointBlksz(u32 *aCkpt){ return (int)aCkpt[CKPT_HDR_BLKSZ]; }

void lsmCheckpointLogoffset(
  u32 *aCkpt,
  DbLog *pLog
){ 
  pLog->aRegion[2].iStart = (lsmCheckpointLogOffset(aCkpt) >> 1);

  pLog->cksum0 = aCkpt[CKPT_HDR_LO_CKSUM1];
  pLog->cksum1 = aCkpt[CKPT_HDR_LO_CKSUM2];
  pLog->iSnapshotId = lsmCheckpointId(aCkpt, 0);
}

void lsmCheckpointZeroLogoffset(lsm_db *pDb){
  u32 nCkpt;

  nCkpt = pDb->aSnapshot[CKPT_HDR_NCKPT];
  assert( nCkpt>CKPT_HDR_NCKPT );
  assert( nCkpt==pDb->pShmhdr->aSnap1[CKPT_HDR_NCKPT] );
  assert( 0==memcmp(pDb->aSnapshot, pDb->pShmhdr->aSnap1, nCkpt*sizeof(u32)) );
  assert( 0==memcmp(pDb->aSnapshot, pDb->pShmhdr->aSnap2, nCkpt*sizeof(u32)) );

  pDb->aSnapshot[CKPT_HDR_LO_MSW] = 0;
  pDb->aSnapshot[CKPT_HDR_LO_LSW] = 0;
  ckptChecksum(pDb->aSnapshot, nCkpt, 
      &pDb->aSnapshot[nCkpt-2], &pDb->aSnapshot[nCkpt-1]
  );

  memcpy(pDb->pShmhdr->aSnap1, pDb->aSnapshot, nCkpt*sizeof(u32));
  memcpy(pDb->pShmhdr->aSnap2, pDb->aSnapshot, nCkpt*sizeof(u32));
}

/*
** Set the output variable to the number of KB of data written into the
** database file since the most recent checkpoint.
*/
int lsmCheckpointSize(lsm_db *db, int *pnKB){
  int rc = LSM_OK;
  u32 nSynced;

  /* Set nSynced to the number of pages that had been written when the 
  ** database was last checkpointed. */
  rc = lsmCheckpointSynced(db, 0, 0, &nSynced);

  if( rc==LSM_OK ){
    u32 nPgsz = db->pShmhdr->aSnap1[CKPT_HDR_PGSZ];
    u32 nWrite = db->pShmhdr->aSnap1[CKPT_HDR_NWRITE];
    *pnKB = (int)(( ((i64)(nWrite - nSynced) * nPgsz) + 1023) / 1024);
  }

  return rc;
}

Added ext/lsm1/lsm_file.c.

























































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
1388
1389
1390
1391
1392
1393
1394
1395
1396
1397
1398
1399
1400
1401
1402
1403
1404
1405
1406
1407
1408
1409
1410
1411
1412
1413
1414
1415
1416
1417
1418
1419
1420
1421
1422
1423
1424
1425
1426
1427
1428
1429
1430
1431
1432
1433
1434
1435
1436
1437
1438
1439
1440
1441
1442
1443
1444
1445
1446
1447
1448
1449
1450
1451
1452
1453
1454
1455
1456
1457
1458
1459
1460
1461
1462
1463
1464
1465
1466
1467
1468
1469
1470
1471
1472
1473
1474
1475
1476
1477
1478
1479
1480
1481
1482
1483
1484
1485
1486
1487
1488
1489
1490
1491
1492
1493
1494
1495
1496
1497
1498
1499
1500
1501
1502
1503
1504
1505
1506
1507
1508
1509
1510
1511
1512
1513
1514
1515
1516
1517
1518
1519
1520
1521
1522
1523
1524
1525
1526
1527
1528
1529
1530
1531
1532
1533
1534
1535
1536
1537
1538
1539
1540
1541
1542
1543
1544
1545
1546
1547
1548
1549
1550
1551
1552
1553
1554
1555
1556
1557
1558
1559
1560
1561
1562
1563
1564
1565
1566
1567
1568
1569
1570
1571
1572
1573
1574
1575
1576
1577
1578
1579
1580
1581
1582
1583
1584
1585
1586
1587
1588
1589
1590
1591
1592
1593
1594
1595
1596
1597
1598
1599
1600
1601
1602
1603
1604
1605
1606
1607
1608
1609
1610
1611
1612
1613
1614
1615
1616
1617
1618
1619
1620
1621
1622
1623
1624
1625
1626
1627
1628
1629
1630
1631
1632
1633
1634
1635
1636
1637
1638
1639
1640
1641
1642
1643
1644
1645
1646
1647
1648
1649
1650
1651
1652
1653
1654
1655
1656
1657
1658
1659
1660
1661
1662
1663
1664
1665
1666
1667
1668
1669
1670
1671
1672
1673
1674
1675
1676
1677
1678
1679
1680
1681
1682
1683
1684
1685
1686
1687
1688
1689
1690
1691
1692
1693
1694
1695
1696
1697
1698
1699
1700
1701
1702
1703
1704
1705
1706
1707
1708
1709
1710
1711
1712
1713
1714
1715
1716
1717
1718
1719
1720
1721
1722
1723
1724
1725
1726
1727
1728
1729
1730
1731
1732
1733
1734
1735
1736
1737
1738
1739
1740
1741
1742
1743
1744
1745
1746
1747
1748
1749
1750
1751
1752
1753
1754
1755
1756
1757
1758
1759
1760
1761
1762
1763
1764
1765
1766
1767
1768
1769
1770
1771
1772
1773
1774
1775
1776
1777
1778
1779
1780
1781
1782
1783
1784
1785
1786
1787
1788
1789
1790
1791
1792
1793
1794
1795
1796
1797
1798
1799
1800
1801
1802
1803
1804
1805
1806
1807
1808
1809
1810
1811
1812
1813
1814
1815
1816
1817
1818
1819
1820
1821
1822
1823
1824
1825
1826
1827
1828
1829
1830
1831
1832
1833
1834
1835
1836
1837
1838
1839
1840
1841
1842
1843
1844
1845
1846
1847
1848
1849
1850
1851
1852
1853
1854
1855
1856
1857
1858
1859
1860
1861
1862
1863
1864
1865
1866
1867
1868
1869
1870
1871
1872
1873
1874
1875
1876
1877
1878
1879
1880
1881
1882
1883
1884
1885
1886
1887
1888
1889
1890
1891
1892
1893
1894
1895
1896
1897
1898
1899
1900
1901
1902
1903
1904
1905
1906
1907
1908
1909
1910
1911
1912
1913
1914
1915
1916
1917
1918
1919
1920
1921
1922
1923
1924
1925
1926
1927
1928
1929
1930
1931
1932
1933
1934
1935
1936
1937
1938
1939
1940
1941
1942
1943
1944
1945
1946
1947
1948
1949
1950
1951
1952
1953
1954
1955
1956
1957
1958
1959
1960
1961
1962
1963
1964
1965
1966
1967
1968
1969
1970
1971
1972
1973
1974
1975
1976
1977
1978
1979
1980
1981
1982
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
2018
2019
2020
2021
2022
2023
2024
2025
2026
2027
2028
2029
2030
2031
2032
2033
2034
2035
2036
2037
2038
2039
2040
2041
2042
2043
2044
2045
2046
2047
2048
2049
2050
2051
2052
2053
2054
2055
2056
2057
2058
2059
2060
2061
2062
2063
2064
2065
2066
2067
2068
2069
2070
2071
2072
2073
2074
2075
2076
2077
2078
2079
2080
2081
2082
2083
2084
2085
2086
2087
2088
2089
2090
2091
2092
2093
2094
2095
2096
2097
2098
2099
2100
2101
2102
2103
2104
2105
2106
2107
2108
2109
2110
2111
2112
2113
2114
2115
2116
2117
2118
2119
2120
2121
2122
2123
2124
2125
2126
2127
2128
2129
2130
2131
2132
2133
2134
2135
2136
2137
2138
2139
2140
2141
2142
2143
2144
2145
2146
2147
2148
2149
2150
2151
2152
2153
2154
2155
2156
2157
2158
2159
2160
2161
2162
2163
2164
2165
2166
2167
2168
2169
2170
2171
2172
2173
2174
2175
2176
2177
2178
2179
2180
2181
2182
2183
2184
2185
2186
2187
2188
2189
2190
2191
2192
2193
2194
2195
2196
2197
2198
2199
2200
2201
2202
2203
2204
2205
2206
2207
2208
2209
2210
2211
2212
2213
2214
2215
2216
2217
2218
2219
2220
2221
2222
2223
2224
2225
2226
2227
2228
2229
2230
2231
2232
2233
2234
2235
2236
2237
2238
2239
2240
2241
2242
2243
2244
2245
2246
2247
2248
2249
2250
2251
2252
2253
2254
2255
2256
2257
2258
2259
2260
2261
2262
2263
2264
2265
2266
2267
2268
2269
2270
2271
2272
2273
2274
2275
2276
2277
2278
2279
2280
2281
2282
2283
2284
2285
2286
2287
2288
2289
2290
2291
2292
2293
2294
2295
2296
2297
2298
2299
2300
2301
2302
2303
2304
2305
2306
2307
2308
2309
2310
2311
2312
2313
2314
2315
2316
2317
2318
2319
2320
2321
2322
2323
2324
2325
2326
2327
2328
2329
2330
2331
2332
2333
2334
2335
2336
2337
2338
2339
2340
2341
2342
2343
2344
2345
2346
2347
2348
2349
2350
2351
2352
2353
2354
2355
2356
2357
2358
2359
2360
2361
2362
2363
2364
2365
2366
2367
2368
2369
2370
2371
2372
2373
2374
2375
2376
2377
2378
2379
2380
2381
2382
2383
2384
2385
2386
2387
2388
2389
2390
2391
2392
2393
2394
2395
2396
2397
2398
2399
2400
2401
2402
2403
2404
2405
2406
2407
2408
2409
2410
2411
2412
2413
2414
2415
2416
2417
2418
2419
2420
2421
2422
2423
2424
2425
2426
2427
2428
2429
2430
2431
2432
2433
2434
2435
2436
2437
2438
2439
2440
2441
2442
2443
2444
2445
2446
2447
2448
2449
2450
2451
2452
2453
2454
2455
2456
2457
2458
2459
2460
2461
2462
2463
2464
2465
2466
2467
2468
2469
2470
2471
2472
2473
2474
2475
2476
2477
2478
2479
2480
2481
2482
2483
2484
2485
2486
2487
2488
2489
2490
2491
2492
2493
2494
2495
2496
2497
2498
2499
2500
2501
2502
2503
2504
2505
2506
2507
2508
2509
2510
2511
2512
2513
2514
2515
2516
2517
2518
2519
2520
2521
2522
2523
2524
2525
2526
2527
2528
2529
2530
2531
2532
2533
2534
2535
2536
2537
2538
2539
2540
2541
2542
2543
2544
2545
2546
2547
2548
2549
2550
2551
2552
2553
2554
2555
2556
2557
2558
2559
2560
2561
2562
2563
2564
2565
2566
2567
2568
2569
2570
2571
2572
2573
2574
2575
2576
2577
2578
2579
2580
2581
2582
2583
2584
2585
2586
2587
2588
2589
2590
2591
2592
2593
2594
2595
2596
2597
2598
2599
2600
2601
2602
2603
2604
2605
2606
2607
2608
2609
2610
2611
2612
2613
2614
2615
2616
2617
2618
2619
2620
2621
2622
2623
2624
2625
2626
2627
2628
2629
2630
2631
2632
2633
2634
2635
2636
2637
2638
2639
2640
2641
2642
2643
2644
2645
2646
2647
2648
2649
2650
2651
2652
2653
2654
2655
2656
2657
2658
2659
2660
2661
2662
2663
2664
2665
2666
2667
2668
2669
2670
2671
2672
2673
2674
2675
2676
2677
2678
2679
2680
2681
2682
2683
2684
2685
2686
2687
2688
2689
2690
2691
2692
2693
2694
2695
2696
2697
2698
2699
2700
2701
2702
2703
2704
2705
2706
2707
2708
2709
2710
2711
2712
2713
2714
2715
2716
2717
2718
2719
2720
2721
2722
2723
2724
2725
2726
2727
2728
2729
2730
2731
2732
2733
2734
2735
2736
2737
2738
2739
2740
2741
2742
2743
2744
2745
2746
2747
2748
2749
2750
2751
2752
2753
2754
2755
2756
2757
2758
2759
2760
2761
2762
2763
2764
2765
2766
2767
2768
2769
2770
2771
2772
2773
2774
2775
2776
2777
2778
2779
2780
2781
2782
2783
2784
2785
2786
2787
2788
2789
2790
2791
2792
2793
2794
2795
2796
2797
2798
2799
2800
2801
2802
2803
2804
2805
2806
2807
2808
2809
2810
2811
2812
2813
2814
2815
2816
2817
2818
2819
2820
2821
2822
2823
2824
2825
2826
2827
2828
2829
2830
2831
2832
2833
2834
2835
2836
2837
2838
2839
2840
2841
2842
2843
2844
2845
2846
2847
2848
2849
2850
2851
2852
2853
2854
2855
2856
2857
2858
2859
2860
2861
2862
2863
2864
2865
2866
2867
2868
2869
2870
2871
2872
2873
2874
2875
2876
2877
2878
2879
2880
2881
2882
2883
2884
2885
2886
2887
2888
2889
2890
2891
2892
2893
2894
2895
2896
2897
2898
2899
2900
2901
2902
2903
2904
2905
2906
2907
2908
2909
2910
2911
2912
2913
2914
2915
2916
2917
2918
2919
2920
2921
2922
2923
2924
2925
2926
2927
2928
2929
2930
2931
2932
2933
2934
2935
2936
2937
2938
2939
2940
2941
2942
2943
2944
2945
2946
2947
2948
2949
2950
2951
2952
2953
2954
2955
2956
2957
2958
2959
2960
2961
2962
2963
2964
2965
2966
2967
2968
2969
2970
2971
2972
2973
2974
2975
2976
2977
2978
2979
2980
2981
2982
2983
2984
2985
2986
2987
2988
2989
2990
2991
2992
2993
2994
2995
2996
2997
2998
2999
3000
3001
3002
3003
3004
3005
3006
3007
3008
3009
3010
3011
3012
3013
3014
3015
3016
3017
3018
3019
3020
3021
3022
3023
3024
3025
3026
3027
3028
3029
3030
3031
3032
3033
3034
3035
3036
3037
3038
3039
3040
3041
3042
3043
3044
3045
3046
3047
3048
3049
3050
3051
3052
3053
3054
3055
3056
3057
3058
3059
3060
3061
3062
3063
3064
3065
3066
3067
3068
3069
3070
3071
3072
3073
3074
3075
3076
3077
3078
3079
3080
3081
3082
3083
3084
3085
3086
3087
3088
3089
3090
3091
3092
3093
3094
3095
3096
3097
3098
3099
3100
3101
3102
3103
3104
3105
3106
3107
3108
3109
3110
3111
3112
3113
3114
3115
3116
3117
3118
3119
3120
3121
3122
3123
3124
3125
3126
3127
3128
3129
3130
3131
3132
3133
3134
3135
3136
3137
3138
3139
3140
3141
3142
3143
3144
3145
3146
3147
3148
3149
3150
3151
3152
3153
3154
3155
3156
3157
3158
3159
3160
3161
3162
3163
3164
3165
3166
3167
3168
3169
3170
3171
3172
3173
3174
3175
3176
3177
3178
3179
3180
3181
3182
3183
3184
3185
3186
3187
3188
3189
3190
3191
3192
3193
3194
3195
3196
3197
3198
3199
3200
3201
3202
3203
3204
3205
3206
3207
3208
3209
3210
3211
3212
3213
3214
3215
3216
3217
3218
3219
3220
3221
3222
3223
3224
3225
3226
3227
3228
3229
3230
3231
3232
3233
3234
3235
3236
3237
3238
3239
3240
3241
3242
3243
3244
3245
3246
3247
3248
3249
3250
3251
3252
3253
3254
3255
3256
3257
3258
3259
3260
3261
3262
3263
3264
3265
3266
3267
3268
3269
3270
3271
3272
3273
3274
3275
3276
3277
3278
3279
3280
3281
3282
3283
3284
3285
3286
3287
3288
3289
3290
3291
3292
/*
** 2011-08-26
**
** The author disclaims copyright to this source code.  In place of
** a legal notice, here is a blessing:
**
**    May you do good and not evil.
**    May you find forgiveness for yourself and forgive others.
**    May you share freely, never taking more than you give.
**
*************************************************************************
** 
** NORMAL DATABASE FILE FORMAT
**
** The following database file format concepts are used by the code in
** this file to read and write the database file.
**
** Pages:
**
**   A database file is divided into pages. The first 8KB of the file consists
**   of two 4KB meta-pages. The meta-page size is not configurable. The 
**   remainder of the file is made up of database pages. The default database
**   page size is 4KB. Database pages are aligned to page-size boundaries,
**   so if the database page size is larger than 8KB there is a gap between
**   the end of the meta pages and the start of the database pages.
**
**   Database pages are numbered based on their position in the file. Page N
**   begins at byte offset ((N-1)*pgsz). This means that page 1 does not 
**   exist - since it would always overlap with the meta pages. If the 
**   page-size is (say) 512 bytes, then the first usable page in the database
**   is page 33.
**
**   It is assumed that the first two meta pages and the data that follows
**   them are located on different disk sectors. So that if a power failure 
**   while writing to a meta page there is no risk of damage to the other
**   meta page or any other part of the database file. TODO: This may need
**   to be revisited.
**
** Blocks:
**
**   The database file is also divided into blocks. The default block size is
**   1MB. When writing to the database file, an attempt is made to write data
**   in contiguous block-sized chunks.
**
**   The first and last page on each block are special in that they are 4 
**   bytes smaller than all other pages. This is because the last four bytes 
**   of space on the first and last pages of each block are reserved for
**   pointers to other blocks (i.e. a 32-bit block number).
**
** Runs:
**
**   A run is a sequence of pages that the upper layer uses to store a 
**   sorted array of database keys (and accompanying data - values, FC 
**   pointers and so on). Given a page within a run, it is possible to
**   navigate to the next page in the run as follows:
**
**     a) if the current page is not the last in a block, the next page 
**        in the run is located immediately after the current page, OR
**
**     b) if the current page is the last page in a block, the next page 
**        in the run is the first page on the block identified by the
**        block pointer stored in the last 4 bytes of the current block.
**
**   It is possible to navigate to the previous page in a similar fashion,
**   using the block pointer embedded in the last 4 bytes of the first page
**   of each block as required.
**
**   The upper layer is responsible for identifying by page number the 
**   first and last page of any run that it needs to navigate - there are
**   no "end-of-run" markers stored or identified by this layer. This is
**   necessary as clients reading different database snapshots may access 
**   different subsets of a run.
**
** THE LOG FILE 
**
** This file opens and closes the log file. But it does not contain any
** logic related to the log file format. Instead, it exports the following
** functions that are used by the code in lsm_log.c to read and write the
** log file:
**
**     lsmFsOpenLog
**     lsmFsWriteLog
**     lsmFsSyncLog
**     lsmFsReadLog
**     lsmFsTruncateLog
**     lsmFsCloseAndDeleteLog
**
** COMPRESSED DATABASE FILE FORMAT
**
** The compressed database file format is very similar to the normal format.
** The file still begins with two 4KB meta-pages (which are never compressed).
** It is still divided into blocks.
**
** The first and last four bytes of each block are reserved for 32-bit 
** pointer values. Similar to the way four bytes are carved from the end of 
** the first and last page of each block in uncompressed databases. From
** the point of view of the upper layer, all pages are the same size - this
** is different from the uncompressed format where the first and last pages
** on each block are 4 bytes smaller than the others.
**
** Pages are stored in variable length compressed form, as follows:
**
**     * 3-byte size field containing the size of the compressed page image
**       in bytes. The most significant bit of each byte of the size field
**       is always set. The remaining 7 bits are used to store a 21-bit
**       integer value (in big-endian order - the first byte in the field
**       contains the most significant 7 bits). Since the maximum allowed 
**       size of a compressed page image is (2^17 - 1) bytes, there are
**       actually 4 unused bits in the size field.
**
**       In other words, if the size of the compressed page image is nSz,
**       the header can be serialized as follows:
**
**         u8 aHdr[3]
**         aHdr[0] = 0x80 | (u8)(nSz >> 14);
**         aHdr[1] = 0x80 | (u8)(nSz >>  7);
**         aHdr[2] = 0x80 | (u8)(nSz >>  0);
**
**     * Compressed page image.
**
**     * A second copy of the 3-byte record header.
**
** A page number is a byte offset into the database file. So the smallest
** possible page number is 8192 (immediately after the two meta-pages).
** The first and root page of a segment are identified by a page number
** corresponding to the byte offset of the first byte in the corresponding
** page record. The last page of a segment is identified by the byte offset
** of the last byte in its record.
**
** Unlike uncompressed pages, compressed page records may span blocks.
**
** Sometimes, in order to avoid touching sectors that contain synced data
** when writing, it is necessary to insert unused space between compressed
** page records. This can be done as follows:
**
**     * For less than 6 bytes of empty space, the first and last byte
**       of the free space contain the total number of free bytes. For
**       example:
**
**         Block of 4 free bytes: 0x04 0x?? 0x?? 0x04
**         Block of 2 free bytes: 0x02 0x02
**         A single free byte:    0x01
**
**     * For 6 or more bytes of empty space, a record similar to a 
**       compressed page record is added to the segment. A padding record
**       is distinguished from a compressed page record by the most 
**       significant bit of the second byte of the size field, which is
**       cleared instead of set. 
*/
#include "lsmInt.h"

#include <sys/types.h>
#include <sys/stat.h>
#include <fcntl.h>

/*
** File-system object. Each database connection allocates a single instance
** of the following structure. It is used for all access to the database and
** log files.
**
** The database file may be accessed via two methods - using mmap() or using
** read() and write() calls. In the general case both methods are used - a
** prefix of the file is mapped into memory and the remainder accessed using
** read() and write(). This is helpful when accessing very large files (or
** files that may grow very large during the lifetime of a database
** connection) on systems with 32-bit address spaces. However, it also requires
** that this object manage two distinct types of Page objects simultaneously -
** those that carry pointers to the mapped file and those that carry arrays
** populated by read() calls.
**
** pFree:
**   The head of a singly-linked list that containing currently unused Page 
**   structures suitable for use as mmap-page handles. Connected by the
**   Page.pFreeNext pointers.
**
** pMapped:
**   The head of a singly-linked list that contains all pages that currently
**   carry pointers to the mapped region. This is used if the region is
**   every remapped - the pointers carried by existing pages can be adjusted
**   to account for the remapping. Connected by the Page.pMappedNext pointers.
**
** pWaiting:
**   When the upper layer wishes to append a new b-tree page to a segment,
**   it allocates a Page object that carries a malloc'd block of memory -
**   regardless of the mmap-related configuration. The page is not assigned
**   a page number at first. When the upper layer has finished constructing
**   the page contents, it calls lsmFsPagePersist() to assign a page number
**   to it. At this point it is likely that N pages have been written to the
**   segment, the (N+1)th page is still outstanding and the b-tree page is
**   assigned page number (N+2). To avoid writing page (N+2) before page 
**   (N+1), the recently completed b-tree page is held in the singly linked
**   list headed by pWaiting until page (N+1) has been written. 
**
**   Function lsmFsFlushWaiting() is responsible for eventually writing 
**   waiting pages to disk.
**
** apHash/nHash:
**   Hash table used to store all Page objects that carry malloc'd arrays,
**   except those b-tree pages that have not yet been assigned page numbers.
**   Once they have been assigned page numbers - they are added to this
**   hash table.
**
**   Hash table overflow chains are connected using the Page.pHashNext
**   pointers.
**
** pLruFirst, pLruLast:
**   The first and last entries in a doubly-linked list of pages. This
**   list contains all pages with malloc'd data that are present in the
**   hash table and have a ref-count of zero.
*/
struct FileSystem {
  lsm_db *pDb;                    /* Database handle that owns this object */
  lsm_env *pEnv;                  /* Environment pointer */
  char *zDb;                      /* Database file name */
  char *zLog;                     /* Database file name */
  int nMetasize;                  /* Size of meta pages in bytes */
  int nPagesize;                  /* Database page-size in bytes */
  int nBlocksize;                 /* Database block-size in bytes */

  /* r/w file descriptors for both files. */
  LsmFile *pLsmFile;              /* Used after lsm_close() to link into list */
  lsm_file *fdDb;                 /* Database file */
  lsm_file *fdLog;                /* Log file */
  int szSector;                   /* Database file sector size */

  /* If this is a compressed database, a pointer to the compression methods.
  ** For an uncompressed database, a NULL pointer.  */
  lsm_compress *pCompress;
  u8 *aIBuffer;                   /* Buffer to compress to */
  u8 *aOBuffer;                   /* Buffer to uncompress from */
  int nBuffer;                    /* Allocated size of above buffers in bytes */

  /* mmap() page related things */
  i64 nMapLimit;                  /* Maximum bytes of file to map */
  void *pMap;                     /* Current mapping of database file */
  i64 nMap;                       /* Bytes mapped at pMap */
  Page *pFree;                    /* Unused Page structures */
  Page *pMapped;                  /* List of Page structs that point to pMap */

  /* Page cache parameters for non-mmap() pages */
  int nCacheMax;                  /* Configured cache size (in pages) */
  int nCacheAlloc;                /* Current cache size (in pages) */
  Page *pLruFirst;                /* Head of the LRU list */
  Page *pLruLast;                 /* Tail of the LRU list */
  int nHash;                      /* Number of hash slots in hash table */
  Page **apHash;                  /* nHash Hash slots */
  Page *pWaiting;                 /* b-tree pages waiting to be written */

  /* Statistics */
  int nOut;                       /* Number of outstanding pages */
  int nWrite;                     /* Total number of pages written */
  int nRead;                      /* Total number of pages read */
};

/*
** Database page handle.
**
** pSeg:
**   When lsmFsSortedAppend() is called on a compressed database, the new
**   page is not assigned a page number or location in the database file
**   immediately. Instead, these are assigned by the lsmFsPagePersist() call
**   right before it writes the compressed page image to disk.
**
**   The lsmFsSortedAppend() function sets the pSeg pointer to point to the
**   segment that the new page will be a part of. It is unset by
**   lsmFsPagePersist() after the page is written to disk.
*/
struct Page {
  u8 *aData;                      /* Buffer containing page data */
  int nData;                      /* Bytes of usable data at aData[] */
  Pgno iPg;                       /* Page number */
  int nRef;                       /* Number of outstanding references */
  int flags;                      /* Combination of PAGE_XXX flags */
  Page *pHashNext;                /* Next page in hash table slot */
  Page *pLruNext;                 /* Next page in LRU list */
  Page *pLruPrev;                 /* Previous page in LRU list */
  FileSystem *pFS;                /* File system that owns this page */

  /* Only used in compressed database mode: */
  int nCompress;                  /* Compressed size (or 0 for uncomp. db) */
  int nCompressPrev;              /* Compressed size of prev page */
  Segment *pSeg;                  /* Segment this page will be written to */

  /* Pointers for singly linked lists */
  Page *pWaitingNext;             /* Next page in FileSystem.pWaiting list */
  Page *pFreeNext;                /* Next page in FileSystem.pFree list */
  Page *pMappedNext;              /* Next page in FileSystem.pMapped list */
};

/*
** Meta-data page handle. There are two meta-data pages at the start of
** the database file, each FileSystem.nMetasize bytes in size.
*/
struct MetaPage {
  int iPg;                        /* Either 1 or 2 */
  int bWrite;                     /* Write back to db file on release */
  u8 *aData;                      /* Pointer to buffer */
  FileSystem *pFS;                /* FileSystem that owns this page */
};

/* 
** Values for LsmPage.flags 
*/
#define PAGE_DIRTY   0x00000001   /* Set if page is dirty */
#define PAGE_FREE    0x00000002   /* Set if Page.aData requires lsmFree() */
#define PAGE_HASPREV 0x00000004   /* Set if page is first on uncomp. block */

/*
** Number of pgsz byte pages omitted from the start of block 1. The start
** of block 1 contains two 4096 byte meta pages (8192 bytes in total).
*/
#define BLOCK1_HDR_SIZE(pgsz)  LSM_MAX(1, 8192/(pgsz))

/*
** If NDEBUG is not defined, set a breakpoint in function lsmIoerrBkpt()
** to catch IO errors (any error returned by a VFS method). 
*/
#ifndef NDEBUG
static void lsmIoerrBkpt(void){
  static int nErr = 0;
  nErr++;
}
static int IOERR_WRAPPER(int rc){
  if( rc!=LSM_OK ) lsmIoerrBkpt();
  return rc;
}
#else
# define IOERR_WRAPPER(rc) (rc)
#endif

#ifdef NDEBUG
# define assert_lists_are_ok(x)
#else
static Page *fsPageFindInHash(FileSystem *pFS, Pgno iPg, int *piHash);

static void assert_lists_are_ok(FileSystem *pFS){
#if 0
  Page *p;

  assert( pFS->nMapLimit>=0 );

  /* Check that all pages in the LRU list have nRef==0, pointers to buffers
  ** in heap memory, and corresponding entries in the hash table.  */
  for(p=pFS->pLruFirst; p; p=p->pLruNext){
    assert( p==pFS->pLruFirst || p->pLruPrev!=0 );
    assert( p==pFS->pLruLast || p->pLruNext!=0 );
    assert( p->pLruPrev==0 || p->pLruPrev->pLruNext==p );
    assert( p->pLruNext==0 || p->pLruNext->pLruPrev==p );
    assert( p->nRef==0 );
    assert( p->flags & PAGE_FREE );
    assert( p==fsPageFindInHash(pFS, p->iPg, 0) );
  }
#endif
}
#endif

/*
** Wrappers around the VFS methods of the lsm_env object:
**
**     lsmEnvOpen()
**     lsmEnvRead()
**     lsmEnvWrite()
**     lsmEnvSync()
**     lsmEnvSectorSize()
**     lsmEnvClose()
**     lsmEnvTruncate()
**     lsmEnvUnlink()
**     lsmEnvRemap()
*/
int lsmEnvOpen(lsm_env *pEnv, const char *zFile, int flags, lsm_file **ppNew){
  return pEnv->xOpen(pEnv, zFile, flags, ppNew);
}

static int lsmEnvRead(
  lsm_env *pEnv, 
  lsm_file *pFile, 
  lsm_i64 iOff, 
  void *pRead, 
  int nRead
){
  return IOERR_WRAPPER( pEnv->xRead(pFile, iOff, pRead, nRead) );
}

static int lsmEnvWrite(
  lsm_env *pEnv, 
  lsm_file *pFile, 
  lsm_i64 iOff, 
  const void *pWrite, 
  int nWrite
){
  return IOERR_WRAPPER( pEnv->xWrite(pFile, iOff, (void *)pWrite, nWrite) );
}

static int lsmEnvSync(lsm_env *pEnv, lsm_file *pFile){
  return IOERR_WRAPPER( pEnv->xSync(pFile) );
}

static int lsmEnvSectorSize(lsm_env *pEnv, lsm_file *pFile){
  return pEnv->xSectorSize(pFile);
}

int lsmEnvClose(lsm_env *pEnv, lsm_file *pFile){
  return IOERR_WRAPPER( pEnv->xClose(pFile) );
}

static int lsmEnvTruncate(lsm_env *pEnv, lsm_file *pFile, lsm_i64 nByte){
  return IOERR_WRAPPER( pEnv->xTruncate(pFile, nByte) );
}

static int lsmEnvUnlink(lsm_env *pEnv, const char *zDel){
  return IOERR_WRAPPER( pEnv->xUnlink(pEnv, zDel) );
}

static int lsmEnvRemap(
  lsm_env *pEnv, 
  lsm_file *pFile, 
  i64 szMin,
  void **ppMap,
  i64 *pszMap
){
  return pEnv->xRemap(pFile, szMin, ppMap, pszMap);
}

int lsmEnvLock(lsm_env *pEnv, lsm_file *pFile, int iLock, int eLock){
  if( pFile==0 ) return LSM_OK;
  return pEnv->xLock(pFile, iLock, eLock);
}

int lsmEnvTestLock(
  lsm_env *pEnv, 
  lsm_file *pFile, 
  int iLock, 
  int nLock, 
  int eLock
){
  return pEnv->xTestLock(pFile, iLock, nLock, eLock);
}

int lsmEnvShmMap(
  lsm_env *pEnv, 
  lsm_file *pFile, 
  int iChunk, 
  int sz, 
  void **ppOut
){
  return pEnv->xShmMap(pFile, iChunk, sz, ppOut);
}

void lsmEnvShmBarrier(lsm_env *pEnv){
  pEnv->xShmBarrier();
}

void lsmEnvShmUnmap(lsm_env *pEnv, lsm_file *pFile, int bDel){
  pEnv->xShmUnmap(pFile, bDel);
}

void lsmEnvSleep(lsm_env *pEnv, int nUs){
  pEnv->xSleep(pEnv, nUs);
}


/*
** Write the contents of string buffer pStr into the log file, starting at
** offset iOff.
*/
int lsmFsWriteLog(FileSystem *pFS, i64 iOff, LsmString *pStr){
  assert( pFS->fdLog );
  return lsmEnvWrite(pFS->pEnv, pFS->fdLog, iOff, pStr->z, pStr->n);
}

/*
** fsync() the log file.
*/
int lsmFsSyncLog(FileSystem *pFS){
  assert( pFS->fdLog );
  return lsmEnvSync(pFS->pEnv, pFS->fdLog);
}

/*
** Read nRead bytes of data starting at offset iOff of the log file. Append
** the results to string buffer pStr.
*/
int lsmFsReadLog(FileSystem *pFS, i64 iOff, int nRead, LsmString *pStr){
  int rc;                         /* Return code */
  assert( pFS->fdLog );
  rc = lsmStringExtend(pStr, nRead);
  if( rc==LSM_OK ){
    rc = lsmEnvRead(pFS->pEnv, pFS->fdLog, iOff, &pStr->z[pStr->n], nRead);
    pStr->n += nRead;
  }
  return rc;
}

/*
** Truncate the log file to nByte bytes in size.
*/
int lsmFsTruncateLog(FileSystem *pFS, i64 nByte){
  if( pFS->fdLog==0 ) return LSM_OK;
  return lsmEnvTruncate(pFS->pEnv, pFS->fdLog, nByte);
}

/*
** Truncate the db file to nByte bytes in size.
*/
int lsmFsTruncateDb(FileSystem *pFS, i64 nByte){
  if( pFS->fdDb==0 ) return LSM_OK;
  return lsmEnvTruncate(pFS->pEnv, pFS->fdDb, nByte);
}

/*
** Close the log file. Then delete it from the file-system. This function
** is called during database shutdown only.
*/
int lsmFsCloseAndDeleteLog(FileSystem *pFS){
  char *zDel;

  if( pFS->fdLog ){
    lsmEnvClose(pFS->pEnv, pFS->fdLog );
    pFS->fdLog = 0;
  }

  zDel = lsmMallocPrintf(pFS->pEnv, "%s-log", pFS->zDb);
  if( zDel ){
    lsmEnvUnlink(pFS->pEnv, zDel);
    lsmFree(pFS->pEnv, zDel);
  }
  return LSM_OK;
}

/*
** Return true if page iReal of the database should be accessed using mmap.
** False otherwise.
*/
static int fsMmapPage(FileSystem *pFS, Pgno iReal){
  return ((i64)iReal*pFS->nPagesize <= pFS->nMapLimit);
}

/*
** Given that there are currently nHash slots in the hash table, return 
** the hash key for file iFile, page iPg.
*/
static int fsHashKey(int nHash, int iPg){
  return (iPg % nHash);
}

/*
** This is a helper function for lsmFsOpen(). It opens a single file on
** disk (either the database or log file).
*/
static lsm_file *fsOpenFile(
  FileSystem *pFS,                /* File system object */
  int bReadonly,                  /* True to open this file read-only */
  int bLog,                       /* True for log, false for db */
  int *pRc                        /* IN/OUT: Error code */
){
  lsm_file *pFile = 0;
  if( *pRc==LSM_OK ){
    int flags = (bReadonly ? LSM_OPEN_READONLY : 0);
    const char *zPath = (bLog ? pFS->zLog : pFS->zDb);

    *pRc = lsmEnvOpen(pFS->pEnv, zPath, flags, &pFile);
  }
  return pFile;
}

/*
** If it is not already open, this function opens the log file. It returns
** LSM_OK if successful (or if the log file was already open) or an LSM
** error code otherwise.
**
** The log file must be opened before any of the following may be called:
**
**     lsmFsWriteLog
**     lsmFsSyncLog
**     lsmFsReadLog
*/
int lsmFsOpenLog(lsm_db *db, int *pbOpen){
  int rc = LSM_OK;
  FileSystem *pFS = db->pFS;

  if( 0==pFS->fdLog ){ 
    pFS->fdLog = fsOpenFile(pFS, db->bReadonly, 1, &rc); 

    if( rc==LSM_IOERR_NOENT && db->bReadonly ){
      rc = LSM_OK;
    }
  }

  if( pbOpen ) *pbOpen = (pFS->fdLog!=0);
  return rc;
}

/*
** Close the log file, if it is open.
*/
void lsmFsCloseLog(lsm_db *db){
  FileSystem *pFS = db->pFS;
  if( pFS->fdLog ){
    lsmEnvClose(pFS->pEnv, pFS->fdLog);
    pFS->fdLog = 0;
  }
}

/*
** Open a connection to a database stored within the file-system.
**
** If parameter bReadonly is true, then open a read-only file-descriptor
** on the database file. It is possible that bReadonly will be false even
** if the user requested that pDb be opened read-only. This is because the
** file-descriptor may later on be recycled by a read-write connection.
** If the db file can be opened for read-write access, it always is. Parameter
** bReadonly is only ever true if it has already been determined that the
** db can only be opened for read-only access.
**
** Return LSM_OK if successful or an lsm error code otherwise.
*/
int lsmFsOpen(
  lsm_db *pDb,                    /* Database connection to open fd for */
  const char *zDb,                /* Full path to database file */
  int bReadonly                   /* True to open db file read-only */
){
  FileSystem *pFS;
  int rc = LSM_OK;
  int nDb = strlen(zDb);
  int nByte;

  assert( pDb->pFS==0 );
  assert( pDb->pWorker==0 && pDb->pClient==0 );

  nByte = sizeof(FileSystem) + nDb+1 + nDb+4+1;
  pFS = (FileSystem *)lsmMallocZeroRc(pDb->pEnv, nByte, &rc);
  if( pFS ){
    LsmFile *pLsmFile;
    pFS->zDb = (char *)&pFS[1];
    pFS->zLog = &pFS->zDb[nDb+1];
    pFS->nPagesize = LSM_DFLT_PAGE_SIZE;
    pFS->nBlocksize = LSM_DFLT_BLOCK_SIZE;
    pFS->nMetasize = 4 * 1024;
    pFS->pDb = pDb;
    pFS->pEnv = pDb->pEnv;

    /* Make a copy of the database and log file names. */
    memcpy(pFS->zDb, zDb, nDb+1);
    memcpy(pFS->zLog, zDb, nDb);
    memcpy(&pFS->zLog[nDb], "-log", 5);

    /* Allocate the hash-table here. At some point, it should be changed
    ** so that it can grow dynamicly. */
    pFS->nCacheMax = 2048*1024 / pFS->nPagesize;
    pFS->nHash = 4096;
    pFS->apHash = lsmMallocZeroRc(pDb->pEnv, sizeof(Page *) * pFS->nHash, &rc);

    /* Open the database file */
    pLsmFile = lsmDbRecycleFd(pDb);
    if( pLsmFile ){
      pFS->pLsmFile = pLsmFile;
      pFS->fdDb = pLsmFile->pFile;
      memset(pLsmFile, 0, sizeof(LsmFile));
    }else{
      pFS->pLsmFile = lsmMallocZeroRc(pDb->pEnv, sizeof(LsmFile), &rc);
      if( rc==LSM_OK ){
        pFS->fdDb = fsOpenFile(pFS, bReadonly, 0, &rc);
      }
    }

    if( rc!=LSM_OK ){
      lsmFsClose(pFS);
      pFS = 0;
    }else{
      pFS->szSector = lsmEnvSectorSize(pFS->pEnv, pFS->fdDb);
    }
  }

  pDb->pFS = pFS;
  return rc;
}

/*
** Configure the file-system object according to the current values of
** the LSM_CONFIG_MMAP and LSM_CONFIG_SET_COMPRESSION options.
*/
int lsmFsConfigure(lsm_db *db){
  FileSystem *pFS = db->pFS;
  if( pFS ){
    lsm_env *pEnv = pFS->pEnv;
    Page *pPg;

    assert( pFS->nOut==0 );
    assert( pFS->pWaiting==0 );
    assert( pFS->pMapped==0 );

    /* Reset any compression/decompression buffers already allocated */
    lsmFree(pEnv, pFS->aIBuffer);
    lsmFree(pEnv, pFS->aOBuffer);
    pFS->nBuffer = 0;

    /* Unmap the file, if it is currently mapped */
    if( pFS->pMap ){
      lsmEnvRemap(pEnv, pFS->fdDb, -1, &pFS->pMap, &pFS->nMap);
      pFS->nMapLimit = 0;
    }

    /* Free all allocated page structures */
    pPg = pFS->pLruFirst;
    while( pPg ){
      Page *pNext = pPg->pLruNext;
      assert( pPg->flags & PAGE_FREE );
      lsmFree(pEnv, pPg->aData);
      lsmFree(pEnv, pPg);
      pPg = pNext;
    }

    pPg = pFS->pFree;
    while( pPg ){
      Page *pNext = pPg->pFreeNext;
      lsmFree(pEnv, pPg);
      pPg = pNext;
    }

    /* Zero pointers that point to deleted page objects */
    pFS->nCacheAlloc = 0;
    pFS->pLruFirst = 0;
    pFS->pLruLast = 0;
    pFS->pFree = 0;
    if( pFS->apHash ){
      memset(pFS->apHash, 0, pFS->nHash*sizeof(pFS->apHash[0]));
    }

    /* Configure the FileSystem object */
    if( db->compress.xCompress ){
      pFS->pCompress = &db->compress;
      pFS->nMapLimit = 0;
    }else{
      pFS->pCompress = 0;
      if( db->iMmap==1 ){
        /* Unlimited */
        pFS->nMapLimit = (i64)1 << 60;
      }else{
        /* iMmap is a limit in KB. Set nMapLimit to the same value in bytes. */
        pFS->nMapLimit = (i64)db->iMmap * 1024;
      }
    }
  }

  return LSM_OK;
}

/*
** Close and destroy a FileSystem object.
*/
void lsmFsClose(FileSystem *pFS){
  if( pFS ){
    Page *pPg;
    lsm_env *pEnv = pFS->pEnv;

    assert( pFS->nOut==0 );
    pPg = pFS->pLruFirst;
    while( pPg ){
      Page *pNext = pPg->pLruNext;
      if( pPg->flags & PAGE_FREE ) lsmFree(pEnv, pPg->aData);
      lsmFree(pEnv, pPg);
      pPg = pNext;
    }

    pPg = pFS->pFree;
    while( pPg ){
      Page *pNext = pPg->pFreeNext;
      if( pPg->flags & PAGE_FREE ) lsmFree(pEnv, pPg->aData);
      lsmFree(pEnv, pPg);
      pPg = pNext;
    }

    if( pFS->fdDb ) lsmEnvClose(pFS->pEnv, pFS->fdDb );
    if( pFS->fdLog ) lsmEnvClose(pFS->pEnv, pFS->fdLog );
    lsmFree(pEnv, pFS->pLsmFile);
    lsmFree(pEnv, pFS->apHash);
    lsmFree(pEnv, pFS->aIBuffer);
    lsmFree(pEnv, pFS->aOBuffer);
    lsmFree(pEnv, pFS);
  }
}

/*
** This function is called when closing a database handle (i.e. lsm_close()) 
** if there exist other connections to the same database within this process.
** In that case the file-descriptor open on the database file is not closed
** when the FileSystem object is destroyed, as this would cause any POSIX
** locks held by the other connections to be silently dropped (see "man close"
** for details). Instead, the file-descriptor is stored in a list by the
** lsm_shared.c module until it is either closed or reused.
**
** This function returns a pointer to an object that can be linked into
** the list described above. The returned object now 'owns' the database
** file descriptr, so that when the FileSystem object is destroyed, it
** will not be closed. 
**
** This function may be called at most once in the life-time of a 
** FileSystem object. The results of any operations involving the database 
** file descriptor are undefined once this function has been called.
**
** None of this is necessary on non-POSIX systems. But we do it anyway in
** the name of using as similar code as possible on all platforms.
*/
LsmFile *lsmFsDeferClose(FileSystem *pFS){
  LsmFile *p = pFS->pLsmFile;
  assert( p->pNext==0 );
  p->pFile = pFS->fdDb;
  pFS->fdDb = 0;
  pFS->pLsmFile = 0;
  return p;
}

/*
** Allocate a buffer and populate it with the output of the xFileid() 
** method of the database file handle. If successful, set *ppId to point 
** to the buffer and *pnId to the number of bytes in the buffer and return
** LSM_OK. Otherwise, set *ppId and *pnId to zero and return an LSM
** error code.
*/
int lsmFsFileid(lsm_db *pDb, void **ppId, int *pnId){
  lsm_env *pEnv = pDb->pEnv;
  FileSystem *pFS = pDb->pFS;
  int rc;
  int nId = 0;
  void *pId;

  rc = pEnv->xFileid(pFS->fdDb, 0, &nId);
  pId = lsmMallocZeroRc(pEnv, nId, &rc);
  if( rc==LSM_OK ) rc = pEnv->xFileid(pFS->fdDb, pId, &nId);

  if( rc!=LSM_OK ){
    lsmFree(pEnv, pId);
    pId = 0;
    nId = 0;
  }

  *ppId = pId;
  *pnId = nId;
  return rc;
}

/*
** Return the nominal page-size used by this file-system. Actual pages
** may be smaller or larger than this value.
*/
int lsmFsPageSize(FileSystem *pFS){
  return pFS->nPagesize;
}

/*
** Return the block-size used by this file-system.
*/
int lsmFsBlockSize(FileSystem *pFS){
  return pFS->nBlocksize;
}

/*
** Configure the nominal page-size used by this file-system. Actual 
** pages may be smaller or larger than this value.
*/
void lsmFsSetPageSize(FileSystem *pFS, int nPgsz){
  pFS->nPagesize = nPgsz;
  pFS->nCacheMax = 2048*1024 / pFS->nPagesize;
}

/*
** Configure the block-size used by this file-system. 
*/
void lsmFsSetBlockSize(FileSystem *pFS, int nBlocksize){
  pFS->nBlocksize = nBlocksize;
}

/*
** Return the page number of the first page on block iBlock. Blocks are
** numbered starting from 1.
**
** For a compressed database, page numbers are byte offsets. The first
** page on each block is the byte offset immediately following the 4-byte
** "previous block" pointer at the start of each block.
*/
static Pgno fsFirstPageOnBlock(FileSystem *pFS, int iBlock){
  Pgno iPg;
  if( pFS->pCompress ){
    if( iBlock==1 ){
      iPg = pFS->nMetasize * 2 + 4;
    }else{
      iPg = pFS->nBlocksize * (Pgno)(iBlock-1) + 4;
    }
  }else{
    const int nPagePerBlock = (pFS->nBlocksize / pFS->nPagesize);
    if( iBlock==1 ){
      iPg = 1 + ((pFS->nMetasize*2 + pFS->nPagesize - 1) / pFS->nPagesize);
    }else{
      iPg = 1 + (iBlock-1) * nPagePerBlock;
    }
  }
  return iPg;
}

/*
** Return the page number of the last page on block iBlock. Blocks are
** numbered starting from 1.
**
** For a compressed database, page numbers are byte offsets. The first
** page on each block is the byte offset of the byte immediately before 
** the 4-byte "next block" pointer at the end of each block.
*/
static Pgno fsLastPageOnBlock(FileSystem *pFS, int iBlock){
  if( pFS->pCompress ){
    return pFS->nBlocksize * (Pgno)iBlock - 1 - 4;
  }else{
    const int nPagePerBlock = (pFS->nBlocksize / pFS->nPagesize);
    return iBlock * nPagePerBlock;
  }
}

/*
** Return the block number of the block that page iPg is located on. 
** Blocks are numbered starting from 1.
*/
static int fsPageToBlock(FileSystem *pFS, Pgno iPg){
  if( pFS->pCompress ){
    return (iPg / pFS->nBlocksize) + 1;
  }else{
    return 1 + ((iPg-1) / (pFS->nBlocksize / pFS->nPagesize));
  }
}

/*
** Return true if page iPg is the last page on its block.
**
** This function is only called in non-compressed database mode.
*/
static int fsIsLast(FileSystem *pFS, Pgno iPg){
  const int nPagePerBlock = (pFS->nBlocksize / pFS->nPagesize);
  assert( !pFS->pCompress );
  return ( iPg && (iPg % nPagePerBlock)==0 );
}

/*
** Return true if page iPg is the first page on its block.
**
** This function is only called in non-compressed database mode.
*/
static int fsIsFirst(FileSystem *pFS, Pgno iPg){
  const int nPagePerBlock = (pFS->nBlocksize / pFS->nPagesize);
  assert( !pFS->pCompress );
  return ( (iPg % nPagePerBlock)==1
        || (iPg<nPagePerBlock && iPg==fsFirstPageOnBlock(pFS, 1))
  );
}

/*
** Given a page reference, return a pointer to the buffer containing the 
** pages contents. If parameter pnData is not NULL, set *pnData to the size
** of the buffer in bytes before returning.
*/
u8 *lsmFsPageData(Page *pPage, int *pnData){
  if( pnData ){
    *pnData = pPage->nData;
  }
  return pPage->aData;
}

/*
** Return the page number of a page.
*/
Pgno lsmFsPageNumber(Page *pPage){
  /* assert( (pPage->flags & PAGE_DIRTY)==0 ); */
  return pPage ? pPage->iPg : 0;
}

/*
** Page pPg is currently part of the LRU list belonging to pFS. Remove
** it from the list. pPg->pLruNext and pPg->pLruPrev are cleared by this
** operation.
*/
static void fsPageRemoveFromLru(FileSystem *pFS, Page *pPg){
  assert( pPg->pLruNext || pPg==pFS->pLruLast );
  assert( pPg->pLruPrev || pPg==pFS->pLruFirst );
  if( pPg->pLruNext ){
    pPg->pLruNext->pLruPrev = pPg->pLruPrev;
  }else{
    pFS->pLruLast = pPg->pLruPrev;
  }
  if( pPg->pLruPrev ){
    pPg->pLruPrev->pLruNext = pPg->pLruNext;
  }else{
    pFS->pLruFirst = pPg->pLruNext;
  }
  pPg->pLruPrev = 0;
  pPg->pLruNext = 0;
}

/*
** Page pPg is not currently part of the LRU list belonging to pFS. Add it.
*/
static void fsPageAddToLru(FileSystem *pFS, Page *pPg){
  assert( pPg->pLruNext==0 && pPg->pLruPrev==0 );
  pPg->pLruPrev = pFS->pLruLast;
  if( pPg->pLruPrev ){
    pPg->pLruPrev->pLruNext = pPg;
  }else{
    pFS->pLruFirst = pPg;
  }
  pFS->pLruLast = pPg;
}

/*
** Page pPg is currently stored in the apHash/nHash hash table. Remove it.
*/
static void fsPageRemoveFromHash(FileSystem *pFS, Page *pPg){
  int iHash;
  Page **pp;

  iHash = fsHashKey(pFS->nHash, pPg->iPg);
  for(pp=&pFS->apHash[iHash]; *pp!=pPg; pp=&(*pp)->pHashNext);
  *pp = pPg->pHashNext;
  pPg->pHashNext = 0;
}

/*
** Free a Page object allocated by fsPageBuffer().
*/
static void fsPageBufferFree(Page *pPg){
  pPg->pFS->nCacheAlloc--;
  lsmFree(pPg->pFS->pEnv, pPg->aData);
  lsmFree(pPg->pFS->pEnv, pPg);
}


/*
** Purge the cache of all non-mmap pages with nRef==0.
*/
void lsmFsPurgeCache(FileSystem *pFS){
  Page *pPg;

  pPg = pFS->pLruFirst;
  while( pPg ){
    Page *pNext = pPg->pLruNext;
    assert( pPg->flags & PAGE_FREE );
    fsPageRemoveFromHash(pFS, pPg);
    fsPageBufferFree(pPg);
    pPg = pNext;
  }
  pFS->pLruFirst = 0;
  pFS->pLruLast = 0;

  assert( pFS->nCacheAlloc<=pFS->nOut && pFS->nCacheAlloc>=0 );
}

/*
** Search the hash-table for page iPg. If an entry is round, return a pointer
** to it. Otherwise, return NULL.
**
** Either way, if argument piHash is not NULL set *piHash to the hash slot
** number that page iPg would be stored in before returning.
*/
static Page *fsPageFindInHash(FileSystem *pFS, Pgno iPg, int *piHash){
  Page *p;                        /* Return value */
  int iHash = fsHashKey(pFS->nHash, iPg);

  if( piHash ) *piHash = iHash;
  for(p=pFS->apHash[iHash]; p; p=p->pHashNext){
    if( p->iPg==iPg) break;
  }
  return p;
}

/*
** Allocate and return a non-mmap Page object. If there are already 
** nCacheMax such Page objects outstanding, try to recycle an existing 
** Page instead.
*/
static int fsPageBuffer(
  FileSystem *pFS, 
  Page **ppOut
){
  int rc = LSM_OK;
  Page *pPage = 0;
  if( pFS->pLruFirst==0 || pFS->nCacheAlloc<pFS->nCacheMax ){
    /* Allocate a new Page object */
    pPage = lsmMallocZero(pFS->pEnv, sizeof(Page));
    if( !pPage ){
      rc = LSM_NOMEM_BKPT;
    }else{
      pPage->aData = (u8 *)lsmMalloc(pFS->pEnv, pFS->nPagesize);
      if( !pPage->aData ){
        lsmFree(pFS->pEnv, pPage);
        rc = LSM_NOMEM_BKPT;
        pPage = 0;
      }else{
        pFS->nCacheAlloc++;
      }
    }
  }else{
    /* Reuse an existing Page object */
    u8 *aData;
    pPage = pFS->pLruFirst;
    aData = pPage->aData;
    fsPageRemoveFromLru(pFS, pPage);
    fsPageRemoveFromHash(pFS, pPage);

    memset(pPage, 0, sizeof(Page));
    pPage->aData = aData;
  }

  if( pPage ){
    pPage->flags = PAGE_FREE;
  }
  *ppOut = pPage;
  return rc;
}

/*
** Assuming *pRc is initially LSM_OK, attempt to ensure that the 
** memory-mapped region is at least iSz bytes in size. If it is not already,
** iSz bytes in size, extend it and update the pointers associated with any
** outstanding Page objects.
**
** If *pRc is not LSM_OK when this function is called, it is a no-op. 
** Otherwise, *pRc is set to an lsm error code if an error occurs, or
** left unmodified otherwise.
**
** This function is never called in compressed database mode.
*/
static void fsGrowMapping(
  FileSystem *pFS,                /* File system object */
  i64 iSz,                        /* Minimum size to extend mapping to */
  int *pRc                        /* IN/OUT: Error code */
){
  assert( pFS->pCompress==0 );
  assert( PAGE_HASPREV==4 );

  if( *pRc==LSM_OK && iSz>pFS->nMap ){
    int rc;
    u8 *aOld = pFS->pMap;
    rc = lsmEnvRemap(pFS->pEnv, pFS->fdDb, iSz, &pFS->pMap, &pFS->nMap);
    if( rc==LSM_OK && pFS->pMap!=aOld ){
      Page *pFix;
      i64 iOff = (u8 *)pFS->pMap - aOld;
      for(pFix=pFS->pMapped; pFix; pFix=pFix->pMappedNext){
        pFix->aData += iOff;
      }
      lsmSortedRemap(pFS->pDb);
    }
    *pRc = rc;
  }
}

/*
** fsync() the database file.
*/
int lsmFsSyncDb(FileSystem *pFS, int nBlock){
  return lsmEnvSync(pFS->pEnv, pFS->fdDb);
}

/*
** If block iBlk has been redirected according to the redirections in the
** object passed as the first argument, return the destination block to
** which it is redirected. Otherwise, return a copy of iBlk.
*/
static int fsRedirectBlock(Redirect *p, int iBlk){
  if( p ){
    int i;
    for(i=0; i<p->n; i++){
      if( iBlk==p->a[i].iFrom ) return p->a[i].iTo;
    }
  }
  assert( iBlk!=0 );
  return iBlk;
}

/*
** If page iPg has been redirected according to the redirections in the
** object passed as the second argument, return the destination page to
** which it is redirected. Otherwise, return a copy of iPg.
*/
Pgno lsmFsRedirectPage(FileSystem *pFS, Redirect *pRedir, Pgno iPg){
  Pgno iReal = iPg;

  if( pRedir ){
    const int nPagePerBlock = (
        pFS->pCompress ? pFS->nBlocksize : (pFS->nBlocksize / pFS->nPagesize)
    );
    int iBlk = fsPageToBlock(pFS, iPg);
    int i;
    for(i=0; i<pRedir->n; i++){
      int iFrom = pRedir->a[i].iFrom;
      if( iFrom>iBlk ) break;
      if( iFrom==iBlk ){
        int iTo = pRedir->a[i].iTo;
        iReal = iPg - (Pgno)(iFrom - iTo) * nPagePerBlock;
        if( iTo==1 ){
          iReal += (fsFirstPageOnBlock(pFS, 1)-1);
        }
        break;
      }
    }
  }

  assert( iReal!=0 );
  return iReal;
}

/* Required by the circular fsBlockNext<->fsPageGet dependency. */
static int fsPageGet(FileSystem *, Segment *, Pgno, int, Page **, int *);

/*
** Parameter iBlock is a database file block. This function reads the value 
** stored in the blocks "next block" pointer and stores it in *piNext.
** LSM_OK is returned if everything is successful, or an LSM error code
** otherwise.
*/
static int fsBlockNext(
  FileSystem *pFS,                /* File-system object handle */
  Segment *pSeg,                  /* Use this segment for block redirects */
  int iBlock,                     /* Read field from this block */
  int *piNext                     /* OUT: Next block in linked list */
){
  int rc;
  int iRead;                      /* Read block from here */
  
  if( pSeg ){
    iRead = fsRedirectBlock(pSeg->pRedirect, iBlock);
  }else{
    iRead = iBlock;
  }

  assert( pFS->nMapLimit==0 || pFS->pCompress==0 );
  if( pFS->pCompress ){
    i64 iOff;                     /* File offset to read data from */
    u8 aNext[4];                  /* 4-byte pointer read from db file */

    iOff = (i64)iRead * pFS->nBlocksize - sizeof(aNext);
    rc = lsmEnvRead(pFS->pEnv, pFS->fdDb, iOff, aNext, sizeof(aNext));
    if( rc==LSM_OK ){
      *piNext = (int)lsmGetU32(aNext);
    }
  }else{
    const int nPagePerBlock = (pFS->nBlocksize / pFS->nPagesize);
    Page *pLast;
    rc = fsPageGet(pFS, 0, iRead*nPagePerBlock, 0, &pLast, 0);
    if( rc==LSM_OK ){
      *piNext = lsmGetU32(&pLast->aData[pFS->nPagesize-4]);
      lsmFsPageRelease(pLast);
    }
  }

  if( pSeg ){
    *piNext = fsRedirectBlock(pSeg->pRedirect, *piNext);
  }
  return rc;
}

/*
** Return the page number of the last page on the same block as page iPg.
*/
Pgno fsLastPageOnPagesBlock(FileSystem *pFS, Pgno iPg){
  return fsLastPageOnBlock(pFS, fsPageToBlock(pFS, iPg));
}

/*
** Read nData bytes of data from offset iOff of the database file into
** buffer aData. If this means reading past the end of a block, follow
** the block pointer to the next block and continue reading.
**
** Offset iOff is an absolute offset - not subject to any block redirection.
** However any block pointer followed is. Use pSeg->pRedirect in this case.
**
** This function is only called in compressed database mode.
*/
static int fsReadData(
  FileSystem *pFS,                /* File-system handle */
  Segment *pSeg,                  /* Block redirection */
  i64 iOff,                       /* Read data from this offset */
  u8 *aData,                      /* Buffer to read data into */
  int nData                       /* Number of bytes to read */
){
  i64 iEob;                       /* End of block */
  int nRead;
  int rc;

  assert( pFS->pCompress );

  iEob = fsLastPageOnPagesBlock(pFS, iOff) + 1;
  nRead = LSM_MIN(iEob - iOff, nData);

  rc = lsmEnvRead(pFS->pEnv, pFS->fdDb, iOff, aData, nRead);
  if( rc==LSM_OK && nRead!=nData ){
    int iBlk;

    rc = fsBlockNext(pFS, pSeg, fsPageToBlock(pFS, iOff), &iBlk);
    if( rc==LSM_OK ){
      i64 iOff2 = fsFirstPageOnBlock(pFS, iBlk);
      rc = lsmEnvRead(pFS->pEnv, pFS->fdDb, iOff2, &aData[nRead], nData-nRead);
    }
  }

  return rc;
}

/*
** Parameter iBlock is a database file block. This function reads the value 
** stored in the blocks "previous block" pointer and stores it in *piPrev.
** LSM_OK is returned if everything is successful, or an LSM error code
** otherwise.
*/
static int fsBlockPrev(
  FileSystem *pFS,                /* File-system object handle */
  Segment *pSeg,                  /* Use this segment for block redirects */
  int iBlock,                     /* Read field from this block */
  int *piPrev                     /* OUT: Previous block in linked list */
){
  int rc = LSM_OK;                /* Return code */

  assert( pFS->nMapLimit==0 || pFS->pCompress==0 );
  assert( iBlock>0 );

  if( pFS->pCompress ){
    i64 iOff = fsFirstPageOnBlock(pFS, iBlock) - 4;
    u8 aPrev[4];                  /* 4-byte pointer read from db file */
    rc = lsmEnvRead(pFS->pEnv, pFS->fdDb, iOff, aPrev, sizeof(aPrev));
    if( rc==LSM_OK ){
      Redirect *pRedir = (pSeg ? pSeg->pRedirect : 0);
      *piPrev = fsRedirectBlock(pRedir, (int)lsmGetU32(aPrev));
    }
  }else{
    assert( 0 );
  }
  return rc;
}

/*
** Encode and decode routines for record size fields.
*/
static void putRecordSize(u8 *aBuf, int nByte, int bFree){
  aBuf[0] = (u8)(nByte >> 14) | 0x80;
  aBuf[1] = ((u8)(nByte >>  7) & 0x7F) | (bFree ? 0x00 : 0x80);
  aBuf[2] = (u8)nByte | 0x80;
}
static int getRecordSize(u8 *aBuf, int *pbFree){
  int nByte;
  nByte  = (aBuf[0] & 0x7F) << 14;
  nByte += (aBuf[1] & 0x7F) << 7;
  nByte += (aBuf[2] & 0x7F);
  *pbFree = !(aBuf[1] & 0x80);
  return nByte;
}

/*
** Subtract iSub from database file offset iOff and set *piRes to the
** result. If doing so means passing the start of a block, follow the
** block pointer stored in the first 4 bytes of the block.
**
** Offset iOff is an absolute offset - not subject to any block redirection.
** However any block pointer followed is. Use pSeg->pRedirect in this case.
**
** Return LSM_OK if successful or an lsm error code if an error occurs.
*/
static int fsSubtractOffset(
  FileSystem *pFS, 
  Segment *pSeg,
  i64 iOff, 
  int iSub, 
  i64 *piRes
){
  i64 iStart;
  int iBlk = 0;
  int rc;

  assert( pFS->pCompress );

  iStart = fsFirstPageOnBlock(pFS, fsPageToBlock(pFS, iOff));
  if( (iOff-iSub)>=iStart ){
    *piRes = (iOff-iSub);
    return LSM_OK;
  }

  rc = fsBlockPrev(pFS, pSeg, fsPageToBlock(pFS, iOff), &iBlk);
  *piRes = fsLastPageOnBlock(pFS, iBlk) - iSub + (iOff - iStart + 1);
  return rc;
}

/*
** Add iAdd to database file offset iOff and set *piRes to the
** result. If doing so means passing the end of a block, follow the
** block pointer stored in the last 4 bytes of the block.
**
** Offset iOff is an absolute offset - not subject to any block redirection.
** However any block pointer followed is. Use pSeg->pRedirect in this case.
**
** Return LSM_OK if successful or an lsm error code if an error occurs.
*/
static int fsAddOffset(
  FileSystem *pFS, 
  Segment *pSeg,
  i64 iOff, 
  int iAdd, 
  i64 *piRes
){
  i64 iEob;
  int iBlk;
  int rc;

  assert( pFS->pCompress );

  iEob = fsLastPageOnPagesBlock(pFS, iOff);
  if( (iOff+iAdd)<=iEob ){
    *piRes = (iOff+iAdd);
    return LSM_OK;
  }

  rc = fsBlockNext(pFS, pSeg, fsPageToBlock(pFS, iOff), &iBlk);
  *piRes = fsFirstPageOnBlock(pFS, iBlk) + iAdd - (iEob - iOff + 1);
  return rc;
}

/*
** If it is not already allocated, allocate either the FileSystem.aOBuffer (if
** bWrite is true) or the FileSystem.aIBuffer (if bWrite is false). Return
** LSM_OK if successful if the attempt to allocate memory fails.
*/
static int fsAllocateBuffer(FileSystem *pFS, int bWrite){
  u8 **pp;                        /* Pointer to either aIBuffer or aOBuffer */

  assert( pFS->pCompress );

  /* If neither buffer has been allocated, figure out how large they
  ** should be. Store this value in FileSystem.nBuffer.  */
  if( pFS->nBuffer==0 ){
    assert( pFS->aIBuffer==0 && pFS->aOBuffer==0 );
    pFS->nBuffer = pFS->pCompress->xBound(pFS->pCompress->pCtx, pFS->nPagesize);
    if( pFS->nBuffer<(pFS->szSector+6) ){
      pFS->nBuffer = pFS->szSector+6;
    }
  }

  pp = (bWrite ? &pFS->aOBuffer : &pFS->aIBuffer);
  if( *pp==0 ){
    *pp = lsmMalloc(pFS->pEnv, LSM_MAX(pFS->nBuffer, pFS->nPagesize));
    if( *pp==0 ) return LSM_NOMEM_BKPT;
  }

  return LSM_OK;
}

/*
** This function is only called in compressed database mode. It reads and
** uncompresses the compressed data for page pPg from the database and
** populates the pPg->aData[] buffer and pPg->nCompress field.
**
** It is possible that instead of a page record, there is free space
** at offset pPg->iPgno. In this case no data is read from the file, but
** output variable *pnSpace is set to the total number of free bytes.
**
** LSM_OK is returned if successful, or an LSM error code otherwise.
*/
static int fsReadPagedata(
  FileSystem *pFS,                /* File-system handle */
  Segment *pSeg,                  /* pPg is part of this segment */
  Page *pPg,                      /* Page to read and uncompress data for */
  int *pnSpace                    /* OUT: Total bytes of free space */
){
  lsm_compress *p = pFS->pCompress;
  i64 iOff = pPg->iPg;
  u8 aSz[3];
  int rc;

  assert( p && pPg->nCompress==0 );

  if( fsAllocateBuffer(pFS, 0) ) return LSM_NOMEM;

  rc = fsReadData(pFS, pSeg, iOff, aSz, sizeof(aSz));

  if( rc==LSM_OK ){
    int bFree;
    if( aSz[0] & 0x80 ){
      pPg->nCompress = (int)getRecordSize(aSz, &bFree);
    }else{
      pPg->nCompress = (int)aSz[0] - sizeof(aSz)*2;
      bFree = 1;
    }
    if( bFree ){
      if( pnSpace ){
        *pnSpace = pPg->nCompress + sizeof(aSz)*2;
      }else{
        rc = LSM_CORRUPT_BKPT;
      }
    }else{
      rc = fsAddOffset(pFS, pSeg, iOff, 3, &iOff);
      if( rc==LSM_OK ){
        if( pPg->nCompress>pFS->nBuffer ){
          rc = LSM_CORRUPT_BKPT;
        }else{
          rc = fsReadData(pFS, pSeg, iOff, pFS->aIBuffer, pPg->nCompress);
        }
        if( rc==LSM_OK ){
          int n = pFS->nPagesize;
          rc = p->xUncompress(p->pCtx, 
              (char *)pPg->aData, &n, 
              (const char *)pFS->aIBuffer, pPg->nCompress
          );
          if( rc==LSM_OK && n!=pPg->pFS->nPagesize ){
            rc = LSM_CORRUPT_BKPT;
          }
        }
      }
    }
  }
  return rc;
}

/*
** Return a handle for a database page.
**
** If this file-system object is accessing a compressed database it may be
** that there is no page record at database file offset iPg. Instead, there
** may be a free space record. In this case, set *ppPg to NULL and *pnSpace
** to the total number of free bytes before returning.
**
** If no error occurs, LSM_OK is returned. Otherwise, an lsm error code.
*/
static int fsPageGet(
  FileSystem *pFS,                /* File-system handle */
  Segment *pSeg,                  /* Block redirection to use (or NULL) */
  Pgno iPg,                       /* Page id */
  int noContent,                  /* True to not load content from disk */
  Page **ppPg,                    /* OUT: New page handle */
  int *pnSpace                    /* OUT: Bytes of free space */
){
  Page *p;
  int iHash;
  int rc = LSM_OK;

  /* In most cases iReal is the same as iPg. Except, if pSeg->pRedirect is 
  ** not NULL, and the block containing iPg has been redirected, then iReal
  ** is the page number after redirection.  */
  Pgno iReal = lsmFsRedirectPage(pFS, (pSeg ? pSeg->pRedirect : 0), iPg);

  assert_lists_are_ok(pFS);
  assert( iPg>=fsFirstPageOnBlock(pFS, 1) );
  assert( iReal>=fsFirstPageOnBlock(pFS, 1) );
  *ppPg = 0;

  /* Search the hash-table for the page */
  p = fsPageFindInHash(pFS, iReal, &iHash);

  if( p ){
    assert( p->flags & PAGE_FREE );
    if( p->nRef==0 ) fsPageRemoveFromLru(pFS, p);
  }else{

    if( fsMmapPage(pFS, iReal) ){
      i64 iEnd = (i64)iReal * pFS->nPagesize;
      fsGrowMapping(pFS, iEnd, &rc);
      if( rc!=LSM_OK ) return rc;

      if( pFS->pFree ){
        p = pFS->pFree;
        pFS->pFree = p->pFreeNext;
        assert( p->nRef==0 );
      }else{
        p = lsmMallocZeroRc(pFS->pEnv, sizeof(Page), &rc);
        if( rc ) return rc;
        p->pFS = pFS;
      }
      p->aData = &((u8 *)pFS->pMap)[pFS->nPagesize * (iReal-1)];
      p->iPg = iReal;

      /* This page now carries a pointer to the mapping. Link it in to
      ** the FileSystem.pMapped list.  */
      assert( p->pMappedNext==0 );
      p->pMappedNext = pFS->pMapped;
      pFS->pMapped = p;

      assert( pFS->pCompress==0 );
      assert( (p->flags & PAGE_FREE)==0 );
    }else{
      rc = fsPageBuffer(pFS, &p);
      if( rc==LSM_OK ){
        int nSpace = 0;
        p->iPg = iReal;
        p->nRef = 0;
        p->pFS = pFS;
        assert( p->flags==0 || p->flags==PAGE_FREE );

#ifdef LSM_DEBUG
        memset(p->aData, 0x56, pFS->nPagesize);
#endif
        assert( p->pLruNext==0 && p->pLruPrev==0 );
        if( noContent==0 ){
          if( pFS->pCompress ){
            rc = fsReadPagedata(pFS, pSeg, p, &nSpace);
          }else{
            int nByte = pFS->nPagesize;
            i64 iOff = (i64)(iReal-1) * pFS->nPagesize;
            rc = lsmEnvRead(pFS->pEnv, pFS->fdDb, iOff, p->aData, nByte);
          }
          pFS->nRead++;
        }

        /* If the xRead() call was successful (or not attempted), link the
        ** page into the page-cache hash-table. Otherwise, if it failed,
        ** free the buffer. */
        if( rc==LSM_OK && nSpace==0 ){
          p->pHashNext = pFS->apHash[iHash];
          pFS->apHash[iHash] = p;
        }else{
          fsPageBufferFree(p);
          p = 0;
          if( pnSpace ) *pnSpace = nSpace;
        }
      }
    }

    assert( (rc==LSM_OK && (p || (pnSpace && *pnSpace)))
         || (rc!=LSM_OK && p==0) 
    );
  }

  if( rc==LSM_OK && p ){
    if( pFS->pCompress==0 && (fsIsLast(pFS, iReal) || fsIsFirst(pFS, iReal)) ){
      p->nData = pFS->nPagesize - 4;
      if( fsIsFirst(pFS, iReal) && p->nRef==0 ){
        p->aData += 4;
        p->flags |= PAGE_HASPREV;
      }
    }else{
      p->nData = pFS->nPagesize;
    }
    pFS->nOut += (p->nRef==0);
    p->nRef++;
  }
  *ppPg = p;
  return rc;
}

/*
** Read the 64-bit checkpoint id of the checkpoint currently stored on meta
** page iMeta of the database file. If no error occurs, store the id value
** in *piVal and return LSM_OK. Otherwise, return an LSM error code and leave
** *piVal unmodified.
**
** If a checkpointer connection is currently updating meta-page iMeta, or an
** earlier checkpointer crashed while doing so, the value read into *piVal
** may be garbage. It is the callers responsibility to deal with this.
*/
int lsmFsReadSyncedId(lsm_db *db, int iMeta, i64 *piVal){
  FileSystem *pFS = db->pFS;
  int rc = LSM_OK;

  assert( iMeta==1 || iMeta==2 );
  if( pFS->nMapLimit>0 ){
    fsGrowMapping(pFS, iMeta*LSM_META_PAGE_SIZE, &rc);
    if( rc==LSM_OK ){
      *piVal = (i64)lsmGetU64(&((u8 *)pFS->pMap)[(iMeta-1)*LSM_META_PAGE_SIZE]);
    }
  }else{
    MetaPage *pMeta = 0;
    rc = lsmFsMetaPageGet(pFS, 0, iMeta, &pMeta);
    if( rc==LSM_OK ){
      *piVal = (i64)lsmGetU64(pMeta->aData);
      lsmFsMetaPageRelease(pMeta);
    }
  }

  return rc;
}


/*
** Return true if the first or last page of segment pRun falls between iFirst
** and iLast, inclusive, and pRun is not equal to pIgnore.
*/
static int fsRunEndsBetween(
  Segment *pRun, 
  Segment *pIgnore, 
  Pgno iFirst, 
  Pgno iLast
){
  return (pRun!=pIgnore && (
        (pRun->iFirst>=iFirst && pRun->iFirst<=iLast)
     || (pRun->iLastPg>=iFirst && pRun->iLastPg<=iLast)
  ));
}

/*
** Return true if level pLevel contains a segment other than pIgnore for
** which the first or last page is between iFirst and iLast, inclusive.
*/
static int fsLevelEndsBetween(
  Level *pLevel, 
  Segment *pIgnore, 
  Pgno iFirst, 
  Pgno iLast
){
  int i;

  if( fsRunEndsBetween(&pLevel->lhs, pIgnore, iFirst, iLast) ){
    return 1;
  }
  for(i=0; i<pLevel->nRight; i++){
    if( fsRunEndsBetween(&pLevel->aRhs[i], pIgnore, iFirst, iLast) ){
      return 1;
    }
  }

  return 0;
}

/*
** Block iBlk is no longer in use by segment pIgnore. If it is not in use
** by any other segment, move it to the free block list.
*/
static int fsFreeBlock(
  FileSystem *pFS,                /* File system object */
  Snapshot *pSnapshot,            /* Worker snapshot */
  Segment *pIgnore,               /* Ignore this run when searching */
  int iBlk                        /* Block number of block to free */
){
  int rc = LSM_OK;                /* Return code */
  int iFirst;                     /* First page on block iBlk */
  int iLast;                      /* Last page on block iBlk */
  Level *pLevel;                  /* Used to iterate through levels */

  int iIn;                        /* Used to iterate through append points */
  int iOut = 0;                   /* Used to output append points */
  Pgno *aApp = pSnapshot->aiAppend;

  iFirst = fsFirstPageOnBlock(pFS, iBlk);
  iLast = fsLastPageOnBlock(pFS, iBlk);

  /* Check if any other run in the snapshot has a start or end page 
  ** within this block. If there is such a run, return early. */
  for(pLevel=lsmDbSnapshotLevel(pSnapshot); pLevel; pLevel=pLevel->pNext){
    if( fsLevelEndsBetween(pLevel, pIgnore, iFirst, iLast) ){
      return LSM_OK;
    }
  }

  /* Remove any entries that lie on this block from the append-list. */
  for(iIn=0; iIn<LSM_APPLIST_SZ; iIn++){
    if( aApp[iIn]<iFirst || aApp[iIn]>iLast ){
      aApp[iOut++] = aApp[iIn];
    }
  }
  while( iOut<LSM_APPLIST_SZ ) aApp[iOut++] = 0;

  if( rc==LSM_OK ){
    rc = lsmBlockFree(pFS->pDb, iBlk);
  }
  return rc;
}

/*
** Delete or otherwise recycle the blocks currently occupied by run pDel.
*/
int lsmFsSortedDelete(
  FileSystem *pFS, 
  Snapshot *pSnapshot,
  int bZero,                      /* True to zero the Segment structure */
  Segment *pDel
){
  if( pDel->iFirst ){
    int rc = LSM_OK;

    int iBlk;
    int iLastBlk;

    iBlk = fsPageToBlock(pFS, pDel->iFirst);
    iLastBlk = fsPageToBlock(pFS, pDel->iLastPg);

    /* Mark all blocks currently used by this sorted run as free */
    while( iBlk && rc==LSM_OK ){
      int iNext = 0;
      if( iBlk!=iLastBlk ){
        rc = fsBlockNext(pFS, pDel, iBlk, &iNext);
      }else if( bZero==0 && pDel->iLastPg!=fsLastPageOnBlock(pFS, iLastBlk) ){
        break;
      }
      rc = fsFreeBlock(pFS, pSnapshot, pDel, iBlk);
      iBlk = iNext;
    }

    if( pDel->pRedirect ){
      assert( pDel->pRedirect==&pSnapshot->redirect );
      pSnapshot->redirect.n = 0;
    }

    if( bZero ) memset(pDel, 0, sizeof(Segment));
  }
  return LSM_OK;
}

/*
** aPgno is an array containing nPgno page numbers. Return the smallest page
** number from the array that falls on block iBlk. Or, if none of the pages
** in aPgno[] fall on block iBlk, return 0.
*/
static Pgno firstOnBlock(FileSystem *pFS, int iBlk, Pgno *aPgno, int nPgno){
  Pgno iRet = 0;
  int i;
  for(i=0; i<nPgno; i++){
    Pgno iPg = aPgno[i];
    if( fsPageToBlock(pFS, iPg)==iBlk && (iRet==0 || iPg<iRet) ){
      iRet = iPg;
    }
  }
  return iRet;
}

#ifndef NDEBUG
/*
** Return true if page iPg, which is a part of segment p, lies on
** a redirected block. 
*/
static int fsPageRedirects(FileSystem *pFS, Segment *p, Pgno iPg){
  return (iPg!=0 && iPg!=lsmFsRedirectPage(pFS, p->pRedirect, iPg));
}

/*
** Return true if the second argument is not NULL and any of the first
** last or root pages lie on a redirected block. 
*/
static int fsSegmentRedirects(FileSystem *pFS, Segment *p){
  return (p && (
      fsPageRedirects(pFS, p, p->iFirst)
   || fsPageRedirects(pFS, p, p->iRoot)
   || fsPageRedirects(pFS, p, p->iLastPg)
  ));
}
#endif

/*
** Argument aPgno is an array of nPgno page numbers. All pages belong to
** the segment pRun. This function gobbles from the start of the run to the
** first page that appears in aPgno[] (i.e. so that the aPgno[] entry is
** the new first page of the run).
*/
void lsmFsGobble(
  lsm_db *pDb,
  Segment *pRun, 
  Pgno *aPgno,
  int nPgno
){
  int rc = LSM_OK;
  FileSystem *pFS = pDb->pFS;
  Snapshot *pSnapshot = pDb->pWorker;
  int iBlk;

  assert( pRun->nSize>0 );
  assert( 0==fsSegmentRedirects(pFS, pRun) );
  assert( nPgno>0 && 0==fsPageRedirects(pFS, pRun, aPgno[0]) );

  iBlk = fsPageToBlock(pFS, pRun->iFirst);
  pRun->nSize += (pRun->iFirst - fsFirstPageOnBlock(pFS, iBlk));

  while( rc==LSM_OK ){
    int iNext = 0;
    Pgno iFirst = firstOnBlock(pFS, iBlk, aPgno, nPgno);
    if( iFirst ){
      pRun->iFirst = iFirst;
      break;
    }
    rc = fsBlockNext(pFS, pRun, iBlk, &iNext);
    if( rc==LSM_OK ) rc = fsFreeBlock(pFS, pSnapshot, pRun, iBlk);
    pRun->nSize -= (
        1 + fsLastPageOnBlock(pFS, iBlk) - fsFirstPageOnBlock(pFS, iBlk)
    );
    iBlk = iNext;
  }

  pRun->nSize -= (pRun->iFirst - fsFirstPageOnBlock(pFS, iBlk));
  assert( pRun->nSize>0 );
}

/*
** This function is only used in compressed database mode.
**
** Argument iPg is the page number (byte offset) of a page within segment
** pSeg. The page record, including all headers, is nByte bytes in size.
** Before returning, set *piNext to the page number of the next page in
** the segment, or to zero if iPg is the last.
**
** In other words, do:
**
**   *piNext = iPg + nByte;
**
** But take block overflow and redirection into account.
*/
static int fsNextPageOffset(
  FileSystem *pFS,                /* File system object */
  Segment *pSeg,                  /* Segment to move within */
  Pgno iPg,                       /* Offset of current page */
  int nByte,                      /* Size of current page including headers */
  Pgno *piNext                    /* OUT: Offset of next page. Or zero (EOF) */
){
  Pgno iNext;
  int rc;

  assert( pFS->pCompress );

  rc = fsAddOffset(pFS, pSeg, iPg, nByte-1, &iNext);
  if( pSeg && iNext==pSeg->iLastPg ){
    iNext = 0;
  }else if( rc==LSM_OK ){
    rc = fsAddOffset(pFS, pSeg, iNext, 1, &iNext);
  }

  *piNext = iNext;
  return rc;
}

/*
** This function is only used in compressed database mode.
**
** Argument iPg is the page number of a pagethat appears in segment pSeg.
** This function determines the page number of the previous page in the
** same run. *piPrev is set to the previous page number before returning.
**
** LSM_OK is returned if no error occurs. Otherwise, an lsm error code.
** If any value other than LSM_OK is returned, then the final value of
** *piPrev is undefined.
*/
static int fsGetPageBefore(
  FileSystem *pFS, 
  Segment *pSeg, 
  Pgno iPg, 
  Pgno *piPrev
){
  u8 aSz[3];
  int rc;
  i64 iRead;

  assert( pFS->pCompress );

  rc = fsSubtractOffset(pFS, pSeg, iPg, sizeof(aSz), &iRead);
  if( rc==LSM_OK ) rc = fsReadData(pFS, pSeg, iRead, aSz, sizeof(aSz));

  if( rc==LSM_OK ){
    int bFree;
    int nSz;
    if( aSz[2] & 0x80 ){
      nSz = getRecordSize(aSz, &bFree) + sizeof(aSz)*2;
    }else{
      nSz = (int)(aSz[2] & 0x7F);
      bFree = 1;
    }
    rc = fsSubtractOffset(pFS, pSeg, iPg, nSz, piPrev);
  }

  return rc;
}

/*
** The first argument to this function is a valid reference to a database
** file page that is part of a sorted run. If parameter eDir is -1, this 
** function attempts to locate and load the previous page in the same run. 
** Or, if eDir is +1, it attempts to find the next page in the same run.
** The results of passing an eDir value other than positive or negative one
** are undefined.
**
** If parameter pRun is not NULL then it must point to the run that page
** pPg belongs to. In this case, if pPg is the first or last page of the
** run, and the request is for the previous or next page, respectively,
** *ppNext is set to NULL before returning LSM_OK. If pRun is NULL, then it
** is assumed that the next or previous page, as requested, exists.
**
** If the previous/next page does exist and is successfully loaded, *ppNext
** is set to point to it and LSM_OK is returned. Otherwise, if an error 
** occurs, *ppNext is set to NULL and and lsm error code returned.
**
** Page references returned by this function should be released by the 
** caller using lsmFsPageRelease().
*/
int lsmFsDbPageNext(Segment *pRun, Page *pPg, int eDir, Page **ppNext){
  int rc = LSM_OK;
  FileSystem *pFS = pPg->pFS;
  Pgno iPg = pPg->iPg;

  assert( 0==fsSegmentRedirects(pFS, pRun) );
  if( pFS->pCompress ){
    int nSpace = pPg->nCompress + 2*3;

    do {
      if( eDir>0 ){
        rc = fsNextPageOffset(pFS, pRun, iPg, nSpace, &iPg);
      }else{
        if( iPg==pRun->iFirst ){
          iPg = 0;
        }else{
          rc = fsGetPageBefore(pFS, pRun, iPg, &iPg);
        }
      }

      nSpace = 0;
      if( iPg!=0 ){
        rc = fsPageGet(pFS, pRun, iPg, 0, ppNext, &nSpace);
        assert( (*ppNext==0)==(rc!=LSM_OK || nSpace>0) );
      }else{
        *ppNext = 0;
      }
    }while( nSpace>0 && rc==LSM_OK );

  }else{
    Redirect *pRedir = pRun ? pRun->pRedirect : 0;
    assert( eDir==1 || eDir==-1 );
    if( eDir<0 ){
      if( pRun && iPg==pRun->iFirst ){
        *ppNext = 0;
        return LSM_OK;
      }else if( fsIsFirst(pFS, iPg) ){
        assert( pPg->flags & PAGE_HASPREV );
        iPg = fsLastPageOnBlock(pFS, lsmGetU32(&pPg->aData[-4]));
      }else{
        iPg--;
      }
    }else{
      if( pRun ){
        if( iPg==pRun->iLastPg ){
          *ppNext = 0;
          return LSM_OK;
        }
      }

      if( fsIsLast(pFS, iPg) ){
        int iBlk = fsRedirectBlock(
            pRedir, lsmGetU32(&pPg->aData[pFS->nPagesize-4])
        );
        iPg = fsFirstPageOnBlock(pFS, iBlk);
      }else{
        iPg++;
      }
    }
    rc = fsPageGet(pFS, pRun, iPg, 0, ppNext, 0);
  }

  return rc;
}

/*
** This function is called when creating a new segment to determine if the
** first part of it can be written following an existing segment on an
** already allocated block. If it is possible, the page number of the first
** page to use for the new segment is returned. Otherwise zero.
**
** If argument pLvl is not NULL, then this function will not attempt to
** start the new segment immediately following any segment that is part
** of the right-hand-side of pLvl.
*/
static Pgno findAppendPoint(FileSystem *pFS, Level *pLvl){
  int i;
  Pgno *aiAppend = pFS->pDb->pWorker->aiAppend;
  Pgno iRet = 0;

  for(i=LSM_APPLIST_SZ-1; iRet==0 && i>=0; i--){
    if( (iRet = aiAppend[i]) ){
      if( pLvl ){
        int iBlk = fsPageToBlock(pFS, iRet);
        int j;
        for(j=0; iRet && j<pLvl->nRight; j++){
          if( fsPageToBlock(pFS, pLvl->aRhs[j].iLastPg)==iBlk ){
            iRet = 0;
          }
        }
      }
      if( iRet ) aiAppend[i] = 0;
    }
  }
  return iRet;
}

/*
** Append a page to the left-hand-side of pLvl. Set the ref-count to 1 and
** return a pointer to it. The page is writable until either 
** lsmFsPagePersist() is called on it or the ref-count drops to zero.
*/
int lsmFsSortedAppend(
  FileSystem *pFS, 
  Snapshot *pSnapshot,
  Level *pLvl,
  int bDefer,
  Page **ppOut
){
  int rc = LSM_OK;
  Page *pPg = 0;
  int iApp = 0;
  int iNext = 0;
  Segment *p = &pLvl->lhs;
  int iPrev = p->iLastPg;

  *ppOut = 0;
  assert( p->pRedirect==0 );

  if( pFS->pCompress || bDefer ){
    /* In compressed database mode the page is not assigned a page number
    ** or location in the database file at this point. This will be done
    ** by the lsmFsPagePersist() call.  */
    rc = fsPageBuffer(pFS, &pPg);
    if( rc==LSM_OK ){
      pPg->pFS = pFS;
      pPg->pSeg = p;
      pPg->iPg = 0;
      pPg->flags |= PAGE_DIRTY;
      pPg->nData = pFS->nPagesize;
      assert( pPg->aData );
      if( pFS->pCompress==0 ) pPg->nData -= 4;

      pPg->nRef = 1;
      pFS->nOut++;
    }
  }else{
    if( iPrev==0 ){
      iApp = findAppendPoint(pFS, pLvl);
    }else if( fsIsLast(pFS, iPrev) ){
      int iNext;
      rc = fsBlockNext(pFS, 0, fsPageToBlock(pFS, iPrev), &iNext);
      if( rc!=LSM_OK ) return rc;
      iApp = fsFirstPageOnBlock(pFS, iNext);
    }else{
      iApp = iPrev + 1;
    }

    /* If this is the first page allocated, or if the page allocated is the
    ** last in the block, also allocate the next block here.  */
    if( iApp==0 || fsIsLast(pFS, iApp) ){
      int iNew;                     /* New block number */

      rc = lsmBlockAllocate(pFS->pDb, 0, &iNew);
      if( rc!=LSM_OK ) return rc;
      if( iApp==0 ){
        iApp = fsFirstPageOnBlock(pFS, iNew);
      }else{
        iNext = fsFirstPageOnBlock(pFS, iNew);
      }
    }

    /* Grab the new page. */
    pPg = 0;
    rc = fsPageGet(pFS, 0, iApp, 1, &pPg, 0);
    assert( rc==LSM_OK || pPg==0 );

    /* If this is the first or last page of a block, fill in the pointer 
     ** value at the end of the new page. */
    if( rc==LSM_OK ){
      p->nSize++;
      p->iLastPg = iApp;
      if( p->iFirst==0 ) p->iFirst = iApp;
      pPg->flags |= PAGE_DIRTY;

      if( fsIsLast(pFS, iApp) ){
        lsmPutU32(&pPg->aData[pFS->nPagesize-4], fsPageToBlock(pFS, iNext));
      }else if( fsIsFirst(pFS, iApp) ){
        lsmPutU32(&pPg->aData[-4], fsPageToBlock(pFS, iPrev));
      }
    }
  }

  *ppOut = pPg;
  return rc;
}

/*
** Mark the segment passed as the second argument as finished. Once a segment
** is marked as finished it is not possible to append any further pages to 
** it.
**
** Return LSM_OK if successful or an lsm error code if an error occurs.
*/
int lsmFsSortedFinish(FileSystem *pFS, Segment *p){
  int rc = LSM_OK;
  if( p && p->iLastPg ){
    assert( p->pRedirect==0 );

    /* Check if the last page of this run happens to be the last of a block.
    ** If it is, then an extra block has already been allocated for this run.
    ** Shift this extra block back to the free-block list. 
    **
    ** Otherwise, add the first free page in the last block used by the run
    ** to the lAppend list.
    */
    if( fsLastPageOnPagesBlock(pFS, p->iLastPg)!=p->iLastPg ){
      int i;
      Pgno *aiAppend = pFS->pDb->pWorker->aiAppend;
      for(i=0; i<LSM_APPLIST_SZ; i++){
        if( aiAppend[i]==0 ){
          aiAppend[i] = p->iLastPg+1;
          break;
        }
      }
    }else if( pFS->pCompress==0 ){
      Page *pLast;
      rc = fsPageGet(pFS, 0, p->iLastPg, 0, &pLast, 0);
      if( rc==LSM_OK ){
        int iBlk = (int)lsmGetU32(&pLast->aData[pFS->nPagesize-4]);
        lsmBlockRefree(pFS->pDb, iBlk);
        lsmFsPageRelease(pLast);
      }
    }else{
      int iBlk = 0;
      rc = fsBlockNext(pFS, p, fsPageToBlock(pFS, p->iLastPg), &iBlk);
      if( rc==LSM_OK ){
        lsmBlockRefree(pFS->pDb, iBlk);
      }
    }
  }
  return rc;
}

/*
** Obtain a reference to page number iPg.
**
** Return LSM_OK if successful, or an lsm error code if an error occurs.
*/
int lsmFsDbPageGet(FileSystem *pFS, Segment *pSeg, Pgno iPg, Page **ppPg){
  return fsPageGet(pFS, pSeg, iPg, 0, ppPg, 0);
}

/*
** Obtain a reference to the last page in the segment passed as the 
** second argument.
**
** Return LSM_OK if successful, or an lsm error code if an error occurs.
*/
int lsmFsDbPageLast(FileSystem *pFS, Segment *pSeg, Page **ppPg){
  int rc;
  Pgno iPg = pSeg->iLastPg;
  if( pFS->pCompress ){
    int nSpace;
    iPg++;
    do {
      nSpace = 0;
      rc = fsGetPageBefore(pFS, pSeg, iPg, &iPg);
      if( rc==LSM_OK ){
        rc = fsPageGet(pFS, pSeg, iPg, 0, ppPg, &nSpace);
      }
    }while( rc==LSM_OK && nSpace>0 );

  }else{
    rc = fsPageGet(pFS, pSeg, iPg, 0, ppPg, 0);
  }
  return rc;
}

/*
** Return a reference to meta-page iPg. If successful, LSM_OK is returned
** and *ppPg populated with the new page reference. The reference should
** be released by the caller using lsmFsPageRelease().
**
** Otherwise, if an error occurs, *ppPg is set to NULL and an LSM error 
** code is returned.
*/
int lsmFsMetaPageGet(
  FileSystem *pFS,                /* File-system connection */
  int bWrite,                     /* True for write access, false for read */
  int iPg,                        /* Either 1 or 2 */
  MetaPage **ppPg                 /* OUT: Pointer to MetaPage object */
){
  int rc = LSM_OK;
  MetaPage *pPg;
  assert( iPg==1 || iPg==2 );

  pPg = lsmMallocZeroRc(pFS->pEnv, sizeof(Page), &rc);

  if( pPg ){
    i64 iOff = (iPg-1) * pFS->nMetasize;
    if( pFS->nMapLimit>0 ){
      fsGrowMapping(pFS, 2*pFS->nMetasize, &rc);
      pPg->aData = (u8 *)(pFS->pMap) + iOff;
    }else{
      pPg->aData = lsmMallocRc(pFS->pEnv, pFS->nMetasize, &rc);
      if( rc==LSM_OK && bWrite==0 ){
        rc = lsmEnvRead(pFS->pEnv, pFS->fdDb, iOff, pPg->aData, pFS->nMetasize);
      }
#ifndef NDEBUG
      /* pPg->aData causes an uninitialized access via a downstreadm write().
         After discussion on this list, this memory should not, for performance
         reasons, be memset. However, tracking down "real" misuse is more
         difficult with this "false" positive, so it is set when NDEBUG.
      */
      else if( rc==LSM_OK ){
        memset( pPg->aData, 0x77, pFS->nMetasize );
      }
#endif
    }

    if( rc!=LSM_OK ){
      if( pFS->nMapLimit==0 ) lsmFree(pFS->pEnv, pPg->aData);
      lsmFree(pFS->pEnv, pPg);
      pPg = 0;
    }else{
      pPg->iPg = iPg;
      pPg->bWrite = bWrite;
      pPg->pFS = pFS;
    }
  }

  *ppPg = pPg;
  return rc;
}

/*
** Release a meta-page reference obtained via a call to lsmFsMetaPageGet().
*/
int lsmFsMetaPageRelease(MetaPage *pPg){
  int rc = LSM_OK;
  if( pPg ){
    FileSystem *pFS = pPg->pFS;

    if( pFS->nMapLimit==0 ){
      if( pPg->bWrite ){
        i64 iOff = (pPg->iPg==2 ? pFS->nMetasize : 0);
        int nWrite = pFS->nMetasize;
        rc = lsmEnvWrite(pFS->pEnv, pFS->fdDb, iOff, pPg->aData, nWrite);
      }
      lsmFree(pFS->pEnv, pPg->aData);
    }

    lsmFree(pFS->pEnv, pPg);
  }
  return rc;
}

/*
** Return a pointer to a buffer containing the data associated with the
** meta-page passed as the first argument. If parameter pnData is not NULL,
** set *pnData to the size of the meta-page in bytes before returning.
*/
u8 *lsmFsMetaPageData(MetaPage *pPg, int *pnData){
  if( pnData ) *pnData = pPg->pFS->nMetasize;
  return pPg->aData;
}

/*
** Return true if page is currently writable. This is used in assert() 
** statements only.
*/
#ifndef NDEBUG
int lsmFsPageWritable(Page *pPg){
  return (pPg->flags & PAGE_DIRTY) ? 1 : 0;
}
#endif

/*
** This is called when block iFrom is being redirected to iTo. If page 
** number (*piPg) lies on block iFrom, then calculate the equivalent
** page on block iTo and set *piPg to this value before returning.
*/
static void fsMovePage(
  FileSystem *pFS,                /* File system object */
  int iTo,                        /* Destination block */
  int iFrom,                      /* Source block */
  Pgno *piPg                      /* IN/OUT: Page number */
){
  Pgno iPg = *piPg;
  if( iFrom==fsPageToBlock(pFS, iPg) ){
    const int nPagePerBlock = (
        pFS->pCompress ? pFS ->nBlocksize : (pFS->nBlocksize / pFS->nPagesize)
    );
    *piPg = iPg - (Pgno)(iFrom - iTo) * nPagePerBlock;
  }
}

/*
** Copy the contents of block iFrom to block iTo. 
**
** It is safe to assume that there are no outstanding references to pages 
** on block iTo. And that block iFrom is not currently being written. In
** other words, the data can be read and written directly.
*/
int lsmFsMoveBlock(FileSystem *pFS, Segment *pSeg, int iTo, int iFrom){
  Snapshot *p = pFS->pDb->pWorker;
  int rc = LSM_OK;
  int i;
  i64 nMap;

  i64 iFromOff = (i64)(iFrom-1) * pFS->nBlocksize;
  i64 iToOff = (i64)(iTo-1) * pFS->nBlocksize;
  
  assert( iTo!=1 );
  assert( iFrom>iTo );

  /* Grow the mapping as required. */
  nMap = LSM_MIN(pFS->nMapLimit, (i64)iFrom * pFS->nBlocksize);
  fsGrowMapping(pFS, nMap, &rc);

  if( rc==LSM_OK ){
    const int nPagePerBlock = (pFS->nBlocksize / pFS->nPagesize);
    int nSz = pFS->nPagesize;
    u8 *aBuf = 0;
    u8 *aData = 0;

    for(i=0; rc==LSM_OK && i<nPagePerBlock; i++){
      i64 iOff = iFromOff + i*nSz;

      /* Set aData to point to a buffer containing the from page */
      if( (iOff+nSz)<=pFS->nMapLimit ){
        u8 *aMap = (u8 *)(pFS->pMap);
        aData = &aMap[iOff];
      }else{
        if( aBuf==0 ){
          aBuf = (u8 *)lsmMallocRc(pFS->pEnv, nSz, &rc);
          if( aBuf==0 ) break;
        }
        aData = aBuf;
        rc = lsmEnvRead(pFS->pEnv, pFS->fdDb, iOff, aData, nSz);
      }

      /* Copy aData to the to page */
      if( rc==LSM_OK ){
        iOff = iToOff + i*nSz;
        if( (iOff+nSz)<=pFS->nMapLimit ){
          u8 *aMap = (u8 *)(pFS->pMap);
          memcpy(&aMap[iOff], aData, nSz);
        }else{
          rc = lsmEnvWrite(pFS->pEnv, pFS->fdDb, iOff, aData, nSz);
        }
      }
    }
    lsmFree(pFS->pEnv, aBuf);
    lsmFsPurgeCache(pFS);
  }

  /* Update append-point list if necessary */
  for(i=0; i<LSM_APPLIST_SZ; i++){
    fsMovePage(pFS, iTo, iFrom, &p->aiAppend[i]);
  }

  /* Update the Segment structure itself */
  fsMovePage(pFS, iTo, iFrom, &pSeg->iFirst);
  fsMovePage(pFS, iTo, iFrom, &pSeg->iLastPg);
  fsMovePage(pFS, iTo, iFrom, &pSeg->iRoot);

  return rc;
}

/*
** Append raw data to a segment. Return the database file offset that the
** data is written to (this may be used as the page number if the data
** being appended is a new page record).
**
** This function is only used in compressed database mode.
*/
static Pgno fsAppendData(
  FileSystem *pFS,                /* File-system handle */
  Segment *pSeg,                  /* Segment to append to */
  const u8 *aData,                /* Buffer containing data to write */
  int nData,                      /* Size of buffer aData[] in bytes */
  int *pRc                        /* IN/OUT: Error code */
){
  Pgno iRet = 0;
  int rc = *pRc;
  assert( pFS->pCompress );
  if( rc==LSM_OK ){
    int nRem;
    int nWrite;
    Pgno iLastOnBlock;
    Pgno iApp = pSeg->iLastPg+1;

    /* If this is the first data written into the segment, find an append-point
    ** or allocate a new block.  */
    if( iApp==1 ){
      pSeg->iFirst = iApp = findAppendPoint(pFS, 0);
      if( iApp==0 ){
        int iBlk;
        rc = lsmBlockAllocate(pFS->pDb, 0, &iBlk);
        pSeg->iFirst = iApp = fsFirstPageOnBlock(pFS, iBlk);
      }
    }
    iRet = iApp;

    /* Write as much data as is possible at iApp (usually all of it). */
    iLastOnBlock = fsLastPageOnPagesBlock(pFS, iApp);
    if( rc==LSM_OK ){
      int nSpace = iLastOnBlock - iApp + 1;
      nWrite = LSM_MIN(nData, nSpace);
      nRem = nData - nWrite;
      assert( nWrite>=0 );
      if( nWrite!=0 ){
        rc = lsmEnvWrite(pFS->pEnv, pFS->fdDb, iApp, aData, nWrite);
      }
      iApp += nWrite;
    }

    /* If required, allocate a new block and write the rest of the data
    ** into it. Set the next and previous block pointers to link the new
    ** block to the old.  */
    assert( nRem<=0 || (iApp-1)==iLastOnBlock );
    if( rc==LSM_OK && (iApp-1)==iLastOnBlock ){
      u8 aPtr[4];                 /* Space to serialize a u32 */
      int iBlk;                   /* New block number */

      if( nWrite>0 ){
        /* Allocate a new block. */
        rc = lsmBlockAllocate(pFS->pDb, 0, &iBlk);

        /* Set the "next" pointer on the old block */
        if( rc==LSM_OK ){
          assert( iApp==(fsPageToBlock(pFS, iApp)*pFS->nBlocksize)-4 );
          lsmPutU32(aPtr, iBlk);
          rc = lsmEnvWrite(pFS->pEnv, pFS->fdDb, iApp, aPtr, sizeof(aPtr));
        }

        /* Set the "prev" pointer on the new block */
        if( rc==LSM_OK ){
          Pgno iWrite;
          lsmPutU32(aPtr, fsPageToBlock(pFS, iApp));
          iWrite = fsFirstPageOnBlock(pFS, iBlk);
          rc = lsmEnvWrite(pFS->pEnv, pFS->fdDb, iWrite-4, aPtr, sizeof(aPtr));
          if( nRem>0 ) iApp = iWrite;
        }
      }else{
        /* The next block is already allocated. */
        assert( nRem>0 );
        assert( pSeg->pRedirect==0 );
        rc = fsBlockNext(pFS, 0, fsPageToBlock(pFS, iApp), &iBlk);
        iRet = iApp = fsFirstPageOnBlock(pFS, iBlk);
      }

      /* Write the remaining data into the new block */
      if( rc==LSM_OK && nRem>0 ){
        rc = lsmEnvWrite(pFS->pEnv, pFS->fdDb, iApp, &aData[nWrite], nRem);
        iApp += nRem;
      }
    }

    pSeg->iLastPg = iApp-1;
    *pRc = rc;
  }

  return iRet;
}

/*
** This function is only called in compressed database mode. It 
** compresses the contents of page pPg and writes the result to the 
** buffer at pFS->aOBuffer. The size of the compressed data is stored in
** pPg->nCompress.
**
** If buffer pFS->aOBuffer[] has not been allocated then this function
** allocates it. If this fails, LSM_NOMEM is returned. Otherwise, LSM_OK.
*/
static int fsCompressIntoBuffer(FileSystem *pFS, Page *pPg){
  lsm_compress *p = pFS->pCompress;

  if( fsAllocateBuffer(pFS, 1) ) return LSM_NOMEM;
  assert( pPg->nData==pFS->nPagesize );

  pPg->nCompress = pFS->nBuffer;
  return p->xCompress(p->pCtx, 
      (char *)pFS->aOBuffer, &pPg->nCompress, 
      (const char *)pPg->aData, pPg->nData
  );
}

/*
** Append a new page to segment pSeg. Set output variable *piNew to the
** page number of the new page before returning.
**
** If the new page is the last on its block, then the 'next' block that
** will be used by the segment is allocated here too. In this case output
** variable *piNext is set to the block number of the next block.
**
** If the new page is the first on its block but not the first in the
** entire segment, set output variable *piPrev to the block number of
** the previous block in the segment.
**
** LSM_OK is returned if successful, or an lsm error code otherwise. If
** any value other than LSM_OK is returned, then the final value of all
** output variables is undefined.
*/
static int fsAppendPage(
  FileSystem *pFS, 
  Segment *pSeg,
  Pgno *piNew,
  int *piPrev,
  int *piNext
){
  Pgno iPrev = pSeg->iLastPg;
  int rc;
  assert( iPrev!=0 );

  *piPrev = 0;
  *piNext = 0;

  if( fsIsLast(pFS, iPrev) ){
    /* Grab the first page on the next block (which has already be
    ** allocated). In this case set *piPrev to tell the caller to set
    ** the "previous block" pointer in the first 4 bytes of the page.
    */
    int iNext;
    int iBlk = fsPageToBlock(pFS, iPrev);
    assert( pSeg->pRedirect==0 );
    rc = fsBlockNext(pFS, 0, iBlk, &iNext);
    if( rc!=LSM_OK ) return rc;
    *piNew = fsFirstPageOnBlock(pFS, iNext);
    *piPrev = iBlk;
  }else{
    *piNew = iPrev+1;
    if( fsIsLast(pFS, *piNew) ){
      /* Allocate the next block here. */
      int iBlk;
      rc = lsmBlockAllocate(pFS->pDb, 0, &iBlk);
      if( rc!=LSM_OK ) return rc;
      *piNext = iBlk;
    }
  }

  pSeg->nSize++;
  pSeg->iLastPg = *piNew;
  return LSM_OK;
}

/*
** Flush all pages in the FileSystem.pWaiting list to disk.
*/
void lsmFsFlushWaiting(FileSystem *pFS, int *pRc){
  int rc = *pRc;
  Page *pPg;

  pPg = pFS->pWaiting;
  pFS->pWaiting = 0;

  while( pPg ){
    Page *pNext = pPg->pWaitingNext;
    if( rc==LSM_OK ) rc = lsmFsPagePersist(pPg);
    assert( pPg->nRef==1 );
    lsmFsPageRelease(pPg);
    pPg = pNext;
  }
  *pRc = rc;
}

/*
** If there exists a hash-table entry associated with page iPg, remove it.
*/
static void fsRemoveHashEntry(FileSystem *pFS, Pgno iPg){
  Page *p;
  int iHash = fsHashKey(pFS->nHash, iPg);

  for(p=pFS->apHash[iHash]; p && p->iPg!=iPg; p=p->pHashNext);

  if( p ){
    assert( p->nRef==0 || (p->flags & PAGE_FREE)==0 );
    fsPageRemoveFromHash(pFS, p);
    p->iPg = 0;
    iHash = fsHashKey(pFS->nHash, 0);
    p->pHashNext = pFS->apHash[iHash];
    pFS->apHash[iHash] = p;
  }
}

/*
** If the page passed as an argument is dirty, update the database file
** (or mapping of the database file) with its current contents and mark
** the page as clean.
**
** Return LSM_OK if the operation is a success, or an LSM error code
** otherwise.
*/
int lsmFsPagePersist(Page *pPg){
  int rc = LSM_OK;
  if( pPg && (pPg->flags & PAGE_DIRTY) ){
    FileSystem *pFS = pPg->pFS;

    if( pFS->pCompress ){
      int iHash;                  /* Hash key of assigned page number */
      u8 aSz[3];                  /* pPg->nCompress as a 24-bit big-endian */
      assert( pPg->pSeg && pPg->iPg==0 && pPg->nCompress==0 );

      /* Compress the page image. */
      rc = fsCompressIntoBuffer(pFS, pPg);

      /* Serialize the compressed size into buffer aSz[] */
      putRecordSize(aSz, pPg->nCompress, 0);

      /* Write the serialized page record into the database file. */
      pPg->iPg = fsAppendData(pFS, pPg->pSeg, aSz, sizeof(aSz), &rc);
      fsAppendData(pFS, pPg->pSeg, pFS->aOBuffer, pPg->nCompress, &rc);
      fsAppendData(pFS, pPg->pSeg, aSz, sizeof(aSz), &rc);

      /* Now that it has a page number, insert the page into the hash table */
      iHash = fsHashKey(pFS->nHash, pPg->iPg);
      pPg->pHashNext = pFS->apHash[iHash];
      pFS->apHash[iHash] = pPg;

      pPg->pSeg->nSize += (sizeof(aSz) * 2) + pPg->nCompress;

      pPg->flags &= ~PAGE_DIRTY;
      pFS->nWrite++;
    }else{

      if( pPg->iPg==0 ){
        /* No page number has been assigned yet. This occurs with pages used
        ** in the b-tree hierarchy. They were not assigned page numbers when
        ** they were created as doing so would cause this call to
        ** lsmFsPagePersist() to write an out-of-order page. Instead a page 
        ** number is assigned here so that the page data will be appended
        ** to the current segment.
        */
        Page **pp;
        int iPrev = 0;
        int iNext = 0;
        int iHash;

        assert( pPg->pSeg->iFirst );
        assert( pPg->flags & PAGE_FREE );
        assert( (pPg->flags & PAGE_HASPREV)==0 );
        assert( pPg->nData==pFS->nPagesize-4 );

        rc = fsAppendPage(pFS, pPg->pSeg, &pPg->iPg, &iPrev, &iNext);
        if( rc!=LSM_OK ) return rc;

        assert( pPg->flags & PAGE_FREE );
        iHash = fsHashKey(pFS->nHash, pPg->iPg);
        fsRemoveHashEntry(pFS, pPg->iPg);
        pPg->pHashNext = pFS->apHash[iHash];
        pFS->apHash[iHash] = pPg;
        assert( pPg->pHashNext==0 || pPg->pHashNext->iPg!=pPg->iPg );

        if( iPrev ){
          assert( iNext==0 );
          memmove(&pPg->aData[4], pPg->aData, pPg->nData);
          lsmPutU32(pPg->aData, iPrev);
          pPg->flags |= PAGE_HASPREV;
          pPg->aData += 4;
        }else if( iNext ){
          assert( iPrev==0 );
          lsmPutU32(&pPg->aData[pPg->nData], iNext);
        }else{
          int nData = pPg->nData;
          pPg->nData += 4;
          lsmSortedExpandBtreePage(pPg, nData);
        }

        pPg->nRef++;
        for(pp=&pFS->pWaiting; *pp; pp=&(*pp)->pWaitingNext);
        *pp = pPg;
        assert( pPg->pWaitingNext==0 );

      }else{
        i64 iOff;                   /* Offset to write within database file */

        iOff = (i64)pFS->nPagesize * (i64)(pPg->iPg-1);
        if( fsMmapPage(pFS, pPg->iPg)==0 ){
          u8 *aData = pPg->aData - (pPg->flags & PAGE_HASPREV);
          rc = lsmEnvWrite(pFS->pEnv, pFS->fdDb, iOff, aData, pFS->nPagesize);
        }else if( pPg->flags & PAGE_FREE ){
          fsGrowMapping(pFS, iOff + pFS->nPagesize, &rc);
          if( rc==LSM_OK ){
            u8 *aTo = &((u8 *)(pFS->pMap))[iOff];
            u8 *aFrom = pPg->aData - (pPg->flags & PAGE_HASPREV);
            memcpy(aTo, aFrom, pFS->nPagesize);
            lsmFree(pFS->pEnv, aFrom);
            pFS->nCacheAlloc--;
            pPg->aData = aTo + (pPg->flags & PAGE_HASPREV);
            pPg->flags &= ~PAGE_FREE;
            fsPageRemoveFromHash(pFS, pPg);
            pPg->pMappedNext = pFS->pMapped;
            pFS->pMapped = pPg;
          }
        }

        lsmFsFlushWaiting(pFS, &rc);
        pPg->flags &= ~PAGE_DIRTY;
        pFS->nWrite++;
      }
    }
  }

  return rc;
}

/*
** For non-compressed databases, this function is a no-op. For compressed
** databases, it adds a padding record to the segment passed as the third
** argument.
**
** The size of the padding records is selected so that the last byte 
** written is the last byte of a disk sector. This means that if a 
** snapshot is taken and checkpointed, subsequent worker processes will
** not write to any sector that contains checkpointed data.
*/
int lsmFsSortedPadding(
  FileSystem *pFS, 
  Snapshot *pSnapshot,
  Segment *pSeg
){
  int rc = LSM_OK;
  if( pFS->pCompress ){
    Pgno iLast2;
    Pgno iLast = pSeg->iLastPg;     /* Current last page of segment */
    int nPad;                       /* Bytes of padding required */
    u8 aSz[3];

    iLast2 = (1 + iLast/pFS->szSector) * pFS->szSector - 1;
    assert( fsPageToBlock(pFS, iLast)==fsPageToBlock(pFS, iLast2) );
    nPad = iLast2 - iLast;

    if( iLast2>fsLastPageOnPagesBlock(pFS, iLast) ){
      nPad -= 4;
    }
    assert( nPad>=0 );

    if( nPad>=6 ){
      pSeg->nSize += nPad;
      nPad -= 6;
      putRecordSize(aSz, nPad, 1);
      fsAppendData(pFS, pSeg, aSz, sizeof(aSz), &rc);
      memset(pFS->aOBuffer, 0, nPad);
      fsAppendData(pFS, pSeg, pFS->aOBuffer, nPad, &rc);
      fsAppendData(pFS, pSeg, aSz, sizeof(aSz), &rc);
    }else if( nPad>0 ){
      u8 aBuf[5] = {0,0,0,0,0};
      aBuf[0] = (u8)nPad;
      aBuf[nPad-1] = (u8)nPad;
      fsAppendData(pFS, pSeg, aBuf, nPad, &rc);
    }

    assert( rc!=LSM_OK 
        || pSeg->iLastPg==fsLastPageOnPagesBlock(pFS, pSeg->iLastPg)
        || ((pSeg->iLastPg + 1) % pFS->szSector)==0
    );
  }

  return rc;
}


/*
** Increment the reference count on the page object passed as the first
** argument.
*/
void lsmFsPageRef(Page *pPg){
  if( pPg ){
    pPg->nRef++;
  }
}

/*
** Release a page-reference obtained using fsPageGet().
*/
int lsmFsPageRelease(Page *pPg){
  int rc = LSM_OK;
  if( pPg ){
    assert( pPg->nRef>0 );
    pPg->nRef--;
    if( pPg->nRef==0 ){
      FileSystem *pFS = pPg->pFS;
      rc = lsmFsPagePersist(pPg);
      pFS->nOut--;

      assert( pPg->pFS->pCompress 
           || fsIsFirst(pPg->pFS, pPg->iPg)==0 
           || (pPg->flags & PAGE_HASPREV)
      );
      pPg->aData -= (pPg->flags & PAGE_HASPREV);
      pPg->flags &= ~PAGE_HASPREV;

      if( (pPg->flags & PAGE_FREE)==0 ){
        /* Removed from mapped list */
        Page **pp;
        for(pp=&pFS->pMapped; (*pp)!=pPg; pp=&(*pp)->pMappedNext);
        *pp = pPg->pMappedNext;
        pPg->pMappedNext = 0;

        /* Add to free list */
        pPg->pFreeNext = pFS->pFree;
        pFS->pFree = pPg;
      }else{
        fsPageAddToLru(pFS, pPg);
      }
    }
  }

  return rc;
}

/*
** Return the total number of pages read from the database file.
*/
int lsmFsNRead(FileSystem *pFS){ return pFS->nRead; }

/*
** Return the total number of pages written to the database file.
*/
int lsmFsNWrite(FileSystem *pFS){ return pFS->nWrite; }

/*
** Return a copy of the environment pointer used by the file-system object.
*/
lsm_env *lsmFsEnv(FileSystem *pFS){ 
  return pFS->pEnv; 
}

/*
** Return a copy of the environment pointer used by the file-system object
** to which this page belongs.
*/
lsm_env *lsmPageEnv(Page *pPg) { 
  return pPg->pFS->pEnv; 
}

/*
** Return a pointer to the file-system object associated with the Page
** passed as the only argument.
*/
FileSystem *lsmPageFS(Page *pPg){
  return pPg->pFS;
}

/*
** Return the sector-size as reported by the log file handle.
*/
int lsmFsSectorSize(FileSystem *pFS){
  return pFS->szSector;
}

/*
** Helper function for lsmInfoArrayStructure().
*/
static Segment *startsWith(Segment *pRun, Pgno iFirst){
  return (iFirst==pRun->iFirst) ? pRun : 0;
}

/*
** Return the segment that starts with page iFirst, if any. If no such segment
** can be found, return NULL.
*/
static Segment *findSegment(Snapshot *pWorker, Pgno iFirst){
  Level *pLvl;                    /* Used to iterate through db levels */
  Segment *pSeg = 0;              /* Pointer to segment to return */

  for(pLvl=lsmDbSnapshotLevel(pWorker); pLvl && pSeg==0; pLvl=pLvl->pNext){
    if( 0==(pSeg = startsWith(&pLvl->lhs, iFirst)) ){
      int i;
      for(i=0; i<pLvl->nRight; i++){
        if( (pSeg = startsWith(&pLvl->aRhs[i], iFirst)) ) break;
      }
    }
  }

  return pSeg;
}

/*
** This function implements the lsm_info(LSM_INFO_ARRAY_STRUCTURE) request.
** If successful, *pzOut is set to point to a nul-terminated string 
** containing the array structure and LSM_OK is returned. The caller should
** eventually free the string using lsmFree().
**
** If an error occurs, *pzOut is set to NULL and an LSM error code returned.
*/
int lsmInfoArrayStructure(
  lsm_db *pDb, 
  int bBlock,                     /* True for block numbers only */
  Pgno iFirst,
  char **pzOut
){
  int rc = LSM_OK;
  Snapshot *pWorker;              /* Worker snapshot */
  Segment *pArray = 0;            /* Array to report on */
  int bUnlock = 0;

  *pzOut = 0;
  if( iFirst==0 ) return LSM_ERROR;

  /* Obtain the worker snapshot */
  pWorker = pDb->pWorker;
  if( !pWorker ){
    rc = lsmBeginWork(pDb);
    if( rc!=LSM_OK ) return rc;
    pWorker = pDb->pWorker;
    bUnlock = 1;
  }

  /* Search for the array that starts on page iFirst */
  pArray = findSegment(pWorker, iFirst);

  if( pArray==0 ){
    /* Could not find the requested array. This is an error. */
    rc = LSM_ERROR;
  }else{
    FileSystem *pFS = pDb->pFS;
    LsmString str;
    int iBlk;
    int iLastBlk;
   
    iBlk = fsPageToBlock(pFS, pArray->iFirst);
    iLastBlk = fsPageToBlock(pFS, pArray->iLastPg);

    lsmStringInit(&str, pDb->pEnv);
    if( bBlock ){
      lsmStringAppendf(&str, "%d", iBlk);
      while( iBlk!=iLastBlk ){
        fsBlockNext(pFS, pArray, iBlk, &iBlk);
        lsmStringAppendf(&str, " %d", iBlk);
      }
    }else{
      lsmStringAppendf(&str, "%d", pArray->iFirst);
      while( iBlk!=iLastBlk ){
        lsmStringAppendf(&str, " %d", fsLastPageOnBlock(pFS, iBlk));
        fsBlockNext(pFS, pArray, iBlk, &iBlk);
        lsmStringAppendf(&str, " %d", fsFirstPageOnBlock(pFS, iBlk));
      }
      lsmStringAppendf(&str, " %d", pArray->iLastPg);
    }

    *pzOut = str.z;
  }

  if( bUnlock ){
    int rcwork = LSM_BUSY;
    lsmFinishWork(pDb, 0, &rcwork);
  }
  return rc;
}

int lsmFsSegmentContainsPg(
  FileSystem *pFS, 
  Segment *pSeg, 
  Pgno iPg, 
  int *pbRes
){
  Redirect *pRedir = pSeg->pRedirect;
  int rc = LSM_OK;
  int iBlk;
  int iLastBlk;
  int iPgBlock;                   /* Block containing page iPg */

  iPgBlock = fsPageToBlock(pFS, pSeg->iFirst);
  iBlk = fsRedirectBlock(pRedir, fsPageToBlock(pFS, pSeg->iFirst));
  iLastBlk = fsRedirectBlock(pRedir, fsPageToBlock(pFS, pSeg->iLastPg));

  while( iBlk!=iLastBlk && iBlk!=iPgBlock && rc==LSM_OK ){
    rc = fsBlockNext(pFS, pSeg, iBlk, &iBlk);
  }

  *pbRes = (iBlk==iPgBlock);
  return rc;
}

/*
** This function implements the lsm_info(LSM_INFO_ARRAY_PAGES) request.
** If successful, *pzOut is set to point to a nul-terminated string 
** containing the array structure and LSM_OK is returned. The caller should
** eventually free the string using lsmFree().
**
** If an error occurs, *pzOut is set to NULL and an LSM error code returned.
*/
int lsmInfoArrayPages(lsm_db *pDb, Pgno iFirst, char **pzOut){
  int rc = LSM_OK;
  Snapshot *pWorker;              /* Worker snapshot */
  Segment *pSeg = 0;              /* Array to report on */
  int bUnlock = 0;

  *pzOut = 0;
  if( iFirst==0 ) return LSM_ERROR;

  /* Obtain the worker snapshot */
  pWorker = pDb->pWorker;
  if( !pWorker ){
    rc = lsmBeginWork(pDb);
    if( rc!=LSM_OK ) return rc;
    pWorker = pDb->pWorker;
    bUnlock = 1;
  }

  /* Search for the array that starts on page iFirst */
  pSeg = findSegment(pWorker, iFirst);

  if( pSeg==0 ){
    /* Could not find the requested array. This is an error. */
    rc = LSM_ERROR;
  }else{
    Page *pPg = 0;
    FileSystem *pFS = pDb->pFS;
    LsmString str;

    lsmStringInit(&str, pDb->pEnv);
    rc = lsmFsDbPageGet(pFS, pSeg, iFirst, &pPg);
    while( rc==LSM_OK && pPg ){
      Page *pNext = 0;
      lsmStringAppendf(&str, " %lld", lsmFsPageNumber(pPg));
      rc = lsmFsDbPageNext(pSeg, pPg, 1, &pNext);
      lsmFsPageRelease(pPg);
      pPg = pNext;
    }

    if( rc!=LSM_OK ){
      lsmFree(pDb->pEnv, str.z);
    }else{
      *pzOut = str.z;
    }
  }

  if( bUnlock ){
    int rcwork = LSM_BUSY;
    lsmFinishWork(pDb, 0, &rcwork);
  }
  return rc;
}

/*
** The following macros are used by the integrity-check code. Associated with
** each block in the database is an 8-bit bit mask (the entry in the aUsed[]
** array). As the integrity-check meanders through the database, it sets the
** following bits to indicate how each block is used.
**
** INTEGRITY_CHECK_FIRST_PG:
**   First page of block is in use by sorted run.
**
** INTEGRITY_CHECK_LAST_PG:
**   Last page of block is in use by sorted run.
**
** INTEGRITY_CHECK_USED:
**   At least one page of the block is in use by a sorted run.
**
** INTEGRITY_CHECK_FREE:
**   The free block list contains an entry corresponding to this block.
*/
#define INTEGRITY_CHECK_FIRST_PG 0x01
#define INTEGRITY_CHECK_LAST_PG  0x02
#define INTEGRITY_CHECK_USED     0x04
#define INTEGRITY_CHECK_FREE     0x08

/*
** Helper function for lsmFsIntegrityCheck()
*/
static void checkBlocks(
  FileSystem *pFS, 
  Segment *pSeg,
  int bExtra,                     /* If true, count the "next" block if any */
  int nUsed,
  u8 *aUsed
){
  if( pSeg ){
    if( pSeg && pSeg->nSize>0 ){
      int rc;
      int iBlk;                   /* Current block (during iteration) */
      int iLastBlk;               /* Last block of segment */
      int iFirstBlk;              /* First block of segment */
      int bLastIsLastOnBlock;     /* True iLast is the last on its block */

      assert( 0==fsSegmentRedirects(pFS, pSeg) );
      iBlk = iFirstBlk = fsPageToBlock(pFS, pSeg->iFirst);
      iLastBlk = fsPageToBlock(pFS, pSeg->iLastPg);

      bLastIsLastOnBlock = (fsLastPageOnBlock(pFS, iLastBlk)==pSeg->iLastPg);
      assert( iBlk>0 );

      do {
        /* iBlk is a part of this sorted run. */
        aUsed[iBlk-1] |= INTEGRITY_CHECK_USED;

        /* If the first page of this block is also part of the segment,
        ** set the flag to indicate that the first page of iBlk is in use.  
        */
        if( fsFirstPageOnBlock(pFS, iBlk)==pSeg->iFirst || iBlk!=iFirstBlk ){
          assert( (aUsed[iBlk-1] & INTEGRITY_CHECK_FIRST_PG)==0 );
          aUsed[iBlk-1] |= INTEGRITY_CHECK_FIRST_PG;
        }

        /* Unless the sorted run finishes before the last page on this block, 
        ** the last page of this block is also in use.  */
        if( iBlk!=iLastBlk || bLastIsLastOnBlock ){
          assert( (aUsed[iBlk-1] & INTEGRITY_CHECK_LAST_PG)==0 );
          aUsed[iBlk-1] |= INTEGRITY_CHECK_LAST_PG;
        }

        /* Special case. The sorted run being scanned is the output run of
        ** a level currently undergoing an incremental merge. The sorted
        ** run ends on the last page of iBlk, but the next block has already
        ** been allocated. So mark it as in use as well.  */
        if( iBlk==iLastBlk && bLastIsLastOnBlock && bExtra ){
          int iExtra = 0;
          rc = fsBlockNext(pFS, pSeg, iBlk, &iExtra);
          assert( rc==LSM_OK );

          assert( aUsed[iExtra-1]==0 );
          aUsed[iExtra-1] |= INTEGRITY_CHECK_USED;
          aUsed[iExtra-1] |= INTEGRITY_CHECK_FIRST_PG;
          aUsed[iExtra-1] |= INTEGRITY_CHECK_LAST_PG;
        }

        /* Move on to the next block in the sorted run. Or set iBlk to zero
        ** in order to break out of the loop if this was the last block in
        ** the run.  */
        if( iBlk==iLastBlk ){
          iBlk = 0;
        }else{
          rc = fsBlockNext(pFS, pSeg, iBlk, &iBlk);
          assert( rc==LSM_OK );
        }
      }while( iBlk );
    }
  }
}

typedef struct CheckFreelistCtx CheckFreelistCtx;
struct CheckFreelistCtx {
  u8 *aUsed;
  int nBlock;
};
static int checkFreelistCb(void *pCtx, int iBlk, i64 iSnapshot){
  CheckFreelistCtx *p = (CheckFreelistCtx *)pCtx;

  assert( iBlk>=1 );
  assert( iBlk<=p->nBlock );
  assert( p->aUsed[iBlk-1]==0 );
  p->aUsed[iBlk-1] = INTEGRITY_CHECK_FREE;
  return 0;
}

/*
** This function checks that all blocks in the database file are accounted
** for. For each block, exactly one of the following must be true:
**
**   + the block is part of a sorted run, or
**   + the block is on the free-block list
**
** This function also checks that there are no references to blocks with
** out-of-range block numbers.
**
** If no errors are found, non-zero is returned. If an error is found, an
** assert() fails.
*/
int lsmFsIntegrityCheck(lsm_db *pDb){
  CheckFreelistCtx ctx;
  FileSystem *pFS = pDb->pFS;
  int i;
  int rc;
  Freelist freelist = {0, 0, 0};
  u8 *aUsed;
  Level *pLevel;
  Snapshot *pWorker = pDb->pWorker;
  int nBlock = pWorker->nBlock;

#if 0 
  static int nCall = 0;
  nCall++;
  printf("%d calls\n", nCall);
#endif

  aUsed = lsmMallocZero(pDb->pEnv, nBlock);
  if( aUsed==0 ){
    /* Malloc has failed. Since this function is only called within debug
    ** builds, this probably means the user is running an OOM injection test.
    ** Regardless, it will not be possible to run the integrity-check at this
    ** time, so assume the database is Ok and return non-zero. */
    return 1;
  }

  for(pLevel=pWorker->pLevel; pLevel; pLevel=pLevel->pNext){
    int i;
    checkBlocks(pFS, &pLevel->lhs, (pLevel->nRight!=0), nBlock, aUsed);
    for(i=0; i<pLevel->nRight; i++){
      checkBlocks(pFS, &pLevel->aRhs[i], 0, nBlock, aUsed);
    }
  }

  /* Mark all blocks in the free-list as used */
  ctx.aUsed = aUsed;
  ctx.nBlock = nBlock;
  rc = lsmWalkFreelist(pDb, 0, checkFreelistCb, (void *)&ctx);

  if( rc==LSM_OK ){
    for(i=0; i<nBlock; i++) assert( aUsed[i]!=0 );
  }

  lsmFree(pDb->pEnv, aUsed);
  lsmFree(pDb->pEnv, freelist.aEntry);

  return 1;
}

#ifndef NDEBUG
/*
** Return true if pPg happens to be the last page in segment pSeg. Or false
** otherwise. This function is only invoked as part of assert() conditions.
*/
int lsmFsDbPageIsLast(Segment *pSeg, Page *pPg){
  if( pPg->pFS->pCompress ){
    Pgno iNext = 0;
    int rc;
    rc = fsNextPageOffset(pPg->pFS, pSeg, pPg->iPg, pPg->nCompress+6, &iNext);
    return (rc!=LSM_OK || iNext==0);
  }
  return (pPg->iPg==pSeg->iLastPg);
}
#endif

Added ext/lsm1/lsm_log.c.





























































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
/*
** 2011-08-13
**
** The author disclaims copyright to this source code.  In place of
** a legal notice, here is a blessing:
**
**    May you do good and not evil.
**    May you find forgiveness for yourself and forgive others.
**    May you share freely, never taking more than you give.
**
*************************************************************************
**
** This file contains the implementation of LSM database logging. Logging
** has one purpose in LSM - to make transactions durable.
**
** When data is written to an LSM database, it is initially stored in an
** in-memory tree structure. Since this structure is in volatile memory,
** if a power failure or application crash occurs it may be lost. To
** prevent loss of data in this case, each time a record is written to the
** in-memory tree an equivalent record is appended to the log on disk.
** If a power failure or application crash does occur, data can be recovered
** by reading the log.
**
** A log file consists of the following types of records representing data
** written into the database:
**
**   LOG_WRITE:  A key-value pair written to the database.
**   LOG_DELETE: A delete key issued to the database.
**   LOG_COMMIT: A transaction commit.
**
** And the following types of records for ancillary purposes..
**
**   LOG_EOF:    A record indicating the end of a log file.
**   LOG_PAD1:   A single byte padding record.
**   LOG_PAD2:   An N byte padding record (N>1).
**   LOG_JUMP:   A pointer to another offset within the log file.
**
** Each transaction written to the log contains one or more LOG_WRITE and/or
** LOG_DELETE records, followed by a LOG_COMMIT record. The LOG_COMMIT record
** contains an 8-byte checksum based on all previous data written to the
** log file.
**
** LOG CHECKSUMS & RECOVERY
**
**   Checksums are found in two types of log records: LOG_COMMIT and
**   LOG_CKSUM records. In order to recover content from a log, a client
**   reads each record from the start of the log, calculating a checksum as
**   it does. Each time a LOG_COMMIT or LOG_CKSUM is encountered, the 
**   recovery process verifies that the checksum stored in the log 
**   matches the calculated checksum. If it does not, the recovery process
**   can stop reading the log.
**
**   If a recovery process reads records (other than COMMIT or CKSUM) 
**   consisting of at least LSM_CKSUM_MAXDATA bytes, then the next record in
**   the log must be either a LOG_CKSUM or LOG_COMMIT record. If it is
**   not, the recovery process also stops reading the log.
**
**   To recover the log file, it must be read twice. The first time to 
**   determine the location of the last valid commit record. And the second
**   time to load data into the in-memory tree.
**
**   Todo: Surely there is a better way...
**
** LOG WRAPPING
**
**   If the log file were never deleted or wrapped, it would be possible to
**   read it from start to end each time is required recovery (i.e each time
**   the number of database clients changes from 0 to 1). Effectively reading
**   the entire history of the database each time. This would quickly become 
**   inefficient. Additionally, since the log file would grow without bound,
**   it wastes storage space.
**
**   Instead, part of each checkpoint written into the database file contains 
**   a log offset (and other information required to read the log starting at
**   at this offset) at which to begin recovery. Offset $O.
**
**   Once a checkpoint has been written and synced into the database file, it
**   is guaranteed that no recovery process will need to read any data before
**   offset $O of the log file. It is therefore safe to begin overwriting
**   any data that occurs before offset $O.
**
**   This implementation separates the log into three regions mapped into
**   the log file - regions 0, 1 and 2. During recovery, regions are read
**   in ascending order (i.e. 0, then 1, then 2). Each region is zero or
**   more bytes in size.
**
**     |---1---|..|--0--|.|--2--|....
**
**   New records are always appended to the end of region 2.
**
**   Initially (when it is empty), all three regions are zero bytes in size.
**   Each of them are located at the beginning of the file. As records are
**   added to the log, region 2 grows, so that the log consists of a zero
**   byte region 1, followed by a zero byte region 0, followed by an N byte
**   region 2. After one or more checkpoints have been written to disk, 
**   the start point of region 2 is moved to $O. For example:
**
**     A) ||.........|--2--|....
**   
**   (both regions 0 and 1 are 0 bytes in size at offset 0).
**
**   Eventually, the log wraps around to write new records into the start.
**   At this point, region 2 is renamed to region 0. Region 0 is renamed
**   to region 2. After appending a few records to the new region 2, the
**   log file looks like this:
**
**     B) ||--2--|...|--0--|....
**
**   (region 1 is still 0 bytes in size, located at offset 0).
**
**   Any checkpoints made at this point may reduce the size of region 0.
**   However, if they do not, and region 2 expands so that it is about to
**   overwrite the start of region 0, then region 2 is renamed to region 1,
**   and a new region 2 created at the end of the file following the existing
**   region 0.
**
**     C) |---1---|..|--0--|.|-2-|
**
**   In this state records are appended to region 2 until checkpoints have
**   contracted regions 0 AND 1 UNTil they are both zero bytes in size. They 
**   are then shifted to the start of the log file, leaving the system in 
**   the equivalent of state A above.
**
**   Alternatively, state B may transition directly to state A if the size
**   of region 0 is reduced to zero bytes before region 2 threatens to 
**   encroach upon it.
**
** LOG_PAD1 & LOG_PAD2 RECORDS
**
**   PAD1 and PAD2 records may appear in a log file at any point. They allow
**   a process writing the log file align the beginning of transactions with 
**   the beginning of disk sectors, which increases robustness.
**
** RECORD FORMATS:
**
**   LOG_EOF:    * A single 0x00 byte.
**
**   LOG_PAD1:   * A single 0x01 byte.
**
**   LOG_PAD2:   * A single 0x02 byte, followed by
**               * The number of unused bytes (N) as a varint,
**               * An N byte block of unused space.
**
**   LOG_COMMIT: * A single 0x03 byte.
**               * An 8-byte checksum.
**
**   LOG_JUMP:   * A single 0x04 byte.
**               * Absolute file offset to jump to, encoded as a varint.
**
**   LOG_WRITE:  * A single 0x06 or 0x07 byte, 
**               * The number of bytes in the key, encoded as a varint, 
**               * The number of bytes in the value, encoded as a varint, 
**               * If the first byte was 0x07, an 8 byte checksum.
**               * The key data,
**               * The value data.
**
**   LOG_DELETE: * A single 0x08 or 0x09 byte, 
**               * The number of bytes in the key, encoded as a varint, 
**               * If the first byte was 0x09, an 8 byte checksum.
**               * The key data.
**
**   Varints are as described in lsm_varint.c (SQLite 4 format).
**
** CHECKSUMS:
**
**   The checksum is calculated using two 32-bit unsigned integers, s0 and
**   s1. The initial value for both is 42. It is updated each time a record
**   is written into the log file by treating the encoded (binary) record as 
**   an array of 32-bit little-endian integers. Then, if x[] is the integer
**   array, updating the checksum accumulators as follows:
**
**     for i from 0 to n-1 step 2:
**       s0 += x[i] + s1;
**       s1 += x[i+1] + s0;
**     endfor
**
**   If the record is not an even multiple of 8-bytes in size it is padded
**   with zeroes to make it so before the checksum is updated.
**
**   The checksum stored in a COMMIT, WRITE or DELETE is based on all bytes
**   up to the start of the 8-byte checksum itself, including the COMMIT,
**   WRITE or DELETE fields that appear before the checksum in the record.
**
** VARINT FORMAT
**
** See lsm_varint.c.
*/

#ifndef _LSM_INT_H
# include "lsmInt.h"
#endif

/* Log record types */
#define LSM_LOG_EOF          0x00
#define LSM_LOG_PAD1         0x01
#define LSM_LOG_PAD2         0x02
#define LSM_LOG_COMMIT       0x03
#define LSM_LOG_JUMP         0x04

#define LSM_LOG_WRITE        0x06
#define LSM_LOG_WRITE_CKSUM  0x07
#define LSM_LOG_DELETE       0x08
#define LSM_LOG_DELETE_CKSUM 0x09

/* Require a checksum every 32KB. */
#define LSM_CKSUM_MAXDATA (32*1024)

/* Do not wrap a log file smaller than this in bytes. */
#define LSM_MIN_LOGWRAP      (128*1024)

/*
** szSector:
**   Commit records must be aligned to end on szSector boundaries. If
**   the safety-mode is set to NORMAL or OFF, this value is 1. Otherwise,
**   if the safety-mode is set to FULL, it is the size of the file-system
**   sectors as reported by lsmFsSectorSize().
*/
struct LogWriter {
  u32 cksum0;                     /* Checksum 0 at offset iOff */
  u32 cksum1;                     /* Checksum 1 at offset iOff */
  int iCksumBuf;                  /* Bytes of buf that have been checksummed */
  i64 iOff;                       /* Offset at start of buffer buf */
  int szSector;                   /* Sector size for this transaction */
  LogRegion jump;                 /* Avoid writing to this region */
  i64 iRegion1End;                /* End of first region written by trans */
  i64 iRegion2Start;              /* Start of second regions written by trans */
  LsmString buf;                  /* Buffer containing data not yet written */
};

/*
** Return the result of interpreting the first 4 bytes in buffer aIn as 
** a 32-bit unsigned little-endian integer.
*/
static u32 getU32le(u8 *aIn){
  return ((u32)aIn[3] << 24) 
       + ((u32)aIn[2] << 16) 
       + ((u32)aIn[1] << 8) 
       + ((u32)aIn[0]);
}


/*
** This function is the same as logCksum(), except that pointer "a" need
** not be aligned to an 8-byte boundary or padded with zero bytes. This
** version is slower, but sometimes more convenient to use.
*/
static void logCksumUnaligned(
  char *z,                        /* Input buffer */
  int n,                          /* Size of input buffer in bytes */
  u32 *pCksum0,                   /* IN/OUT: Checksum value 1 */
  u32 *pCksum1                    /* IN/OUT: Checksum value 2 */
){
  u8 *a = (u8 *)z;
  u32 cksum0 = *pCksum0;
  u32 cksum1 = *pCksum1;
  int nIn = (n/8) * 8;
  int i;

  assert( n>0 );
  for(i=0; i<nIn; i+=8){
    cksum0 += getU32le(&a[i]) + cksum1;
    cksum1 += getU32le(&a[i+4]) + cksum0;
  }

  if( nIn!=n ){
    u8 aBuf[8] = {0, 0, 0, 0, 0, 0, 0, 0};
    assert( (n-nIn)<8 && n>nIn );
    memcpy(aBuf, &a[nIn], n-nIn);
    cksum0 += getU32le(aBuf) + cksum1;
    cksum1 += getU32le(&aBuf[4]) + cksum0;
  }

  *pCksum0 = cksum0;
  *pCksum1 = cksum1;
}

/*
** Update pLog->cksum0 and pLog->cksum1 so that the first nBuf bytes in the 
** write buffer (pLog->buf) are included in the checksum.
*/
static void logUpdateCksum(LogWriter *pLog, int nBuf){
  assert( (pLog->iCksumBuf % 8)==0 );
  assert( pLog->iCksumBuf<=nBuf );
  assert( (nBuf % 8)==0 || nBuf==pLog->buf.n );
  if( nBuf>pLog->iCksumBuf ){
    logCksumUnaligned(
        &pLog->buf.z[pLog->iCksumBuf], nBuf-pLog->iCksumBuf, 
        &pLog->cksum0, &pLog->cksum1
    );
  }
  pLog->iCksumBuf = nBuf;
}

static i64 firstByteOnSector(LogWriter *pLog, i64 iOff){
  return (iOff / pLog->szSector) * pLog->szSector;
}
static i64 lastByteOnSector(LogWriter *pLog, i64 iOff){
  return firstByteOnSector(pLog, iOff) + pLog->szSector - 1;
}

/*
** If possible, reclaim log file space. Log file space is reclaimed after
** a snapshot that points to the same data in the database file is synced
** into the db header.
*/
static int logReclaimSpace(lsm_db *pDb){
  int rc;
  int iMeta;
  int bRotrans;                   /* True if there exists some ro-trans */

  /* Test if there exists some other connection with a read-only transaction
  ** open. If there does, then log file space may not be reclaimed.  */
  rc = lsmDetectRoTrans(pDb, &bRotrans);
  if( rc!=LSM_OK || bRotrans ) return rc;

  iMeta = (int)pDb->pShmhdr->iMetaPage;
  if( iMeta==1 || iMeta==2 ){
    DbLog *pLog = &pDb->treehdr.log;
    i64 iSyncedId;

    /* Read the snapshot-id of the snapshot stored on meta-page iMeta. Note
    ** that in theory, the value read is untrustworthy (due to a race 
    ** condition - see comments above lsmFsReadSyncedId()). So it is only 
    ** ever used to conclude that no log space can be reclaimed. If it seems
    ** to indicate that it may be possible to reclaim log space, a
    ** second call to lsmCheckpointSynced() (which does return trustworthy
    ** values) is made below to confirm.  */
    rc = lsmFsReadSyncedId(pDb, iMeta, &iSyncedId);

    if( rc==LSM_OK && pLog->iSnapshotId!=iSyncedId ){
      i64 iSnapshotId = 0;
      i64 iOff = 0;
      rc = lsmCheckpointSynced(pDb, &iSnapshotId, &iOff, 0);
      if( rc==LSM_OK && pLog->iSnapshotId<iSnapshotId ){
        int iRegion;
        for(iRegion=0; iRegion<3; iRegion++){
          LogRegion *p = &pLog->aRegion[iRegion];
          if( iOff>=p->iStart && iOff<=p->iEnd ) break;
          p->iStart = 0;
          p->iEnd = 0;
        }
        assert( iRegion<3 );
        pLog->aRegion[iRegion].iStart = iOff;
        pLog->iSnapshotId = iSnapshotId;
      }
    }
  }
  return rc;
}

/*
** This function is called when a write-transaction is first opened. It
** is assumed that the caller is holding the client-mutex when it is 
** called.
**
** Before returning, this function allocates the LogWriter object that
** will be used to write to the log file during the write transaction.
** LSM_OK is returned if no error occurs, otherwise an LSM error code.
*/
int lsmLogBegin(lsm_db *pDb){
  int rc = LSM_OK;
  LogWriter *pNew;
  LogRegion *aReg;

  if( pDb->bUseLog==0 ) return LSM_OK;

  /* If the log file has not yet been opened, open it now. Also allocate
  ** the LogWriter structure, if it has not already been allocated.  */
  rc = lsmFsOpenLog(pDb, 0);
  if( pDb->pLogWriter==0 ){
    pNew = lsmMallocZeroRc(pDb->pEnv, sizeof(LogWriter), &rc);
    if( pNew ){
      lsmStringInit(&pNew->buf, pDb->pEnv);
      rc = lsmStringExtend(&pNew->buf, 2);
    }
  }else{
    pNew = pDb->pLogWriter;
    assert( (u8 *)(&pNew[1])==(u8 *)(&((&pNew->buf)[1])) );
    memset(pNew, 0, ((u8 *)&pNew->buf) - (u8 *)pNew);
    pNew->buf.n = 0;
  }

  if( rc==LSM_OK ){
    /* The following call detects whether or not a new snapshot has been 
    ** synced into the database file. If so, it updates the contents of
    ** the pDb->treehdr.log structure to reclaim any space in the log
    ** file that is no longer required. 
    **
    ** TODO: Calling this every transaction is overkill. And since the 
    ** call has to read and checksum a snapshot from the database file,
    ** it is expensive. It would be better to figure out a way so that
    ** this is only called occasionally - say for every 32KB written to 
    ** the log file.
    */
    rc = logReclaimSpace(pDb);
  }
  if( rc!=LSM_OK ){
    lsmLogClose(pDb);
    return rc;
  }

  /* Set the effective sector-size for this transaction. Sectors are assumed
  ** to be one byte in size if the safety-mode is OFF or NORMAL, or as
  ** reported by lsmFsSectorSize if it is FULL.  */
  if( pDb->eSafety==LSM_SAFETY_FULL ){
    pNew->szSector = lsmFsSectorSize(pDb->pFS);
    assert( pNew->szSector>0 );
  }else{
    pNew->szSector = 1;
  }

  /* There are now three scenarios:
  **
  **   1) Regions 0 and 1 are both zero bytes in size and region 2 begins
  **      at a file offset greater than LSM_MIN_LOGWRAP. In this case, wrap
  **      around to the start and write data into the start of the log file. 
  **
  **   2) Region 1 is zero bytes in size and region 2 occurs earlier in the 
  **      file than region 0. In this case, append data to region 2, but
  **      remember to jump over region 1 if required.
  **
  **   3) Region 2 is the last in the file. Append to it.
  */
  aReg = &pDb->treehdr.log.aRegion[0];

  assert( aReg[0].iEnd==0 || aReg[0].iEnd>aReg[0].iStart );
  assert( aReg[1].iEnd==0 || aReg[1].iEnd>aReg[1].iStart );

  pNew->cksum0 = pDb->treehdr.log.cksum0;
  pNew->cksum1 = pDb->treehdr.log.cksum1;

  if( aReg[0].iEnd==0 && aReg[1].iEnd==0 && aReg[2].iStart>=LSM_MIN_LOGWRAP ){
    /* Case 1. Wrap around to the start of the file. Write an LSM_LOG_JUMP 
    ** into the log file in this case. Pad it out to 8 bytes using a PAD2
    ** record so that the checksums can be updated immediately.  */
    u8 aJump[] = { 
      LSM_LOG_PAD2, 0x04, 0x00, 0x00, 0x00, 0x00, LSM_LOG_JUMP, 0x00 
    };

    lsmStringBinAppend(&pNew->buf, aJump, sizeof(aJump));
    logUpdateCksum(pNew, pNew->buf.n);
    rc = lsmFsWriteLog(pDb->pFS, aReg[2].iEnd, &pNew->buf);
    pNew->iCksumBuf = pNew->buf.n = 0;

    aReg[2].iEnd += 8;
    pNew->jump = aReg[0] = aReg[2];
    aReg[2].iStart = aReg[2].iEnd = 0;
  }else if( aReg[1].iEnd==0 && aReg[2].iEnd<aReg[0].iEnd ){
    /* Case 2. */
    pNew->iOff = aReg[2].iEnd;
    pNew->jump = aReg[0];
  }else{
    /* Case 3. */
    assert( aReg[2].iStart>=aReg[0].iEnd && aReg[2].iStart>=aReg[1].iEnd );
    pNew->iOff = aReg[2].iEnd;
  }

  if( pNew->jump.iStart ){
    i64 iRound;
    assert( pNew->jump.iStart>pNew->iOff );

    iRound = firstByteOnSector(pNew, pNew->jump.iStart);
    if( iRound>pNew->iOff ) pNew->jump.iStart = iRound;
    pNew->jump.iEnd = lastByteOnSector(pNew, pNew->jump.iEnd);
  }

  pDb->pLogWriter = pNew;
  return rc;
}

/*
** This function is called when a write-transaction is being closed.
** Parameter bCommit is true if the transaction is being committed,
** or false otherwise. The caller must hold the client-mutex to call
** this function.
**
** A call to this function deletes the LogWriter object allocated by
** lsmLogBegin(). If the transaction is being committed, the shared state
** in *pLog is updated before returning.
*/
void lsmLogEnd(lsm_db *pDb, int bCommit){
  DbLog *pLog;
  LogWriter *p;
  p = pDb->pLogWriter;

  if( p==0 ) return;
  pLog = &pDb->treehdr.log;

  if( bCommit ){
    pLog->aRegion[2].iEnd = p->iOff;
    pLog->cksum0 = p->cksum0;
    pLog->cksum1 = p->cksum1;
    if( p->iRegion1End ){
      /* This happens when the transaction had to jump over some other
      ** part of the log.  */
      assert( pLog->aRegion[1].iEnd==0 );
      assert( pLog->aRegion[2].iStart<p->iRegion1End );
      pLog->aRegion[1].iStart = pLog->aRegion[2].iStart;
      pLog->aRegion[1].iEnd = p->iRegion1End;
      pLog->aRegion[2].iStart = p->iRegion2Start;
    }
  }
}

static int jumpIfRequired(
  lsm_db *pDb,
  LogWriter *pLog,
  int nReq,
  int *pbJump
){
  /* Determine if it is necessary to add an LSM_LOG_JUMP to jump over the
  ** jump region before writing the LSM_LOG_WRITE or DELETE record. This
  ** is necessary if there is insufficient room between the current offset
  ** and the jump region to fit the new WRITE/DELETE record and the largest
  ** possible JUMP record with up to 7 bytes of padding (a total of 17 
  ** bytes).  */
  if( (pLog->jump.iStart > (pLog->iOff + pLog->buf.n))
   && (pLog->jump.iStart < (pLog->iOff + pLog->buf.n + (nReq + 17))) 
  ){
    int rc;                       /* Return code */
    i64 iJump;                    /* Offset to jump to */
    u8 aJump[10];                 /* Encoded jump record */
    int nJump;                    /* Valid bytes in aJump[] */
    int nPad;                     /* Bytes of padding required */

    /* Serialize the JUMP record */
    iJump = pLog->jump.iEnd+1;
    aJump[0] = LSM_LOG_JUMP;
    nJump = 1 + lsmVarintPut64(&aJump[1], iJump);

    /* Adding padding to the contents of the buffer so that it will be a 
    ** multiple of 8 bytes in size after the JUMP record is appended. This
    ** is not strictly required, it just makes the keeping the running 
    ** checksum up to date in this file a little simpler.  */
    nPad = (pLog->buf.n + nJump) % 8;
    if( nPad ){
      u8 aPad[7] = {0,0,0,0,0,0,0};
      nPad = 8-nPad;
      if( nPad==1 ){
        aPad[0] = LSM_LOG_PAD1;
      }else{
        aPad[0] = LSM_LOG_PAD2;
        aPad[1] = (nPad-2);
      }
      rc = lsmStringBinAppend(&pLog->buf, aPad, nPad);
      if( rc!=LSM_OK ) return rc;
    }

    /* Append the JUMP record to the buffer. Then flush the buffer to disk
    ** and update the checksums. The next write to the log file (assuming
    ** there is no transaction rollback) will be to offset iJump (just past
    ** the jump region).  */
    rc = lsmStringBinAppend(&pLog->buf, aJump, nJump);
    if( rc!=LSM_OK ) return rc;
    assert( (pLog->buf.n % 8)==0 );
    rc = lsmFsWriteLog(pDb->pFS, pLog->iOff, &pLog->buf);
    if( rc!=LSM_OK ) return rc;
    logUpdateCksum(pLog, pLog->buf.n);
    pLog->iRegion1End = (pLog->iOff + pLog->buf.n);
    pLog->iRegion2Start = iJump;
    pLog->iOff = iJump;
    pLog->iCksumBuf = pLog->buf.n = 0;
    if( pbJump ) *pbJump = 1;
  }

  return LSM_OK;
}

static int logCksumAndFlush(lsm_db *pDb){
  int rc;                         /* Return code */
  LogWriter *pLog = pDb->pLogWriter;

  /* Calculate the checksum value. Append it to the buffer. */
  logUpdateCksum(pLog, pLog->buf.n);
  lsmPutU32((u8 *)&pLog->buf.z[pLog->buf.n], pLog->cksum0);
  pLog->buf.n += 4;
  lsmPutU32((u8 *)&pLog->buf.z[pLog->buf.n], pLog->cksum1);
  pLog->buf.n += 4;

  /* Write the contents of the buffer to disk. */
  rc = lsmFsWriteLog(pDb->pFS, pLog->iOff, &pLog->buf);
  pLog->iOff += pLog->buf.n;
  pLog->iCksumBuf = pLog->buf.n = 0;

  return rc;
}

/*
** Write the contents of the log-buffer to disk. Then write either a CKSUM
** or COMMIT record, depending on the value of parameter eType.
*/
static int logFlush(lsm_db *pDb, int eType){
  int rc;
  int nReq;
  LogWriter *pLog = pDb->pLogWriter;
  
  assert( eType==LSM_LOG_COMMIT );
  assert( pLog );

  /* Commit record is always 9 bytes in size. */
  nReq = 9;
  if( eType==LSM_LOG_COMMIT && pLog->szSector>1 ) nReq += pLog->szSector + 17;
  rc = jumpIfRequired(pDb, pLog, nReq, 0);

  /* If this is a COMMIT, add padding to the log so that the COMMIT record
  ** is aligned against the end of a disk sector. In other words, add padding
  ** so that the first byte following the COMMIT record lies on a different
  ** sector.  */
  if( eType==LSM_LOG_COMMIT && pLog->szSector>1 ){
    int nPad;                     /* Bytes of padding to add */

    /* Determine the value of nPad. */
    nPad = ((pLog->iOff + pLog->buf.n + 9) % pLog->szSector);
    if( nPad ) nPad = pLog->szSector - nPad;
    rc = lsmStringExtend(&pLog->buf, nPad);
    if( rc!=LSM_OK ) return rc;

    while( nPad ){
      if( nPad==1 ){
        pLog->buf.z[pLog->buf.n++] = LSM_LOG_PAD1;
        nPad = 0;
      }else{
        int n = LSM_MIN(200, nPad-2);
        pLog->buf.z[pLog->buf.n++] = LSM_LOG_PAD2;
        pLog->buf.z[pLog->buf.n++] = n;
        nPad -= 2;
        memset(&pLog->buf.z[pLog->buf.n], 0x2B, n);
        pLog->buf.n += n;
        nPad -= n;
      }
    }
  }

  /* Make sure there is room in the log-buffer to add the CKSUM or COMMIT
  ** record. Then add the first byte of it.  */
  rc = lsmStringExtend(&pLog->buf, 9);
  if( rc!=LSM_OK ) return rc;
  pLog->buf.z[pLog->buf.n++] = eType;
  memset(&pLog->buf.z[pLog->buf.n], 0, 8);

  rc = logCksumAndFlush(pDb);

  /* If this is a commit and synchronous=full, sync the log to disk. */
  if( rc==LSM_OK && eType==LSM_LOG_COMMIT && pDb->eSafety==LSM_SAFETY_FULL ){
    rc = lsmFsSyncLog(pDb->pFS);
  }
  return rc;
}

/*
** Append an LSM_LOG_WRITE (if nVal>=0) or LSM_LOG_DELETE (if nVal<0) 
** record to the database log.
*/
int lsmLogWrite(
  lsm_db *pDb,                    /* Database handle */
  void *pKey, int nKey,           /* Database key to write to log */
  void *pVal, int nVal            /* Database value (or nVal<0) to write */
){
  int rc = LSM_OK;
  LogWriter *pLog;                /* Log object to write to */
  int nReq;                       /* Bytes of space required in log */
  int bCksum = 0;                 /* True to embed a checksum in this record */

  if( pDb->bUseLog==0 ) return LSM_OK;
  pLog = pDb->pLogWriter;

  /* Determine how many bytes of space are required, assuming that a checksum
  ** will be embedded in this record (even though it may not be).  */
  nReq = 1 + lsmVarintLen32(nKey) + 8 + nKey;
  if( nVal>=0 ) nReq += lsmVarintLen32(nVal) + nVal;

  /* Jump over the jump region if required. Set bCksum to true to tell the
  ** code below to include a checksum in the record if either (a) writing
  ** this record would mean that more than LSM_CKSUM_MAXDATA bytes of data
  ** have been written to the log since the last checksum, or (b) the jump
  ** is taken.  */
  rc = jumpIfRequired(pDb, pLog, nReq, &bCksum);
  if( (pLog->buf.n+nReq) > LSM_CKSUM_MAXDATA ) bCksum = 1;

  if( rc==LSM_OK ){
    rc = lsmStringExtend(&pLog->buf, nReq);
  }
  if( rc==LSM_OK ){
    u8 *a = (u8 *)&pLog->buf.z[pLog->buf.n];
    
    /* Write the record header - the type byte followed by either 1 (for
    ** DELETE) or 2 (for WRITE) varints.  */
    assert( LSM_LOG_WRITE_CKSUM == (LSM_LOG_WRITE | 0x0001) );
    assert( LSM_LOG_DELETE_CKSUM == (LSM_LOG_DELETE | 0x0001) );
    *(a++) = (nVal>=0 ? LSM_LOG_WRITE : LSM_LOG_DELETE) | (u8)bCksum;
    a += lsmVarintPut32(a, nKey);
    if( nVal>=0 ) a += lsmVarintPut32(a, nVal);

    if( bCksum ){
      pLog->buf.n = (a - (u8 *)pLog->buf.z);
      rc = logCksumAndFlush(pDb);
      a = (u8 *)&pLog->buf.z[pLog->buf.n];
    }

    memcpy(a, pKey, nKey);
    a += nKey;
    if( nVal>=0 ){
      memcpy(a, pVal, nVal);
      a += nVal;
    }
    pLog->buf.n = a - (u8 *)pLog->buf.z;
    assert( pLog->buf.n<=pLog->buf.nAlloc );
  }

  return rc;
}

/*
** Append an LSM_LOG_COMMIT record to the database log.
*/
int lsmLogCommit(lsm_db *pDb){
  if( pDb->bUseLog==0 ) return LSM_OK;
  return logFlush(pDb, LSM_LOG_COMMIT);
}

/*
** Store the current offset and other checksum related information in the
** structure *pMark. Later, *pMark can be passed to lsmLogSeek() to "rewind"
** the LogWriter object to the current log file offset. This is used when
** rolling back savepoint transactions.
*/
void lsmLogTell(
  lsm_db *pDb,                    /* Database handle */
  LogMark *pMark                  /* Populate this object with current offset */
){
  LogWriter *pLog;
  int nCksum;

  if( pDb->bUseLog==0 ) return;
  pLog = pDb->pLogWriter;
  nCksum = pLog->buf.n & 0xFFFFFFF8;
  logUpdateCksum(pLog, nCksum);
  assert( pLog->iCksumBuf==nCksum );
  pMark->nBuf = pLog->buf.n - nCksum;
  memcpy(pMark->aBuf, &pLog->buf.z[nCksum], pMark->nBuf);

  pMark->iOff = pLog->iOff + pLog->buf.n;
  pMark->cksum0 = pLog->cksum0;
  pMark->cksum1 = pLog->cksum1;
}

/*
** Seek (rewind) back to the log file offset stored by an ealier call to
** lsmLogTell() in *pMark.
*/
void lsmLogSeek(
  lsm_db *pDb,                    /* Database handle */
  LogMark *pMark                  /* Object containing log offset to seek to */
){
  LogWriter *pLog;

  if( pDb->bUseLog==0 ) return;
  pLog = pDb->pLogWriter;

  assert( pMark->iOff<=pLog->iOff+pLog->buf.n );
  if( (pMark->iOff & 0xFFFFFFF8)>=pLog->iOff ){
    pLog->buf.n = pMark->iOff - pLog->iOff;
    pLog->iCksumBuf = (pLog->buf.n & 0xFFFFFFF8);
  }else{
    pLog->buf.n = pMark->nBuf;
    memcpy(pLog->buf.z, pMark->aBuf, pMark->nBuf);
    pLog->iCksumBuf = 0;
    pLog->iOff = pMark->iOff - pMark->nBuf;
  }
  pLog->cksum0 = pMark->cksum0;
  pLog->cksum1 = pMark->cksum1;

  if( pMark->iOff > pLog->iRegion1End ) pLog->iRegion1End = 0;
  if( pMark->iOff > pLog->iRegion2Start ) pLog->iRegion2Start = 0;
}

/*
** This function does the work for an lsm_info(LOG_STRUCTURE) request.
*/
int lsmInfoLogStructure(lsm_db *pDb, char **pzVal){
  int rc = LSM_OK;
  char *zVal = 0;

  /* If there is no read or write transaction open, read the latest 
  ** tree-header from shared-memory to report on. If necessary, update
  ** it based on the contents of the database header.  
  **
  ** No locks are taken here - these are passive read operations only.
  */
  if( pDb->pCsr==0 && pDb->nTransOpen==0 ){
    rc = lsmTreeLoadHeader(pDb, 0);
    if( rc==LSM_OK ) rc = logReclaimSpace(pDb);
  }

  if( rc==LSM_OK ){
    DbLog *pLog = &pDb->treehdr.log;
    zVal = lsmMallocPrintf(pDb->pEnv, 
        "%d %d %d %d %d %d", 
        (int)pLog->aRegion[0].iStart, (int)pLog->aRegion[0].iEnd,
        (int)pLog->aRegion[1].iStart, (int)pLog->aRegion[1].iEnd,
        (int)pLog->aRegion[2].iStart, (int)pLog->aRegion[2].iEnd
    );
    if( !zVal ) rc = LSM_NOMEM_BKPT;
  }

  *pzVal = zVal;
  return rc;
}

/*************************************************************************
** Begin code for log recovery.
*/

typedef struct LogReader LogReader;
struct LogReader {
  FileSystem *pFS;                /* File system to read from */
  i64 iOff;                       /* File offset at end of buf content */
  int iBuf;                       /* Current read offset in buf */
  LsmString buf;                  /* Buffer containing file content */

  int iCksumBuf;                  /* Offset in buf corresponding to cksum[01] */
  u32 cksum0;                     /* Checksum 0 at offset iCksumBuf */
  u32 cksum1;                     /* Checksum 1 at offset iCksumBuf */
};

static void logReaderBlob(
  LogReader *p,                   /* Log reader object */
  LsmString *pBuf,                /* Dynamic storage, if required */
  int nBlob,                      /* Number of bytes to read */
  u8 **ppBlob,                    /* OUT: Pointer to blob read */
  int *pRc                        /* IN/OUT: Error code */
){
  static const int LOG_READ_SIZE = 512;
  int rc = *pRc;                  /* Return code */
  int nReq = nBlob;               /* Bytes required */

  while( rc==LSM_OK && nReq>0 ){
    int nAvail;                   /* Bytes of data available in p->buf */
    if( p->buf.n==p->iBuf ){
      int nCksum;                 /* Total bytes requiring checksum */
      int nCarry = 0;             /* Total bytes requiring checksum */

      nCksum = p->iBuf - p->iCksumBuf;
      if( nCksum>0 ){
        nCarry = nCksum % 8;
        nCksum = ((nCksum / 8) * 8);
        if( nCksum>0 ){
          logCksumUnaligned(
              &p->buf.z[p->iCksumBuf], nCksum, &p->cksum0, &p->cksum1
          );
        }
      }
      if( nCarry>0 ) memcpy(p->buf.z, &p->buf.z[p->iBuf-nCarry], nCarry);
      p->buf.n = nCarry;
      p->iBuf = nCarry;

      rc = lsmFsReadLog(p->pFS, p->iOff, LOG_READ_SIZE, &p->buf);
      if( rc!=LSM_OK ) break;
      p->iCksumBuf = 0;
      p->iOff += LOG_READ_SIZE;
    }

    nAvail = p->buf.n - p->iBuf;
    if( ppBlob && nReq==nBlob && nBlob<=nAvail ){
      *ppBlob = (u8 *)&p->buf.z[p->iBuf];
      p->iBuf += nBlob;
      nReq = 0;
    }else{
      int nCopy = LSM_MIN(nAvail, nReq);
      if( nBlob==nReq ){
        if( ppBlob ) *ppBlob = (u8 *)pBuf->z;
        pBuf->n = 0;
      }
      rc = lsmStringBinAppend(pBuf, (u8 *)&p->buf.z[p->iBuf], nCopy);
      nReq -= nCopy;
      p->iBuf += nCopy;
    }
  }

  *pRc = rc;
}

static void logReaderVarint(
  LogReader *p, 
  LsmString *pBuf,
  int *piVal,                     /* OUT: Value read from log */
  int *pRc                        /* IN/OUT: Error code */
){
  if( *pRc==LSM_OK ){
    u8 *aVarint;
    if( p->buf.n==p->iBuf ){
      logReaderBlob(p, 0, 10, &aVarint, pRc);
      if( LSM_OK==*pRc ) p->iBuf -= (10 - lsmVarintGet32(aVarint, piVal));
    }else{
      logReaderBlob(p, pBuf, lsmVarintSize(p->buf.z[p->iBuf]), &aVarint, pRc);
      if( LSM_OK==*pRc ) lsmVarintGet32(aVarint, piVal);
    }
  }
}

static void logReaderByte(LogReader *p, u8 *pByte, int *pRc){
  u8 *pPtr = 0;
  logReaderBlob(p, 0, 1, &pPtr, pRc);
  if( pPtr ) *pByte = *pPtr;
}

static void logReaderCksum(LogReader *p, LsmString *pBuf, int *pbEof, int *pRc){
  if( *pRc==LSM_OK ){
    u8 *pPtr = 0;
    u32 cksum0, cksum1;
    int nCksum = p->iBuf - p->iCksumBuf;

    /* Update in-memory (expected) checksums */
    assert( nCksum>=0 );
    logCksumUnaligned(&p->buf.z[p->iCksumBuf], nCksum, &p->cksum0, &p->cksum1);
    p->iCksumBuf = p->iBuf + 8;
    logReaderBlob(p, pBuf, 8, &pPtr, pRc);

    /* Read the checksums from the log file. Set *pbEof if they do not match. */
    if( pPtr ){
      cksum0 = lsmGetU32(pPtr);
      cksum1 = lsmGetU32(&pPtr[4]);
      *pbEof = (cksum0!=p->cksum0 || cksum1!=p->cksum1);
      p->iCksumBuf = p->iBuf;
    }
  }
}

static void logReaderInit(
  lsm_db *pDb,                    /* Database handle */
  DbLog *pLog,                    /* Log object associated with pDb */
  int bInitBuf,                   /* True if p->buf is uninitialized */
  LogReader *p                    /* Initialize this LogReader object */
){
  p->pFS = pDb->pFS;
  p->iOff = pLog->aRegion[2].iStart;
  p->cksum0 = pLog->cksum0;
  p->cksum1 = pLog->cksum1;
  if( bInitBuf ){ lsmStringInit(&p->buf, pDb->pEnv); }
  p->buf.n = 0;
  p->iCksumBuf = 0;
  p->iBuf = 0;
}

/*
** This function is called after reading the header of a LOG_DELETE or
** LOG_WRITE record. Parameter nByte is the total size of the key and
** value that follow the header just read. Return true if the size and
** position of the record indicate that it should contain a checksum.
*/
static int logRequireCksum(LogReader *p, int nByte){
  return ((p->iBuf + nByte - p->iCksumBuf) > LSM_CKSUM_MAXDATA);
}

/*
** Recover the contents of the log file.
*/
int lsmLogRecover(lsm_db *pDb){
  LsmString buf1;                 /* Key buffer */
  LsmString buf2;                 /* Value buffer */
  LogReader reader;               /* Log reader object */
  int rc = LSM_OK;                /* Return code */
  int nCommit = 0;                /* Number of transactions to recover */
  int iPass;
  int nJump = 0;                  /* Number of LSM_LOG_JUMP records in pass 0 */
  DbLog *pLog;
  int bOpen;

  rc = lsmFsOpenLog(pDb, &bOpen);
  if( rc!=LSM_OK ) return rc;

  rc = lsmTreeInit(pDb);
  if( rc!=LSM_OK ) return rc;

  pLog = &pDb->treehdr.log;
  lsmCheckpointLogoffset(pDb->pShmhdr->aSnap2, pLog);

  logReaderInit(pDb, pLog, 1, &reader);
  lsmStringInit(&buf1, pDb->pEnv);
  lsmStringInit(&buf2, pDb->pEnv);

  /* The outer for() loop runs at most twice. The first iteration is to 
  ** count the number of committed transactions in the log. The second 
  ** iterates through those transactions and updates the in-memory tree 
  ** structure with their contents.  */
  if( bOpen ){
    for(iPass=0; iPass<2 && rc==LSM_OK; iPass++){
      int bEof = 0;

      while( rc==LSM_OK && !bEof ){
        u8 eType = 0;
        logReaderByte(&reader, &eType, &rc);

        switch( eType ){
          case LSM_LOG_PAD1:
            break;

          case LSM_LOG_PAD2: {
            int nPad;
            logReaderVarint(&reader, &buf1, &nPad, &rc);
            logReaderBlob(&reader, &buf1, nPad, 0, &rc);
            break;
          }

          case LSM_LOG_WRITE:
          case LSM_LOG_WRITE_CKSUM: {
            int nKey;
            int nVal;
            u8 *aVal;
            logReaderVarint(&reader, &buf1, &nKey, &rc);
            logReaderVarint(&reader, &buf2, &nVal, &rc);

            if( eType==LSM_LOG_WRITE_CKSUM ){
              logReaderCksum(&reader, &buf1, &bEof, &rc);
            }else{
              bEof = logRequireCksum(&reader, nKey+nVal);
            }
            if( bEof ) break;

            logReaderBlob(&reader, &buf1, nKey, 0, &rc);
            logReaderBlob(&reader, &buf2, nVal, &aVal, &rc);
            if( iPass==1 && rc==LSM_OK ){ 
              rc = lsmTreeInsert(pDb, (u8 *)buf1.z, nKey, aVal, nVal);
            }
            break;
          }

          case LSM_LOG_DELETE:
          case LSM_LOG_DELETE_CKSUM: {
            int nKey; u8 *aKey;
            logReaderVarint(&reader, &buf1, &nKey, &rc);

            if( eType==LSM_LOG_DELETE_CKSUM ){
              logReaderCksum(&reader, &buf1, &bEof, &rc);
            }else{
              bEof = logRequireCksum(&reader, nKey);
            }
            if( bEof ) break;

            logReaderBlob(&reader, &buf1, nKey, &aKey, &rc);
            if( iPass==1 && rc==LSM_OK ){ 
              rc = lsmTreeInsert(pDb, aKey, nKey, NULL, -1);
            }
            break;
          }

          case LSM_LOG_COMMIT:
            logReaderCksum(&reader, &buf1, &bEof, &rc);
            if( bEof==0 ){
              nCommit++;
              assert( nCommit>0 || iPass==1 );
              if( nCommit==0 ) bEof = 1;
            }
            break;

          case LSM_LOG_JUMP: {
            int iOff = 0;
            logReaderVarint(&reader, &buf1, &iOff, &rc);
            if( rc==LSM_OK ){
              if( iPass==1 ){
                if( pLog->aRegion[2].iStart==0 ){
                  assert( pLog->aRegion[1].iStart==0 );
                  pLog->aRegion[1].iEnd = reader.iOff;
                }else{
                  assert( pLog->aRegion[0].iStart==0 );
                  pLog->aRegion[0].iStart = pLog->aRegion[2].iStart;
                  pLog->aRegion[0].iEnd = reader.iOff-reader.buf.n+reader.iBuf;
                }
                pLog->aRegion[2].iStart = iOff;
              }else{
                if( (nJump++)==2 ){
                  bEof = 1;
                }
              }

              reader.iOff = iOff;
              reader.buf.n = reader.iBuf;
            }
            break;
          }

          default:
            /* Including LSM_LOG_EOF */
            bEof = 1;
            break;
        }
      }

      if( rc==LSM_OK && iPass==0 ){
        if( nCommit==0 ){
          if( pLog->aRegion[2].iStart==0 ){
            iPass = 1;
          }else{
            pLog->aRegion[2].iStart = 0;
            iPass = -1;
            lsmCheckpointZeroLogoffset(pDb);
          }
        }
        logReaderInit(pDb, pLog, 0, &reader);
        nCommit = nCommit * -1;
      }
    }
  }

  /* Initialize DbLog object */
  if( rc==LSM_OK ){
    pLog->aRegion[2].iEnd = reader.iOff - reader.buf.n + reader.iBuf;
    pLog->cksum0 = reader.cksum0;
    pLog->cksum1 = reader.cksum1;
  }

  if( rc==LSM_OK ){
    rc = lsmFinishRecovery(pDb);
  }else{
    lsmFinishRecovery(pDb);
  }

  if( pDb->bRoTrans ){
    lsmFsCloseLog(pDb);
  }

  lsmStringClear(&buf1);
  lsmStringClear(&buf2);
  lsmStringClear(&reader.buf);
  return rc;
}

void lsmLogClose(lsm_db *db){
  if( db->pLogWriter ){
    lsmFree(db->pEnv, db->pLogWriter->buf.z);
    lsmFree(db->pEnv, db->pLogWriter);
    db->pLogWriter = 0;
  }
}

Added ext/lsm1/lsm_main.c.



































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
/*
** 2011-08-18
**
** The author disclaims copyright to this source code.  In place of
** a legal notice, here is a blessing:
**
**    May you do good and not evil.
**    May you find forgiveness for yourself and forgive others.
**    May you share freely, never taking more than you give.
**
*************************************************************************
**
** The main interface to the LSM module.
*/
#include "lsmInt.h"


#ifdef LSM_DEBUG
/*
** This function returns a copy of its only argument.
**
** When the library is built with LSM_DEBUG defined, this function is called
** whenever an error code is generated (not propagated - generated). So
** if the library is mysteriously returning (say) LSM_IOERR, a breakpoint
** may be set in this function to determine why.
*/
int lsmErrorBkpt(int rc){
  /* Set breakpoint here! */
  return rc;
}

/*
** This function contains various assert() statements that test that the
** lsm_db structure passed as an argument is internally consistent.
*/
static void assert_db_state(lsm_db *pDb){

  /* If there is at least one cursor or a write transaction open, the database
  ** handle must be holding a pointer to a client snapshot. And the reverse 
  ** - if there are no open cursors and no write transactions then there must 
  ** not be a client snapshot.  */
  
  assert( (pDb->pCsr!=0||pDb->nTransOpen>0)==(pDb->iReader>=0||pDb->bRoTrans) );

  assert( (pDb->iReader<0 && pDb->bRoTrans==0) || pDb->pClient!=0 );

  assert( pDb->nTransOpen>=0 );
}
#else
# define assert_db_state(x) 
#endif

/*
** The default key-compare function.
*/
static int xCmp(void *p1, int n1, void *p2, int n2){
  int res;
  res = memcmp(p1, p2, LSM_MIN(n1, n2));
  if( res==0 ) res = (n1-n2);
  return res;
}

static void xLog(void *pCtx, int rc, const char *z){
  (void)(rc);
  (void)(pCtx);
  fprintf(stderr, "%s\n", z);
  fflush(stderr);
}

/*
** Allocate a new db handle.
*/
int lsm_new(lsm_env *pEnv, lsm_db **ppDb){
  lsm_db *pDb;

  /* If the user did not provide an environment, use the default. */
  if( pEnv==0 ) pEnv = lsm_default_env();
  assert( pEnv );

  /* Allocate the new database handle */
  *ppDb = pDb = (lsm_db *)lsmMallocZero(pEnv, sizeof(lsm_db));
  if( pDb==0 ) return LSM_NOMEM_BKPT;

  /* Initialize the new object */
  pDb->pEnv = pEnv;
  pDb->nTreeLimit = LSM_DFLT_AUTOFLUSH;
  pDb->nAutockpt = LSM_DFLT_AUTOCHECKPOINT;
  pDb->bAutowork = LSM_DFLT_AUTOWORK;
  pDb->eSafety = LSM_DFLT_SAFETY;
  pDb->xCmp = xCmp;
  pDb->nDfltPgsz = LSM_DFLT_PAGE_SIZE;
  pDb->nDfltBlksz = LSM_DFLT_BLOCK_SIZE;
  pDb->nMerge = LSM_DFLT_AUTOMERGE;
  pDb->nMaxFreelist = LSM_MAX_FREELIST_ENTRIES;
  pDb->bUseLog = LSM_DFLT_USE_LOG;
  pDb->iReader = -1;
  pDb->iRwclient = -1;
  pDb->bMultiProc = LSM_DFLT_MULTIPLE_PROCESSES;
  pDb->iMmap = LSM_DFLT_MMAP;
  pDb->xLog = xLog;
  pDb->compress.iId = LSM_COMPRESSION_NONE;
  return LSM_OK;
}

lsm_env *lsm_get_env(lsm_db *pDb){
  assert( pDb->pEnv );
  return pDb->pEnv;
}

/*
** If database handle pDb is currently holding a client snapshot, but does
** not have any open cursors or write transactions, release it.
*/
static void dbReleaseClientSnapshot(lsm_db *pDb){
  if( pDb->nTransOpen==0 && pDb->pCsr==0 ){
    lsmFinishReadTrans(pDb);
  }
}

static int getFullpathname(
  lsm_env *pEnv, 
  const char *zRel,
  char **pzAbs
){
  int nAlloc = 0;
  char *zAlloc = 0;
  int nReq = 0;
  int rc;

  do{
    nAlloc = nReq;
    rc = pEnv->xFullpath(pEnv, zRel, zAlloc, &nReq);
    if( nReq>nAlloc ){
      zAlloc = lsmReallocOrFreeRc(pEnv, zAlloc, nReq, &rc);
    }
  }while( nReq>nAlloc && rc==LSM_OK );

  if( rc!=LSM_OK ){
    lsmFree(pEnv, zAlloc);
    zAlloc = 0;
  }
  *pzAbs = zAlloc;
  return rc;
}

/*
** Check that the bits in the db->mLock mask are consistent with the
** value stored in db->iRwclient. An assert shall fail otherwise.
*/
static void assertRwclientLockValue(lsm_db *db){
#ifndef NDEBUG
  u64 msk;                        /* Mask of mLock bits for RWCLIENT locks */
  u64 rwclient = 0;               /* Bit corresponding to db->iRwclient */

  if( db->iRwclient>=0 ){
    rwclient = ((u64)1 << (LSM_LOCK_RWCLIENT(db->iRwclient)-1));
  }
  msk  = ((u64)1 << (LSM_LOCK_RWCLIENT(LSM_LOCK_NRWCLIENT)-1)) - 1;
  msk -= (((u64)1 << (LSM_LOCK_RWCLIENT(0)-1)) - 1);

  assert( (db->mLock & msk)==rwclient );
#endif
}

/*
** Open a new connection to database zFilename.
*/
int lsm_open(lsm_db *pDb, const char *zFilename){
  int rc;

  if( pDb->pDatabase ){
    rc = LSM_MISUSE;
  }else{
    char *zFull;

    /* Translate the possibly relative pathname supplied by the user into
    ** an absolute pathname. This is required because the supplied path
    ** is used (either directly or with "-log" appended to it) for more 
    ** than one purpose - to open both the database and log files, and 
    ** perhaps to unlink the log file during disconnection. An absolute
    ** path is required to ensure that the correct files are operated
    ** on even if the application changes the cwd.  */
    rc = getFullpathname(pDb->pEnv, zFilename, &zFull);
    assert( rc==LSM_OK || zFull==0 );

    /* Connect to the database. */
    if( rc==LSM_OK ){
      rc = lsmDbDatabaseConnect(pDb, zFull);
    }

    if( pDb->bReadonly==0 ){
      /* Configure the file-system connection with the page-size and block-size
      ** of this database. Even if the database file is zero bytes in size
      ** on disk, these values have been set in shared-memory by now, and so 
      ** are guaranteed not to change during the lifetime of this connection.  
      */
      if( rc==LSM_OK && LSM_OK==(rc = lsmCheckpointLoad(pDb, 0)) ){
        lsmFsSetPageSize(pDb->pFS, lsmCheckpointPgsz(pDb->aSnapshot));
        lsmFsSetBlockSize(pDb->pFS, lsmCheckpointBlksz(pDb->aSnapshot));
      }
    }

    lsmFree(pDb->pEnv, zFull);
    assertRwclientLockValue(pDb);
  }

  assert( pDb->bReadonly==0 || pDb->bReadonly==1 );
  assert( rc!=LSM_OK || (pDb->pShmhdr==0)==(pDb->bReadonly==1) );

  return rc;
}

int lsm_close(lsm_db *pDb){
  int rc = LSM_OK;
  if( pDb ){
    assert_db_state(pDb);
    if( pDb->pCsr || pDb->nTransOpen ){
      rc = LSM_MISUSE_BKPT;
    }else{
      lsmMCursorFreeCache(pDb);
      lsmFreeSnapshot(pDb->pEnv, pDb->pClient);
      pDb->pClient = 0;

      assertRwclientLockValue(pDb);

      lsmDbDatabaseRelease(pDb);
      lsmLogClose(pDb);
      lsmFsClose(pDb->pFS);
      assert( pDb->mLock==0 );
      
      /* Invoke any destructors registered for the compression or 
      ** compression factory callbacks.  */
      if( pDb->factory.xFree ) pDb->factory.xFree(pDb->factory.pCtx);
      if( pDb->compress.xFree ) pDb->compress.xFree(pDb->compress.pCtx);

      lsmFree(pDb->pEnv, pDb->rollback.aArray);
      lsmFree(pDb->pEnv, pDb->aTrans);
      lsmFree(pDb->pEnv, pDb->apShm);
      lsmFree(pDb->pEnv, pDb);
    }
  }
  return rc;
}

int lsm_config(lsm_db *pDb, int eParam, ...){
  int rc = LSM_OK;
  va_list ap;
  va_start(ap, eParam);

  switch( eParam ){
    case LSM_CONFIG_AUTOFLUSH: {
      /* This parameter is read and written in KB. But all internal 
      ** processing is done in bytes.  */
      int *piVal = va_arg(ap, int *);
      int iVal = *piVal;
      if( iVal>=0 && iVal<=(1024*1024) ){
        pDb->nTreeLimit = iVal*1024;
      }
      *piVal = (pDb->nTreeLimit / 1024);
      break;
    }

    case LSM_CONFIG_AUTOWORK: {
      int *piVal = va_arg(ap, int *);
      if( *piVal>=0 ){
        pDb->bAutowork = *piVal;
      }
      *piVal = pDb->bAutowork;
      break;
    }

    case LSM_CONFIG_AUTOCHECKPOINT: {
      /* This parameter is read and written in KB. But all internal processing
      ** (including the lsm_db.nAutockpt variable) is done in bytes.  */
      int *piVal = va_arg(ap, int *);
      if( *piVal>=0 ){
        int iVal = *piVal;
        pDb->nAutockpt = (i64)iVal * 1024;
      }
      *piVal = (int)(pDb->nAutockpt / 1024);
      break;
    }

    case LSM_CONFIG_PAGE_SIZE: {
      int *piVal = va_arg(ap, int *);
      if( pDb->pDatabase ){
        /* If lsm_open() has been called, this is a read-only parameter. 
        ** Set the output variable to the page-size according to the 
        ** FileSystem object.  */
        *piVal = lsmFsPageSize(pDb->pFS);
      }else{
        if( *piVal>=256 && *piVal<=65536 && ((*piVal-1) & *piVal)==0 ){
          pDb->nDfltPgsz = *piVal;
        }else{
          *piVal = pDb->nDfltPgsz;
        }
      }
      break;
    }

    case LSM_CONFIG_BLOCK_SIZE: {
      /* This parameter is read and written in KB. But all internal 
      ** processing is done in bytes.  */
      int *piVal = va_arg(ap, int *);
      if( pDb->pDatabase ){
        /* If lsm_open() has been called, this is a read-only parameter. 
        ** Set the output variable to the block-size in KB according to the 
        ** FileSystem object.  */
        *piVal = lsmFsBlockSize(pDb->pFS) / 1024;
      }else{
        int iVal = *piVal;
        if( iVal>=64 && iVal<=65536 && ((iVal-1) & iVal)==0 ){
          pDb->nDfltBlksz = iVal * 1024;
        }else{
          *piVal = pDb->nDfltBlksz / 1024;
        }
      }
      break;
    }

    case LSM_CONFIG_SAFETY: {
      int *piVal = va_arg(ap, int *);
      if( *piVal>=0 && *piVal<=2 ){
        pDb->eSafety = *piVal;
      }
      *piVal = pDb->eSafety;
      break;
    }

    case LSM_CONFIG_MMAP: {
      int *piVal = va_arg(ap, int *);
      if( pDb->iReader<0 && *piVal>=0 ){
        pDb->iMmap = *piVal;
        rc = lsmFsConfigure(pDb);
      }
      *piVal = pDb->iMmap;
      break;
    }

    case LSM_CONFIG_USE_LOG: {
      int *piVal = va_arg(ap, int *);
      if( pDb->nTransOpen==0 && (*piVal==0 || *piVal==1) ){
        pDb->bUseLog = *piVal;
      }
      *piVal = pDb->bUseLog;
      break;
    }

    case LSM_CONFIG_AUTOMERGE: {
      int *piVal = va_arg(ap, int *);
      if( *piVal>1 ) pDb->nMerge = *piVal;
      *piVal = pDb->nMerge;
      break;
    }

    case LSM_CONFIG_MAX_FREELIST: {
      int *piVal = va_arg(ap, int *);
      if( *piVal>=2 && *piVal<=LSM_MAX_FREELIST_ENTRIES ){
        pDb->nMaxFreelist = *piVal;
      }
      *piVal = pDb->nMaxFreelist;
      break;
    }

    case LSM_CONFIG_MULTIPLE_PROCESSES: {
      int *piVal = va_arg(ap, int *);
      if( pDb->pDatabase ){
        /* If lsm_open() has been called, this is a read-only parameter. 
        ** Set the output variable to true if this connection is currently
        ** in multi-process mode.  */
        *piVal = lsmDbMultiProc(pDb);
      }else{
        pDb->bMultiProc = *piVal = (*piVal!=0);
      }
      break;
    }

    case LSM_CONFIG_READONLY: {
      int *piVal = va_arg(ap, int *);
      /* If lsm_open() has been called, this is a read-only parameter. */
      if( pDb->pDatabase==0 && *piVal>=0 ){
        pDb->bReadonly = *piVal = (*piVal!=0);
      }
      *piVal = pDb->bReadonly;
      break;
    }

    case LSM_CONFIG_SET_COMPRESSION: {
      lsm_compress *p = va_arg(ap, lsm_compress *);
      if( pDb->iReader>=0 && pDb->bInFactory==0 ){
        /* May not change compression schemes with an open transaction */
        rc = LSM_MISUSE_BKPT;
      }else{
        if( pDb->compress.xFree ){
          /* Invoke any destructor belonging to the current compression. */
          pDb->compress.xFree(pDb->compress.pCtx);
        }
        if( p->xBound==0 ){
          memset(&pDb->compress, 0, sizeof(lsm_compress));
          pDb->compress.iId = LSM_COMPRESSION_NONE;
        }else{
          memcpy(&pDb->compress, p, sizeof(lsm_compress));
        }
        rc = lsmFsConfigure(pDb);
      }
      break;
    }

    case LSM_CONFIG_SET_COMPRESSION_FACTORY: {
      lsm_compress_factory *p = va_arg(ap, lsm_compress_factory *);
      if( pDb->factory.xFree ){
        /* Invoke any destructor belonging to the current factory. */
        pDb->factory.xFree(pDb->factory.pCtx);
      }
      memcpy(&pDb->factory, p, sizeof(lsm_compress_factory));
      break;
    }

    case LSM_CONFIG_GET_COMPRESSION: {
      lsm_compress *p = va_arg(ap, lsm_compress *);
      memcpy(p, &pDb->compress, sizeof(lsm_compress));
      break;
    }

    default:
      rc = LSM_MISUSE;
      break;
  }

  va_end(ap);
  return rc;
}

void lsmAppendSegmentList(LsmString *pStr, char *zPre, Segment *pSeg){
  lsmStringAppendf(pStr, "%s{%d %d %d %d}", zPre, 
        pSeg->iFirst, pSeg->iLastPg, pSeg->iRoot, pSeg->nSize
  );
}

static int infoGetWorker(lsm_db *pDb, Snapshot **pp, int *pbUnlock){
  int rc = LSM_OK;

  assert( *pbUnlock==0 );
  if( !pDb->pWorker ){
    rc = lsmBeginWork(pDb);
    if( rc!=LSM_OK ) return rc;
    *pbUnlock = 1;
  }
  if( pp ) *pp = pDb->pWorker;
  return rc;
}

static void infoFreeWorker(lsm_db *pDb, int bUnlock){
  if( bUnlock ){
    int rcdummy = LSM_BUSY;
    lsmFinishWork(pDb, 0, &rcdummy);
  }
}

int lsmStructList(
  lsm_db *pDb,                    /* Database handle */
  char **pzOut                    /* OUT: Nul-terminated string (tcl list) */
){
  Level *pTopLevel = 0;           /* Top level of snapshot to report on */
  int rc = LSM_OK;
  Level *p;
  LsmString s;
  Snapshot *pWorker;              /* Worker snapshot */
  int bUnlock = 0;

  /* Obtain the worker snapshot */
  rc = infoGetWorker(pDb, &pWorker, &bUnlock);
  if( rc!=LSM_OK ) return rc;

  /* Format the contents of the snapshot as text */
  pTopLevel = lsmDbSnapshotLevel(pWorker);
  lsmStringInit(&s, pDb->pEnv);
  for(p=pTopLevel; rc==LSM_OK && p; p=p->pNext){
    int i;
    lsmStringAppendf(&s, "%s{%d", (s.n ? " " : ""), (int)p->iAge);
    lsmAppendSegmentList(&s, " ", &p->lhs);
    for(i=0; rc==LSM_OK && i<p->nRight; i++){
      lsmAppendSegmentList(&s, " ", &p->aRhs[i]);
    }
    lsmStringAppend(&s, "}", 1);
  }
  rc = s.n>=0 ? LSM_OK : LSM_NOMEM;

  /* Release the snapshot and return */
  infoFreeWorker(pDb, bUnlock);
  *pzOut = s.z;
  return rc;
}

static int infoFreelistCb(void *pCtx, int iBlk, i64 iSnapshot){
  LsmString *pStr = (LsmString *)pCtx;
  lsmStringAppendf(pStr, "%s{%d %lld}", (pStr->n?" ":""), iBlk, iSnapshot);
  return 0;
}

int lsmInfoFreelist(lsm_db *pDb, char **pzOut){
  Snapshot *pWorker;              /* Worker snapshot */
  int bUnlock = 0;
  LsmString s;
  int rc;

  /* Obtain the worker snapshot */
  rc = infoGetWorker(pDb, &pWorker, &bUnlock);
  if( rc!=LSM_OK ) return rc;

  lsmStringInit(&s, pDb->pEnv);
  rc = lsmWalkFreelist(pDb, 0, infoFreelistCb, &s);
  if( rc!=LSM_OK ){
    lsmFree(pDb->pEnv, s.z);
  }else{
    *pzOut = s.z;
  }

  /* Release the snapshot and return */
  infoFreeWorker(pDb, bUnlock);
  return rc;
}

static int infoTreeSize(lsm_db *db, int *pnOldKB, int *pnNewKB){
  ShmHeader *pShm = db->pShmhdr;
  TreeHeader *p = &pShm->hdr1;

  /* The following code suffers from two race conditions, as it accesses and
  ** trusts the contents of shared memory without verifying checksums:
  **
  **   * The two values read - TreeHeader.root.nByte and oldroot.nByte - are 
  **     32-bit fields. It is assumed that reading from one of these
  **     is atomic - that it is not possible to read a partially written
  **     garbage value. However the two values may be mutually inconsistent. 
  **
  **   * TreeHeader.iLogOff is a 64-bit value. And lsmCheckpointLogOffset()
  **     reads a 64-bit value from a snapshot stored in shared memory. It
  **     is assumed that in each case it is possible to read a partially
  **     written garbage value. If this occurs, then the value returned
  **     for the size of the "old" tree may reflect the size of an "old"
  **     tree that was recently flushed to disk.
  **
  ** Given the context in which this function is called (as a result of an
  ** lsm_info(LSM_INFO_TREE_SIZE) request), neither of these are considered to
  ** be problems.
  */
  *pnNewKB = ((int)p->root.nByte + 1023) / 1024;
  if( p->iOldShmid ){
    if( p->iOldLog==lsmCheckpointLogOffset(pShm->aSnap1) ){
      *pnOldKB = 0;
    }else{
      *pnOldKB = ((int)p->oldroot.nByte + 1023) / 1024;
    }
  }else{
    *pnOldKB = 0;
  }

  return LSM_OK;
}

int lsm_info(lsm_db *pDb, int eParam, ...){
  int rc = LSM_OK;
  va_list ap;
  va_start(ap, eParam);

  switch( eParam ){
    case LSM_INFO_NWRITE: {
      int *piVal = va_arg(ap, int *);
      *piVal = lsmFsNWrite(pDb->pFS);
      break;
    }

    case LSM_INFO_NREAD: {
      int *piVal = va_arg(ap, int *);
      *piVal = lsmFsNRead(pDb->pFS);
      break;
    }

    case LSM_INFO_DB_STRUCTURE: {
      char **pzVal = va_arg(ap, char **);
      rc = lsmStructList(pDb, pzVal);
      break;
    }

    case LSM_INFO_ARRAY_STRUCTURE: {
      Pgno pgno = va_arg(ap, Pgno);
      char **pzVal = va_arg(ap, char **);
      rc = lsmInfoArrayStructure(pDb, 0, pgno, pzVal);
      break;
    }

    case LSM_INFO_ARRAY_PAGES: {
      Pgno pgno = va_arg(ap, Pgno);
      char **pzVal = va_arg(ap, char **);
      rc = lsmInfoArrayPages(pDb, pgno, pzVal);
      break;
    }

    case LSM_INFO_PAGE_HEX_DUMP:
    case LSM_INFO_PAGE_ASCII_DUMP: {
      Pgno pgno = va_arg(ap, Pgno);
      char **pzVal = va_arg(ap, char **);
      int bUnlock = 0;
      rc = infoGetWorker(pDb, 0, &bUnlock);
      if( rc==LSM_OK ){
        int bHex = (eParam==LSM_INFO_PAGE_HEX_DUMP);
        rc = lsmInfoPageDump(pDb, pgno, bHex, pzVal);
      }
      infoFreeWorker(pDb, bUnlock);
      break;
    }

    case LSM_INFO_LOG_STRUCTURE: {
      char **pzVal = va_arg(ap, char **);
      rc = lsmInfoLogStructure(pDb, pzVal);
      break;
    }

    case LSM_INFO_FREELIST: {
      char **pzVal = va_arg(ap, char **);
      rc = lsmInfoFreelist(pDb, pzVal);
      break;
    }

    case LSM_INFO_CHECKPOINT_SIZE: {
      int *pnKB = va_arg(ap, int *);
      rc = lsmCheckpointSize(pDb, pnKB);
      break;
    }

    case LSM_INFO_TREE_SIZE: {
      int *pnOld = va_arg(ap, int *);
      int *pnNew = va_arg(ap, int *);
      rc = infoTreeSize(pDb, pnOld, pnNew);
      break;
    }

    case LSM_INFO_COMPRESSION_ID: {
      unsigned int *piOut = va_arg(ap, unsigned int *);
      if( pDb->pClient ){
        *piOut = pDb->pClient->iCmpId;
      }else{
        rc = lsmInfoCompressionId(pDb, piOut);
      }
      break;
    }

    default:
      rc = LSM_MISUSE;
      break;
  }

  va_end(ap);
  return rc;
}

static int doWriteOp(
  lsm_db *pDb,
  int bDeleteRange,
  const void *pKey, int nKey,     /* Key to write or delete */
  const void *pVal, int nVal      /* Value to write. Or nVal==-1 for a delete */
){
  int rc = LSM_OK;                /* Return code */
  int bCommit = 0;                /* True to commit before returning */

  if( pDb->nTransOpen==0 ){
    bCommit = 1;
    rc = lsm_begin(pDb, 1);
  }

  if( rc==LSM_OK ){
    if( bDeleteRange==0 ){
      rc = lsmLogWrite(pDb, (void *)pKey, nKey, (void *)pVal, nVal);
    }else{
      /* TODO */
    }
  }

  lsmSortedSaveTreeCursors(pDb);

  if( rc==LSM_OK ){
    int pgsz = lsmFsPageSize(pDb->pFS);
    int nQuant = LSM_AUTOWORK_QUANT * pgsz;
    int nBefore;
    int nAfter;
    int nDiff;

    if( nQuant>pDb->nTreeLimit ){
      nQuant = pDb->nTreeLimit;
    }

    nBefore = lsmTreeSize(pDb);
    if( bDeleteRange ){
      rc = lsmTreeDelete(pDb, (void *)pKey, nKey, (void *)pVal, nVal);
    }else{
      rc = lsmTreeInsert(pDb, (void *)pKey, nKey, (void *)pVal, nVal);
    }

    nAfter = lsmTreeSize(pDb);
    nDiff = (nAfter/nQuant) - (nBefore/nQuant);
    if( rc==LSM_OK && pDb->bAutowork && nDiff!=0 ){
      rc = lsmSortedAutoWork(pDb, nDiff * LSM_AUTOWORK_QUANT);
    }
  }

  /* If a transaction was opened at the start of this function, commit it. 
  ** Or, if an error has occurred, roll it back.  */
  if( bCommit ){
    if( rc==LSM_OK ){
      rc = lsm_commit(pDb, 0);
    }else{
      lsm_rollback(pDb, 0);
    }
  }

  return rc;
}

/* 
** Write a new value into the database.
*/
int lsm_insert(
  lsm_db *db,                     /* Database connection */
  const void *pKey, int nKey,     /* Key to write or delete */
  const void *pVal, int nVal      /* Value to write. Or nVal==-1 for a delete */
){
  return doWriteOp(db, 0, pKey, nKey, pVal, nVal);
}

/*
** Delete a value from the database. 
*/
int lsm_delete(lsm_db *db, const void *pKey, int nKey){
  return doWriteOp(db, 0, pKey, nKey, 0, -1);
}

/*
** Delete a range of database keys.
*/
int lsm_delete_range(
  lsm_db *db,                     /* Database handle */
  const void *pKey1, int nKey1,   /* Lower bound of range to delete */
  const void *pKey2, int nKey2    /* Upper bound of range to delete */
){
  int rc = LSM_OK;
  if( db->xCmp((void *)pKey1, nKey1, (void *)pKey2, nKey2)<0 ){
    rc = doWriteOp(db, 1, pKey1, nKey1, pKey2, nKey2);
  }
  return rc;
}

/*
** Open a new cursor handle. 
**
** If there are currently no other open cursor handles, and no open write
** transaction, open a read transaction here.
*/
int lsm_csr_open(lsm_db *pDb, lsm_cursor **ppCsr){
  int rc = LSM_OK;                /* Return code */
  MultiCursor *pCsr = 0;          /* New cursor object */

  /* Open a read transaction if one is not already open. */
  assert_db_state(pDb);

  if( pDb->pShmhdr==0 ){
    assert( pDb->bReadonly );
    rc = lsmBeginRoTrans(pDb);
  }else if( pDb->iReader<0 ){
    rc = lsmBeginReadTrans(pDb);
  }

  /* Allocate the multi-cursor. */
  if( rc==LSM_OK ){
    rc = lsmMCursorNew(pDb, &pCsr);
  }

  /* If an error has occured, set the output to NULL and delete any partially
  ** allocated cursor. If this means there are no open cursors, release the
  ** client snapshot.  */
  if( rc!=LSM_OK ){
    lsmMCursorClose(pCsr, 0);
    dbReleaseClientSnapshot(pDb);
  }

  assert_db_state(pDb);
  *ppCsr = (lsm_cursor *)pCsr;
  return rc;
}

/*
** Close a cursor opened using lsm_csr_open().
*/
int lsm_csr_close(lsm_cursor *p){
  if( p ){
    lsm_db *pDb = lsmMCursorDb((MultiCursor *)p);
    assert_db_state(pDb);
    lsmMCursorClose((MultiCursor *)p, 1);
    dbReleaseClientSnapshot(pDb);
    assert_db_state(pDb);
  }
  return LSM_OK;
}

/*
** Attempt to seek the cursor to the database entry specified by pKey/nKey.
** If an error occurs (e.g. an OOM or IO error), return an LSM error code.
** Otherwise, return LSM_OK.
*/
int lsm_csr_seek(lsm_cursor *pCsr, const void *pKey, int nKey, int eSeek){
  return lsmMCursorSeek((MultiCursor *)pCsr, 0, (void *)pKey, nKey, eSeek);
}

int lsm_csr_next(lsm_cursor *pCsr){
  return lsmMCursorNext((MultiCursor *)pCsr);
}

int lsm_csr_prev(lsm_cursor *pCsr){
  return lsmMCursorPrev((MultiCursor *)pCsr);
}

int lsm_csr_first(lsm_cursor *pCsr){
  return lsmMCursorFirst((MultiCursor *)pCsr);
}

int lsm_csr_last(lsm_cursor *pCsr){
  return lsmMCursorLast((MultiCursor *)pCsr);
}

int lsm_csr_valid(lsm_cursor *pCsr){
  return lsmMCursorValid((MultiCursor *)pCsr);
}

int lsm_csr_key(lsm_cursor *pCsr, const void **ppKey, int *pnKey){
  return lsmMCursorKey((MultiCursor *)pCsr, (void **)ppKey, pnKey);
}

int lsm_csr_value(lsm_cursor *pCsr, const void **ppVal, int *pnVal){
  return lsmMCursorValue((MultiCursor *)pCsr, (void **)ppVal, pnVal);
}

void lsm_config_log(
  lsm_db *pDb, 
  void (*xLog)(void *, int, const char *), 
  void *pCtx
){
  pDb->xLog = xLog;
  pDb->pLogCtx = pCtx;
}

void lsm_config_work_hook(
  lsm_db *pDb, 
  void (*xWork)(lsm_db *, void *), 
  void *pCtx
){
  pDb->xWork = xWork;
  pDb->pWorkCtx = pCtx;
}

void lsmLogMessage(lsm_db *pDb, int rc, const char *zFormat, ...){
  if( pDb->xLog ){
    LsmString s;
    va_list ap, ap2;
    lsmStringInit(&s, pDb->pEnv);
    va_start(ap, zFormat);
    va_start(ap2, zFormat);
    lsmStringVAppendf(&s, zFormat, ap, ap2);
    va_end(ap);
    va_end(ap2);
    pDb->xLog(pDb->pLogCtx, rc, s.z);
    lsmStringClear(&s);
  }
}

int lsm_begin(lsm_db *pDb, int iLevel){
  int rc;

  assert_db_state( pDb );
  rc = (pDb->bReadonly ? LSM_READONLY : LSM_OK);

  /* A value less than zero means open one more transaction. */
  if( iLevel<0 ) iLevel = pDb->nTransOpen + 1;
  if( iLevel>pDb->nTransOpen ){
    int i;

    /* Extend the pDb->aTrans[] array if required. */
    if( rc==LSM_OK && pDb->nTransAlloc<iLevel ){
      TransMark *aNew;            /* New allocation */
      int nByte = sizeof(TransMark) * (iLevel+1);
      aNew = (TransMark *)lsmRealloc(pDb->pEnv, pDb->aTrans, nByte);
      if( !aNew ){
        rc = LSM_NOMEM;
      }else{
        nByte = sizeof(TransMark) * (iLevel+1 - pDb->nTransAlloc);
        memset(&aNew[pDb->nTransAlloc], 0, nByte);
        pDb->nTransAlloc = iLevel+1;
        pDb->aTrans = aNew;
      }
    }

    if( rc==LSM_OK && pDb->nTransOpen==0 ){
      rc = lsmBeginWriteTrans(pDb);
    }

    if( rc==LSM_OK ){
      for(i=pDb->nTransOpen; i<iLevel; i++){
        lsmTreeMark(pDb, &pDb->aTrans[i].tree);
        lsmLogTell(pDb, &pDb->aTrans[i].log);
      }
      pDb->nTransOpen = iLevel;
    }
  }

  return rc;
}

int lsm_commit(lsm_db *pDb, int iLevel){
  int rc = LSM_OK;

  assert_db_state( pDb );

  /* A value less than zero means close the innermost nested transaction. */
  if( iLevel<0 ) iLevel = LSM_MAX(0, pDb->nTransOpen - 1);

  if( iLevel<pDb->nTransOpen ){
    if( iLevel==0 ){
      /* Commit the transaction to disk. */
      if( rc==LSM_OK ) rc = lsmLogCommit(pDb);
      if( rc==LSM_OK && pDb->eSafety==LSM_SAFETY_FULL ){
        rc = lsmFsSyncLog(pDb->pFS);
      }
      lsmFinishWriteTrans(pDb, (rc==LSM_OK));
    }
    pDb->nTransOpen = iLevel;
  }
  dbReleaseClientSnapshot(pDb);
  return rc;
}

int lsm_rollback(lsm_db *pDb, int iLevel){
  int rc = LSM_OK;
  assert_db_state( pDb );

  if( pDb->nTransOpen ){
    /* A value less than zero means close the innermost nested transaction. */
    if( iLevel<0 ) iLevel = LSM_MAX(0, pDb->nTransOpen - 1);

    if( iLevel<=pDb->nTransOpen ){
      TransMark *pMark = &pDb->aTrans[(iLevel==0 ? 0 : iLevel-1)];
      lsmTreeRollback(pDb, &pMark->tree);
      if( iLevel ) lsmLogSeek(pDb, &pMark->log);
      pDb->nTransOpen = iLevel;
    }

    if( pDb->nTransOpen==0 ){
      lsmFinishWriteTrans(pDb, 0);
    }
    dbReleaseClientSnapshot(pDb);
  }

  return rc;
}

int lsm_get_user_version(lsm_db *pDb, unsigned int *piUsr){
  int rc = LSM_OK;                /* Return code */

  /* Open a read transaction if one is not already open. */
  assert_db_state(pDb);
  if( pDb->pShmhdr==0 ){
    assert( pDb->bReadonly );
    rc = lsmBeginRoTrans(pDb);
  }else if( pDb->iReader<0 ){
    rc = lsmBeginReadTrans(pDb);
  }

  /* Allocate the multi-cursor. */
  if( rc==LSM_OK ){
    *piUsr = pDb->treehdr.iUsrVersion;
  }

  dbReleaseClientSnapshot(pDb);
  assert_db_state(pDb);
  return rc;
}

int lsm_set_user_version(lsm_db *pDb, unsigned int iUsr){
  int rc = LSM_OK;                /* Return code */
  int bCommit = 0;                /* True to commit before returning */

  if( pDb->nTransOpen==0 ){
    bCommit = 1;
    rc = lsm_begin(pDb, 1);
  }

  if( rc==LSM_OK ){
    pDb->treehdr.iUsrVersion = iUsr;
  }

  /* If a transaction was opened at the start of this function, commit it. 
  ** Or, if an error has occurred, roll it back.  */
  if( bCommit ){
    if( rc==LSM_OK ){
      rc = lsm_commit(pDb, 0);
    }else{
      lsm_rollback(pDb, 0);
    }
  }

  return rc;
}

Added ext/lsm1/lsm_mem.c.

















































































































































































































>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
/*
** 2011-08-18
**
** The author disclaims copyright to this source code.  In place of
** a legal notice, here is a blessing:
**
**    May you do good and not evil.
**    May you find forgiveness for yourself and forgive others.
**    May you share freely, never taking more than you give.
**
*************************************************************************
**
** Helper routines for memory allocation.
*/
#include "lsmInt.h"

/*
** The following routines are called internally by LSM sub-routines. In
** this case a valid environment pointer must be supplied.
*/
void *lsmMalloc(lsm_env *pEnv, size_t N){
  assert( pEnv );
  return pEnv->xMalloc(pEnv, N);
}
void lsmFree(lsm_env *pEnv, void *p){
  assert( pEnv );
  pEnv->xFree(pEnv, p);
}
void *lsmRealloc(lsm_env *pEnv, void *p, size_t N){
  assert( pEnv );
  return pEnv->xRealloc(pEnv, p, N);
}

/*
** Core memory allocation routines for LSM.
*/
void *lsm_malloc(lsm_env *pEnv, size_t N){
  return lsmMalloc(pEnv ? pEnv : lsm_default_env(), N);
}
void lsm_free(lsm_env *pEnv, void *p){
  lsmFree(pEnv ? pEnv : lsm_default_env(), p);
}
void *lsm_realloc(lsm_env *pEnv, void *p, size_t N){
  return lsmRealloc(pEnv ? pEnv : lsm_default_env(), p, N);
}

void *lsmMallocZero(lsm_env *pEnv, size_t N){
  void *pRet;
  assert( pEnv );
  pRet = lsmMalloc(pEnv, N);
  if( pRet ) memset(pRet, 0, N);
  return pRet;
}

void *lsmMallocRc(lsm_env *pEnv, size_t N, int *pRc){
  void *pRet = 0;
  if( *pRc==LSM_OK ){
    pRet = lsmMalloc(pEnv, N);
    if( pRet==0 ){
      *pRc = LSM_NOMEM_BKPT;
    }
  }
  return pRet;
}

void *lsmMallocZeroRc(lsm_env *pEnv, size_t N, int *pRc){
  void *pRet = 0;
  if( *pRc==LSM_OK ){
    pRet = lsmMallocZero(pEnv, N);
    if( pRet==0 ){
      *pRc = LSM_NOMEM_BKPT;
    }
  }
  return pRet;
}

void *lsmReallocOrFree(lsm_env *pEnv, void *p, size_t N){
  void *pNew;
  pNew = lsm_realloc(pEnv, p, N);
  if( !pNew ) lsm_free(pEnv, p);
  return pNew;
}

void *lsmReallocOrFreeRc(lsm_env *pEnv, void *p, size_t N, int *pRc){
  void *pRet = 0;
  if( *pRc ){
    lsmFree(pEnv, p);
  }else{
    pRet = lsmReallocOrFree(pEnv, p, N);
    if( !pRet ) *pRc = LSM_NOMEM_BKPT;
  }
  return pRet;
}

char *lsmMallocStrdup(lsm_env *pEnv, const char *zIn){
  int nByte;
  char *zRet;
  nByte = strlen(zIn);
  zRet = lsmMalloc(pEnv, nByte+1);
  if( zRet ){
    memcpy(zRet, zIn, nByte+1);
  }
  return zRet;
}

Added ext/lsm1/lsm_mutex.c.

















































































































































































>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
/*
** 2012-01-30
**
** The author disclaims copyright to this source code.  In place of
** a legal notice, here is a blessing:
**
**    May you do good and not evil.
**    May you find forgiveness for yourself and forgive others.
**    May you share freely, never taking more than you give.
**
*************************************************************************
**
** Mutex functions for LSM.
*/
#include "lsmInt.h"

/*
** Allocate a new mutex.
*/
int lsmMutexNew(lsm_env *pEnv, lsm_mutex **ppNew){
  return pEnv->xMutexNew(pEnv, ppNew);
}

/*
** Return a handle for one of the static mutexes.
*/
int lsmMutexStatic(lsm_env *pEnv, int iMutex, lsm_mutex **ppStatic){
  return pEnv->xMutexStatic(pEnv, iMutex, ppStatic);
}

/*
** Free a mutex allocated by lsmMutexNew().
*/
void lsmMutexDel(lsm_env *pEnv, lsm_mutex *pMutex){
  if( pMutex ) pEnv->xMutexDel(pMutex);
}

/*
** Enter a mutex.
*/
void lsmMutexEnter(lsm_env *pEnv, lsm_mutex *pMutex){
  pEnv->xMutexEnter(pMutex);
}

/*
** Attempt to enter a mutex, but do not block. If successful, return zero.
** Otherwise, if the mutex is already held by some other thread and is not
** entered, return non zero.
**
** Each successful call to this function must be matched by a call to
** lsmMutexLeave().
*/
int lsmMutexTry(lsm_env *pEnv, lsm_mutex *pMutex){
  return pEnv->xMutexTry(pMutex);
}

/*
** Leave a mutex.
*/
void lsmMutexLeave(lsm_env *pEnv, lsm_mutex *pMutex){
  pEnv->xMutexLeave(pMutex);
}

#ifndef NDEBUG
/*
** Return non-zero if the mutex passed as the second argument is held
** by the calling thread, or zero otherwise. If the implementation is not 
** able to tell if the mutex is held by the caller, it should return
** non-zero.
**
** This function is only used as part of assert() statements.
*/
int lsmMutexHeld(lsm_env *pEnv, lsm_mutex *pMutex){
  return pEnv->xMutexHeld ? pEnv->xMutexHeld(pMutex) : 1;
}

/*
** Return non-zero if the mutex passed as the second argument is not 
** held by the calling thread, or zero otherwise. If the implementation 
** is not able to tell if the mutex is held by the caller, it should 
** return non-zero.
**
** This function is only used as part of assert() statements.
*/
int lsmMutexNotHeld(lsm_env *pEnv, lsm_mutex *pMutex){
  return pEnv->xMutexNotHeld ? pEnv->xMutexNotHeld(pMutex) : 1;
}
#endif

Added ext/lsm1/lsm_shared.c.





































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
1388
1389
1390
1391
1392
1393
1394
1395
1396
1397
1398
1399
1400
1401
1402
1403
1404
1405
1406
1407
1408
1409
1410
1411
1412
1413
1414
1415
1416
1417
1418
1419
1420
1421
1422
1423
1424
1425
1426
1427
1428
1429
1430
1431
1432
1433
1434
1435
1436
1437
1438
1439
1440
1441
1442
1443
1444
1445
1446
1447
1448
1449
1450
1451
1452
1453
1454
1455
1456
1457
1458
1459
1460
1461
1462
1463
1464
1465
1466
1467
1468
1469
1470
1471
1472
1473
1474
1475
1476
1477
1478
1479
1480
1481
1482
1483
1484
1485
1486
1487
1488
1489
1490
1491
1492
1493
1494
1495
1496
1497
1498
1499
1500
1501
1502
1503
1504
1505
1506
1507
1508
1509
1510
1511
1512
1513
1514
1515
1516
1517
1518
1519
1520
1521
1522
1523
1524
1525
1526
1527
1528
1529
1530
1531
1532
1533
1534
1535
1536
1537
1538
1539
1540
1541
1542
1543
1544
1545
1546
1547
1548
1549
1550
1551
1552
1553
1554
1555
1556
1557
1558
1559
1560
1561
1562
1563
1564
1565
1566
1567
1568
1569
1570
1571
1572
1573
1574
1575
1576
1577
1578
1579
1580
1581
1582
1583
1584
1585
1586
1587
1588
1589
1590
1591
1592
1593
1594
1595
1596
1597
1598
1599
1600
1601
1602
1603
1604
1605
1606
1607
1608
1609
1610
1611
1612
1613
1614
1615
1616
1617
1618
1619
1620
1621
1622
1623
1624
1625
1626
1627
1628
1629
1630
1631
1632
1633
1634
1635
1636
1637
1638
1639
1640
1641
1642
1643
1644
1645
1646
1647
1648
1649
1650
1651
1652
1653
1654
1655
1656
1657
1658
1659
1660
1661
1662
1663
1664
1665
1666
1667
1668
1669
1670
1671
1672
1673
1674
1675
1676
1677
1678
1679
1680
1681
1682
1683
1684
1685
1686
1687
1688
1689
1690
1691
1692
1693
1694
1695
1696
1697
1698
1699
1700
1701
1702
1703
1704
1705
1706
1707
1708
1709
1710
1711
1712
1713
1714
1715
1716
1717
1718
1719
1720
1721
1722
1723
1724
1725
1726
1727
1728
1729
1730
1731
1732
1733
1734
1735
1736
1737
1738
1739
1740
1741
1742
1743
1744
1745
1746
1747
1748
1749
1750
1751
1752
1753
1754
1755
1756
1757
1758
1759
1760
1761
1762
1763
1764
1765
1766
1767
1768
1769
1770
1771
1772
1773
1774
1775
1776
1777
1778
1779
1780
1781
1782
1783
1784
1785
1786
1787
1788
1789
1790
1791
1792
1793
1794
1795
1796
1797
1798
1799
1800
1801
1802
1803
1804
1805
1806
1807
1808
1809
1810
1811
1812
1813
1814
1815
1816
1817
1818
1819
1820
1821
1822
1823
1824
1825
1826
1827
1828
1829
1830
1831
1832
1833
1834
1835
1836
1837
1838
1839
1840
1841
1842
1843
1844
1845
1846
1847
1848
1849
1850
1851
1852
1853
1854
1855
1856
1857
1858
1859
1860
1861
1862
1863
1864
1865
1866
1867
1868
1869
1870
1871
1872
1873
1874
1875
1876
1877
1878
1879
1880
1881
1882
1883
1884
1885
1886
1887
1888
1889
1890
1891
1892
1893
1894
1895
1896
1897
1898
1899
1900
1901
1902
1903
1904
1905
1906
1907
1908
1909
1910
1911
1912
1913
1914
1915
1916
1917
1918
1919
1920
1921
1922
1923
1924
1925
1926
1927
1928
1929
1930
1931
1932
1933
1934
1935
1936
1937
1938
1939
1940
1941
1942
1943
1944
1945
1946
1947
1948
1949
1950
1951
1952
1953
1954
1955
1956
1957
1958
1959
1960
1961
1962
1963
1964
1965
1966
1967
1968
1969
1970
/*
** 2012-01-23
**
** The author disclaims copyright to this source code.  In place of
** a legal notice, here is a blessing:
**
**    May you do good and not evil.
**    May you find forgiveness for yourself and forgive others.
**    May you share freely, never taking more than you give.
**
*************************************************************************
**
** Utilities used to help multiple LSM clients to coexist within the
** same process space.
*/
#include "lsmInt.h"

/*
** Global data. All global variables used by code in this file are grouped
** into the following structure instance.
**
** pDatabase:
**   Linked list of all Database objects allocated within this process.
**   This list may not be traversed without holding the global mutex (see
**   functions enterGlobalMutex() and leaveGlobalMutex()).
*/
static struct SharedData {
  Database *pDatabase;            /* Linked list of all Database objects */
} gShared;

/*
** Database structure. There is one such structure for each distinct 
** database accessed by this process. They are stored in the singly linked 
** list starting at global variable gShared.pDatabase. Database objects are 
** reference counted. Once the number of connections to the associated
** database drops to zero, they are removed from the linked list and deleted.
**
** pFile:
**   In multi-process mode, this file descriptor is used to obtain locks 
**   and to access shared-memory. In single process mode, its only job is
**   to hold the exclusive lock on the file.
**   
*/
struct Database {
  /* Protected by the global mutex (enterGlobalMutex/leaveGlobalMutex): */
  char *zName;                    /* Canonical path to database file */
  int nName;                      /* strlen(zName) */
  int nDbRef;                     /* Number of associated lsm_db handles */
  Database *pDbNext;              /* Next Database structure in global list */

  /* Protected by the local mutex (pClientMutex) */
  int bReadonly;                  /* True if Database.pFile is read-only */
  int bMultiProc;                 /* True if running in multi-process mode */
  lsm_file *pFile;                /* Used for locks/shm in multi-proc mode */
  LsmFile *pLsmFile;              /* List of deferred closes */
  lsm_mutex *pClientMutex;        /* Protects the apShmChunk[] and pConn */
  int nShmChunk;                  /* Number of entries in apShmChunk[] array */
  void **apShmChunk;              /* Array of "shared" memory regions */
  lsm_db *pConn;                  /* List of connections to this db. */
};

/*
** Functions to enter and leave the global mutex. This mutex is used
** to protect the global linked-list headed at gShared.pDatabase.
*/
static int enterGlobalMutex(lsm_env *pEnv){
  lsm_mutex *p;
  int rc = lsmMutexStatic(pEnv, LSM_MUTEX_GLOBAL, &p);
  if( rc==LSM_OK ) lsmMutexEnter(pEnv, p);
  return rc;
}
static void leaveGlobalMutex(lsm_env *pEnv){
  lsm_mutex *p;
  lsmMutexStatic(pEnv, LSM_MUTEX_GLOBAL, &p);
  lsmMutexLeave(pEnv, p);
}

#ifdef LSM_DEBUG
static int holdingGlobalMutex(lsm_env *pEnv){
  lsm_mutex *p;
  lsmMutexStatic(pEnv, LSM_MUTEX_GLOBAL, &p);
  return lsmMutexHeld(pEnv, p);
}
#endif

#if 0
static void assertNotInFreelist(Freelist *p, int iBlk){
  int i; 
  for(i=0; i<p->nEntry; i++){
    assert( p->aEntry[i].iBlk!=iBlk );
  }
}
#else
# define assertNotInFreelist(x,y)
#endif

/*
** Append an entry to the free-list. If (iId==-1), this is a delete.
*/
int freelistAppend(lsm_db *db, int iBlk, i64 iId){
  lsm_env *pEnv = db->pEnv;
  Freelist *p;
  int i; 

  assert( iId==-1 || iId>=0 );
  p = db->bUseFreelist ? db->pFreelist : &db->pWorker->freelist;

  /* Extend the space allocated for the freelist, if required */
  assert( p->nAlloc>=p->nEntry );
  if( p->nAlloc==p->nEntry ){
    int nNew; 
    int nByte; 
    FreelistEntry *aNew;

    nNew = (p->nAlloc==0 ? 4 : p->nAlloc*2);
    nByte = sizeof(FreelistEntry) * nNew;
    aNew = (FreelistEntry *)lsmRealloc(pEnv, p->aEntry, nByte);
    if( !aNew ) return LSM_NOMEM_BKPT;
    p->nAlloc = nNew;
    p->aEntry = aNew;
  }

  for(i=0; i<p->nEntry; i++){
    assert( i==0 || p->aEntry[i].iBlk > p->aEntry[i-1].iBlk );
    if( p->aEntry[i].iBlk>=iBlk ) break;
  }

  if( i<p->nEntry && p->aEntry[i].iBlk==iBlk ){
    /* Clobber an existing entry */
    p->aEntry[i].iId = iId;
  }else{
    /* Insert a new entry into the list */
    int nByte = sizeof(FreelistEntry)*(p->nEntry-i);
    memmove(&p->aEntry[i+1], &p->aEntry[i], nByte);
    p->aEntry[i].iBlk = iBlk;
    p->aEntry[i].iId = iId;
    p->nEntry++;
  }

  return LSM_OK;
}

/*
** This function frees all resources held by the Database structure passed
** as the only argument.
*/
static void freeDatabase(lsm_env *pEnv, Database *p){
  assert( holdingGlobalMutex(pEnv) );
  if( p ){
    /* Free the mutexes */
    lsmMutexDel(pEnv, p->pClientMutex);

    if( p->pFile ){
      lsmEnvClose(pEnv, p->pFile);
    }

    /* Free the array of shm pointers */
    lsmFree(pEnv, p->apShmChunk);

    /* Free the memory allocated for the Database struct itself */
    lsmFree(pEnv, p);
  }
}

typedef struct DbTruncateCtx DbTruncateCtx;
struct DbTruncateCtx {
  int nBlock;
  i64 iInUse;
};

static int dbTruncateCb(void *pCtx, int iBlk, i64 iSnapshot){
  DbTruncateCtx *p = (DbTruncateCtx *)pCtx;
  if( iBlk!=p->nBlock || (p->iInUse>=0 && iSnapshot>=p->iInUse) ) return 1;
  p->nBlock--;
  return 0;
}

static int dbTruncate(lsm_db *pDb, i64 iInUse){
  int rc = LSM_OK;
#if 0
  int i;
  DbTruncateCtx ctx;

  assert( pDb->pWorker );
  ctx.nBlock = pDb->pWorker->nBlock;
  ctx.iInUse = iInUse;

  rc = lsmWalkFreelist(pDb, 1, dbTruncateCb, (void *)&ctx);
  for(i=ctx.nBlock+1; rc==LSM_OK && i<=pDb->pWorker->nBlock; i++){
    rc = freelistAppend(pDb, i, -1);
  }

  if( rc==LSM_OK ){
#ifdef LSM_LOG_FREELIST
    if( ctx.nBlock!=pDb->pWorker->nBlock ){
      lsmLogMessage(pDb, 0, 
          "dbTruncate(): truncated db to %d blocks",ctx.nBlock
      );
    }
#endif
    pDb->pWorker->nBlock = ctx.nBlock;
  }
#endif
  return rc;
}


/*
** This function is called during database shutdown (when the number of
** connections drops from one to zero). It truncates the database file
** to as small a size as possible without truncating away any blocks that
** contain data.
*/
static int dbTruncateFile(lsm_db *pDb){
  int rc;

  assert( pDb->pWorker==0 );
  assert( lsmShmAssertLock(pDb, LSM_LOCK_DMS1, LSM_LOCK_EXCL) );
  rc = lsmCheckpointLoadWorker(pDb);

  if( rc==LSM_OK ){
    DbTruncateCtx ctx;

    /* Walk the database free-block-list in reverse order. Set ctx.nBlock
    ** to the block number of the last block in the database that actually
    ** contains data. */
    ctx.nBlock = pDb->pWorker->nBlock;
    ctx.iInUse = -1;
    rc = lsmWalkFreelist(pDb, 1, dbTruncateCb, (void *)&ctx);

    /* If the last block that contains data is not already the last block in
    ** the database file, truncate the database file so that it is. */
    if( rc==LSM_OK && ctx.nBlock!=pDb->pWorker->nBlock ){
      rc = lsmFsTruncateDb(
          pDb->pFS, (i64)ctx.nBlock*lsmFsBlockSize(pDb->pFS)
      );
    }
  }

  lsmFreeSnapshot(pDb->pEnv, pDb->pWorker);
  pDb->pWorker = 0;
  return rc;
}

static void doDbDisconnect(lsm_db *pDb){
  int rc;

  if( pDb->bReadonly ){
    lsmShmLock(pDb, LSM_LOCK_DMS3, LSM_LOCK_UNLOCK, 0);
  }else{
    /* Block for an exclusive lock on DMS1. This lock serializes all calls
    ** to doDbConnect() and doDbDisconnect() across all processes.  */
    rc = lsmShmLock(pDb, LSM_LOCK_DMS1, LSM_LOCK_EXCL, 1);
    if( rc==LSM_OK ){

      /* Try an exclusive lock on DMS2. If successful, this is the last
      ** connection to the database. In this case flush the contents of the
      ** in-memory tree to disk and write a checkpoint.  */
      rc = lsmShmTestLock(pDb, LSM_LOCK_DMS2, 1, LSM_LOCK_EXCL);
      if( rc==LSM_OK ){
        rc = lsmShmTestLock(pDb, LSM_LOCK_CHECKPOINTER, 1, LSM_LOCK_EXCL);
      }
      if( rc==LSM_OK ){
        int bReadonly = 0;        /* True if there exist read-only conns. */

        /* Flush the in-memory tree, if required. If there is data to flush,
        ** this will create a new client snapshot in Database.pClient. The
        ** checkpoint (serialization) of this snapshot may be written to disk
        ** by the following block.  
        **
        ** There is no need to take a WRITER lock here. That there are no 
        ** other locks on DMS2 guarantees that there are no other read-write
        ** connections at this time (and the lock on DMS1 guarantees that
        ** no new ones may appear).
        */
        rc = lsmTreeLoadHeader(pDb, 0);
        if( rc==LSM_OK && (lsmTreeHasOld(pDb) || lsmTreeSize(pDb)>0) ){
          rc = lsmFlushTreeToDisk(pDb);
        }

        /* Now check if there are any read-only connections. If there are,
        ** then do not truncate the db file or unlink the shared-memory 
        ** region.  */
        if( rc==LSM_OK ){
          rc = lsmShmTestLock(pDb, LSM_LOCK_DMS3, 1, LSM_LOCK_EXCL);
          if( rc==LSM_BUSY ){
            bReadonly = 1;
            rc = LSM_OK;
          }
        }

        /* Write a checkpoint to disk. */
        if( rc==LSM_OK ){
          rc = lsmCheckpointWrite(pDb, (bReadonly==0), 0);
        }

        /* If the checkpoint was written successfully, delete the log file
        ** and, if possible, truncate the database file.  */
        if( rc==LSM_OK ){
          int bRotrans = 0;
          Database *p = pDb->pDatabase;

          /* The log file may only be deleted if there are no clients 
          ** read-only clients running rotrans transactions.  */
          rc = lsmDetectRoTrans(pDb, &bRotrans);
          if( rc==LSM_OK && bRotrans==0 ){
            lsmFsCloseAndDeleteLog(pDb->pFS);
          }

          /* The database may only be truncated if there exist no read-only
          ** clients - either connected or running rotrans transactions. */
          if( bReadonly==0 && bRotrans==0 ){
            dbTruncateFile(pDb);
            if( p->pFile && p->bMultiProc ){
              lsmEnvShmUnmap(pDb->pEnv, p->pFile, 1);
            }
          }
        }
      }
    }

    if( pDb->iRwclient>=0 ){
      lsmShmLock(pDb, LSM_LOCK_RWCLIENT(pDb->iRwclient), LSM_LOCK_UNLOCK, 0);
      pDb->iRwclient = -1;
    }

    lsmShmLock(pDb, LSM_LOCK_DMS2, LSM_LOCK_UNLOCK, 0);
    lsmShmLock(pDb, LSM_LOCK_DMS1, LSM_LOCK_UNLOCK, 0);
  }
  pDb->pShmhdr = 0;
}

static int doDbConnect(lsm_db *pDb){
  const int nUsMax = 100000;      /* Max value for nUs */
  int nUs = 1000;                 /* us to wait between DMS1 attempts */
  int rc;

  /* Obtain a pointer to the shared-memory header */
  assert( pDb->pShmhdr==0 );
  assert( pDb->bReadonly==0 );
  rc = lsmShmCacheChunks(pDb, 1);
  if( rc!=LSM_OK ) return rc;
  pDb->pShmhdr = (ShmHeader *)pDb->apShm[0];

  /* Block for an exclusive lock on DMS1. This lock serializes all calls
  ** to doDbConnect() and doDbDisconnect() across all processes.  */
  while( 1 ){
    rc = lsmShmLock(pDb, LSM_LOCK_DMS1, LSM_LOCK_EXCL, 1);
    if( rc!=LSM_BUSY ) break;
    lsmEnvSleep(pDb->pEnv, nUs);
    nUs = nUs * 2;
    if( nUs>nUsMax ) nUs = nUsMax;
  }
  if( rc!=LSM_OK ){
    pDb->pShmhdr = 0;
    return rc;
  }

  /* Try an exclusive lock on DMS2/DMS3. If successful, this is the first 
  ** and only connection to the database. In this case initialize the 
  ** shared-memory and run log file recovery.  */
  assert( LSM_LOCK_DMS3==1+LSM_LOCK_DMS2 );
  rc = lsmShmTestLock(pDb, LSM_LOCK_DMS2, 2, LSM_LOCK_EXCL);
  if( rc==LSM_OK ){
    memset(pDb->pShmhdr, 0, sizeof(ShmHeader));
    rc = lsmCheckpointRecover(pDb);
    if( rc==LSM_OK ){
      rc = lsmLogRecover(pDb);
    }
    if( rc==LSM_OK ){
      ShmHeader *pShm = pDb->pShmhdr;
      pShm->aReader[0].iLsmId = lsmCheckpointId(pShm->aSnap1, 0);
      pShm->aReader[0].iTreeId = pDb->treehdr.iUsedShmid;
    }
  }else if( rc==LSM_BUSY ){
    rc = LSM_OK;
  }

  /* Take a shared lock on DMS2. In multi-process mode this lock "cannot" 
  ** fail, as connections may only hold an exclusive lock on DMS2 if they 
  ** first hold an exclusive lock on DMS1. And this connection is currently 
  ** holding the exclusive lock on DSM1. 
  **
  ** However, if some other connection has the database open in single-process
  ** mode, this operation will fail. In this case, return the error to the
  ** caller - the attempt to connect to the db has failed.
  */
  if( rc==LSM_OK ){
    rc = lsmShmLock(pDb, LSM_LOCK_DMS2, LSM_LOCK_SHARED, 0);
  }

  /* If anything went wrong, unlock DMS2. Otherwise, try to take an exclusive
  ** lock on one of the LSM_LOCK_RWCLIENT() locks. Unlock DMS1 in any case. */
  if( rc!=LSM_OK ){
    pDb->pShmhdr = 0;
  }else{
    int i;
    for(i=0; i<LSM_LOCK_NRWCLIENT; i++){
      int rc2 = lsmShmLock(pDb, LSM_LOCK_RWCLIENT(i), LSM_LOCK_EXCL, 0);
      if( rc2==LSM_OK ) pDb->iRwclient = i;
      if( rc2!=LSM_BUSY ){
        rc = rc2;
        break;
      }
    }
  }
  lsmShmLock(pDb, LSM_LOCK_DMS1, LSM_LOCK_UNLOCK, 0);

  return rc;
}

static int dbOpenSharedFd(lsm_env *pEnv, Database *p, int bRoOk){
  int rc;

  rc = lsmEnvOpen(pEnv, p->zName, 0, &p->pFile);
  if( rc==LSM_IOERR && bRoOk ){
    rc = lsmEnvOpen(pEnv, p->zName, LSM_OPEN_READONLY, &p->pFile);
    p->bReadonly = 1;
  }

  return rc;
}

/*
** Return a reference to the shared Database handle for the database 
** identified by canonical path zName. If this is the first connection to
** the named database, a new Database object is allocated. Otherwise, a
** pointer to an existing object is returned.
**
** If successful, *ppDatabase is set to point to the shared Database 
** structure and LSM_OK returned. Otherwise, *ppDatabase is set to NULL
** and and LSM error code returned.
**
** Each successful call to this function should be (eventually) matched
** by a call to lsmDbDatabaseRelease().
*/
int lsmDbDatabaseConnect(
  lsm_db *pDb,                    /* Database handle */
  const char *zName               /* Full-path to db file */
){
  lsm_env *pEnv = pDb->pEnv;
  int rc;                         /* Return code */
  Database *p = 0;                /* Pointer returned via *ppDatabase */
  int nName = lsmStrlen(zName);

  assert( pDb->pDatabase==0 );
  rc = enterGlobalMutex(pEnv);
  if( rc==LSM_OK ){

    /* Search the global list for an existing object. TODO: Need something
    ** better than the memcmp() below to figure out if a given Database
    ** object represents the requested file.  */
    for(p=gShared.pDatabase; p; p=p->pDbNext){
      if( nName==p->nName && 0==memcmp(zName, p->zName, nName) ) break;
    }

    /* If no suitable Database object was found, allocate a new one. */
    if( p==0 ){
      p = (Database *)lsmMallocZeroRc(pEnv, sizeof(Database)+nName+1, &rc);

      /* If the allocation was successful, fill in other fields and
      ** allocate the client mutex. */ 
      if( rc==LSM_OK ){
        p->bMultiProc = pDb->bMultiProc;
        p->zName = (char *)&p[1];
        p->nName = nName;
        memcpy((void *)p->zName, zName, nName+1);
        rc = lsmMutexNew(pEnv, &p->pClientMutex);
      }

      /* If nothing has gone wrong so far, open the shared fd. And if that
      ** succeeds and this connection requested single-process mode, 
      ** attempt to take the exclusive lock on DMS2.  */
      if( rc==LSM_OK ){
        int bReadonly = (pDb->bReadonly && pDb->bMultiProc);
        rc = dbOpenSharedFd(pDb->pEnv, p, bReadonly);
      }

      if( rc==LSM_OK && p->bMultiProc==0 ){
        assert( p->bReadonly==0 );
        rc = lsmEnvLock(pDb->pEnv, p->pFile, LSM_LOCK_DMS2, LSM_LOCK_EXCL);
      }

      if( rc==LSM_OK ){
        p->pDbNext = gShared.pDatabase;
        gShared.pDatabase = p;
      }else{
        freeDatabase(pEnv, p);
        p = 0;
      }
    }

    if( p ){
      p->nDbRef++;
    }
    leaveGlobalMutex(pEnv);

    if( p ){
      lsmMutexEnter(pDb->pEnv, p->pClientMutex);
      pDb->pNext = p->pConn;
      p->pConn = pDb;
      lsmMutexLeave(pDb->pEnv, p->pClientMutex);
    }
  }

  pDb->pDatabase = p;
  if( rc==LSM_OK ){
    assert( p );
    rc = lsmFsOpen(pDb, zName, p->bReadonly);
  }

  /* If the db handle is read-write, then connect to the system now. Run
  ** recovery as necessary. Or, if this is a read-only database handle,
  ** defer attempting to connect to the system until a read-transaction
  ** is opened.  */
  if( pDb->bReadonly==0 ){
    if( rc==LSM_OK ){
      rc = lsmFsConfigure(pDb);
    }
    if( rc==LSM_OK ){
      rc = doDbConnect(pDb);
    }
  }

  return rc;
}

static void dbDeferClose(lsm_db *pDb){
  if( pDb->pFS ){
    LsmFile *pLsmFile;
    Database *p = pDb->pDatabase;
    pLsmFile = lsmFsDeferClose(pDb->pFS);
    pLsmFile->pNext = p->pLsmFile;
    p->pLsmFile = pLsmFile;
  }
}

LsmFile *lsmDbRecycleFd(lsm_db *db){
  LsmFile *pRet;
  Database *p = db->pDatabase;
  lsmMutexEnter(db->pEnv, p->pClientMutex);
  if( (pRet = p->pLsmFile)!=0 ){
    p->pLsmFile = pRet->pNext;
  }
  lsmMutexLeave(db->pEnv, p->pClientMutex);
  return pRet;
}

/*
** Release a reference to a Database object obtained from 
** lsmDbDatabaseConnect(). There should be exactly one call to this function 
** for each successful call to Find().
*/
void lsmDbDatabaseRelease(lsm_db *pDb){
  Database *p = pDb->pDatabase;
  if( p ){
    lsm_db **ppDb;

    if( pDb->pShmhdr ){
      doDbDisconnect(pDb);
    }

    lsmMutexEnter(pDb->pEnv, p->pClientMutex);
    for(ppDb=&p->pConn; *ppDb!=pDb; ppDb=&((*ppDb)->pNext));
    *ppDb = pDb->pNext;
    dbDeferClose(pDb);
    lsmMutexLeave(pDb->pEnv, p->pClientMutex);

    enterGlobalMutex(pDb->pEnv);
    p->nDbRef--;
    if( p->nDbRef==0 ){
      LsmFile *pIter;
      LsmFile *pNext;
      Database **pp;

      /* Remove the Database structure from the linked list. */
      for(pp=&gShared.pDatabase; *pp!=p; pp=&((*pp)->pDbNext));
      *pp = p->pDbNext;

      /* If they were allocated from the heap, free the shared memory chunks */
      if( p->bMultiProc==0 ){
        int i;
        for(i=0; i<p->nShmChunk; i++){
          lsmFree(pDb->pEnv, p->apShmChunk[i]);
        }
      }

      /* Close any outstanding file descriptors */
      for(pIter=p->pLsmFile; pIter; pIter=pNext){
        pNext = pIter->pNext;
        lsmEnvClose(pDb->pEnv, pIter->pFile);
        lsmFree(pDb->pEnv, pIter);
      }
      freeDatabase(pDb->pEnv, p);
    }
    leaveGlobalMutex(pDb->pEnv);
  }
}

Level *lsmDbSnapshotLevel(Snapshot *pSnapshot){
  return pSnapshot->pLevel;
}

void lsmDbSnapshotSetLevel(Snapshot *pSnap, Level *pLevel){
  pSnap->pLevel = pLevel;
}

/* TODO: Shuffle things around to get rid of this */
static int firstSnapshotInUse(lsm_db *, i64 *);

/* 
** Context object used by the lsmWalkFreelist() utility. 
*/
typedef struct WalkFreelistCtx WalkFreelistCtx;
struct WalkFreelistCtx {
  lsm_db *pDb;
  int bReverse;
  Freelist *pFreelist;
  int iFree;
  int (*xUsr)(void *, int, i64);  /* User callback function */
  void *pUsrctx;                  /* User callback context */
  int bDone;                      /* Set to true after xUsr() returns true */
};

/* 
** Callback used by lsmWalkFreelist().
*/
static int walkFreelistCb(void *pCtx, int iBlk, i64 iSnapshot){
  WalkFreelistCtx *p = (WalkFreelistCtx *)pCtx;
  const int iDir = (p->bReverse ? -1 : 1);
  Freelist *pFree = p->pFreelist;

  assert( p->bDone==0 );
  if( pFree ){
    while( (p->iFree < pFree->nEntry) && p->iFree>=0 ){
      FreelistEntry *pEntry = &pFree->aEntry[p->iFree];
      if( (p->bReverse==0 && pEntry->iBlk>iBlk)
       || (p->bReverse!=0 && pEntry->iBlk<iBlk)
      ){
        break;
      }else{
        p->iFree += iDir;
        if( pEntry->iId>=0 
            && p->xUsr(p->pUsrctx, pEntry->iBlk, pEntry->iId) 
          ){
          p->bDone = 1;
          return 1;
        }
        if( pEntry->iBlk==iBlk ) return 0;
      }
    }
  }

  if( p->xUsr(p->pUsrctx, iBlk, iSnapshot) ){
    p->bDone = 1;
    return 1;
  }
  return 0;
}

/*
** The database handle passed as the first argument must be the worker
** connection. This function iterates through the contents of the current
** free block list, invoking the supplied callback once for each list
** element.
**
** The difference between this function and lsmSortedWalkFreelist() is
** that lsmSortedWalkFreelist() only considers those free-list elements
** stored within the LSM. This function also merges in any in-memory 
** elements.
*/
int lsmWalkFreelist(
  lsm_db *pDb,                    /* Database handle (must be worker) */
  int bReverse,                   /* True to iterate from largest to smallest */
  int (*x)(void *, int, i64),     /* Callback function */
  void *pCtx                      /* First argument to pass to callback */
){
  const int iDir = (bReverse ? -1 : 1);
  int rc;
  int iCtx;

  WalkFreelistCtx ctx[2];

  ctx[0].pDb = pDb;
  ctx[0].bReverse = bReverse;
  ctx[0].pFreelist = &pDb->pWorker->freelist;
  if( ctx[0].pFreelist && bReverse ){
    ctx[0].iFree = ctx[0].pFreelist->nEntry-1;
  }else{
    ctx[0].iFree = 0;
  }
  ctx[0].xUsr = walkFreelistCb;
  ctx[0].pUsrctx = (void *)&ctx[1];
  ctx[0].bDone = 0;

  ctx[1].pDb = pDb;
  ctx[1].bReverse = bReverse;
  ctx[1].pFreelist = pDb->pFreelist;
  if( ctx[1].pFreelist && bReverse ){
    ctx[1].iFree = ctx[1].pFreelist->nEntry-1;
  }else{
    ctx[1].iFree = 0;
  }
  ctx[1].xUsr = x;
  ctx[1].pUsrctx = pCtx;
  ctx[1].bDone = 0;

  rc = lsmSortedWalkFreelist(pDb, bReverse, walkFreelistCb, (void *)&ctx[0]);

  if( ctx[0].bDone==0 ){
    for(iCtx=0; iCtx<2; iCtx++){
      int i;
      WalkFreelistCtx *p = &ctx[iCtx];
      for(i=p->iFree; 
          p->pFreelist && rc==LSM_OK && i<p->pFreelist->nEntry && i>=0;
          i += iDir
         ){
        FreelistEntry *pEntry = &p->pFreelist->aEntry[i];
        if( pEntry->iId>=0 && p->xUsr(p->pUsrctx, pEntry->iBlk, pEntry->iId) ){
          return LSM_OK;
        }
      }
    }
  }

  return rc;
}


typedef struct FindFreeblockCtx FindFreeblockCtx;
struct FindFreeblockCtx {
  i64 iInUse;
  int iRet;
  int bNotOne;
};

static int findFreeblockCb(void *pCtx, int iBlk, i64 iSnapshot){
  FindFreeblockCtx *p = (FindFreeblockCtx *)pCtx;
  if( iSnapshot<p->iInUse && (iBlk!=1 || p->bNotOne==0) ){
    p->iRet = iBlk;
    return 1;
  }
  return 0;
}

static int findFreeblock(lsm_db *pDb, i64 iInUse, int bNotOne, int *piRet){
  int rc;                         /* Return code */
  FindFreeblockCtx ctx;           /* Context object */

  ctx.iInUse = iInUse;
  ctx.iRet = 0;
  ctx.bNotOne = bNotOne;
  rc = lsmWalkFreelist(pDb, 0, findFreeblockCb, (void *)&ctx);
  *piRet = ctx.iRet;

  return rc;
}

/*
** Allocate a new database file block to write data to, either by extending
** the database file or by recycling a free-list entry. The worker snapshot 
** must be held in order to call this function.
**
** If successful, *piBlk is set to the block number allocated and LSM_OK is
** returned. Otherwise, *piBlk is zeroed and an lsm error code returned.
*/
int lsmBlockAllocate(lsm_db *pDb, int iBefore, int *piBlk){
  Snapshot *p = pDb->pWorker;
  int iRet = 0;                   /* Block number of allocated block */
  int rc = LSM_OK;
  i64 iInUse = 0;                 /* Snapshot id still in use */
  i64 iSynced = 0;                /* Snapshot id synced to disk */

  assert( p );

#ifdef LSM_LOG_FREELIST
  {
    static int nCall = 0;
    char *zFree = 0;
    nCall++;
    rc = lsmInfoFreelist(pDb, &zFree);
    if( rc!=LSM_OK ) return rc;
    lsmLogMessage(pDb, 0, "lsmBlockAllocate(): %d freelist: %s", nCall, zFree);
    lsmFree(pDb->pEnv, zFree);
  }
#endif

  /* Set iInUse to the smallest snapshot id that is either:
  **
  **   * Currently in use by a database client,
  **   * May be used by a database client in the future, or
  **   * Is the most recently checkpointed snapshot (i.e. the one that will
  **     be used following recovery if a failure occurs at this point).
  */
  rc = lsmCheckpointSynced(pDb, &iSynced, 0, 0);
  if( rc==LSM_OK && iSynced==0 ) iSynced = p->iId;
  iInUse = iSynced;
  if( rc==LSM_OK && pDb->iReader>=0 ){
    assert( pDb->pClient );
    iInUse = LSM_MIN(iInUse, pDb->pClient->iId);
  }
  if( rc==LSM_OK ) rc = firstSnapshotInUse(pDb, &iInUse);

#ifdef LSM_LOG_FREELIST
  {
    lsmLogMessage(pDb, 0, "lsmBlockAllocate(): "
        "snapshot-in-use: %lld (iSynced=%lld) (client-id=%lld)", 
        iInUse, iSynced, (pDb->iReader>=0 ? pDb->pClient->iId : 0)
    );
  }
#endif


  /* Unless there exists a read-only transaction (which prevents us from
  ** recycling any blocks regardless, query the free block list for a 
  ** suitable block to reuse. 
  **
  ** It might seem more natural to check for a read-only transaction at
  ** the start of this function. However, it is better do wait until after
  ** the call to lsmCheckpointSynced() to do so.
  */
  if( rc==LSM_OK ){
    int bRotrans;
    rc = lsmDetectRoTrans(pDb, &bRotrans);

    if( rc==LSM_OK && bRotrans==0 ){
      rc = findFreeblock(pDb, iInUse, (iBefore>0), &iRet);
    }
  }

  if( iBefore>0 && (iRet<=0 || iRet>=iBefore) ){
    iRet = 0;

  }else if( rc==LSM_OK ){
    /* If a block was found in the free block list, use it and remove it from 
    ** the list. Otherwise, if no suitable block was found, allocate one from
    ** the end of the file.  */
    if( iRet>0 ){
#ifdef LSM_LOG_FREELIST
      lsmLogMessage(pDb, 0, 
          "reusing block %d (snapshot-in-use=%lld)", iRet, iInUse);
#endif
      rc = freelistAppend(pDb, iRet, -1);
      if( rc==LSM_OK ){
        rc = dbTruncate(pDb, iInUse);
      }
    }else{
      iRet = ++(p->nBlock);
#ifdef LSM_LOG_FREELIST
      lsmLogMessage(pDb, 0, "extending file to %d blocks", iRet);
#endif
    }
  }

  assert( iBefore>0 || iRet>0 || rc!=LSM_OK );
  *piBlk = iRet;
  return rc;
}

/*
** Free a database block. The worker snapshot must be held in order to call 
** this function.
**
** If successful, LSM_OK is returned. Otherwise, an lsm error code (e.g. 
** LSM_NOMEM).
*/
int lsmBlockFree(lsm_db *pDb, int iBlk){
  Snapshot *p = pDb->pWorker;
  assert( lsmShmAssertWorker(pDb) );

#ifdef LSM_LOG_FREELIST
  lsmLogMessage(pDb, LSM_OK, "lsmBlockFree(): Free block %d", iBlk);
#endif

  return freelistAppend(pDb, iBlk, p->iId);
}

/*
** Refree a database block. The worker snapshot must be held in order to call 
** this function.
**
** Refreeing is required when a block is allocated using lsmBlockAllocate()
** but then not used. This function is used to push the block back onto
** the freelist. Refreeing a block is different from freeing is, as a refreed
** block may be reused immediately. Whereas a freed block can not be reused 
** until (at least) after the next checkpoint.
*/
int lsmBlockRefree(lsm_db *pDb, int iBlk){
  int rc = LSM_OK;                /* Return code */

#ifdef LSM_LOG_FREELIST
  lsmLogMessage(pDb, LSM_OK, "lsmBlockRefree(): Refree block %d", iBlk);
#endif

  rc = freelistAppend(pDb, iBlk, 0);
  return rc;
}

/*
** If required, copy a database checkpoint from shared memory into the
** database itself.
**
** The WORKER lock must not be held when this is called. This is because
** this function may indirectly call fsync(). And the WORKER lock should
** not be held that long (in case it is required by a client flushing an
** in-memory tree to disk).
*/
int lsmCheckpointWrite(lsm_db *pDb, int bTruncate, u32 *pnWrite){
  int rc;                         /* Return Code */
  u32 nWrite = 0;

  assert( pDb->pWorker==0 );
  assert( 1 || pDb->pClient==0 );
  assert( lsmShmAssertLock(pDb, LSM_LOCK_WORKER, LSM_LOCK_UNLOCK) );

  rc = lsmShmLock(pDb, LSM_LOCK_CHECKPOINTER, LSM_LOCK_EXCL, 0);
  if( rc!=LSM_OK ) return rc;

  rc = lsmCheckpointLoad(pDb, 0);
  if( rc==LSM_OK ){
    int nBlock = lsmCheckpointNBlock(pDb->aSnapshot);
    ShmHeader *pShm = pDb->pShmhdr;
    int bDone = 0;                /* True if checkpoint is already stored */

    /* Check if this checkpoint has already been written to the database
    ** file. If so, set variable bDone to true.  */
    if( pShm->iMetaPage ){
      MetaPage *pPg;              /* Meta page */
      u8 *aData;                  /* Meta-page data buffer */
      int nData;                  /* Size of aData[] in bytes */
      i64 iCkpt;                  /* Id of checkpoint just loaded */
      i64 iDisk;                  /* Id of checkpoint already stored in db */
      iCkpt = lsmCheckpointId(pDb->aSnapshot, 0);
      rc = lsmFsMetaPageGet(pDb->pFS, 0, pShm->iMetaPage, &pPg);
      if( rc==LSM_OK ){
        aData = lsmFsMetaPageData(pPg, &nData);
        iDisk = lsmCheckpointId((u32 *)aData, 1);
        nWrite = lsmCheckpointNWrite((u32 *)aData, 1);
        lsmFsMetaPageRelease(pPg);
      }
      bDone = (iDisk>=iCkpt);
    }

    if( rc==LSM_OK && bDone==0 ){
      int iMeta = (pShm->iMetaPage % 2) + 1;
      if( pDb->eSafety!=LSM_SAFETY_OFF ){
        rc = lsmFsSyncDb(pDb->pFS, nBlock);
      }
      if( rc==LSM_OK ) rc = lsmCheckpointStore(pDb, iMeta);
      if( rc==LSM_OK && pDb->eSafety!=LSM_SAFETY_OFF){
        rc = lsmFsSyncDb(pDb->pFS, 0);
      }
      if( rc==LSM_OK ){
        pShm->iMetaPage = iMeta;
        nWrite = lsmCheckpointNWrite(pDb->aSnapshot, 0) - nWrite;
      }
#ifdef LSM_LOG_WORK
      lsmLogMessage(pDb, 0, "finish checkpoint %d", 
          (int)lsmCheckpointId(pDb->aSnapshot, 0)
      );
#endif
    }

    if( rc==LSM_OK && bTruncate && nBlock>0 ){
      rc = lsmFsTruncateDb(pDb->pFS, (i64)nBlock*lsmFsBlockSize(pDb->pFS));
    }
  }

  lsmShmLock(pDb, LSM_LOCK_CHECKPOINTER, LSM_LOCK_UNLOCK, 0);
  if( pnWrite && rc==LSM_OK ) *pnWrite = nWrite;
  return rc;
}

int lsmBeginWork(lsm_db *pDb){
  int rc;

  /* Attempt to take the WORKER lock */
  rc = lsmShmLock(pDb, LSM_LOCK_WORKER, LSM_LOCK_EXCL, 0);

  /* Deserialize the current worker snapshot */
  if( rc==LSM_OK ){
    rc = lsmCheckpointLoadWorker(pDb);
  }
  return rc;
}

void lsmFreeSnapshot(lsm_env *pEnv, Snapshot *p){
  if( p ){
    lsmSortedFreeLevel(pEnv, p->pLevel);
    lsmFree(pEnv, p->freelist.aEntry);
    lsmFree(pEnv, p->redirect.a);
    lsmFree(pEnv, p);
  }
}

/*
** Attempt to populate one of the read-lock slots to contain lock values
** iLsm/iShm. Or, if such a slot exists already, this function is a no-op.
**
** It is not an error if no slot can be populated because the write-lock
** cannot be obtained. If any other error occurs, return an LSM error code.
** Otherwise, LSM_OK.
**
** This function is called at various points to try to ensure that there
** always exists at least one read-lock slot that can be used by a read-only
** client. And so that, in the usual case, there is an "exact match" available
** whenever a read transaction is opened by any client. At present this
** function is called when:
**
**    * A write transaction that called lsmTreeDiscardOld() is committed, and
**    * Whenever the working snapshot is updated (i.e. lsmFinishWork()).
*/
static int dbSetReadLock(lsm_db *db, i64 iLsm, u32 iShm){
  int rc = LSM_OK;
  ShmHeader *pShm = db->pShmhdr;
  int i;

  /* Check if there is already a slot containing the required values. */
  for(i=0; i<LSM_LOCK_NREADER; i++){
    ShmReader *p = &pShm->aReader[i];
    if( p->iLsmId==iLsm && p->iTreeId==iShm ) return LSM_OK;
  }

  /* Iterate through all read-lock slots, attempting to take a write-lock
  ** on each of them. If a write-lock succeeds, populate the locked slot
  ** with the required values and break out of the loop.  */
  for(i=0; rc==LSM_OK && i<LSM_LOCK_NREADER; i++){
    rc = lsmShmLock(db, LSM_LOCK_READER(i), LSM_LOCK_EXCL, 0);
    if( rc==LSM_BUSY ){
      rc = LSM_OK;
    }else{
      ShmReader *p = &pShm->aReader[i];
      p->iLsmId = iLsm;
      p->iTreeId = iShm;
      lsmShmLock(db, LSM_LOCK_READER(i), LSM_LOCK_UNLOCK, 0);
      break;
    }
  }

  return rc;
}

/*
** Release the read-lock currently held by connection db.
*/
int dbReleaseReadlock(lsm_db *db){
  int rc = LSM_OK;
  if( db->iReader>=0 ){
    rc = lsmShmLock(db, LSM_LOCK_READER(db->iReader), LSM_LOCK_UNLOCK, 0);
    db->iReader = -1;
  }
  db->bRoTrans = 0;
  return rc;
}


/*
** Argument bFlush is true if the contents of the in-memory tree has just
** been flushed to disk. The significance of this is that once the snapshot
** created to hold the updated state of the database is synced to disk, log
** file space can be recycled.
*/
void lsmFinishWork(lsm_db *pDb, int bFlush, int *pRc){
  int rc = *pRc;
  assert( rc!=0 || pDb->pWorker );
  if( pDb->pWorker ){
    /* If no error has occurred, serialize the worker snapshot and write
    ** it to shared memory.  */
    if( rc==LSM_OK ){
      rc = lsmSaveWorker(pDb, bFlush);
    }

    /* Assuming no error has occurred, update a read lock slot with the
    ** new snapshot id (see comments above function dbSetReadLock()).  */
    if( rc==LSM_OK ){
      if( pDb->iReader<0 ){
        rc = lsmTreeLoadHeader(pDb, 0);
      }
      if( rc==LSM_OK ){
        rc = dbSetReadLock(pDb, pDb->pWorker->iId, pDb->treehdr.iUsedShmid);
      }
    }

    /* Free the snapshot object. */
    lsmFreeSnapshot(pDb->pEnv, pDb->pWorker);
    pDb->pWorker = 0;
  }

  lsmShmLock(pDb, LSM_LOCK_WORKER, LSM_LOCK_UNLOCK, 0);
  *pRc = rc;
}

/*
** Called when recovery is finished.
*/
int lsmFinishRecovery(lsm_db *pDb){
  lsmTreeEndTransaction(pDb, 1);
  return LSM_OK;
}

/*
** Check if the currently configured compression functions
** (LSM_CONFIG_SET_COMPRESSION) are compatible with a database that has its
** compression id set to iReq. Compression routines are compatible if iReq
** is zero (indicating the database is empty), or if it is equal to the 
** compression id of the configured compression routines.
**
** If the check shows that the current compression are incompatible and there
** is a compression factory registered, give it a chance to install new
** compression routines.
**
** If, after any registered factory is invoked, the compression functions
** are still incompatible, return LSM_MISMATCH. Otherwise, LSM_OK.
*/
int lsmCheckCompressionId(lsm_db *pDb, u32 iReq){
  if( iReq!=LSM_COMPRESSION_EMPTY && pDb->compress.iId!=iReq ){
    if( pDb->factory.xFactory ){
      pDb->bInFactory = 1;
      pDb->factory.xFactory(pDb->factory.pCtx, pDb, iReq);
      pDb->bInFactory = 0;
    }
    if( pDb->compress.iId!=iReq ){
      /* Incompatible */
      return LSM_MISMATCH;
    }
  }
  /* Compatible */
  return LSM_OK;
}

/*
** Begin a read transaction. This function is a no-op if the connection
** passed as the only argument already has an open read transaction.
*/
int lsmBeginReadTrans(lsm_db *pDb){
  const int MAX_READLOCK_ATTEMPTS = 10;
  const int nMaxAttempt = (pDb->bRoTrans ? 1 : MAX_READLOCK_ATTEMPTS);

  int rc = LSM_OK;                /* Return code */
  int iAttempt = 0;

  assert( pDb->pWorker==0 );

  while( rc==LSM_OK && pDb->iReader<0 && (iAttempt++)<nMaxAttempt ){
    int iTreehdr = 0;
    int iSnap = 0;
    assert( pDb->pCsr==0 && pDb->nTransOpen==0 );

    /* Load the in-memory tree header. */
    rc = lsmTreeLoadHeader(pDb, &iTreehdr);

    /* Load the database snapshot */
    if( rc==LSM_OK ){
      if( lsmCheckpointClientCacheOk(pDb)==0 ){
        lsmFreeSnapshot(pDb->pEnv, pDb->pClient);
        pDb->pClient = 0;
        lsmMCursorFreeCache(pDb);
        lsmFsPurgeCache(pDb->pFS);
        rc = lsmCheckpointLoad(pDb, &iSnap);
      }else{
        iSnap = 1;
      }
    }

    /* Take a read-lock on the tree and snapshot just loaded. Then check
    ** that the shared-memory still contains the same values. If so, proceed.
    ** Otherwise, relinquish the read-lock and retry the whole procedure
    ** (starting with loading the in-memory tree header).  */
    if( rc==LSM_OK ){
      u32 iShmMax = pDb->treehdr.iUsedShmid;
      u32 iShmMin = pDb->treehdr.iNextShmid+1-LSM_MAX_SHMCHUNKS;
      rc = lsmReadlock(
          pDb, lsmCheckpointId(pDb->aSnapshot, 0), iShmMin, iShmMax
      );
      if( rc==LSM_OK ){
        if( lsmTreeLoadHeaderOk(pDb, iTreehdr)
         && lsmCheckpointLoadOk(pDb, iSnap)
        ){
          /* Read lock has been successfully obtained. Deserialize the 
          ** checkpoint just loaded. TODO: This will be removed after 
          ** lsm_sorted.c is changed to work directly from the serialized
          ** version of the snapshot.  */
          if( pDb->pClient==0 ){
            rc = lsmCheckpointDeserialize(pDb, 0, pDb->aSnapshot,&pDb->pClient);
          }
          assert( (rc==LSM_OK)==(pDb->pClient!=0) );
          assert( pDb->iReader>=0 );

          /* Check that the client has the right compression hooks loaded.
          ** If not, set rc to LSM_MISMATCH.  */
          if( rc==LSM_OK ){
            rc = lsmCheckCompressionId(pDb, pDb->pClient->iCmpId);
          }
        }else{
          rc = dbReleaseReadlock(pDb);
        }
      }

      if( rc==LSM_BUSY ){
        rc = LSM_OK;
      }
    }
#if 0
if( rc==LSM_OK && pDb->pClient ){
  fprintf(stderr, 
      "reading %p: snapshot:%d used-shmid:%d trans-id:%d iOldShmid=%d\n",
      (void *)pDb,
      (int)pDb->pClient->iId, (int)pDb->treehdr.iUsedShmid, 
      (int)pDb->treehdr.root.iTransId,
      (int)pDb->treehdr.iOldShmid
  );
}
#endif
  }

  if( rc==LSM_OK ){
    rc = lsmShmCacheChunks(pDb, pDb->treehdr.nChunk);
  }
  if( rc!=LSM_OK ){
    dbReleaseReadlock(pDb);
  }
  if( pDb->pClient==0 && rc==LSM_OK ) rc = LSM_BUSY;
  return rc;
}

/*
** This function is used by a read-write connection to determine if there
** are currently one or more read-only transactions open on the database
** (in this context a read-only transaction is one opened by a read-only
** connection on a non-live database).
**
** If no error occurs, LSM_OK is returned and *pbExists is set to true if
** some other connection has a read-only transaction open, or false 
** otherwise. If an error occurs an LSM error code is returned and the final
** value of *pbExist is undefined.
*/
int lsmDetectRoTrans(lsm_db *db, int *pbExist){
  int rc;

  /* Only a read-write connection may use this function. */
  assert( db->bReadonly==0 );

  rc = lsmShmTestLock(db, LSM_LOCK_ROTRANS, 1, LSM_LOCK_EXCL);
  if( rc==LSM_BUSY ){
    *pbExist = 1;
    rc = LSM_OK;
  }else{
    *pbExist = 0;
  }

  return rc;
}

/*
** db is a read-only database handle in the disconnected state. This function
** attempts to open a read-transaction on the database. This may involve
** connecting to the database system (opening shared memory etc.).
*/
int lsmBeginRoTrans(lsm_db *db){
  int rc = LSM_OK;

  assert( db->bReadonly && db->pShmhdr==0 );
  assert( db->iReader<0 );

  if( db->bRoTrans==0 ){

    /* Attempt a shared-lock on DMS1. */
    rc = lsmShmLock(db, LSM_LOCK_DMS1, LSM_LOCK_SHARED, 0);
    if( rc!=LSM_OK ) return rc;

    rc = lsmShmTestLock(
        db, LSM_LOCK_RWCLIENT(0), LSM_LOCK_NREADER, LSM_LOCK_SHARED
    );
    if( rc==LSM_OK ){
      /* System is not live. Take a SHARED lock on the ROTRANS byte and
      ** release DMS1. Locking ROTRANS tells all read-write clients that they
      ** may not recycle any disk space from within the database or log files,
      ** as a read-only client may be using it.  */
      rc = lsmShmLock(db, LSM_LOCK_ROTRANS, LSM_LOCK_SHARED, 0);
      lsmShmLock(db, LSM_LOCK_DMS1, LSM_LOCK_UNLOCK, 0);

      if( rc==LSM_OK ){
        db->bRoTrans = 1;
        rc = lsmShmCacheChunks(db, 1);
        if( rc==LSM_OK ){
          db->pShmhdr = (ShmHeader *)db->apShm[0];
          memset(db->pShmhdr, 0, sizeof(ShmHeader));
          rc = lsmCheckpointRecover(db);
          if( rc==LSM_OK ){
            rc = lsmLogRecover(db);
          }
        }
      }
    }else if( rc==LSM_BUSY ){
      /* System is live! */
      rc = lsmShmLock(db, LSM_LOCK_DMS3, LSM_LOCK_SHARED, 0);
      lsmShmLock(db, LSM_LOCK_DMS1, LSM_LOCK_UNLOCK, 0);
      if( rc==LSM_OK ){
        rc = lsmShmCacheChunks(db, 1);
        if( rc==LSM_OK ){
          db->pShmhdr = (ShmHeader *)db->apShm[0];
        }
      }
    }

    if( rc==LSM_OK ){
      rc = lsmBeginReadTrans(db);
    }
  }

  return rc;
}

/*
** Close the currently open read transaction.
*/
void lsmFinishReadTrans(lsm_db *pDb){

  /* Worker connections should not be closing read transactions. And
  ** read transactions should only be closed after all cursors and write
  ** transactions have been closed. Finally pClient should be non-NULL
  ** only iff pDb->iReader>=0.  */
  assert( pDb->pWorker==0 );
  assert( pDb->pCsr==0 && pDb->nTransOpen==0 );

  if( pDb->bRoTrans ){
    int i;
    for(i=0; i<pDb->nShm; i++){
      lsmFree(pDb->pEnv, pDb->apShm[i]);
    }
    lsmFree(pDb->pEnv, pDb->apShm);
    pDb->apShm = 0;
    pDb->nShm = 0;
    pDb->pShmhdr = 0;

    lsmShmLock(pDb, LSM_LOCK_ROTRANS, LSM_LOCK_UNLOCK, 0);
  }
  dbReleaseReadlock(pDb);
}

/*
** Open a write transaction.
*/
int lsmBeginWriteTrans(lsm_db *pDb){
  int rc = LSM_OK;                /* Return code */
  ShmHeader *pShm = pDb->pShmhdr; /* Shared memory header */

  assert( pDb->nTransOpen==0 );
  assert( pDb->bDiscardOld==0 );
  assert( pDb->bReadonly==0 );

  /* If there is no read-transaction open, open one now. */
  if( pDb->iReader<0 ){
    rc = lsmBeginReadTrans(pDb);
  }

  /* Attempt to take the WRITER lock */
  if( rc==LSM_OK ){
    rc = lsmShmLock(pDb, LSM_LOCK_WRITER, LSM_LOCK_EXCL, 0);
  }

  /* If the previous writer failed mid-transaction, run emergency rollback. */
  if( rc==LSM_OK && pShm->bWriter ){
    rc = lsmTreeRepair(pDb);
    if( rc==LSM_OK ) pShm->bWriter = 0;
  }

  /* Check that this connection is currently reading from the most recent
  ** version of the database. If not, return LSM_BUSY.  */
  if( rc==LSM_OK && memcmp(&pShm->hdr1, &pDb->treehdr, sizeof(TreeHeader)) ){
    rc = LSM_BUSY;
  }

  if( rc==LSM_OK ){
    rc = lsmLogBegin(pDb);
  }

  /* If everything was successful, set the "transaction-in-progress" flag
  ** and return LSM_OK. Otherwise, if some error occurred, relinquish the 
  ** WRITER lock and return an error code.  */
  if( rc==LSM_OK ){
    TreeHeader *p = &pDb->treehdr;
    pShm->bWriter = 1;
    p->root.iTransId++;
    if( lsmTreeHasOld(pDb) && p->iOldLog==pDb->pClient->iLogOff ){
      lsmTreeDiscardOld(pDb);
      pDb->bDiscardOld = 1;
    }
  }else{
    lsmShmLock(pDb, LSM_LOCK_WRITER, LSM_LOCK_UNLOCK, 0);
    if( pDb->pCsr==0 ) lsmFinishReadTrans(pDb);
  }
  return rc;
}

/*
** End the current write transaction. The connection is left with an open
** read transaction. It is an error to call this if there is no open write 
** transaction.
**
** If the transaction was committed, then a commit record has already been
** written into the log file when this function is called. Or, if the
** transaction was rolled back, both the log file and in-memory tree 
** structure have already been restored. In either case, this function 
** merely releases locks and other resources held by the write-transaction.
**
** LSM_OK is returned if successful, or an LSM error code otherwise.
*/
int lsmFinishWriteTrans(lsm_db *pDb, int bCommit){
  int rc = LSM_OK;
  int bFlush = 0;

  lsmLogEnd(pDb, bCommit);
  if( rc==LSM_OK && bCommit && lsmTreeSize(pDb)>pDb->nTreeLimit ){
    bFlush = 1;
    lsmTreeMakeOld(pDb);
  }
  lsmTreeEndTransaction(pDb, bCommit);

  if( rc==LSM_OK ){
    if( bFlush && pDb->bAutowork ){
      rc = lsmSortedAutoWork(pDb, 1);
    }else if( bCommit && pDb->bDiscardOld ){
      rc = dbSetReadLock(pDb, pDb->pClient->iId, pDb->treehdr.iUsedShmid);
    }
  }
  pDb->bDiscardOld = 0;
  lsmShmLock(pDb, LSM_LOCK_WRITER, LSM_LOCK_UNLOCK, 0);

  if( bFlush && pDb->bAutowork==0 && pDb->xWork ){
    pDb->xWork(pDb, pDb->pWorkCtx);
  }
  return rc;
}


/*
** Return non-zero if the caller is holding the client mutex.
*/
#ifdef LSM_DEBUG
int lsmHoldingClientMutex(lsm_db *pDb){
  return lsmMutexHeld(pDb->pEnv, pDb->pDatabase->pClientMutex);
}
#endif

static int slotIsUsable(ShmReader *p, i64 iLsm, u32 iShmMin, u32 iShmMax){
  return( 
      p->iLsmId && p->iLsmId<=iLsm 
      && shm_sequence_ge(iShmMax, p->iTreeId)
      && shm_sequence_ge(p->iTreeId, iShmMin)
  );
}

/*
** Obtain a read-lock on database version identified by the combination
** of snapshot iLsm and tree iTree. Return LSM_OK if successful, or
** an LSM error code otherwise.
*/
int lsmReadlock(lsm_db *db, i64 iLsm, u32 iShmMin, u32 iShmMax){
  int rc = LSM_OK;
  ShmHeader *pShm = db->pShmhdr;
  int i;

  assert( db->iReader<0 );
  assert( shm_sequence_ge(iShmMax, iShmMin) );

  /* This is a no-op if the read-only transaction flag is set. */
  if( db->bRoTrans ){
    db->iReader = 0;
    return LSM_OK;
  }

  /* Search for an exact match. */
  for(i=0; db->iReader<0 && rc==LSM_OK && i<LSM_LOCK_NREADER; i++){
    ShmReader *p = &pShm->aReader[i];
    if( p->iLsmId==iLsm && p->iTreeId==iShmMax ){
      rc = lsmShmLock(db, LSM_LOCK_READER(i), LSM_LOCK_SHARED, 0);
      if( rc==LSM_OK && p->iLsmId==iLsm && p->iTreeId==iShmMax ){
        db->iReader = i;
      }else if( rc==LSM_BUSY ){
        rc = LSM_OK;
      }
    }
  }

  /* Try to obtain a write-lock on each slot, in order. If successful, set
  ** the slot values to iLsm/iTree.  */
  for(i=0; db->iReader<0 && rc==LSM_OK && i<LSM_LOCK_NREADER; i++){
    rc = lsmShmLock(db, LSM_LOCK_READER(i), LSM_LOCK_EXCL, 0);
    if( rc==LSM_BUSY ){
      rc = LSM_OK;
    }else{
      ShmReader *p = &pShm->aReader[i];
      p->iLsmId = iLsm;
      p->iTreeId = iShmMax;
      rc = lsmShmLock(db, LSM_LOCK_READER(i), LSM_LOCK_SHARED, 0);
      assert( rc!=LSM_BUSY );
      if( rc==LSM_OK ) db->iReader = i;
    }
  }

  /* Search for any usable slot */
  for(i=0; db->iReader<0 && rc==LSM_OK && i<LSM_LOCK_NREADER; i++){
    ShmReader *p = &pShm->aReader[i];
    if( slotIsUsable(p, iLsm, iShmMin, iShmMax) ){
      rc = lsmShmLock(db, LSM_LOCK_READER(i), LSM_LOCK_SHARED, 0);
      if( rc==LSM_OK && slotIsUsable(p, iLsm, iShmMin, iShmMax) ){
        db->iReader = i;
      }else if( rc==LSM_BUSY ){
        rc = LSM_OK;
      }
    }
  }

  if( rc==LSM_OK && db->iReader<0 ){
    rc = LSM_BUSY;
  }
  return rc;
}

/*
** This is used to check if there exists a read-lock locking a particular
** version of either the in-memory tree or database file. 
**
** If iLsmId is non-zero, then it is a snapshot id. If there exists a 
** read-lock using this snapshot or newer, set *pbInUse to true. Or,
** if there is no such read-lock, set it to false.
**
** Or, if iLsmId is zero, then iShmid is a shared-memory sequence id.
** Search for a read-lock using this sequence id or newer. etc.
*/
static int isInUse(lsm_db *db, i64 iLsmId, u32 iShmid, int *pbInUse){
  ShmHeader *pShm = db->pShmhdr;
  int i;
  int rc = LSM_OK;

  for(i=0; rc==LSM_OK && i<LSM_LOCK_NREADER; i++){
    ShmReader *p = &pShm->aReader[i];
    if( p->iLsmId ){
      if( (iLsmId!=0 && p->iLsmId!=0 && iLsmId>=p->iLsmId) 
       || (iLsmId==0 && shm_sequence_ge(p->iTreeId, iShmid))
      ){
        rc = lsmShmLock(db, LSM_LOCK_READER(i), LSM_LOCK_EXCL, 0);
        if( rc==LSM_OK ){
          p->iLsmId = 0;
          lsmShmLock(db, LSM_LOCK_READER(i), LSM_LOCK_UNLOCK, 0);
        }
      }
    }
  }

  if( rc==LSM_BUSY ){
    *pbInUse = 1;
    return LSM_OK;
  }
  *pbInUse = 0;
  return rc;
}

/*
** This function is called by worker connections to determine the smallest
** snapshot id that is currently in use by a database client. The worker
** connection uses this result to determine whether or not it is safe to
** recycle a database block.
*/
static int firstSnapshotInUse(
  lsm_db *db,                     /* Database handle */
  i64 *piInUse                    /* IN/OUT: Smallest snapshot id in use */
){
  ShmHeader *pShm = db->pShmhdr;
  i64 iInUse = *piInUse;
  int i;

  assert( iInUse>0 );
  for(i=0; i<LSM_LOCK_NREADER; i++){
    ShmReader *p = &pShm->aReader[i];
    if( p->iLsmId ){
      i64 iThis = p->iLsmId;
      if( iThis!=0 && iInUse>iThis ){
        int rc = lsmShmLock(db, LSM_LOCK_READER(i), LSM_LOCK_EXCL, 0);
        if( rc==LSM_OK ){
          p->iLsmId = 0;
          lsmShmLock(db, LSM_LOCK_READER(i), LSM_LOCK_UNLOCK, 0);
        }else if( rc==LSM_BUSY ){
          iInUse = iThis;
        }else{
          /* Some error other than LSM_BUSY. Return the error code to
          ** the caller in this case.  */
          return rc;
        }
      }
    }
  }

  *piInUse = iInUse;
  return LSM_OK;
}

int lsmTreeInUse(lsm_db *db, u32 iShmid, int *pbInUse){
  if( db->treehdr.iUsedShmid==iShmid ){
    *pbInUse = 1;
    return LSM_OK;
  }
  return isInUse(db, 0, iShmid, pbInUse);
}

int lsmLsmInUse(lsm_db *db, i64 iLsmId, int *pbInUse){
  if( db->pClient && db->pClient->iId<=iLsmId ){
    *pbInUse = 1;
    return LSM_OK;
  }
  return isInUse(db, iLsmId, 0, pbInUse);
}

/*
** This function may only be called after a successful call to
** lsmDbDatabaseConnect(). It returns true if the connection is in
** multi-process mode, or false otherwise.
*/
int lsmDbMultiProc(lsm_db *pDb){
  return pDb->pDatabase && pDb->pDatabase->bMultiProc;
}


/*************************************************************************
**************************************************************************
**************************************************************************
**************************************************************************
**************************************************************************
*************************************************************************/

/*
** Ensure that database connection db has cached pointers to at least the 
** first nChunk chunks of shared memory.
*/
int lsmShmCacheChunks(lsm_db *db, int nChunk){
  int rc = LSM_OK;
  if( nChunk>db->nShm ){
    static const int NINCR = 16;
    Database *p = db->pDatabase;
    lsm_env *pEnv = db->pEnv;
    int nAlloc;
    int i;

    /* Ensure that the db->apShm[] array is large enough. If an attempt to
    ** allocate memory fails, return LSM_NOMEM immediately. The apShm[] array
    ** is always extended in multiples of 16 entries - so the actual allocated
    ** size can be inferred from nShm.  */ 
    nAlloc = ((db->nShm + NINCR - 1) / NINCR) * NINCR;
    while( nChunk>=nAlloc ){
      void **apShm;
      nAlloc += NINCR;
      apShm = lsmRealloc(pEnv, db->apShm, sizeof(void*)*nAlloc);
      if( !apShm ) return LSM_NOMEM_BKPT;
      db->apShm = apShm;
    }

    if( db->bRoTrans ){
      for(i=db->nShm; rc==LSM_OK && i<nChunk; i++){
        db->apShm[i] = lsmMallocZeroRc(pEnv, LSM_SHM_CHUNK_SIZE, &rc);
        db->nShm++;
      }

    }else{

      /* Enter the client mutex */
      lsmMutexEnter(pEnv, p->pClientMutex);

      /* Extend the Database objects apShmChunk[] array if necessary. Using the
       ** same pattern as for the lsm_db.apShm[] array above.  */
      nAlloc = ((p->nShmChunk + NINCR - 1) / NINCR) * NINCR;
      while( nChunk>=nAlloc ){
        void **apShm;
        nAlloc +=  NINCR;
        apShm = lsmRealloc(pEnv, p->apShmChunk, sizeof(void*)*nAlloc);
        if( !apShm ){
          rc = LSM_NOMEM_BKPT;
          break;
        }
        p->apShmChunk = apShm;
      }

      for(i=db->nShm; rc==LSM_OK && i<nChunk; i++){
        if( i>=p->nShmChunk ){
          void *pChunk = 0;
          if( p->bMultiProc==0 ){
            /* Single process mode */
            pChunk = lsmMallocZeroRc(pEnv, LSM_SHM_CHUNK_SIZE, &rc);
          }else{
            /* Multi-process mode */
            rc = lsmEnvShmMap(pEnv, p->pFile, i, LSM_SHM_CHUNK_SIZE, &pChunk);
          }
          if( rc==LSM_OK ){
            p->apShmChunk[i] = pChunk;
            p->nShmChunk++;
          }
        }
        if( rc==LSM_OK ){
          db->apShm[i] = p->apShmChunk[i];
          db->nShm++;
        }
      }

      /* Release the client mutex */
      lsmMutexLeave(pEnv, p->pClientMutex);
    }
  }

  return rc;
}

static int lockSharedFile(lsm_env *pEnv, Database *p, int iLock, int eOp){
  int rc = LSM_OK;
  if( p->bMultiProc ){
    rc = lsmEnvLock(pEnv, p->pFile, iLock, eOp);
  }
  return rc;
}

/*
** Test if it would be possible for connection db to obtain a lock of type
** eType on the nLock locks starting at iLock. If so, return LSM_OK. If it
** would not be possible to obtain the lock due to a lock held by another
** connection, return LSM_BUSY. If an IO or other error occurs (i.e. in the 
** lsm_env.xTestLock function), return some other LSM error code.
**
** Note that this function never actually locks the database - it merely
** queries the system to see if there exists a lock that would prevent
** it from doing so.
*/
int lsmShmTestLock(
  lsm_db *db,
  int iLock,
  int nLock,
  int eOp
){
  int rc = LSM_OK;
  lsm_db *pIter;
  Database *p = db->pDatabase;
  int i;
  u64 mask = 0;

  for(i=iLock; i<(iLock+nLock); i++){
    mask |= ((u64)1 << (iLock-1));
    if( eOp==LSM_LOCK_EXCL ) mask |= ((u64)1 << (iLock+32-1));
  }

  lsmMutexEnter(db->pEnv, p->pClientMutex);
  for(pIter=p->pConn; pIter; pIter=pIter->pNext){
    if( pIter!=db && (pIter->mLock & mask) ) break;
  }

  if( pIter ){
    rc = LSM_BUSY;
  }else if( p->bMultiProc ){
    rc = lsmEnvTestLock(db->pEnv, p->pFile, iLock, nLock, eOp);
  }

  lsmMutexLeave(db->pEnv, p->pClientMutex);
  return rc;
}

/*
** Attempt to obtain the lock identified by the iLock and bExcl parameters.
** If successful, return LSM_OK. If the lock cannot be obtained because 
** there exists some other conflicting lock, return LSM_BUSY. If some other
** error occurs, return an LSM error code.
**
** Parameter iLock must be one of LSM_LOCK_WRITER, WORKER or CHECKPOINTER,
** or else a value returned by the LSM_LOCK_READER macro.
*/
int lsmShmLock(
  lsm_db *db, 
  int iLock,
  int eOp,                        /* One of LSM_LOCK_UNLOCK, SHARED or EXCL */
  int bBlock                      /* True for a blocking lock */
){
  lsm_db *pIter;
  const u64 me = ((u64)1 << (iLock-1));
  const u64 ms = ((u64)1 << (iLock+32-1));
  int rc = LSM_OK;
  Database *p = db->pDatabase;

  assert( eOp!=LSM_LOCK_EXCL || p->bReadonly==0 );
  assert( iLock>=1 && iLock<=LSM_LOCK_RWCLIENT(LSM_LOCK_NRWCLIENT-1) );
  assert( LSM_LOCK_RWCLIENT(LSM_LOCK_NRWCLIENT-1)<=32 );
  assert( eOp==LSM_LOCK_UNLOCK || eOp==LSM_LOCK_SHARED || eOp==LSM_LOCK_EXCL );

  /* Check for a no-op. Proceed only if this is not one of those. */
  if( (eOp==LSM_LOCK_UNLOCK && (db->mLock & (me|ms))!=0)
   || (eOp==LSM_LOCK_SHARED && (db->mLock & (me|ms))!=ms)
   || (eOp==LSM_LOCK_EXCL   && (db->mLock & me)==0)
  ){
    int nExcl = 0;                /* Number of connections holding EXCLUSIVE */
    int nShared = 0;              /* Number of connections holding SHARED */
    lsmMutexEnter(db->pEnv, p->pClientMutex);

    /* Figure out the locks currently held by this process on iLock, not
    ** including any held by connection db.  */
    for(pIter=p->pConn; pIter; pIter=pIter->pNext){
      assert( (pIter->mLock & me)==0 || (pIter->mLock & ms)!=0 );
      if( pIter!=db ){
        if( pIter->mLock & me ){
          nExcl++;
        }else if( pIter->mLock & ms ){
          nShared++;
        }
      }
    }
    assert( nExcl==0 || nExcl==1 );
    assert( nExcl==0 || nShared==0 );
    assert( nExcl==0 || (db->mLock & (me|ms))==0 );

    switch( eOp ){
      case LSM_LOCK_UNLOCK:
        if( nShared==0 ){
          lockSharedFile(db->pEnv, p, iLock, LSM_LOCK_UNLOCK);
        }
        db->mLock &= ~(me|ms);
        break;

      case LSM_LOCK_SHARED:
        if( nExcl ){
          rc = LSM_BUSY;
        }else{
          if( nShared==0 ){
            rc = lockSharedFile(db->pEnv, p, iLock, LSM_LOCK_SHARED);
          }
          if( rc==LSM_OK ){
            db->mLock |= ms;
            db->mLock &= ~me;
          }
        }
        break;

      default:
        assert( eOp==LSM_LOCK_EXCL );
        if( nExcl || nShared ){
          rc = LSM_BUSY;
        }else{
          rc = lockSharedFile(db->pEnv, p, iLock, LSM_LOCK_EXCL);
          if( rc==LSM_OK ){
            db->mLock |= (me|ms);
          }
        }
        break;
    }

    lsmMutexLeave(db->pEnv, p->pClientMutex);
  }

  return rc;
}

#ifdef LSM_DEBUG

int shmLockType(lsm_db *db, int iLock){
  const u64 me = ((u64)1 << (iLock-1));
  const u64 ms = ((u64)1 << (iLock+32-1));

  if( db->mLock & me ) return LSM_LOCK_EXCL;
  if( db->mLock & ms ) return LSM_LOCK_SHARED;
  return LSM_LOCK_UNLOCK;
}

/*
** The arguments passed to this function are similar to those passed to
** the lsmShmLock() function. However, instead of obtaining a new lock 
** this function returns true if the specified connection already holds 
** (or does not hold) such a lock, depending on the value of eOp. As
** follows:
**
**   (eOp==LSM_LOCK_UNLOCK) -> true if db has no lock on iLock
**   (eOp==LSM_LOCK_SHARED) -> true if db has at least a SHARED lock on iLock.
**   (eOp==LSM_LOCK_EXCL)   -> true if db has an EXCLUSIVE lock on iLock.
*/
int lsmShmAssertLock(lsm_db *db, int iLock, int eOp){
  int ret;
  int eHave;

  assert( iLock>=1 && iLock<=LSM_LOCK_READER(LSM_LOCK_NREADER-1) );
  assert( iLock<=16 );
  assert( eOp==LSM_LOCK_UNLOCK || eOp==LSM_LOCK_SHARED || eOp==LSM_LOCK_EXCL );

  eHave = shmLockType(db, iLock);

  switch( eOp ){
    case LSM_LOCK_UNLOCK:
      ret = (eHave==LSM_LOCK_UNLOCK);
      break;
    case LSM_LOCK_SHARED:
      ret = (eHave!=LSM_LOCK_UNLOCK);
      break;
    case LSM_LOCK_EXCL:
      ret = (eHave==LSM_LOCK_EXCL);
      break;
    default:
      assert( !"bad eOp value passed to lsmShmAssertLock()" );
      break;
  }

  return ret;
}

int lsmShmAssertWorker(lsm_db *db){
  return lsmShmAssertLock(db, LSM_LOCK_WORKER, LSM_LOCK_EXCL) && db->pWorker;
}

/*
** This function does not contribute to library functionality, and is not
** included in release builds. It is intended to be called from within
** an interactive debugger.
**
** When called, this function prints a single line of human readable output
** to stdout describing the locks currently held by the connection. For 
** example:
**
**     (gdb) call print_db_locks(pDb)
**     (shared on dms2) (exclusive on writer) 
*/
void print_db_locks(lsm_db *db){
  int iLock;
  for(iLock=0; iLock<16; iLock++){
    int bOne = 0;
    const char *azLock[] = {0, "shared", "exclusive"};
    const char *azName[] = {
      0, "dms1", "dms2", "writer", "worker", "checkpointer",
      "reader0", "reader1", "reader2", "reader3", "reader4", "reader5"
    };
    int eHave = shmLockType(db, iLock);
    if( azLock[eHave] ){
      printf("%s(%s on %s)", (bOne?" ":""), azLock[eHave], azName[iLock]);
      bOne = 1;
    }
  }
  printf("\n");
}
void print_all_db_locks(lsm_db *db){
  lsm_db *p;
  for(p=db->pDatabase->pConn; p; p=p->pNext){
    printf("%s connection %p ", ((p==db)?"*":""), p);
    print_db_locks(p);
  }
}
#endif

void lsmShmBarrier(lsm_db *db){
  lsmEnvShmBarrier(db->pEnv);
}

int lsm_checkpoint(lsm_db *pDb, int *pnKB){
  int rc;                         /* Return code */
  u32 nWrite = 0;                 /* Number of pages checkpointed */

  /* Attempt the checkpoint. If successful, nWrite is set to the number of
  ** pages written between this and the previous checkpoint.  */
  rc = lsmCheckpointWrite(pDb, 0, &nWrite);

  /* If required, calculate the output variable (KB of data checkpointed). 
  ** Set it to zero if an error occured.  */
  if( pnKB ){
    int nKB = 0;
    if( rc==LSM_OK && nWrite ){
      nKB = (((i64)nWrite * lsmFsPageSize(pDb->pFS)) + 1023) / 1024;
    }
    *pnKB = nKB;
  }

  return rc;
}

Added ext/lsm1/lsm_sorted.c.











































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
1388
1389
1390
1391
1392
1393
1394
1395
1396
1397
1398
1399
1400
1401
1402
1403
1404
1405
1406
1407
1408
1409
1410
1411
1412
1413
1414
1415
1416
1417
1418
1419
1420
1421
1422
1423
1424
1425
1426
1427
1428
1429
1430
1431
1432
1433
1434
1435
1436
1437
1438
1439
1440
1441
1442
1443
1444
1445
1446
1447
1448
1449
1450
1451
1452
1453
1454
1455
1456
1457
1458
1459
1460
1461
1462
1463
1464
1465
1466
1467
1468
1469
1470
1471
1472
1473
1474
1475
1476
1477
1478
1479
1480
1481
1482
1483
1484
1485
1486
1487
1488
1489
1490
1491
1492
1493
1494
1495
1496
1497
1498
1499
1500
1501
1502
1503
1504
1505
1506
1507
1508
1509
1510
1511
1512
1513
1514
1515
1516
1517
1518
1519
1520
1521
1522
1523
1524
1525
1526
1527
1528
1529
1530
1531
1532
1533
1534
1535
1536
1537
1538
1539
1540
1541
1542
1543
1544
1545
1546
1547
1548
1549
1550
1551
1552
1553
1554
1555
1556
1557
1558
1559
1560
1561
1562
1563
1564
1565
1566
1567
1568
1569
1570
1571
1572
1573
1574
1575
1576
1577
1578
1579
1580
1581
1582
1583
1584
1585
1586
1587
1588
1589
1590
1591
1592
1593
1594
1595
1596
1597
1598
1599
1600
1601
1602
1603
1604
1605
1606
1607
1608
1609
1610
1611
1612
1613
1614
1615
1616
1617
1618
1619
1620
1621
1622
1623
1624
1625
1626
1627
1628
1629
1630
1631
1632
1633
1634
1635
1636
1637
1638
1639
1640
1641
1642
1643
1644
1645
1646
1647
1648
1649
1650
1651
1652
1653
1654
1655
1656
1657
1658
1659
1660
1661
1662
1663
1664
1665
1666
1667
1668
1669
1670
1671
1672
1673
1674
1675
1676
1677
1678
1679
1680
1681
1682
1683
1684
1685
1686
1687
1688
1689
1690
1691
1692
1693
1694
1695
1696
1697
1698
1699
1700
1701
1702
1703
1704
1705
1706
1707
1708
1709
1710
1711
1712
1713
1714
1715
1716
1717
1718
1719
1720
1721
1722
1723
1724
1725
1726
1727
1728
1729
1730
1731
1732
1733
1734
1735
1736
1737
1738
1739
1740
1741
1742
1743
1744
1745
1746
1747
1748
1749
1750
1751
1752
1753
1754
1755
1756
1757
1758
1759
1760
1761
1762
1763
1764
1765
1766
1767
1768
1769
1770
1771
1772
1773
1774
1775
1776
1777
1778
1779
1780
1781
1782
1783
1784
1785
1786
1787
1788
1789
1790
1791
1792
1793
1794
1795
1796
1797
1798
1799
1800
1801
1802
1803
1804
1805
1806
1807
1808
1809
1810
1811
1812
1813
1814
1815
1816
1817
1818
1819
1820
1821
1822
1823
1824
1825
1826
1827
1828
1829
1830
1831
1832
1833
1834
1835
1836
1837
1838
1839
1840
1841
1842
1843
1844
1845
1846
1847
1848
1849
1850
1851
1852
1853
1854
1855
1856
1857
1858
1859
1860
1861
1862
1863
1864
1865
1866
1867
1868
1869
1870
1871
1872
1873
1874
1875
1876
1877
1878
1879
1880
1881
1882
1883
1884
1885
1886
1887
1888
1889
1890
1891
1892
1893
1894
1895
1896
1897
1898
1899
1900
1901
1902
1903
1904
1905
1906
1907
1908
1909
1910
1911
1912
1913
1914
1915
1916
1917
1918
1919
1920
1921
1922
1923
1924
1925
1926
1927
1928
1929
1930
1931
1932
1933
1934
1935
1936
1937
1938
1939
1940
1941
1942
1943
1944
1945
1946
1947
1948
1949
1950
1951
1952
1953
1954
1955
1956
1957
1958
1959
1960
1961
1962
1963
1964
1965
1966
1967
1968
1969
1970
1971
1972
1973
1974
1975
1976
1977
1978
1979
1980
1981
1982
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
2018
2019
2020
2021
2022
2023
2024
2025
2026
2027
2028
2029
2030
2031
2032
2033
2034
2035
2036
2037
2038
2039
2040
2041
2042
2043
2044
2045
2046
2047
2048
2049
2050
2051
2052
2053
2054
2055
2056
2057
2058
2059
2060
2061
2062
2063
2064
2065
2066
2067
2068
2069
2070
2071
2072
2073
2074
2075
2076
2077
2078
2079
2080
2081
2082
2083
2084
2085
2086
2087
2088
2089
2090
2091
2092
2093
2094
2095
2096
2097
2098
2099
2100
2101
2102
2103
2104
2105
2106
2107
2108
2109
2110
2111
2112
2113
2114
2115
2116
2117
2118
2119
2120
2121
2122
2123
2124
2125
2126
2127
2128
2129
2130
2131
2132
2133
2134
2135
2136
2137
2138
2139
2140
2141
2142
2143
2144
2145
2146
2147
2148
2149
2150
2151
2152
2153
2154
2155
2156
2157
2158
2159
2160
2161
2162
2163
2164
2165
2166
2167
2168
2169
2170
2171
2172
2173
2174
2175
2176
2177
2178
2179
2180
2181
2182
2183
2184
2185
2186
2187
2188
2189
2190
2191
2192
2193
2194
2195
2196
2197
2198
2199
2200
2201
2202
2203
2204
2205
2206
2207
2208
2209
2210
2211
2212
2213
2214
2215
2216
2217
2218
2219
2220
2221
2222
2223
2224
2225
2226
2227
2228
2229
2230
2231
2232
2233
2234
2235
2236
2237
2238
2239
2240
2241
2242
2243
2244
2245
2246
2247
2248
2249
2250
2251
2252
2253
2254
2255
2256
2257
2258
2259
2260
2261
2262
2263
2264
2265
2266
2267
2268
2269
2270
2271
2272
2273
2274
2275
2276
2277
2278
2279
2280
2281
2282
2283
2284
2285
2286
2287
2288
2289
2290
2291
2292
2293
2294
2295
2296
2297
2298
2299
2300
2301
2302
2303
2304
2305
2306
2307
2308
2309
2310
2311
2312
2313
2314
2315
2316
2317
2318
2319
2320
2321
2322
2323
2324
2325
2326
2327
2328
2329
2330
2331
2332
2333
2334
2335
2336
2337
2338
2339
2340
2341
2342
2343
2344
2345
2346
2347
2348
2349
2350
2351
2352
2353
2354
2355
2356
2357
2358
2359
2360
2361
2362
2363
2364
2365
2366
2367
2368
2369
2370
2371
2372
2373
2374
2375
2376
2377
2378
2379
2380
2381
2382
2383
2384
2385
2386
2387
2388
2389
2390
2391
2392
2393
2394
2395
2396
2397
2398
2399
2400
2401
2402
2403
2404
2405
2406
2407
2408
2409
2410
2411
2412
2413
2414
2415
2416
2417
2418
2419
2420
2421
2422
2423
2424
2425
2426
2427
2428
2429
2430
2431
2432
2433
2434
2435
2436
2437
2438
2439
2440
2441
2442
2443
2444
2445
2446
2447
2448
2449
2450
2451
2452
2453
2454
2455
2456
2457
2458
2459
2460
2461
2462
2463
2464
2465
2466
2467
2468
2469
2470
2471
2472
2473
2474
2475
2476
2477
2478
2479
2480
2481
2482
2483
2484
2485
2486
2487
2488
2489
2490
2491
2492
2493
2494
2495
2496
2497
2498
2499
2500
2501
2502
2503
2504
2505
2506
2507
2508
2509
2510
2511
2512
2513
2514
2515
2516
2517
2518
2519
2520
2521
2522
2523
2524
2525
2526
2527
2528
2529
2530
2531
2532
2533
2534
2535
2536
2537
2538
2539
2540
2541
2542
2543
2544
2545
2546
2547
2548
2549
2550
2551
2552
2553
2554
2555
2556
2557
2558
2559
2560
2561
2562
2563
2564
2565
2566
2567
2568
2569
2570
2571
2572
2573
2574
2575
2576
2577
2578
2579
2580
2581
2582
2583
2584
2585
2586
2587
2588
2589
2590
2591
2592
2593
2594
2595
2596
2597
2598
2599
2600
2601
2602
2603
2604
2605
2606
2607
2608
2609
2610
2611
2612
2613
2614
2615
2616
2617
2618
2619
2620
2621
2622
2623
2624
2625
2626
2627
2628
2629
2630
2631
2632
2633
2634
2635
2636
2637
2638
2639
2640
2641
2642
2643
2644
2645
2646
2647
2648
2649
2650
2651
2652
2653
2654
2655
2656
2657
2658
2659
2660
2661
2662
2663
2664
2665
2666
2667
2668
2669
2670
2671
2672
2673
2674
2675
2676
2677
2678
2679
2680
2681
2682
2683
2684
2685
2686
2687
2688
2689
2690
2691
2692
2693
2694
2695
2696
2697
2698
2699
2700
2701
2702
2703
2704
2705
2706
2707
2708
2709
2710
2711
2712
2713
2714
2715
2716
2717
2718
2719
2720
2721
2722
2723
2724
2725
2726
2727
2728
2729
2730
2731
2732
2733
2734
2735
2736
2737
2738
2739
2740
2741
2742
2743
2744
2745
2746
2747
2748
2749
2750
2751
2752
2753
2754
2755
2756
2757
2758
2759
2760
2761
2762
2763
2764
2765
2766
2767
2768
2769
2770
2771
2772
2773
2774
2775
2776
2777
2778
2779
2780
2781
2782
2783
2784
2785
2786
2787
2788
2789
2790
2791
2792
2793
2794
2795
2796
2797
2798
2799
2800
2801
2802
2803
2804
2805
2806
2807
2808
2809
2810
2811
2812
2813
2814
2815
2816
2817
2818
2819
2820
2821
2822
2823
2824
2825
2826
2827
2828
2829
2830
2831
2832
2833
2834
2835
2836
2837
2838
2839
2840
2841
2842
2843
2844
2845
2846
2847
2848
2849
2850
2851
2852
2853
2854
2855
2856
2857
2858
2859
2860
2861
2862
2863
2864
2865
2866
2867
2868
2869
2870
2871
2872
2873
2874
2875
2876
2877
2878
2879
2880
2881
2882
2883
2884
2885
2886
2887
2888
2889
2890
2891
2892
2893
2894
2895
2896
2897
2898
2899
2900
2901
2902
2903
2904
2905
2906
2907
2908
2909
2910
2911
2912
2913
2914
2915
2916
2917
2918
2919
2920
2921
2922
2923
2924
2925
2926
2927
2928
2929
2930
2931
2932
2933
2934
2935
2936
2937
2938
2939
2940
2941
2942
2943
2944
2945
2946
2947
2948
2949
2950
2951
2952
2953
2954
2955
2956
2957
2958
2959
2960
2961
2962
2963
2964
2965
2966
2967
2968
2969
2970
2971
2972
2973
2974
2975
2976
2977
2978
2979
2980
2981
2982
2983
2984
2985
2986
2987
2988
2989
2990
2991
2992
2993
2994
2995
2996
2997
2998
2999
3000
3001
3002
3003
3004
3005
3006
3007
3008
3009
3010
3011
3012
3013
3014
3015
3016
3017
3018
3019
3020
3021
3022
3023
3024
3025
3026
3027
3028
3029
3030
3031
3032
3033
3034
3035
3036
3037
3038
3039
3040
3041
3042
3043
3044
3045
3046
3047
3048
3049
3050
3051
3052
3053
3054
3055
3056
3057
3058
3059
3060
3061
3062
3063
3064
3065
3066
3067
3068
3069
3070
3071
3072
3073
3074
3075
3076
3077
3078
3079
3080
3081
3082
3083
3084
3085
3086
3087
3088
3089
3090
3091
3092
3093
3094
3095
3096
3097
3098
3099
3100
3101
3102
3103
3104
3105
3106
3107
3108
3109
3110
3111
3112
3113
3114
3115
3116
3117
3118
3119
3120
3121
3122
3123
3124
3125
3126
3127
3128
3129
3130
3131
3132
3133
3134
3135
3136
3137
3138
3139
3140
3141
3142
3143
3144
3145
3146
3147
3148
3149
3150
3151
3152
3153
3154
3155
3156
3157
3158
3159
3160
3161
3162
3163
3164
3165
3166
3167
3168
3169
3170
3171
3172
3173
3174
3175
3176
3177
3178
3179
3180
3181
3182
3183
3184
3185
3186
3187
3188
3189
3190
3191
3192
3193
3194
3195
3196
3197
3198
3199
3200
3201
3202
3203
3204
3205
3206
3207
3208
3209
3210
3211
3212
3213
3214
3215
3216
3217
3218
3219
3220
3221
3222
3223
3224
3225
3226
3227
3228
3229
3230
3231
3232
3233
3234
3235
3236
3237
3238
3239
3240
3241
3242
3243
3244
3245
3246
3247
3248
3249
3250
3251
3252
3253
3254
3255
3256
3257
3258
3259
3260
3261
3262
3263
3264
3265
3266
3267
3268
3269
3270
3271
3272
3273
3274
3275
3276
3277
3278
3279
3280
3281
3282
3283
3284
3285
3286
3287
3288
3289
3290
3291
3292
3293
3294
3295
3296
3297
3298
3299
3300
3301
3302
3303
3304
3305
3306
3307
3308
3309
3310
3311
3312
3313
3314
3315
3316
3317
3318
3319
3320
3321
3322
3323
3324
3325
3326
3327
3328
3329
3330
3331
3332
3333
3334
3335
3336
3337
3338
3339
3340
3341
3342
3343
3344
3345
3346
3347
3348
3349
3350
3351
3352
3353
3354
3355
3356
3357
3358
3359
3360
3361
3362
3363
3364
3365
3366
3367
3368
3369
3370
3371
3372
3373
3374
3375
3376
3377
3378
3379
3380
3381
3382
3383
3384
3385
3386
3387
3388
3389
3390
3391
3392
3393
3394
3395
3396
3397
3398
3399
3400
3401
3402
3403
3404
3405
3406
3407
3408
3409
3410
3411
3412
3413
3414
3415
3416
3417
3418
3419
3420
3421
3422
3423
3424
3425
3426
3427
3428
3429
3430
3431
3432
3433
3434
3435
3436
3437
3438
3439
3440
3441
3442
3443
3444
3445
3446
3447
3448
3449
3450
3451
3452
3453
3454
3455
3456
3457
3458
3459
3460
3461
3462
3463
3464
3465
3466
3467
3468
3469
3470
3471
3472
3473
3474
3475
3476
3477
3478
3479
3480
3481
3482
3483
3484
3485
3486
3487
3488
3489
3490
3491
3492
3493
3494
3495
3496
3497
3498
3499
3500
3501
3502
3503
3504
3505
3506
3507
3508
3509
3510
3511
3512
3513
3514
3515
3516
3517
3518
3519
3520
3521
3522
3523
3524
3525
3526
3527
3528
3529
3530
3531
3532
3533
3534
3535
3536
3537
3538
3539
3540
3541
3542
3543
3544
3545
3546
3547
3548
3549
3550
3551
3552
3553
3554
3555
3556
3557
3558
3559
3560
3561
3562
3563
3564
3565
3566
3567
3568
3569
3570
3571
3572
3573
3574
3575
3576
3577
3578
3579
3580
3581
3582
3583
3584
3585
3586
3587
3588
3589
3590
3591
3592
3593
3594
3595
3596
3597
3598
3599
3600
3601
3602
3603
3604
3605
3606
3607
3608
3609
3610
3611
3612
3613
3614
3615
3616
3617
3618
3619
3620
3621
3622
3623
3624
3625
3626
3627
3628
3629
3630
3631
3632
3633
3634
3635
3636
3637
3638
3639
3640
3641
3642
3643
3644
3645
3646
3647
3648
3649
3650
3651
3652
3653
3654
3655
3656
3657
3658
3659
3660
3661
3662
3663
3664
3665
3666
3667
3668
3669
3670
3671
3672
3673
3674
3675
3676
3677
3678
3679
3680
3681
3682
3683
3684
3685
3686
3687
3688
3689
3690
3691
3692
3693
3694
3695
3696
3697
3698
3699
3700
3701
3702
3703
3704
3705
3706
3707
3708
3709
3710
3711
3712
3713
3714
3715
3716
3717
3718
3719
3720
3721
3722
3723
3724
3725
3726
3727
3728
3729
3730
3731
3732
3733
3734
3735
3736
3737
3738
3739
3740
3741
3742
3743
3744
3745
3746
3747
3748
3749
3750
3751
3752
3753
3754
3755
3756
3757
3758
3759
3760
3761
3762
3763
3764
3765
3766
3767
3768
3769
3770
3771
3772
3773
3774
3775
3776
3777
3778
3779
3780
3781
3782
3783
3784
3785
3786
3787
3788
3789
3790
3791
3792
3793
3794
3795
3796
3797
3798
3799
3800
3801
3802
3803
3804
3805
3806
3807
3808
3809
3810
3811
3812
3813
3814
3815
3816
3817
3818
3819
3820
3821
3822
3823
3824
3825
3826
3827
3828
3829
3830
3831
3832
3833
3834
3835
3836
3837
3838
3839
3840
3841
3842
3843
3844
3845
3846
3847
3848
3849
3850
3851
3852
3853
3854
3855
3856
3857
3858
3859
3860
3861
3862
3863
3864
3865
3866
3867
3868
3869
3870
3871
3872
3873
3874
3875
3876
3877
3878
3879
3880
3881
3882
3883
3884
3885
3886
3887
3888
3889
3890
3891
3892
3893
3894
3895
3896
3897
3898
3899
3900
3901
3902
3903
3904
3905
3906
3907
3908
3909
3910
3911
3912
3913
3914
3915
3916
3917
3918
3919
3920
3921
3922
3923
3924
3925
3926
3927
3928
3929
3930
3931
3932
3933
3934
3935
3936
3937
3938
3939
3940
3941
3942
3943
3944
3945
3946
3947
3948
3949
3950
3951
3952
3953
3954
3955
3956
3957
3958
3959
3960
3961
3962
3963
3964
3965
3966
3967
3968
3969
3970
3971
3972
3973
3974
3975
3976
3977
3978
3979
3980
3981
3982
3983
3984
3985
3986
3987
3988
3989
3990
3991
3992
3993
3994
3995
3996
3997
3998
3999
4000
4001
4002
4003
4004
4005
4006
4007
4008
4009
4010
4011
4012
4013
4014
4015
4016
4017
4018
4019
4020
4021
4022
4023
4024
4025
4026
4027
4028
4029
4030
4031
4032
4033
4034
4035
4036
4037
4038
4039
4040
4041
4042
4043
4044
4045
4046
4047
4048
4049
4050
4051
4052
4053
4054
4055
4056
4057
4058
4059
4060
4061
4062
4063
4064
4065
4066
4067
4068
4069
4070
4071
4072
4073
4074
4075
4076
4077
4078
4079
4080
4081
4082
4083
4084
4085
4086
4087
4088
4089
4090
4091
4092
4093
4094
4095
4096
4097
4098
4099
4100
4101
4102
4103
4104
4105
4106
4107
4108
4109
4110
4111
4112
4113
4114
4115
4116
4117
4118
4119
4120
4121
4122
4123
4124
4125
4126
4127
4128
4129
4130
4131
4132
4133
4134
4135
4136
4137
4138
4139
4140
4141
4142
4143
4144
4145
4146
4147
4148
4149
4150
4151
4152
4153
4154
4155
4156
4157
4158
4159
4160
4161
4162
4163
4164
4165
4166
4167
4168
4169
4170
4171
4172
4173
4174
4175
4176
4177
4178
4179
4180
4181
4182
4183
4184
4185
4186
4187
4188
4189
4190
4191
4192
4193
4194
4195
4196
4197
4198
4199
4200
4201
4202
4203
4204
4205
4206
4207
4208
4209
4210
4211
4212
4213
4214
4215
4216
4217
4218
4219
4220
4221
4222
4223
4224
4225
4226
4227
4228
4229
4230
4231
4232
4233
4234
4235
4236
4237
4238
4239
4240
4241
4242
4243
4244
4245
4246
4247
4248
4249
4250
4251
4252
4253
4254
4255
4256
4257
4258
4259
4260
4261
4262
4263
4264
4265
4266
4267
4268
4269
4270
4271
4272
4273
4274
4275
4276
4277
4278
4279
4280
4281
4282
4283
4284
4285
4286
4287
4288
4289
4290
4291
4292
4293
4294
4295
4296
4297
4298
4299
4300
4301
4302
4303
4304
4305
4306
4307
4308
4309
4310
4311
4312
4313
4314
4315
4316
4317
4318
4319
4320
4321
4322
4323
4324
4325
4326
4327
4328
4329
4330
4331
4332
4333
4334
4335
4336
4337
4338
4339
4340
4341
4342
4343
4344
4345
4346
4347
4348
4349
4350
4351
4352
4353
4354
4355
4356
4357
4358
4359
4360
4361
4362
4363
4364
4365
4366
4367
4368
4369
4370
4371
4372
4373
4374
4375
4376
4377
4378
4379
4380
4381
4382
4383
4384
4385
4386
4387
4388
4389
4390
4391
4392
4393
4394
4395
4396
4397
4398
4399
4400
4401
4402
4403
4404
4405
4406
4407
4408
4409
4410
4411
4412
4413
4414
4415
4416
4417
4418
4419
4420
4421
4422
4423
4424
4425
4426
4427
4428
4429
4430
4431
4432
4433
4434
4435
4436
4437
4438
4439
4440
4441
4442
4443
4444
4445
4446
4447
4448
4449
4450
4451
4452
4453
4454
4455
4456
4457
4458
4459
4460
4461
4462
4463
4464
4465
4466
4467
4468
4469
4470
4471
4472
4473
4474
4475
4476
4477
4478
4479
4480
4481
4482
4483
4484
4485
4486
4487
4488
4489
4490
4491
4492
4493
4494
4495
4496
4497
4498
4499
4500
4501
4502
4503
4504
4505
4506
4507
4508
4509
4510
4511
4512
4513
4514
4515
4516
4517
4518
4519
4520
4521
4522
4523
4524
4525
4526
4527
4528
4529
4530
4531
4532
4533
4534
4535
4536
4537
4538
4539
4540
4541
4542
4543
4544
4545
4546
4547
4548
4549
4550
4551
4552
4553
4554
4555
4556
4557
4558
4559
4560
4561
4562
4563
4564
4565
4566
4567
4568
4569
4570
4571
4572
4573
4574
4575
4576
4577
4578
4579
4580
4581
4582
4583
4584
4585
4586
4587
4588
4589
4590
4591
4592
4593
4594
4595
4596
4597
4598
4599
4600
4601
4602
4603
4604
4605
4606
4607
4608
4609
4610
4611
4612
4613
4614
4615
4616
4617
4618
4619
4620
4621
4622
4623
4624
4625
4626
4627
4628
4629
4630
4631
4632
4633
4634
4635
4636
4637
4638
4639
4640
4641
4642
4643
4644
4645
4646
4647
4648
4649
4650
4651
4652
4653
4654
4655
4656
4657
4658
4659
4660
4661
4662
4663
4664
4665
4666
4667
4668
4669
4670
4671
4672
4673
4674
4675
4676
4677
4678
4679
4680
4681
4682
4683
4684
4685
4686
4687
4688
4689
4690
4691
4692
4693
4694
4695
4696
4697
4698
4699
4700
4701
4702
4703
4704
4705
4706
4707
4708
4709
4710
4711
4712
4713
4714
4715
4716
4717
4718
4719
4720
4721
4722
4723
4724
4725
4726
4727
4728
4729
4730
4731
4732
4733
4734
4735
4736
4737
4738
4739
4740
4741
4742
4743
4744
4745
4746
4747
4748
4749
4750
4751
4752
4753
4754
4755
4756
4757
4758
4759
4760
4761
4762
4763
4764
4765
4766
4767
4768
4769
4770
4771
4772
4773
4774
4775
4776
4777
4778
4779
4780
4781
4782
4783
4784
4785
4786
4787
4788
4789
4790
4791
4792
4793
4794
4795
4796
4797
4798
4799
4800
4801
4802
4803
4804
4805
4806
4807
4808
4809
4810
4811
4812
4813
4814
4815
4816
4817
4818
4819
4820
4821
4822
4823
4824
4825
4826
4827
4828
4829
4830
4831
4832
4833
4834
4835
4836
4837
4838
4839
4840
4841
4842
4843
4844
4845
4846
4847
4848
4849
4850
4851
4852
4853
4854
4855
4856
4857
4858
4859
4860
4861
4862
4863
4864
4865
4866
4867
4868
4869
4870
4871
4872
4873
4874
4875
4876
4877
4878
4879
4880
4881
4882
4883
4884
4885
4886
4887
4888
4889
4890
4891
4892
4893
4894
4895
4896
4897
4898
4899
4900
4901
4902
4903
4904
4905
4906
4907
4908
4909
4910
4911
4912
4913
4914
4915
4916
4917
4918
4919
4920
4921
4922
4923
4924
4925
4926
4927
4928
4929
4930
4931
4932
4933
4934
4935
4936
4937
4938
4939
4940
4941
4942
4943
4944
4945
4946
4947
4948
4949
4950
4951
4952
4953
4954
4955
4956
4957
4958
4959
4960
4961
4962
4963
4964
4965
4966
4967
4968
4969
4970
4971
4972
4973
4974
4975
4976
4977
4978
4979
4980
4981
4982
4983
4984
4985
4986
4987
4988
4989
4990
4991
4992
4993
4994
4995
4996
4997
4998
4999
5000
5001
5002
5003
5004
5005
5006
5007
5008
5009
5010
5011
5012
5013
5014
5015
5016
5017
5018
5019
5020
5021
5022
5023
5024
5025
5026
5027
5028
5029
5030
5031
5032
5033
5034
5035
5036
5037
5038
5039
5040
5041
5042
5043
5044
5045
5046
5047
5048
5049
5050
5051
5052
5053
5054
5055
5056
5057
5058
5059
5060
5061
5062
5063
5064
5065
5066
5067
5068
5069
5070
5071
5072
5073
5074
5075
5076
5077
5078
5079
5080
5081
5082
5083
5084
5085
5086
5087
5088
5089
5090
5091
5092
5093
5094
5095
5096
5097
5098
5099
5100
5101
5102
5103
5104
5105
5106
5107
5108
5109
5110
5111
5112
5113
5114
5115
5116
5117
5118
5119
5120
5121
5122
5123
5124
5125
5126
5127
5128
5129
5130
5131
5132
5133
5134
5135
5136
5137
5138
5139
5140
5141
5142
5143
5144
5145
5146
5147
5148
5149
5150
5151
5152
5153
5154
5155
5156
5157
5158
5159
5160
5161
5162
5163
5164
5165
5166
5167
5168
5169
5170
5171
5172
5173
5174
5175
5176
5177
5178
5179
5180
5181
5182
5183
5184
5185
5186
5187
5188
5189
5190
5191
5192
5193
5194
5195
5196
5197
5198
5199
5200
5201
5202
5203
5204
5205
5206
5207
5208
5209
5210
5211
5212
5213
5214
5215
5216
5217
5218
5219
5220
5221
5222
5223
5224
5225
5226
5227
5228
5229
5230
5231
5232
5233
5234
5235
5236
5237
5238
5239
5240
5241
5242
5243
5244
5245
5246
5247
5248
5249
5250
5251
5252
5253
5254
5255
5256
5257
5258
5259
5260
5261
5262
5263
5264
5265
5266
5267
5268
5269
5270
5271
5272
5273
5274
5275
5276
5277
5278
5279
5280
5281
5282
5283
5284
5285
5286
5287
5288
5289
5290
5291
5292
5293
5294
5295
5296
5297
5298
5299
5300
5301
5302
5303
5304
5305
5306
5307
5308
5309
5310
5311
5312
5313
5314
5315
5316
5317
5318
5319
5320
5321
5322
5323
5324
5325
5326
5327
5328
5329
5330
5331
5332
5333
5334
5335
5336
5337
5338
5339
5340
5341
5342
5343
5344
5345
5346
5347
5348
5349
5350
5351
5352
5353
5354
5355
5356
5357
5358
5359
5360
5361
5362
5363
5364
5365
5366
5367
5368
5369
5370
5371
5372
5373
5374
5375
5376
5377
5378
5379
5380
5381
5382
5383
5384
5385
5386
5387
5388
5389
5390
5391
5392
5393
5394
5395
5396
5397
5398
5399
5400
5401
5402
5403
5404
5405
5406
5407
5408
5409
5410
5411
5412
5413
5414
5415
5416
5417
5418
5419
5420
5421
5422
5423
5424
5425
5426
5427
5428
5429
5430
5431
5432
5433
5434
5435
5436
5437
5438
5439
5440
5441
5442
5443
5444
5445
5446
5447
5448
5449
5450
5451
5452
5453
5454
5455
5456
5457
5458
5459
5460
5461
5462
5463
5464
5465
5466
5467
5468
5469
5470
5471
5472
5473
5474
5475
5476
5477
5478
5479
5480
5481
5482
5483
5484
5485
5486
5487
5488
5489
5490
5491
5492
5493
5494
5495
5496
5497
5498
5499
5500
5501
5502
5503
5504
5505
5506
5507
5508
5509
5510
5511
5512
5513
5514
5515
5516
5517
5518
5519
5520
5521
5522
5523
5524
5525
5526
5527
5528
5529
5530
5531
5532
5533
5534
5535
5536
5537
5538
5539
5540
5541
5542
5543
5544
5545
5546
5547
5548
5549
5550
5551
5552
5553
5554
5555
5556
5557
5558
5559
5560
5561
5562
5563
5564
5565
5566
5567
5568
5569
5570
5571
5572
5573
5574
5575
5576
5577
5578
5579
5580
5581
5582
5583
5584
5585
5586
5587
5588
5589
5590
5591
5592
5593
5594
5595
5596
5597
5598
5599
5600
5601
5602
5603
5604
5605
5606
5607
5608
5609
5610
5611
5612
5613
5614
5615
5616
5617
5618
5619
5620
5621
5622
5623
5624
5625
5626
5627
5628
5629
5630
5631
5632
5633
5634
5635
5636
5637
5638
5639
5640
5641
5642
5643
5644
5645
5646
5647
5648
5649
5650
5651
5652
5653
5654
5655
5656
5657
5658
5659
5660
5661
5662
5663
5664
5665
5666
5667
5668
5669
5670
5671
5672
5673
5674
5675
5676
5677
5678
5679
5680
5681
5682
5683
5684
5685
5686
5687
5688
5689
5690
5691
5692
5693
5694
5695
5696
5697
5698
5699
5700
5701
5702
5703
5704
5705
5706
5707
5708
5709
5710
5711
5712
5713
5714
5715
5716
5717
5718
5719
5720
5721
5722
5723
5724
5725
5726
5727
5728
5729
5730
5731
5732
5733
5734
5735
5736
5737
5738
5739
5740
5741
5742
5743
5744
5745
5746
5747
5748
5749
5750
5751
5752
5753
5754
5755
5756
5757
5758
5759
5760
5761
5762
5763
5764
5765
5766
5767
5768
5769
5770
5771
5772
5773
5774
5775
5776
5777
5778
5779
5780
5781
5782
5783
5784
5785
5786
5787
5788
5789
5790
5791
5792
5793
5794
5795
5796
5797
5798
5799
5800
5801
5802
5803
5804
5805
5806
5807
5808
5809
5810
5811
5812
5813
5814
5815
5816
5817
5818
5819
5820
5821
5822
5823
5824
5825
5826
5827
5828
5829
5830
5831
5832
5833
5834
5835
5836
5837
5838
5839
5840
5841
5842
5843
5844
5845
5846
5847
5848
5849
5850
5851
5852
5853
5854
5855
5856
5857
5858
5859
5860
5861
5862
5863
5864
5865
5866
5867
5868
5869
5870
5871
5872
5873
5874
5875
5876
5877
5878
5879
5880
5881
5882
5883
5884
5885
5886
5887
5888
5889
5890
5891
5892
5893
5894
5895
5896
5897
5898
5899
5900
5901
5902
5903
5904
5905
5906
5907
5908
5909
5910
5911
5912
5913
5914
5915
5916
5917
5918
5919
5920
5921
5922
5923
5924
5925
5926
5927
5928
5929
5930
5931
5932
5933
5934
5935
5936
5937
5938
5939
5940
5941
5942
5943
5944
5945
5946
5947
5948
5949
5950
5951
5952
5953
5954
5955
5956
5957
5958
5959
5960
5961
5962
5963
5964
5965
5966
5967
5968
5969
5970
5971
5972
5973
5974
5975
5976
5977
5978
5979
5980
5981
5982
5983
5984
5985
5986
5987
5988
5989
5990
5991
5992
5993
5994
5995
5996
5997
5998
5999
6000
6001
6002
6003
6004
6005
6006
6007
6008
6009
6010
6011
6012
6013
6014
6015
6016
6017
6018
6019
6020
6021
6022
6023
6024
6025
6026
6027
6028
6029
6030
6031
6032
6033
6034
6035
6036
6037
6038
6039
6040
6041
6042
6043
6044
6045
6046
6047
6048
6049
6050
6051
6052
6053
6054
6055
6056
6057
6058
6059
6060
6061
6062
6063
6064
6065
6066
6067
6068
6069
6070
6071
6072
6073
6074
6075
6076
6077
6078
6079
6080
6081
6082
6083
6084
6085
6086
6087
6088
6089
6090
6091
6092
6093
6094
6095
6096
6097
6098
6099
6100
6101
6102
6103
6104
6105
6106
6107
6108
6109
6110
6111
6112
6113
6114
6115
6116
6117
6118
6119
6120
6121
6122
6123
6124
6125
6126
6127
6128
6129
6130
6131
6132
6133
6134
6135
6136
6137
6138
6139
6140
6141
6142
6143
6144
6145
6146
6147
6148
6149
/*
** 2011-08-14
**
** The author disclaims copyright to this source code.  In place of
** a legal notice, here is a blessing:
**
**    May you do good and not evil.
**    May you find forgiveness for yourself and forgive others.
**    May you share freely, never taking more than you give.
**
*************************************************************************
**
** PAGE FORMAT:
**
**   The maximum page size is 65536 bytes.
**
**   Since all records are equal to or larger than 2 bytes in size, and 
**   some space within the page is consumed by the page footer, there must
**   be less than 2^15 records on each page.
**
**   Each page ends with a footer that describes the pages contents. This
**   footer serves as similar purpose to the page header in an SQLite database.
**   A footer is used instead of a header because it makes it easier to
**   populate a new page based on a sorted list of key/value pairs.
**
**   The footer consists of the following values (starting at the end of
**   the page and continuing backwards towards the start). All values are
**   stored as unsigned big-endian integers.
**
**     * Number of records on page (2 bytes).
**     * Flags field (2 bytes).
**     * Left-hand pointer value (8 bytes).
**     * The starting offset of each record (2 bytes per record).
**
**   Records may span pages. Unless it happens to be an exact fit, the part
**   of the final record that starts on page X that does not fit on page X
**   is stored at the start of page (X+1). This means there may be pages where
**   (N==0). And on most pages the first record that starts on the page will
**   not start at byte offset 0. For example:
**
**      aaaaa bbbbb ccc <footer>    cc eeeee fffff g <footer>    gggg....
**
** RECORD FORMAT:
** 
**   The first byte of the record is a flags byte. It is a combination
**   of the following flags (defined in lsmInt.h):
**
**       LSM_START_DELETE
**       LSM_END_DELETE 
**       LSM_POINT_DELETE
**       LSM_INSERT    
**       LSM_SEPARATOR
**       LSM_SYSTEMKEY
**
**   Immediately following the type byte is a pointer to the smallest key 
**   in the next file that is larger than the key in the current record. The 
**   pointer is encoded as a varint. When added to the 32-bit page number 
**   stored in the footer, it is the page number of the page that contains the
**   smallest key in the next sorted file that is larger than this key. 
**
**   Next is the number of bytes in the key, encoded as a varint.
**
**   If the LSM_INSERT flag is set, the number of bytes in the value, as
**   a varint, is next.
**
**   Finally, the blob of data containing the key, and for LSM_INSERT
**   records, the value as well.
*/

#ifndef _LSM_INT_H
# include "lsmInt.h"
#endif

#define LSM_LOG_STRUCTURE 0
#define LSM_LOG_DATA      0

/*
** Macros to help decode record types.
*/
#define rtTopic(eType)       ((eType) & LSM_SYSTEMKEY)
#define rtIsDelete(eType)    (((eType) & 0x0F)==LSM_POINT_DELETE)

#define rtIsSeparator(eType) (((eType) & LSM_SEPARATOR)!=0)
#define rtIsWrite(eType)     (((eType) & LSM_INSERT)!=0)
#define rtIsSystem(eType)    (((eType) & LSM_SYSTEMKEY)!=0)

/*
** The following macros are used to access a page footer.
*/
#define SEGMENT_NRECORD_OFFSET(pgsz)        ((pgsz) - 2)
#define SEGMENT_FLAGS_OFFSET(pgsz)          ((pgsz) - 2 - 2)
#define SEGMENT_POINTER_OFFSET(pgsz)        ((pgsz) - 2 - 2 - 8)
#define SEGMENT_CELLPTR_OFFSET(pgsz, iCell) ((pgsz) - 2 - 2 - 8 - 2 - (iCell)*2)

#define SEGMENT_EOF(pgsz, nEntry) SEGMENT_CELLPTR_OFFSET(pgsz, nEntry)

#define SEGMENT_BTREE_FLAG     0x0001
#define PGFTR_SKIP_NEXT_FLAG   0x0002
#define PGFTR_SKIP_THIS_FLAG   0x0004

typedef struct SegmentPtr SegmentPtr;
typedef struct Blob Blob;

struct Blob {
  lsm_env *pEnv;
  void *pData;
  int nData;
  int nAlloc;
};

/*
** A SegmentPtr object may be used for one of two purposes:
**
**   * To iterate and/or seek within a single Segment (the combination of a 
**     main run and an optional sorted run).
**
**   * To iterate through the separators array of a segment.
*/
struct SegmentPtr {
  Level *pLevel;                /* Level object segment is part of */
  Segment *pSeg;                /* Segment to access */

  /* Current page. See segmentPtrLoadPage(). */
  Page *pPg;                    /* Current page */
  u16 flags;                    /* Copy of page flags field */
  int nCell;                    /* Number of cells on pPg */
  Pgno iPtr;                    /* Base cascade pointer */

  /* Current cell. See segmentPtrLoadCell() */
  int iCell;                    /* Current record within page pPg */
  int eType;                    /* Type of current record */
  Pgno iPgPtr;                  /* Cascade pointer offset */
  void *pKey; int nKey;         /* Key associated with current record */
  void *pVal; int nVal;         /* Current record value (eType==WRITE only) */

  /* Blobs used to allocate buffers for pKey and pVal as required */
  Blob blob1;
  Blob blob2;
};

/*
** Used to iterate through the keys stored in a b-tree hierarchy from start
** to finish. Only First() and Next() operations are required.
**
**   btreeCursorNew()
**   btreeCursorFirst()
**   btreeCursorNext()
**   btreeCursorFree()
**   btreeCursorPosition()
**   btreeCursorRestore()
*/
typedef struct BtreePg BtreePg;
typedef struct BtreeCursor BtreeCursor;
struct BtreePg {
  Page *pPage;
  int iCell;
};
struct BtreeCursor {
  Segment *pSeg;                  /* Iterate through this segments btree */
  FileSystem *pFS;                /* File system to read pages from */
  int nDepth;                     /* Allocated size of aPg[] */
  int iPg;                        /* Current entry in aPg[]. -1 -> EOF. */
  BtreePg *aPg;                   /* Pages from root to current location */

  /* Cache of current entry. pKey==0 for EOF. */
  void *pKey;
  int nKey;
  int eType;
  Pgno iPtr;

  /* Storage for key, if not local */
  Blob blob;
};


/*
** A cursor used for merged searches or iterations through up to one
** Tree structure and any number of sorted files.
**
**   lsmMCursorNew()
**   lsmMCursorSeek()
**   lsmMCursorNext()
**   lsmMCursorPrev()
**   lsmMCursorFirst()
**   lsmMCursorLast()
**   lsmMCursorKey()
**   lsmMCursorValue()
**   lsmMCursorValid()
**
** iFree:
**   This variable is only used by cursors providing input data for a
**   new top-level segment. Such cursors only ever iterate forwards, not
**   backwards.
*/
struct MultiCursor {
  lsm_db *pDb;                    /* Connection that owns this cursor */
  MultiCursor *pNext;             /* Next cursor owned by connection pDb */
  int flags;                      /* Mask of CURSOR_XXX flags */

  int eType;                      /* Cache of current key type */
  Blob key;                       /* Cache of current key (or NULL) */
  Blob val;                       /* Cache of current value */

  /* All the component cursors: */
  TreeCursor *apTreeCsr[2];       /* Up to two tree cursors */
  int iFree;                      /* Next element of free-list (-ve for eof) */
  SegmentPtr *aPtr;               /* Array of segment pointers */
  int nPtr;                       /* Size of array aPtr[] */
  BtreeCursor *pBtCsr;            /* b-tree cursor (db writes only) */

  /* Comparison results */
  int nTree;                      /* Size of aTree[] array */
  int *aTree;                     /* Array of comparison results */

  /* Used by cursors flushing the in-memory tree only */
  void *pSystemVal;               /* Pointer to buffer to free */

  /* Used by worker cursors only */
  Pgno *pPrevMergePtr;
};

/*
** The following constants are used to assign integers to each component
** cursor of a multi-cursor.
*/
#define CURSOR_DATA_TREE0     0   /* Current tree cursor (apTreeCsr[0]) */
#define CURSOR_DATA_TREE1     1   /* The "old" tree, if any (apTreeCsr[1]) */
#define CURSOR_DATA_SYSTEM    2   /* Free-list entries (new-toplevel only) */
#define CURSOR_DATA_SEGMENT   3   /* First segment pointer (aPtr[0]) */

/*
** CURSOR_IGNORE_DELETE
**   If set, this cursor will not visit SORTED_DELETE keys.
**
** CURSOR_FLUSH_FREELIST
**   This cursor is being used to create a new toplevel. It should also 
**   iterate through the contents of the in-memory free block list.
**
** CURSOR_IGNORE_SYSTEM
**   If set, this cursor ignores system keys.
**
** CURSOR_NEXT_OK
**   Set if it is Ok to call lsm_csr_next().
**
** CURSOR_PREV_OK
**   Set if it is Ok to call lsm_csr_prev().
**
** CURSOR_READ_SEPARATORS
**   Set if this cursor should visit the separator keys in segment 
**   aPtr[nPtr-1].
**
** CURSOR_SEEK_EQ
**   Cursor has undergone a successful lsm_csr_seek(LSM_SEEK_EQ) operation.
**   The key and value are stored in MultiCursor.key and MultiCursor.val
**   respectively.
*/
#define CURSOR_IGNORE_DELETE    0x00000001
#define CURSOR_FLUSH_FREELIST   0x00000002
#define CURSOR_IGNORE_SYSTEM    0x00000010
#define CURSOR_NEXT_OK          0x00000020
#define CURSOR_PREV_OK          0x00000040
#define CURSOR_READ_SEPARATORS  0x00000080
#define CURSOR_SEEK_EQ          0x00000100

typedef struct MergeWorker MergeWorker;
typedef struct Hierarchy Hierarchy;

struct Hierarchy {
  Page **apHier;
  int nHier;
};

/*
** aSave:
**   When mergeWorkerNextPage() is called to advance to the next page in
**   the output segment, if the bStore flag for an element of aSave[] is
**   true, it is cleared and the corresponding iPgno value is set to the 
**   page number of the page just completed.
**
**   aSave[0] is used to record the pointer value to be pushed into the
**   b-tree hierarchy. aSave[1] is used to save the page number of the
**   page containing the indirect key most recently written to the b-tree.
**   see mergeWorkerPushHierarchy() for details.
*/
struct MergeWorker {
  lsm_db *pDb;                    /* Database handle */
  Level *pLevel;                  /* Worker snapshot Level being merged */
  MultiCursor *pCsr;              /* Cursor to read new segment contents from */
  int bFlush;                     /* True if this is an in-memory tree flush */
  Hierarchy hier;                 /* B-tree hierarchy under construction */
  Page *pPage;                    /* Current output page */
  int nWork;                      /* Number of calls to mergeWorkerNextPage() */
  Pgno *aGobble;                  /* Gobble point for each input segment */

  Pgno iIndirect;
  struct SavedPgno {
    Pgno iPgno;
    int bStore;
  } aSave[2];
};

#ifdef LSM_DEBUG_EXPENSIVE
static int assertPointersOk(lsm_db *, Segment *, Segment *, int);
static int assertBtreeOk(lsm_db *, Segment *);
static void assertRunInOrder(lsm_db *pDb, Segment *pSeg);
#else
#define assertRunInOrder(x,y)
#define assertBtreeOk(x,y)
#endif


struct FilePage { u8 *aData; int nData; };
static u8 *fsPageData(Page *pPg, int *pnData){
  *pnData = ((struct FilePage *)(pPg))->nData;
  return ((struct FilePage *)(pPg))->aData;
}
/*UNUSED static u8 *fsPageDataPtr(Page *pPg){
  return ((struct FilePage *)(pPg))->aData;
}*/

/*
** Write nVal as a 16-bit unsigned big-endian integer into buffer aOut.
*/
void lsmPutU16(u8 *aOut, u16 nVal){
  aOut[0] = (u8)((nVal>>8) & 0xFF);
  aOut[1] = (u8)(nVal & 0xFF);
}

void lsmPutU32(u8 *aOut, u32 nVal){
  aOut[0] = (u8)((nVal>>24) & 0xFF);
  aOut[1] = (u8)((nVal>>16) & 0xFF);
  aOut[2] = (u8)((nVal>> 8) & 0xFF);
  aOut[3] = (u8)((nVal    ) & 0xFF);
}

int lsmGetU16(u8 *aOut){
  return (aOut[0] << 8) + aOut[1];
}

u32 lsmGetU32(u8 *aOut){
  return ((u32)aOut[0] << 24) 
       + ((u32)aOut[1] << 16) 
       + ((u32)aOut[2] << 8) 
       + ((u32)aOut[3]);
}

u64 lsmGetU64(u8 *aOut){
  return ((u64)aOut[0] << 56) 
       + ((u64)aOut[1] << 48) 
       + ((u64)aOut[2] << 40) 
       + ((u64)aOut[3] << 32) 
       + ((u64)aOut[4] << 24)
       + ((u32)aOut[5] << 16) 
       + ((u32)aOut[6] << 8) 
       + ((u32)aOut[7]);
}

void lsmPutU64(u8 *aOut, u64 nVal){
  aOut[0] = (u8)((nVal>>56) & 0xFF);
  aOut[1] = (u8)((nVal>>48) & 0xFF);
  aOut[2] = (u8)((nVal>>40) & 0xFF);
  aOut[3] = (u8)((nVal>>32) & 0xFF);
  aOut[4] = (u8)((nVal>>24) & 0xFF);
  aOut[5] = (u8)((nVal>>16) & 0xFF);
  aOut[6] = (u8)((nVal>> 8) & 0xFF);
  aOut[7] = (u8)((nVal    ) & 0xFF);
}

static int sortedBlobGrow(lsm_env *pEnv, Blob *pBlob, int nData){
  assert( pBlob->pEnv==pEnv || (pBlob->pEnv==0 && pBlob->pData==0) );
  if( pBlob->nAlloc<nData ){
    pBlob->pData = lsmReallocOrFree(pEnv, pBlob->pData, nData);
    if( !pBlob->pData ) return LSM_NOMEM_BKPT;
    pBlob->nAlloc = nData;
    pBlob->pEnv = pEnv;
  }
  return LSM_OK;
}

static int sortedBlobSet(lsm_env *pEnv, Blob *pBlob, void *pData, int nData){
  if( sortedBlobGrow(pEnv, pBlob, nData) ) return LSM_NOMEM;
  memcpy(pBlob->pData, pData, nData);
  pBlob->nData = nData;
  return LSM_OK;
}

#if 0
static int sortedBlobCopy(Blob *pDest, Blob *pSrc){
  return sortedBlobSet(pDest, pSrc->pData, pSrc->nData);
}
#endif

static void sortedBlobFree(Blob *pBlob){
  assert( pBlob->pEnv || pBlob->pData==0 );
  if( pBlob->pData ) lsmFree(pBlob->pEnv, pBlob->pData);
  memset(pBlob, 0, sizeof(Blob));
}

static int sortedReadData(
  Segment *pSeg,
  Page *pPg,
  int iOff,
  int nByte,
  void **ppData,
  Blob *pBlob
){
  int rc = LSM_OK;
  int iEnd;
  int nData;
  int nCell;
  u8 *aData;

  aData = fsPageData(pPg, &nData);
  nCell = lsmGetU16(&aData[SEGMENT_NRECORD_OFFSET(nData)]);
  iEnd = SEGMENT_EOF(nData, nCell);
  assert( iEnd>0 && iEnd<nData );

  if( iOff+nByte<=iEnd ){
    *ppData = (void *)&aData[iOff];
  }else{
    int nRem = nByte;
    int i = iOff;
    u8 *aDest;

    /* Make sure the blob is big enough to store the value being loaded. */
    rc = sortedBlobGrow(lsmPageEnv(pPg), pBlob, nByte);
    if( rc!=LSM_OK ) return rc;
    pBlob->nData = nByte;
    aDest = (u8 *)pBlob->pData;
    *ppData = pBlob->pData;

    /* Increment the pointer pages ref-count. */
    lsmFsPageRef(pPg);

    while( rc==LSM_OK ){
      Page *pNext;
      int flags;

      /* Copy data from pPg into the output buffer. */
      int nCopy = LSM_MIN(nRem, iEnd-i);
      if( nCopy>0 ){
        memcpy(&aDest[nByte-nRem], &aData[i], nCopy);
        nRem -= nCopy;
        i += nCopy;
        assert( nRem==0 || i==iEnd );
      }
      assert( nRem>=0 );
      if( nRem==0 ) break;
      i -= iEnd;

      /* Grab the next page in the segment */

      do {
        rc = lsmFsDbPageNext(pSeg, pPg, 1, &pNext);
        if( rc==LSM_OK && pNext==0 ){
          rc = LSM_CORRUPT_BKPT;
        }
        if( rc ) break;
        lsmFsPageRelease(pPg);
        pPg = pNext;
        aData = fsPageData(pPg, &nData);
        flags = lsmGetU16(&aData[SEGMENT_FLAGS_OFFSET(nData)]);
      }while( flags&SEGMENT_BTREE_FLAG );

      iEnd = SEGMENT_EOF(nData, lsmGetU16(&aData[nData-2]));
      assert( iEnd>0 && iEnd<nData );
    }

    lsmFsPageRelease(pPg);
  }

  return rc;
}

static int pageGetNRec(u8 *aData, int nData){
  return (int)lsmGetU16(&aData[SEGMENT_NRECORD_OFFSET(nData)]);
}

static Pgno pageGetPtr(u8 *aData, int nData){
  return (Pgno)lsmGetU64(&aData[SEGMENT_POINTER_OFFSET(nData)]);
}

static int pageGetFlags(u8 *aData, int nData){
  return (int)lsmGetU16(&aData[SEGMENT_FLAGS_OFFSET(nData)]);
}

static u8 *pageGetCell(u8 *aData, int nData, int iCell){
  return &aData[lsmGetU16(&aData[SEGMENT_CELLPTR_OFFSET(nData, iCell)])];
}

/*
** Return the number of cells on page pPg.
*/
static int pageObjGetNRec(Page *pPg){
  int nData;
  u8 *aData = lsmFsPageData(pPg, &nData);
  return pageGetNRec(aData, nData);
}

/*
** Return the decoded (possibly relative) pointer value stored in cell 
** iCell from page aData/nData.
*/
static Pgno pageGetRecordPtr(u8 *aData, int nData, int iCell){
  Pgno iRet;                      /* Return value */
  u8 *aCell;                      /* Pointer to cell iCell */

  assert( iCell<pageGetNRec(aData, nData) && iCell>=0 );
  aCell = pageGetCell(aData, nData, iCell);
  lsmVarintGet64(&aCell[1], &iRet);
  return iRet;
}

static u8 *pageGetKey(
  Segment *pSeg,                  /* Segment pPg belongs to */
  Page *pPg,                      /* Page to read from */
  int iCell,                      /* Index of cell on page to read */
  int *piTopic,                   /* OUT: Topic associated with this key */
  int *pnKey,                     /* OUT: Size of key in bytes */
  Blob *pBlob                     /* If required, use this for dynamic memory */
){
  u8 *pKey;
  int nDummy;
  int eType;
  u8 *aData;
  int nData;

  aData = fsPageData(pPg, &nData);

  assert( !(pageGetFlags(aData, nData) & SEGMENT_BTREE_FLAG) );
  assert( iCell<pageGetNRec(aData, nData) );

  pKey = pageGetCell(aData, nData, iCell);
  eType = *pKey++;
  pKey += lsmVarintGet32(pKey, &nDummy);
  pKey += lsmVarintGet32(pKey, pnKey);
  if( rtIsWrite(eType) ){
    pKey += lsmVarintGet32(pKey, &nDummy);
  }
  *piTopic = rtTopic(eType);

  sortedReadData(pSeg, pPg, pKey-aData, *pnKey, (void **)&pKey, pBlob);
  return pKey;
}

static int pageGetKeyCopy(
  lsm_env *pEnv,                  /* Environment handle */
  Segment *pSeg,                  /* Segment pPg belongs to */
  Page *pPg,                      /* Page to read from */
  int iCell,                      /* Index of cell on page to read */
  int *piTopic,                   /* OUT: Topic associated with this key */
  Blob *pBlob                     /* If required, use this for dynamic memory */
){
  int rc = LSM_OK;
  int nKey;
  u8 *aKey;

  aKey = pageGetKey(pSeg, pPg, iCell, piTopic, &nKey, pBlob);
  assert( (void *)aKey!=pBlob->pData || nKey==pBlob->nData );
  if( (void *)aKey!=pBlob->pData ){
    rc = sortedBlobSet(pEnv, pBlob, aKey, nKey);
  }

  return rc;
}

static Pgno pageGetBtreeRef(Page *pPg, int iKey){
  Pgno iRef;
  u8 *aData;
  int nData;
  u8 *aCell;

  aData = fsPageData(pPg, &nData);
  aCell = pageGetCell(aData, nData, iKey);
  assert( aCell[0]==0 );
  aCell++;
  aCell += lsmVarintGet64(aCell, &iRef);
  lsmVarintGet64(aCell, &iRef);
  assert( iRef>0 );
  return iRef;
}

#define GETVARINT64(a, i) (((i)=((u8*)(a))[0])<=240?1:lsmVarintGet64((a), &(i)))
#define GETVARINT32(a, i) (((i)=((u8*)(a))[0])<=240?1:lsmVarintGet32((a), &(i)))

static int pageGetBtreeKey(
  Segment *pSeg,                  /* Segment page pPg belongs to */
  Page *pPg,
  int iKey, 
  Pgno *piPtr, 
  int *piTopic, 
  void **ppKey,
  int *pnKey,
  Blob *pBlob
){
  u8 *aData;
  int nData;
  u8 *aCell;
  int eType;

  aData = fsPageData(pPg, &nData);
  assert( SEGMENT_BTREE_FLAG & pageGetFlags(aData, nData) );
  assert( iKey>=0 && iKey<pageGetNRec(aData, nData) );

  aCell = pageGetCell(aData, nData, iKey);
  eType = *aCell++;
  aCell += GETVARINT64(aCell, *piPtr);

  if( eType==0 ){
    int rc;
    Pgno iRef;                  /* Page number of referenced page */
    Page *pRef;
    aCell += GETVARINT64(aCell, iRef);
    rc = lsmFsDbPageGet(lsmPageFS(pPg), pSeg, iRef, &pRef);
    if( rc!=LSM_OK ) return rc;
    pageGetKeyCopy(lsmPageEnv(pPg), pSeg, pRef, 0, &eType, pBlob);
    lsmFsPageRelease(pRef);
    *ppKey = pBlob->pData;
    *pnKey = pBlob->nData;
  }else{
    aCell += GETVARINT32(aCell, *pnKey);
    *ppKey = aCell;
  }
  if( piTopic ) *piTopic = rtTopic(eType);

  return LSM_OK;
}

static int btreeCursorLoadKey(BtreeCursor *pCsr){
  int rc = LSM_OK;
  if( pCsr->iPg<0 ){
    pCsr->pKey = 0;
    pCsr->nKey = 0;
    pCsr->eType = 0;
  }else{
    Pgno dummy;
    int iPg = pCsr->iPg;
    int iCell = pCsr->aPg[iPg].iCell;
    while( iCell<0 && (--iPg)>=0 ){
      iCell = pCsr->aPg[iPg].iCell-1;
    }
    if( iPg<0 || iCell<0 ) return LSM_CORRUPT_BKPT;

    rc = pageGetBtreeKey(
        pCsr->pSeg,
        pCsr->aPg[iPg].pPage, iCell,
        &dummy, &pCsr->eType, &pCsr->pKey, &pCsr->nKey, &pCsr->blob
    );
    pCsr->eType |= LSM_SEPARATOR;
  }

  return rc;
}

static int btreeCursorPtr(u8 *aData, int nData, int iCell){
  int nCell;

  nCell = pageGetNRec(aData, nData);
  if( iCell>=nCell ){
    return pageGetPtr(aData, nData);
  }
  return pageGetRecordPtr(aData, nData, iCell);
}

static int btreeCursorNext(BtreeCursor *pCsr){
  int rc = LSM_OK;

  BtreePg *pPg = &pCsr->aPg[pCsr->iPg];
  int nCell; 
  u8 *aData;
  int nData;

  assert( pCsr->iPg>=0 );
  assert( pCsr->iPg==pCsr->nDepth-1 );

  aData = fsPageData(pPg->pPage, &nData);
  nCell = pageGetNRec(aData, nData);
  assert( pPg->iCell<=nCell );
  pPg->iCell++;
  if( pPg->iCell==nCell ){
    Pgno iLoad;

    /* Up to parent. */
    lsmFsPageRelease(pPg->pPage);
    pPg->pPage = 0;
    pCsr->iPg--;
    while( pCsr->iPg>=0 ){
      pPg = &pCsr->aPg[pCsr->iPg];
      aData = fsPageData(pPg->pPage, &nData);
      if( pPg->iCell<pageGetNRec(aData, nData) ) break;
      lsmFsPageRelease(pPg->pPage);
      pCsr->iPg--;
    }

    /* Read the key */
    rc = btreeCursorLoadKey(pCsr);

    /* Unless the cursor is at EOF, descend to cell -1 (yes, negative one) of 
    ** the left-most most descendent. */
    if( pCsr->iPg>=0 ){
      pCsr->aPg[pCsr->iPg].iCell++;

      iLoad = btreeCursorPtr(aData, nData, pPg->iCell);
      do {
        Page *pLoad;
        pCsr->iPg++;
        rc = lsmFsDbPageGet(pCsr->pFS, pCsr->pSeg, iLoad, &pLoad);
        pCsr->aPg[pCsr->iPg].pPage = pLoad;
        pCsr->aPg[pCsr->iPg].iCell = 0;
        if( rc==LSM_OK ){
          if( pCsr->iPg==(pCsr->nDepth-1) ) break;
          aData = fsPageData(pLoad, &nData);
          iLoad = btreeCursorPtr(aData, nData, 0);
        }
      }while( rc==LSM_OK && pCsr->iPg<(pCsr->nDepth-1) );
      pCsr->aPg[pCsr->iPg].iCell = -1;
    }

  }else{
    rc = btreeCursorLoadKey(pCsr);
  }

  if( rc==LSM_OK && pCsr->iPg>=0 ){
    aData = fsPageData(pCsr->aPg[pCsr->iPg].pPage, &nData);
    pCsr->iPtr = btreeCursorPtr(aData, nData, pCsr->aPg[pCsr->iPg].iCell+1);
  }

  return rc;
}

static void btreeCursorFree(BtreeCursor *pCsr){
  if( pCsr ){
    int i;
    lsm_env *pEnv = lsmFsEnv(pCsr->pFS);
    for(i=0; i<=pCsr->iPg; i++){
      lsmFsPageRelease(pCsr->aPg[i].pPage);
    }
    sortedBlobFree(&pCsr->blob);
    lsmFree(pEnv, pCsr->aPg);
    lsmFree(pEnv, pCsr);
  }
}

static int btreeCursorFirst(BtreeCursor *pCsr){
  int rc;

  Page *pPg = 0;
  FileSystem *pFS = pCsr->pFS;
  int iPg = pCsr->pSeg->iRoot;

  do {
    rc = lsmFsDbPageGet(pFS, pCsr->pSeg, iPg, &pPg);
    assert( (rc==LSM_OK)==(pPg!=0) );
    if( rc==LSM_OK ){
      u8 *aData;
      int nData;
      int flags;

      aData = fsPageData(pPg, &nData);
      flags = pageGetFlags(aData, nData);
      if( (flags & SEGMENT_BTREE_FLAG)==0 ) break;

      if( (pCsr->nDepth % 8)==0 ){
        int nNew = pCsr->nDepth + 8;
        pCsr->aPg = (BtreePg *)lsmReallocOrFreeRc(
            lsmFsEnv(pFS), pCsr->aPg, sizeof(BtreePg) * nNew, &rc
        );
        if( rc==LSM_OK ){
          memset(&pCsr->aPg[pCsr->nDepth], 0, sizeof(BtreePg) * 8);
        }
      }

      if( rc==LSM_OK ){
        assert( pCsr->aPg[pCsr->nDepth].iCell==0 );
        pCsr->aPg[pCsr->nDepth].pPage = pPg;
        pCsr->nDepth++;
        iPg = pageGetRecordPtr(aData, nData, 0);
      }
    }
  }while( rc==LSM_OK );
  lsmFsPageRelease(pPg);
  pCsr->iPg = pCsr->nDepth-1;

  if( rc==LSM_OK && pCsr->nDepth ){
    pCsr->aPg[pCsr->iPg].iCell = -1;
    rc = btreeCursorNext(pCsr);
  }

  return rc;
}

static void btreeCursorPosition(BtreeCursor *pCsr, MergeInput *p){
  if( pCsr->iPg>=0 ){
    p->iPg = lsmFsPageNumber(pCsr->aPg[pCsr->iPg].pPage);
    p->iCell = ((pCsr->aPg[pCsr->iPg].iCell + 1) << 8) + pCsr->nDepth;
  }else{
    p->iPg = 0;
    p->iCell = 0;
  }
}

static void btreeCursorSplitkey(BtreeCursor *pCsr, MergeInput *p){
  int iCell = pCsr->aPg[pCsr->iPg].iCell;
  if( iCell>=0 ){
    p->iCell = iCell;
    p->iPg = lsmFsPageNumber(pCsr->aPg[pCsr->iPg].pPage);
  }else{
    int i;
    for(i=pCsr->iPg-1; i>=0; i--){
      if( pCsr->aPg[i].iCell>0 ) break;
    }
    assert( i>=0 );
    p->iCell = pCsr->aPg[i].iCell-1;
    p->iPg = lsmFsPageNumber(pCsr->aPg[i].pPage);
  }
}

static int sortedKeyCompare(
  int (*xCmp)(void *, int, void *, int),
  int iLhsTopic, void *pLhsKey, int nLhsKey,
  int iRhsTopic, void *pRhsKey, int nRhsKey
){
  int res = iLhsTopic - iRhsTopic;
  if( res==0 ){
    res = xCmp(pLhsKey, nLhsKey, pRhsKey, nRhsKey);
  }
  return res;
}

static int btreeCursorRestore(
  BtreeCursor *pCsr, 
  int (*xCmp)(void *, int, void *, int),
  MergeInput *p
){
  int rc = LSM_OK;

  if( p->iPg ){
    lsm_env *pEnv = lsmFsEnv(pCsr->pFS);
    int iCell;                    /* Current cell number on leaf page */
    Pgno iLeaf;                   /* Page number of current leaf page */
    int nDepth;                   /* Depth of b-tree structure */
    Segment *pSeg = pCsr->pSeg;

    /* Decode the MergeInput structure */
    iLeaf = p->iPg;
    nDepth = (p->iCell & 0x00FF);
    iCell = (p->iCell >> 8) - 1;

    /* Allocate the BtreeCursor.aPg[] array */
    assert( pCsr->aPg==0 );
    pCsr->aPg = (BtreePg *)lsmMallocZeroRc(pEnv, sizeof(BtreePg) * nDepth, &rc);

    /* Populate the last entry of the aPg[] array */
    if( rc==LSM_OK ){
      Page **pp = &pCsr->aPg[nDepth-1].pPage;
      pCsr->iPg = nDepth-1;
      pCsr->nDepth = nDepth;
      pCsr->aPg[pCsr->iPg].iCell = iCell;
      rc = lsmFsDbPageGet(pCsr->pFS, pSeg, iLeaf, pp);
    }

    /* Populate any other aPg[] array entries */
    if( rc==LSM_OK && nDepth>1 ){
      Blob blob = {0,0,0};
      void *pSeek;
      int nSeek;
      int iTopicSeek;
      int iPg = 0;
      int iLoad = pSeg->iRoot;
      Page *pPg = pCsr->aPg[nDepth-1].pPage;
 
      if( pageObjGetNRec(pPg)==0 ){
        /* This can happen when pPg is the right-most leaf in the b-tree.
        ** In this case, set the iTopicSeek/pSeek/nSeek key to a value
        ** greater than any real key.  */
        assert( iCell==-1 );
        iTopicSeek = 1000;
        pSeek = 0;
        nSeek = 0;
      }else{
        Pgno dummy;
        rc = pageGetBtreeKey(pSeg, pPg,
            0, &dummy, &iTopicSeek, &pSeek, &nSeek, &pCsr->blob
        );
      }

      do {
        Page *pPg;
        rc = lsmFsDbPageGet(pCsr->pFS, pSeg, iLoad, &pPg);
        assert( rc==LSM_OK || pPg==0 );
        if( rc==LSM_OK ){
          u8 *aData;                  /* Buffer containing page data */
          int nData;                  /* Size of aData[] in bytes */
          int iMin;
          int iMax;
          int iCell;

          aData = fsPageData(pPg, &nData);
          assert( (pageGetFlags(aData, nData) & SEGMENT_BTREE_FLAG) );

          iLoad = pageGetPtr(aData, nData);
          iCell = pageGetNRec(aData, nData); 
          iMax = iCell-1;
          iMin = 0;

          while( iMax>=iMin ){
            int iTry = (iMin+iMax)/2;
            void *pKey; int nKey;         /* Key for cell iTry */
            int iTopic;                   /* Topic for key pKeyT/nKeyT */
            Pgno iPtr;                    /* Pointer for cell iTry */
            int res;                      /* (pSeek - pKeyT) */

            rc = pageGetBtreeKey(
                pSeg, pPg, iTry, &iPtr, &iTopic, &pKey, &nKey, &blob
            );
            if( rc!=LSM_OK ) break;

            res = sortedKeyCompare(
                xCmp, iTopicSeek, pSeek, nSeek, iTopic, pKey, nKey
            );
            assert( res!=0 );

            if( res<0 ){
              iLoad = iPtr;
              iCell = iTry;
              iMax = iTry-1;
            }else{
              iMin = iTry+1;
            }
          }

          pCsr->aPg[iPg].pPage = pPg;
          pCsr->aPg[iPg].iCell = iCell;
          iPg++;
          assert( iPg!=nDepth-1 
               || lsmFsRedirectPage(pCsr->pFS, pSeg->pRedirect, iLoad)==iLeaf
          );
        }
      }while( rc==LSM_OK && iPg<(nDepth-1) );
      sortedBlobFree(&blob);
    }

    /* Load the current key and pointer */
    if( rc==LSM_OK ){
      BtreePg *pBtreePg;
      u8 *aData;
      int nData;

      pBtreePg = &pCsr->aPg[pCsr->iPg];
      aData = fsPageData(pBtreePg->pPage, &nData);
      pCsr->iPtr = btreeCursorPtr(aData, nData, pBtreePg->iCell+1);
      if( pBtreePg->iCell<0 ){
        Pgno dummy;
        int i;
        for(i=pCsr->iPg-1; i>=0; i--){
          if( pCsr->aPg[i].iCell>0 ) break;
        }
        assert( i>=0 );
        rc = pageGetBtreeKey(pSeg,
            pCsr->aPg[i].pPage, pCsr->aPg[i].iCell-1,
            &dummy, &pCsr->eType, &pCsr->pKey, &pCsr->nKey, &pCsr->blob
        );
        pCsr->eType |= LSM_SEPARATOR;

      }else{
        rc = btreeCursorLoadKey(pCsr);
      }
    }
  }
  return rc;
}

static int btreeCursorNew(
  lsm_db *pDb,
  Segment *pSeg,
  BtreeCursor **ppCsr
){
  int rc = LSM_OK;
  BtreeCursor *pCsr;
  
  assert( pSeg->iRoot );
  pCsr = lsmMallocZeroRc(pDb->pEnv, sizeof(BtreeCursor), &rc);
  if( pCsr ){
    pCsr->pFS = pDb->pFS;
    pCsr->pSeg = pSeg;
    pCsr->iPg = -1;
  }

  *ppCsr = pCsr;
  return rc;
}

static void segmentPtrSetPage(SegmentPtr *pPtr, Page *pNext){
  lsmFsPageRelease(pPtr->pPg);
  if( pNext ){
    int nData;
    u8 *aData = fsPageData(pNext, &nData);
    pPtr->nCell = pageGetNRec(aData, nData);
    pPtr->flags = pageGetFlags(aData, nData);
    pPtr->iPtr = pageGetPtr(aData, nData);
  }
  pPtr->pPg = pNext;
}

/*
** Load a new page into the SegmentPtr object pPtr.
*/
static int segmentPtrLoadPage(
  FileSystem *pFS,
  SegmentPtr *pPtr,              /* Load page into this SegmentPtr object */
  int iNew                       /* Page number of new page */
){
  Page *pPg = 0;                 /* The new page */
  int rc;                        /* Return Code */

  rc = lsmFsDbPageGet(pFS, pPtr->pSeg, iNew, &pPg);
  assert( rc==LSM_OK || pPg==0 );
  segmentPtrSetPage(pPtr, pPg);

  return rc;
}

static int segmentPtrReadData(
  SegmentPtr *pPtr,
  int iOff,
  int nByte,
  void **ppData,
  Blob *pBlob
){
  return sortedReadData(pPtr->pSeg, pPtr->pPg, iOff, nByte, ppData, pBlob);
}

static int segmentPtrNextPage(
  SegmentPtr *pPtr,              /* Load page into this SegmentPtr object */
  int eDir                       /* +1 for next(), -1 for prev() */
){
  Page *pNext;                   /* New page to load */
  int rc;                        /* Return code */

  assert( eDir==1 || eDir==-1 );
  assert( pPtr->pPg );
  assert( pPtr->pSeg || eDir>0 );

  rc = lsmFsDbPageNext(pPtr->pSeg, pPtr->pPg, eDir, &pNext);
  assert( rc==LSM_OK || pNext==0 );
  segmentPtrSetPage(pPtr, pNext);
  return rc;
}

static int segmentPtrLoadCell(
  SegmentPtr *pPtr,              /* Load page into this SegmentPtr object */
  int iNew                       /* Cell number of new cell */
){
  int rc = LSM_OK;
  if( pPtr->pPg ){
    u8 *aData;                    /* Pointer to page data buffer */
    int iOff;                     /* Offset in aData[] to read from */
    int nPgsz;                    /* Size of page (aData[]) in bytes */

    assert( iNew<pPtr->nCell );
    pPtr->iCell = iNew;
    aData = fsPageData(pPtr->pPg, &nPgsz);
    iOff = lsmGetU16(&aData[SEGMENT_CELLPTR_OFFSET(nPgsz, pPtr->iCell)]);
    pPtr->eType = aData[iOff];
    iOff++;
    iOff += GETVARINT64(&aData[iOff], pPtr->iPgPtr);
    iOff += GETVARINT32(&aData[iOff], pPtr->nKey);
    if( rtIsWrite(pPtr->eType) ){
      iOff += GETVARINT32(&aData[iOff], pPtr->nVal);
    }
    assert( pPtr->nKey>=0 );

    rc = segmentPtrReadData(
        pPtr, iOff, pPtr->nKey, &pPtr->pKey, &pPtr->blob1
    );
    if( rc==LSM_OK && rtIsWrite(pPtr->eType) ){
      rc = segmentPtrReadData(
          pPtr, iOff+pPtr->nKey, pPtr->nVal, &pPtr->pVal, &pPtr->blob2
      );
    }else{
      pPtr->nVal = 0;
      pPtr->pVal = 0;
    }
  }

  return rc;
}


static Segment *sortedSplitkeySegment(Level *pLevel){
  Merge *pMerge = pLevel->pMerge;
  MergeInput *p = &pMerge->splitkey;
  Segment *pSeg;
  int i;

  for(i=0; i<pMerge->nInput; i++){
    if( p->iPg==pMerge->aInput[i].iPg ) break;
  }
  if( pMerge->nInput==(pLevel->nRight+1) && i>=(pMerge->nInput-1) ){
    pSeg = &pLevel->pNext->lhs;
  }else{
    pSeg = &pLevel->aRhs[i];
  }

  return pSeg;
}

static void sortedSplitkey(lsm_db *pDb, Level *pLevel, int *pRc){
  Segment *pSeg;
  Page *pPg = 0;
  lsm_env *pEnv = pDb->pEnv;      /* Environment handle */
  int rc = *pRc;
  Merge *pMerge = pLevel->pMerge;

  pSeg = sortedSplitkeySegment(pLevel);
  if( rc==LSM_OK ){
    rc = lsmFsDbPageGet(pDb->pFS, pSeg, pMerge->splitkey.iPg, &pPg);
  }
  if( rc==LSM_OK ){
    int iTopic;
    Blob blob = {0, 0, 0, 0};
    u8 *aData;
    int nData;
  
    aData = lsmFsPageData(pPg, &nData);
    if( pageGetFlags(aData, nData) & SEGMENT_BTREE_FLAG ){
      void *pKey;
      int nKey;
      Pgno dummy;
      rc = pageGetBtreeKey(pSeg,
          pPg, pMerge->splitkey.iCell, &dummy, &iTopic, &pKey, &nKey, &blob
      );
      if( rc==LSM_OK && blob.pData!=pKey ){
        rc = sortedBlobSet(pEnv, &blob, pKey, nKey);
      }
    }else{
      rc = pageGetKeyCopy(
          pEnv, pSeg, pPg, pMerge->splitkey.iCell, &iTopic, &blob
      );
    }

    pLevel->iSplitTopic = iTopic;
    pLevel->pSplitKey = blob.pData;
    pLevel->nSplitKey = blob.nData;
    lsmFsPageRelease(pPg);
  }

  *pRc = rc;
}

static void segmentPtrReset(SegmentPtr *pPtr){
  lsmFsPageRelease(pPtr->pPg);
  pPtr->pPg = 0;
  pPtr->nCell = 0;
  pPtr->pKey = 0;
  pPtr->nKey = 0;
  pPtr->pVal = 0;
  pPtr->nVal = 0;
  pPtr->eType = 0;
  pPtr->iCell = 0;
  sortedBlobFree(&pPtr->blob1);
  sortedBlobFree(&pPtr->blob2);
}

static int segmentPtrIgnoreSeparators(MultiCursor *pCsr, SegmentPtr *pPtr){
  return (pCsr->flags & CURSOR_READ_SEPARATORS)==0
      || (pPtr!=&pCsr->aPtr[pCsr->nPtr-1]);
}

static int segmentPtrAdvance(
  MultiCursor *pCsr, 
  SegmentPtr *pPtr,
  int bReverse
){
  int eDir = (bReverse ? -1 : 1);
  Level *pLvl = pPtr->pLevel;
  do {
    int rc;
    int iCell;                    /* Number of new cell in page */
    int svFlags = 0;              /* SegmentPtr.eType before advance */

    iCell = pPtr->iCell + eDir;
    assert( pPtr->pPg );
    assert( iCell<=pPtr->nCell && iCell>=-1 );

    if( bReverse && pPtr->pSeg!=&pPtr->pLevel->lhs ){
      svFlags = pPtr->eType;
      assert( svFlags );
    }

    if( iCell>=pPtr->nCell || iCell<0 ){
      do {
        rc = segmentPtrNextPage(pPtr, eDir); 
      }while( rc==LSM_OK 
           && pPtr->pPg 
           && (pPtr->nCell==0 || (pPtr->flags & SEGMENT_BTREE_FLAG) ) 
      );
      if( rc!=LSM_OK ) return rc;
      iCell = bReverse ? (pPtr->nCell-1) : 0;
    }
    rc = segmentPtrLoadCell(pPtr, iCell);
    if( rc!=LSM_OK ) return rc;

    if( svFlags && pPtr->pPg ){
      int res = sortedKeyCompare(pCsr->pDb->xCmp,
          rtTopic(pPtr->eType), pPtr->pKey, pPtr->nKey,
          pLvl->iSplitTopic, pLvl->pSplitKey, pLvl->nSplitKey
      );
      if( res<0 ) segmentPtrReset(pPtr);
    }

    if( pPtr->pPg==0 && (svFlags & LSM_END_DELETE) ){
      Segment *pSeg = pPtr->pSeg;
      rc = lsmFsDbPageGet(pCsr->pDb->pFS, pSeg, pSeg->iFirst, &pPtr->pPg);
      if( rc!=LSM_OK ) return rc;
      pPtr->eType = LSM_START_DELETE | LSM_POINT_DELETE;
      pPtr->eType |= (pLvl->iSplitTopic ? LSM_SYSTEMKEY : 0);
      pPtr->pKey = pLvl->pSplitKey;
      pPtr->nKey = pLvl->nSplitKey;
    }

  }while( pCsr 
       && pPtr->pPg 
       && segmentPtrIgnoreSeparators(pCsr, pPtr)
       && rtIsSeparator(pPtr->eType)
  );

  return LSM_OK;
}

static void segmentPtrEndPage(
  FileSystem *pFS, 
  SegmentPtr *pPtr, 
  int bLast, 
  int *pRc
){
  if( *pRc==LSM_OK ){
    Segment *pSeg = pPtr->pSeg;
    Page *pNew = 0;
    if( bLast ){
      *pRc = lsmFsDbPageLast(pFS, pSeg, &pNew);
    }else{
      *pRc = lsmFsDbPageGet(pFS, pSeg, pSeg->iFirst, &pNew);
    }
    segmentPtrSetPage(pPtr, pNew);
  }
}


/*
** Try to move the segment pointer passed as the second argument so that it
** points at either the first (bLast==0) or last (bLast==1) cell in the valid
** region of the segment defined by pPtr->iFirst and pPtr->iLast.
**
** Return LSM_OK if successful or an lsm error code if something goes
** wrong (IO error, OOM etc.).
*/
static int segmentPtrEnd(MultiCursor *pCsr, SegmentPtr *pPtr, int bLast){
  Level *pLvl = pPtr->pLevel;
  int rc = LSM_OK;
  FileSystem *pFS = pCsr->pDb->pFS;
  int bIgnore;

  segmentPtrEndPage(pFS, pPtr, bLast, &rc);
  while( rc==LSM_OK && pPtr->pPg 
      && (pPtr->nCell==0 || (pPtr->flags & SEGMENT_BTREE_FLAG))
  ){
    rc = segmentPtrNextPage(pPtr, (bLast ? -1 : 1));
  }

  if( rc==LSM_OK && pPtr->pPg ){
    rc = segmentPtrLoadCell(pPtr, bLast ? (pPtr->nCell-1) : 0);
    if( rc==LSM_OK && bLast && pPtr->pSeg!=&pLvl->lhs ){
      int res = sortedKeyCompare(pCsr->pDb->xCmp,
          rtTopic(pPtr->eType), pPtr->pKey, pPtr->nKey,
          pLvl->iSplitTopic, pLvl->pSplitKey, pLvl->nSplitKey
      );
      if( res<0 ) segmentPtrReset(pPtr);
    }
  }
  
  bIgnore = segmentPtrIgnoreSeparators(pCsr, pPtr);
  if( rc==LSM_OK && pPtr->pPg && bIgnore && rtIsSeparator(pPtr->eType) ){
    rc = segmentPtrAdvance(pCsr, pPtr, bLast);
  }

#if 0
  if( bLast && rc==LSM_OK && pPtr->pPg
   && pPtr->pSeg==&pLvl->lhs 
   && pLvl->nRight && (pPtr->eType & LSM_START_DELETE)
  ){
    pPtr->iCell++;
    pPtr->eType = LSM_END_DELETE | (pLvl->iSplitTopic);
    pPtr->pKey = pLvl->pSplitKey;
    pPtr->nKey = pLvl->nSplitKey;
    pPtr->pVal = 0;
    pPtr->nVal = 0;
  }
#endif

  return rc;
}

static void segmentPtrKey(SegmentPtr *pPtr, void **ppKey, int *pnKey){
  assert( pPtr->pPg );
  *ppKey = pPtr->pKey;
  *pnKey = pPtr->nKey;
}

#if 0 /* NOT USED */
static char *keyToString(lsm_env *pEnv, void *pKey, int nKey){
  int i;
  u8 *aKey = (u8 *)pKey;
  char *zRet = (char *)lsmMalloc(pEnv, nKey+1);

  for(i=0; i<nKey; i++){
    zRet[i] = (char)(isalnum(aKey[i]) ? aKey[i] : '.');
  }
  zRet[nKey] = '\0';
  return zRet;
}
#endif

#if 0 /* NOT USED */
/*
** Check that the page that pPtr currently has loaded is the correct page
** to search for key (pKey/nKey). If it is, return 1. Otherwise, an assert
** fails and this function does not return.
*/
static int assertKeyLocation(
  MultiCursor *pCsr, 
  SegmentPtr *pPtr, 
  void *pKey, int nKey
){
  lsm_env *pEnv = lsmFsEnv(pCsr->pDb->pFS);
  Blob blob = {0, 0, 0};
  int eDir;
  int iTopic = 0;                 /* TODO: Fix me */

  for(eDir=-1; eDir<=1; eDir+=2){
    Page *pTest = pPtr->pPg;

    lsmFsPageRef(pTest);
    while( pTest ){
      Segment *pSeg = pPtr->pSeg;
      Page *pNext;

      int rc = lsmFsDbPageNext(pSeg, pTest, eDir, &pNext);
      lsmFsPageRelease(pTest);
      if( rc ) return 1;
      pTest = pNext;

      if( pTest ){
        int nData;
        u8 *aData = fsPageData(pTest, &nData);
        int nCell = pageGetNRec(aData, nData);
        int flags = pageGetFlags(aData, nData);
        if( nCell && 0==(flags&SEGMENT_BTREE_FLAG) ){
          int nPgKey;
          int iPgTopic;
          u8 *pPgKey;
          int res;
          int iCell;

          iCell = ((eDir < 0) ? (nCell-1) : 0);
          pPgKey = pageGetKey(pSeg, pTest, iCell, &iPgTopic, &nPgKey, &blob);
          res = iTopic - iPgTopic;
          if( res==0 ) res = pCsr->pDb->xCmp(pKey, nKey, pPgKey, nPgKey);
          if( (eDir==1 && res>0) || (eDir==-1 && res<0) ){
            /* Taking this branch means something has gone wrong. */
            char *zMsg = lsmMallocPrintf(pEnv, "Key \"%s\" is not on page %d", 
                keyToString(pEnv, pKey, nKey), lsmFsPageNumber(pPtr->pPg)
            );
            fprintf(stderr, "%s\n", zMsg);
            assert( !"assertKeyLocation() failed" );
          }
          lsmFsPageRelease(pTest);
          pTest = 0;
        }
      }
    }
  }

  sortedBlobFree(&blob);
  return 1;
}
#endif

#ifndef NDEBUG
static int assertSeekResult(
  MultiCursor *pCsr,
  SegmentPtr *pPtr,
  int iTopic,
  void *pKey,
  int nKey,
  int eSeek
){
  if( pPtr->pPg ){
    int res;
    res = sortedKeyCompare(pCsr->pDb->xCmp, iTopic, pKey, nKey,
        rtTopic(pPtr->eType), pPtr->pKey, pPtr->nKey
    );

    if( eSeek==LSM_SEEK_EQ ) return (res==0);
    if( eSeek==LSM_SEEK_LE ) return (res>=0);
    if( eSeek==LSM_SEEK_GE ) return (res<=0);
  }

  return 1;
}
#endif

static int segmentPtrSearchOversized(
  MultiCursor *pCsr,              /* Cursor context */
  SegmentPtr *pPtr,               /* Pointer to seek */
  int iTopic,                     /* Topic of key to search for */
  void *pKey, int nKey            /* Key to seek to */
){
  int (*xCmp)(void *, int, void *, int) = pCsr->pDb->xCmp;
  int rc = LSM_OK;

  /* If the OVERSIZED flag is set, then there is no pointer in the
  ** upper level to the next page in the segment that contains at least
  ** one key. So compare the largest key on the current page with the
  ** key being sought (pKey/nKey). If (pKey/nKey) is larger, advance
  ** to the next page in the segment that contains at least one key. 
  */
  while( rc==LSM_OK && (pPtr->flags & PGFTR_SKIP_NEXT_FLAG) ){
    u8 *pLastKey;
    int nLastKey;
    int iLastTopic;
    int res;                      /* Result of comparison */
    Page *pNext;

    /* Load the last key on the current page. */
    pLastKey = pageGetKey(pPtr->pSeg,
        pPtr->pPg, pPtr->nCell-1, &iLastTopic, &nLastKey, &pPtr->blob1
    );

    /* If the loaded key is >= than (pKey/nKey), break out of the loop.
    ** If (pKey/nKey) is present in this array, it must be on the current 
    ** page.  */
    res = sortedKeyCompare(
        xCmp, iLastTopic, pLastKey, nLastKey, iTopic, pKey, nKey
    );
    if( res>=0 ) break;

    /* Advance to the next page that contains at least one key. */
    pNext = pPtr->pPg;
    lsmFsPageRef(pNext);
    while( 1 ){
      Page *pLoad;
      u8 *aData; int nData;

      rc = lsmFsDbPageNext(pPtr->pSeg, pNext, 1, &pLoad);
      lsmFsPageRelease(pNext);
      pNext = pLoad;
      if( pNext==0 ) break;

      assert( rc==LSM_OK );
      aData = lsmFsPageData(pNext, &nData);
      if( (pageGetFlags(aData, nData) & SEGMENT_BTREE_FLAG)==0
       && pageGetNRec(aData, nData)>0
      ){
        break;
      }
    }
    if( pNext==0 ) break;
    segmentPtrSetPage(pPtr, pNext);

    /* This should probably be an LSM_CORRUPT error. */
    assert( rc!=LSM_OK || (pPtr->flags & PGFTR_SKIP_THIS_FLAG) );
  }

  return rc;
}

static int ptrFwdPointer(
  Page *pPage,
  int iCell,
  Segment *pSeg,
  Pgno *piPtr,
  int *pbFound
){
  Page *pPg = pPage;
  int iFirst = iCell;
  int rc = LSM_OK;

  do {
    Page *pNext = 0;
    u8 *aData;
    int nData;

    aData = lsmFsPageData(pPg, &nData);
    if( (pageGetFlags(aData, nData) & SEGMENT_BTREE_FLAG)==0 ){
      int i;
      int nCell = pageGetNRec(aData, nData);
      for(i=iFirst; i<nCell; i++){
        u8 eType = *pageGetCell(aData, nData, i);
        if( (eType & LSM_START_DELETE)==0 ){
          *pbFound = 1;
          *piPtr = pageGetRecordPtr(aData, nData, i) + pageGetPtr(aData, nData);
          lsmFsPageRelease(pPg);
          return LSM_OK;
        }
      }
    }

    rc = lsmFsDbPageNext(pSeg, pPg, 1, &pNext);
    lsmFsPageRelease(pPg);
    pPg = pNext;
    iFirst = 0;
  }while( pPg && rc==LSM_OK );
  lsmFsPageRelease(pPg);

  *pbFound = 0;
  return rc;
}

static int sortedRhsFirst(MultiCursor *pCsr, Level *pLvl, SegmentPtr *pPtr){
  int rc;
  rc = segmentPtrEnd(pCsr, pPtr, 0);
  while( pPtr->pPg && rc==LSM_OK ){
    int res = sortedKeyCompare(pCsr->pDb->xCmp,
        pLvl->iSplitTopic, pLvl->pSplitKey, pLvl->nSplitKey,
        rtTopic(pPtr->eType), pPtr->pKey, pPtr->nKey
    );
    if( res<=0 ) break;
    rc = segmentPtrAdvance(pCsr, pPtr, 0);
  }
  return rc;
}


/*
** This function is called as part of a SEEK_GE op on a multi-cursor if the 
** FC pointer read from segment *pPtr comes from an entry with the 
** LSM_START_DELETE flag set. In this case the pointer value cannot be 
** trusted. Instead, the pointer that should be followed is that associated
** with the next entry in *pPtr that does not have LSM_START_DELETE set.
**
** Why the pointers can't be trusted:
**
**
**
** TODO: This is a stop-gap solution:
** 
**   At the moment, this function is called from within segmentPtrSeek(), 
**   as part of the initial lsmMCursorSeek() call. However, consider a 
**   database where the following has occurred:
**
**      1. A range delete removes keys 1..9999 using a range delete.
**      2. Keys 1 through 9999 are reinserted.
**      3. The levels containing the ops in 1. and 2. above are merged. Call
**         this level N. Level N contains FC pointers to level N+1.
**
**   Then, if the user attempts to query for (key>=2 LIMIT 10), the 
**   lsmMCursorSeek() call will iterate through 9998 entries searching for a 
**   pointer down to the level N+1 that is never actually used. It would be
**   much better if the multi-cursor could do this lazily - only seek to the
**   level (N+1) page after the user has moved the cursor on level N passed
**   the big range-delete.
*/
static int segmentPtrFwdPointer(
  MultiCursor *pCsr,              /* Multi-cursor pPtr belongs to */
  SegmentPtr *pPtr,               /* Segment-pointer to extract FC ptr from */
  Pgno *piPtr                     /* OUT: FC pointer value */
){
  Level *pLvl = pPtr->pLevel;
  Level *pNext = pLvl->pNext;
  Page *pPg = pPtr->pPg;
  int rc;
  int bFound;
  Pgno iOut = 0;

  if( pPtr->pSeg==&pLvl->lhs || pPtr->pSeg==&pLvl->aRhs[pLvl->nRight-1] ){
    if( pNext==0 
        || (pNext->nRight==0 && pNext->lhs.iRoot)
        || (pNext->nRight!=0 && pNext->aRhs[0].iRoot)
      ){
      /* Do nothing. The pointer will not be used anyway. */
      return LSM_OK;
    }
  }else{
    if( pPtr[1].pSeg->iRoot ){
      return LSM_OK;
    }
  }

  /* Search for a pointer within the current segment. */
  lsmFsPageRef(pPg);
  rc = ptrFwdPointer(pPg, pPtr->iCell, pPtr->pSeg, &iOut, &bFound);

  if( rc==LSM_OK && bFound==0 ){
    /* This case happens when pPtr points to the left-hand-side of a segment
    ** currently undergoing an incremental merge. In this case, jump to the
    ** oldest segment in the right-hand-side of the same level and continue
    ** searching. But - do not consider any keys smaller than the levels
    ** split-key. */
    SegmentPtr ptr;

    if( pPtr->pLevel->nRight==0 || pPtr->pSeg!=&pPtr->pLevel->lhs ){
      return LSM_CORRUPT_BKPT;
    }

    memset(&ptr, 0, sizeof(SegmentPtr));
    ptr.pLevel = pPtr->pLevel;
    ptr.pSeg = &ptr.pLevel->aRhs[ptr.pLevel->nRight-1];
    rc = sortedRhsFirst(pCsr, ptr.pLevel, &ptr);
    if( rc==LSM_OK ){
      rc = ptrFwdPointer(ptr.pPg, ptr.iCell, ptr.pSeg, &iOut, &bFound);
      ptr.pPg = 0;
    }
    segmentPtrReset(&ptr);
  }

  *piPtr = iOut;
  return rc;
}

static int segmentPtrSeek(
  MultiCursor *pCsr,              /* Cursor context */
  SegmentPtr *pPtr,               /* Pointer to seek */
  int iTopic,                     /* Key topic to seek to */
  void *pKey, int nKey,           /* Key to seek to */
  int eSeek,                      /* Search bias - see above */
  int *piPtr,                     /* OUT: FC pointer */
  int *pbStop
){
  int (*xCmp)(void *, int, void *, int) = pCsr->pDb->xCmp;
  int res;                        /* Result of comparison operation */
  int rc = LSM_OK;
  int iMin;
  int iMax;
  Pgno iPtrOut = 0;

  /* If the current page contains an oversized entry, then there are no
  ** pointers to one or more of the subsequent pages in the sorted run.
  ** The following call ensures that the segment-ptr points to the correct 
  ** page in this case.  */
  rc = segmentPtrSearchOversized(pCsr, pPtr, iTopic, pKey, nKey);
  iPtrOut = pPtr->iPtr;

  /* Assert that this page is the right page of this segment for the key
  ** that we are searching for. Do this by loading page (iPg-1) and testing
  ** that pKey/nKey is greater than all keys on that page, and then by 
  ** loading (iPg+1) and testing that pKey/nKey is smaller than all
  ** the keys it houses.  
  **
  ** TODO: With range-deletes in the tree, the test described above may fail.
  */
#if 0
  assert( assertKeyLocation(pCsr, pPtr, pKey, nKey) );
#endif

  assert( pPtr->nCell>0 
       || pPtr->pSeg->nSize==1 
       || lsmFsDbPageIsLast(pPtr->pSeg, pPtr->pPg)
  );
  if( pPtr->nCell==0 ){
    segmentPtrReset(pPtr);
  }else{
    iMin = 0;
    iMax = pPtr->nCell-1;

    while( 1 ){
      int iTry = (iMin+iMax)/2;
      void *pKeyT; int nKeyT;       /* Key for cell iTry */
      int iTopicT;

      assert( iTry<iMax || iMin==iMax );

      rc = segmentPtrLoadCell(pPtr, iTry);
      if( rc!=LSM_OK ) break;

      segmentPtrKey(pPtr, &pKeyT, &nKeyT);
      iTopicT = rtTopic(pPtr->eType);

      res = sortedKeyCompare(xCmp, iTopicT, pKeyT, nKeyT, iTopic, pKey, nKey);
      if( res<=0 ){
        iPtrOut = pPtr->iPtr + pPtr->iPgPtr;
      }

      if( res==0 || iMin==iMax ){
        break;
      }else if( res>0 ){
        iMax = LSM_MAX(iTry-1, iMin);
      }else{
        iMin = iTry+1;
      }
    }

    if( rc==LSM_OK ){
      assert( res==0 || (iMin==iMax && iMin>=0 && iMin<pPtr->nCell) );
      if( res ){
        rc = segmentPtrLoadCell(pPtr, iMin);
      }
      assert( rc!=LSM_OK || res>0 || iPtrOut==(pPtr->iPtr + pPtr->iPgPtr) );

      if( rc==LSM_OK ){
        switch( eSeek ){
          case LSM_SEEK_EQ: {
            int eType = pPtr->eType;
            if( (res<0 && (eType & LSM_START_DELETE))
             || (res>0 && (eType & LSM_END_DELETE))
             || (res==0 && (eType & LSM_POINT_DELETE))
            ){
              *pbStop = 1;
            }else if( res==0 && (eType & LSM_INSERT) ){
              lsm_env *pEnv = pCsr->pDb->pEnv;
              *pbStop = 1;
              pCsr->eType = pPtr->eType;
              rc = sortedBlobSet(pEnv, &pCsr->key, pPtr->pKey, pPtr->nKey);
              if( rc==LSM_OK ){
                rc = sortedBlobSet(pEnv, &pCsr->val, pPtr->pVal, pPtr->nVal);
              }
              pCsr->flags |= CURSOR_SEEK_EQ;
            }
            segmentPtrReset(pPtr);
            break;
          }
          case LSM_SEEK_LE:
            if( res>0 ) rc = segmentPtrAdvance(pCsr, pPtr, 1);
            break;
          case LSM_SEEK_GE: {
            /* Figure out if we need to 'skip' the pointer forward or not */
            if( (res<=0 && (pPtr->eType & LSM_START_DELETE)) 
             || (res>0  && (pPtr->eType & LSM_END_DELETE)) 
            ){
              rc = segmentPtrFwdPointer(pCsr, pPtr, &iPtrOut);
            }
            if( res<0 && rc==LSM_OK ){
              rc = segmentPtrAdvance(pCsr, pPtr, 0);
            }
            break;
          }
        }
      }
    }

    /* If the cursor seek has found a separator key, and this cursor is
    ** supposed to ignore separators keys, advance to the next entry.  */
    if( rc==LSM_OK && pPtr->pPg
     && segmentPtrIgnoreSeparators(pCsr, pPtr) 
     && rtIsSeparator(pPtr->eType)
    ){
      assert( eSeek!=LSM_SEEK_EQ );
      rc = segmentPtrAdvance(pCsr, pPtr, eSeek==LSM_SEEK_LE);
    }
  }

  assert( rc!=LSM_OK || assertSeekResult(pCsr,pPtr,iTopic,pKey,nKey,eSeek) );
  *piPtr = iPtrOut;
  return rc;
}

static int seekInBtree(
  MultiCursor *pCsr,              /* Multi-cursor object */
  Segment *pSeg,                  /* Seek within this segment */
  int iTopic,
  void *pKey, int nKey,           /* Key to seek to */
  Pgno *aPg,                      /* OUT: Page numbers */
  Page **ppPg                     /* OUT: Leaf (sorted-run) page reference */
){
  int i = 0;
  int rc;
  int iPg;
  Page *pPg = 0;
  Blob blob = {0, 0, 0};

  iPg = pSeg->iRoot;
  do {
    Pgno *piFirst = 0;
    if( aPg ){
      aPg[i++] = iPg;
      piFirst = &aPg[i];
    }

    rc = lsmFsDbPageGet(pCsr->pDb->pFS, pSeg, iPg, &pPg);
    assert( rc==LSM_OK || pPg==0 );
    if( rc==LSM_OK ){
      u8 *aData;                  /* Buffer containing page data */
      int nData;                  /* Size of aData[] in bytes */
      int iMin;
      int iMax;
      int nRec;
      int flags;

      aData = fsPageData(pPg, &nData);
      flags = pageGetFlags(aData, nData);
      if( (flags & SEGMENT_BTREE_FLAG)==0 ) break;

      iPg = pageGetPtr(aData, nData);
      nRec = pageGetNRec(aData, nData);

      iMin = 0;
      iMax = nRec-1;
      while( iMax>=iMin ){
        int iTry = (iMin+iMax)/2;
        void *pKeyT; int nKeyT;       /* Key for cell iTry */
        int iTopicT;                  /* Topic for key pKeyT/nKeyT */
        Pgno iPtr;                    /* Pointer associated with cell iTry */
        int res;                      /* (pKey - pKeyT) */

        rc = pageGetBtreeKey(
            pSeg, pPg, iTry, &iPtr, &iTopicT, &pKeyT, &nKeyT, &blob
        );
        if( rc!=LSM_OK ) break;
        if( piFirst && pKeyT==blob.pData ){
          *piFirst = pageGetBtreeRef(pPg, iTry);
          piFirst = 0;
          i++;
        }

        res = sortedKeyCompare(
            pCsr->pDb->xCmp, iTopic, pKey, nKey, iTopicT, pKeyT, nKeyT
        );
        if( res<0 ){
          iPg = iPtr;
          iMax = iTry-1;
        }else{
          iMin = iTry+1;
        }
      }
      lsmFsPageRelease(pPg);
      pPg = 0;
    }
  }while( rc==LSM_OK );

  sortedBlobFree(&blob);
  assert( (rc==LSM_OK)==(pPg!=0) );
  if( ppPg ){
    *ppPg = pPg;
  }else{
    lsmFsPageRelease(pPg);
  }
  return rc;
}

static int seekInSegment(
  MultiCursor *pCsr, 
  SegmentPtr *pPtr,
  int iTopic,
  void *pKey, int nKey,
  int iPg,                        /* Page to search */
  int eSeek,                      /* Search bias - see above */
  int *piPtr,                     /* OUT: FC pointer */
  int *pbStop                     /* OUT: Stop search flag */
){
  int iPtr = iPg;
  int rc = LSM_OK;

  if( pPtr->pSeg->iRoot ){
    Page *pPg;
    assert( pPtr->pSeg->iRoot!=0 );
    rc = seekInBtree(pCsr, pPtr->pSeg, iTopic, pKey, nKey, 0, &pPg);
    if( rc==LSM_OK ) segmentPtrSetPage(pPtr, pPg);
  }else{
    if( iPtr==0 ){
      iPtr = pPtr->pSeg->iFirst;
    }
    if( rc==LSM_OK ){
      rc = segmentPtrLoadPage(pCsr->pDb->pFS, pPtr, iPtr);
    }
  }

  if( rc==LSM_OK ){
    rc = segmentPtrSeek(pCsr, pPtr, iTopic, pKey, nKey, eSeek, piPtr, pbStop);
  }
  return rc;
}

/*
** Seek each segment pointer in the array of (pLvl->nRight+1) at aPtr[].
**
** pbStop:
**   This parameter is only significant if parameter eSeek is set to
**   LSM_SEEK_EQ. In this case, it is set to true before returning if
**   the seek operation is finished. This can happen in two ways:
**   
**     a) A key matching (pKey/nKey) is found, or
**     b) A point-delete or range-delete deleting the key is found.
**
**   In case (a), the multi-cursor CURSOR_SEEK_EQ flag is set and the pCsr->key
**   and pCsr->val blobs populated before returning.
*/
static int seekInLevel(
  MultiCursor *pCsr,              /* Sorted cursor object to seek */
  SegmentPtr *aPtr,               /* Pointer to array of (nRhs+1) SPs */
  int eSeek,                      /* Search bias - see above */
  int iTopic,                     /* Key topic to search for */
  void *pKey, int nKey,           /* Key to search for */
  Pgno *piPgno,                   /* IN/OUT: fraction cascade pointer (or 0) */
  int *pbStop                     /* OUT: See above */
){
  Level *pLvl = aPtr[0].pLevel;   /* Level to seek within */
  int rc = LSM_OK;                /* Return code */
  int iOut = 0;                   /* Pointer to return to caller */
  int res = -1;                   /* Result of xCmp(pKey, split) */
  int nRhs = pLvl->nRight;        /* Number of right-hand-side segments */
  int bStop = 0;

  /* If this is a composite level (one currently undergoing an incremental
  ** merge), figure out if the search key is larger or smaller than the
  ** levels split-key.  */
  if( nRhs ){
    res = sortedKeyCompare(pCsr->pDb->xCmp, iTopic, pKey, nKey, 
        pLvl->iSplitTopic, pLvl->pSplitKey, pLvl->nSplitKey
    );
  }

  /* If (res<0), then key pKey/nKey is smaller than the split-key (or this
  ** is not a composite level and there is no split-key). Search the 
  ** left-hand-side of the level in this case.  */
  if( res<0 ){
    int iPtr = 0;
    if( nRhs==0 ) iPtr = *piPgno;

    rc = seekInSegment(
        pCsr, &aPtr[0], iTopic, pKey, nKey, iPtr, eSeek, &iOut, &bStop
    );
    if( rc==LSM_OK && nRhs>0 && eSeek==LSM_SEEK_GE && aPtr[0].pPg==0 ){
      res = 0;
    }
  }
  
  if( res>=0 ){
    int bHit = 0;                 /* True if at least one rhs is not EOF */
    int iPtr = *piPgno;
    int i;
    for(i=1; rc==LSM_OK && i<=nRhs && bStop==0; i++){
      SegmentPtr *pPtr = &aPtr[i];
      iOut = 0;
      rc = seekInSegment(
          pCsr, pPtr, iTopic, pKey, nKey, iPtr, eSeek, &iOut, &bStop
      );
      iPtr = iOut;

      /* If the segment-pointer has settled on a key that is smaller than
      ** the splitkey, invalidate the segment-pointer.  */
      if( pPtr->pPg ){
        res = sortedKeyCompare(pCsr->pDb->xCmp, 
            rtTopic(pPtr->eType), pPtr->pKey, pPtr->nKey, 
            pLvl->iSplitTopic, pLvl->pSplitKey, pLvl->nSplitKey
        );
        if( res<0 ) segmentPtrReset(pPtr);
      }

      if( aPtr[i].pKey ) bHit = 1;
    }

    if( rc==LSM_OK && eSeek==LSM_SEEK_LE && bHit==0 ){
      rc = segmentPtrEnd(pCsr, &aPtr[0], 1);
    }
  }

  assert( eSeek==LSM_SEEK_EQ || bStop==0 );
  *piPgno = iOut;
  *pbStop = bStop;
  return rc;
}

static void multiCursorGetKey(
  MultiCursor *pCsr, 
  int iKey,
  int *peType,                    /* OUT: Key type (SORTED_WRITE etc.) */
  void **ppKey,                   /* OUT: Pointer to buffer containing key */
  int *pnKey                      /* OUT: Size of *ppKey in bytes */
){
  int nKey = 0;
  void *pKey = 0;
  int eType = 0;

  switch( iKey ){
    case CURSOR_DATA_TREE0:
    case CURSOR_DATA_TREE1: {
      TreeCursor *pTreeCsr = pCsr->apTreeCsr[iKey-CURSOR_DATA_TREE0];
      if( lsmTreeCursorValid(pTreeCsr) ){
        lsmTreeCursorKey(pTreeCsr, &eType, &pKey, &nKey);
      }
      break;
    }

    case CURSOR_DATA_SYSTEM: {
      Snapshot *pWorker = pCsr->pDb->pWorker;
      if( pWorker && (pCsr->flags & CURSOR_FLUSH_FREELIST) ){
        int nEntry = pWorker->freelist.nEntry;
        if( pCsr->iFree < (nEntry*2) ){
          FreelistEntry *aEntry = pWorker->freelist.aEntry;
          int i = nEntry - 1 - (pCsr->iFree / 2);
          u32 iKey = 0;

          if( (pCsr->iFree % 2) ){
            eType = LSM_END_DELETE|LSM_SYSTEMKEY;
            iKey = aEntry[i].iBlk-1;
          }else if( aEntry[i].iId>=0 ){
            eType = LSM_INSERT|LSM_SYSTEMKEY;
            iKey = aEntry[i].iBlk;

            /* If the in-memory entry immediately before this one was a
             ** DELETE, and the block number is one greater than the current
             ** block number, mark this entry as an "end-delete-range". */
            if( i<(nEntry-1) && aEntry[i+1].iBlk==iKey+1 && aEntry[i+1].iId<0 ){
              eType |= LSM_END_DELETE;
            }

          }else{
            eType = LSM_START_DELETE|LSM_SYSTEMKEY;
            iKey = aEntry[i].iBlk + 1;
          }

          /* If the in-memory entry immediately after this one is a
          ** DELETE, and the block number is one less than the current
          ** key, mark this entry as an "start-delete-range".  */
          if( i>0 && aEntry[i-1].iBlk==iKey-1 && aEntry[i-1].iId<0 ){
            eType |= LSM_START_DELETE;
          }

          pKey = pCsr->pSystemVal;
          nKey = 4;
          lsmPutU32(pKey, ~iKey);
        }
      }
      break;
    }

    default: {
      int iPtr = iKey - CURSOR_DATA_SEGMENT;
      assert( iPtr>=0 );
      if( iPtr==pCsr->nPtr ){
        if( pCsr->pBtCsr ){
          pKey = pCsr->pBtCsr->pKey;
          nKey = pCsr->pBtCsr->nKey;
          eType = pCsr->pBtCsr->eType;
        }
      }else if( iPtr<pCsr->nPtr ){
        SegmentPtr *pPtr = &pCsr->aPtr[iPtr];
        if( pPtr->pPg ){
          pKey = pPtr->pKey;
          nKey = pPtr->nKey;
          eType = pPtr->eType;
        }
      }
      break;
    }
  }

  if( peType ) *peType = eType;
  if( pnKey ) *pnKey = nKey;
  if( ppKey ) *ppKey = pKey;
}

static int sortedDbKeyCompare(
  MultiCursor *pCsr,
  int iLhsFlags, void *pLhsKey, int nLhsKey,
  int iRhsFlags, void *pRhsKey, int nRhsKey
){
  int (*xCmp)(void *, int, void *, int) = pCsr->pDb->xCmp;
  int res;

  /* Compare the keys, including the system flag. */
  res = sortedKeyCompare(xCmp, 
    rtTopic(iLhsFlags), pLhsKey, nLhsKey,
    rtTopic(iRhsFlags), pRhsKey, nRhsKey
  );

  /* If a key has the LSM_START_DELETE flag set, but not the LSM_INSERT or
  ** LSM_POINT_DELETE flags, it is considered a delta larger. This prevents
  ** the beginning of an open-ended set from masking a database entry or
  ** delete at a lower level.  */
  if( res==0 && (pCsr->flags & CURSOR_IGNORE_DELETE) ){
    const int m = LSM_POINT_DELETE|LSM_INSERT|LSM_END_DELETE |LSM_START_DELETE;
    int iDel1 = 0;
    int iDel2 = 0;

    if( LSM_START_DELETE==(iLhsFlags & m) ) iDel1 = +1;
    if( LSM_END_DELETE  ==(iLhsFlags & m) ) iDel1 = -1;
    if( LSM_START_DELETE==(iRhsFlags & m) ) iDel2 = +1;
    if( LSM_END_DELETE  ==(iRhsFlags & m) ) iDel2 = -1;

    res = (iDel1 - iDel2);
  }

  return res;
}

static void multiCursorDoCompare(MultiCursor *pCsr, int iOut, int bReverse){
  int i1;
  int i2;
  int iRes;
  void *pKey1; int nKey1; int eType1;
  void *pKey2; int nKey2; int eType2;
  const int mul = (bReverse ? -1 : 1);

  assert( pCsr->aTree && iOut<pCsr->nTree );
  if( iOut>=(pCsr->nTree/2) ){
    i1 = (iOut - pCsr->nTree/2) * 2;
    i2 = i1 + 1;
  }else{
    i1 = pCsr->aTree[iOut*2];
    i2 = pCsr->aTree[iOut*2+1];
  }

  multiCursorGetKey(pCsr, i1, &eType1, &pKey1, &nKey1);
  multiCursorGetKey(pCsr, i2, &eType2, &pKey2, &nKey2);

  if( pKey1==0 ){
    iRes = i2;
  }else if( pKey2==0 ){
    iRes = i1;
  }else{
    int res;

    /* Compare the keys */
    res = sortedDbKeyCompare(pCsr,
        eType1, pKey1, nKey1, eType2, pKey2, nKey2
    );

    res = res * mul;
    if( res==0 ){
      /* The two keys are identical. Normally, this means that the key from
      ** the newer run clobbers the old. However, if the newer key is a
      ** separator key, or a range-delete-boundary only, do not allow it
      ** to clobber an older entry.  */
      int nc1 = (eType1 & (LSM_INSERT|LSM_POINT_DELETE))==0;
      int nc2 = (eType2 & (LSM_INSERT|LSM_POINT_DELETE))==0;
      iRes = (nc1 > nc2) ? i2 : i1;
    }else if( res<0 ){
      iRes = i1;
    }else{
      iRes = i2;
    }
  }

  pCsr->aTree[iOut] = iRes;
}

/*
** This function advances segment pointer iPtr belonging to multi-cursor
** pCsr forward (bReverse==0) or backward (bReverse!=0).
**
** If the segment pointer points to a segment that is part of a composite
** level, then the following special case is handled.
**
**   * If iPtr is the lhs of a composite level, and the cursor is being
**     advanced forwards, and segment iPtr is at EOF, move all pointers
**     that correspond to rhs segments of the same level to the first
**     key in their respective data.
*/
static int segmentCursorAdvance(
  MultiCursor *pCsr, 
  int iPtr,
  int bReverse
){
  int rc;
  SegmentPtr *pPtr = &pCsr->aPtr[iPtr];
  Level *pLvl = pPtr->pLevel;
  int bComposite;                 /* True if pPtr is part of composite level */

  /* Advance the segment-pointer object. */
  rc = segmentPtrAdvance(pCsr, pPtr, bReverse);
  if( rc!=LSM_OK ) return rc;

  bComposite = (pLvl->nRight>0 && pCsr->nPtr>pLvl->nRight);
  if( bComposite && pPtr->pPg==0 ){
    int bFix = 0;
    if( (bReverse==0)==(pPtr->pSeg==&pLvl->lhs) ){
      int i;
      if( bReverse ){
        SegmentPtr *pLhs = &pCsr->aPtr[iPtr - 1 - (pPtr->pSeg - pLvl->aRhs)];
        for(i=0; i<pLvl->nRight; i++){
          if( pLhs[i+1].pPg ) break;
        }
        if( i==pLvl->nRight ){
          bFix = 1;
          rc = segmentPtrEnd(pCsr, pLhs, 1);
        }
      }else{
        bFix = 1;
        for(i=0; rc==LSM_OK && i<pLvl->nRight; i++){
          rc = sortedRhsFirst(pCsr, pLvl, &pCsr->aPtr[iPtr+1+i]);
        }
      }
    }

    if( bFix ){
      int i;
      for(i=pCsr->nTree-1; i>0; i--){
        multiCursorDoCompare(pCsr, i, bReverse);
      }
    }
  }

#if 0
  if( bComposite && pPtr->pSeg==&pLvl->lhs       /* lhs of composite level */
   && bReverse==0                                /* csr advanced forwards */
   && pPtr->pPg==0                               /* segment at EOF */
  ){
    int i;
    for(i=0; rc==LSM_OK && i<pLvl->nRight; i++){
      rc = sortedRhsFirst(pCsr, pLvl, &pCsr->aPtr[iPtr+1+i]);
    }
    for(i=pCsr->nTree-1; i>0; i--){
      multiCursorDoCompare(pCsr, i, 0);
    }
  }
#endif

  return rc;
}

static void mcursorFreeComponents(MultiCursor *pCsr){
  int i;
  lsm_env *pEnv = pCsr->pDb->pEnv;

  /* Close the tree cursor, if any. */
  lsmTreeCursorDestroy(pCsr->apTreeCsr[0]);
  lsmTreeCursorDestroy(pCsr->apTreeCsr[1]);

  /* Reset the segment pointers */
  for(i=0; i<pCsr->nPtr; i++){
    segmentPtrReset(&pCsr->aPtr[i]);
  }

  /* And the b-tree cursor, if any */
  btreeCursorFree(pCsr->pBtCsr);

  /* Free allocations */
  lsmFree(pEnv, pCsr->aPtr);
  lsmFree(pEnv, pCsr->aTree);
  lsmFree(pEnv, pCsr->pSystemVal);

  /* Zero fields */
  pCsr->nPtr = 0;
  pCsr->aPtr = 0;
  pCsr->nTree = 0;
  pCsr->aTree = 0;
  pCsr->pSystemVal = 0;
  pCsr->apTreeCsr[0] = 0;
  pCsr->apTreeCsr[1] = 0;
  pCsr->pBtCsr = 0;
}

void lsmMCursorFreeCache(lsm_db *pDb){
  MultiCursor *p;
  MultiCursor *pNext;
  for(p=pDb->pCsrCache; p; p=pNext){
    pNext = p->pNext;
    lsmMCursorClose(p, 0);
  }
  pDb->pCsrCache = 0;
}

/*
** Close the cursor passed as the first argument.
**
** If the bCache parameter is true, then shift the cursor to the pCsrCache
** list for possible reuse instead of actually deleting it.
*/
void lsmMCursorClose(MultiCursor *pCsr, int bCache){
  if( pCsr ){
    lsm_db *pDb = pCsr->pDb;
    MultiCursor **pp;             /* Iterator variable */

    /* The cursor may or may not be currently part of the linked list 
    ** starting at lsm_db.pCsr. If it is, extract it.  */
    for(pp=&pDb->pCsr; *pp; pp=&((*pp)->pNext)){
      if( *pp==pCsr ){
        *pp = pCsr->pNext;
        break;
      }
    }

    if( bCache ){
      int i;                      /* Used to iterate through segment-pointers */

      /* Release any page references held by this cursor. */
      assert( !pCsr->pBtCsr );
      for(i=0; i<pCsr->nPtr; i++){
        SegmentPtr *pPtr = &pCsr->aPtr[i];
        lsmFsPageRelease(pPtr->pPg);
        pPtr->pPg = 0;
      }

      /* Reset the tree cursors */
      lsmTreeCursorReset(pCsr->apTreeCsr[0]);
      lsmTreeCursorReset(pCsr->apTreeCsr[1]);

      /* Add the cursor to the pCsrCache list */
      pCsr->pNext = pDb->pCsrCache;
      pDb->pCsrCache = pCsr;
    }else{
      /* Free the allocation used to cache the current key, if any. */
      sortedBlobFree(&pCsr->key);
      sortedBlobFree(&pCsr->val);

      /* Free the component cursors */
      mcursorFreeComponents(pCsr);

      /* Free the cursor structure itself */
      lsmFree(pDb->pEnv, pCsr);
    }
  }
}

#define TREE_NONE 0
#define TREE_OLD  1
#define TREE_BOTH 2

/*
** Parameter eTree is one of TREE_OLD or TREE_BOTH.
*/
static int multiCursorAddTree(MultiCursor *pCsr, Snapshot *pSnap, int eTree){
  int rc = LSM_OK;
  lsm_db *db = pCsr->pDb;

  /* Add a tree cursor on the 'old' tree, if it exists. */
  if( eTree!=TREE_NONE 
   && lsmTreeHasOld(db) 
   && db->treehdr.iOldLog!=pSnap->iLogOff 
  ){
    rc = lsmTreeCursorNew(db, 1, &pCsr->apTreeCsr[1]);
  }

  /* Add a tree cursor on the 'current' tree, if required. */
  if( rc==LSM_OK && eTree==TREE_BOTH ){
    rc = lsmTreeCursorNew(db, 0, &pCsr->apTreeCsr[0]);
  }

  return rc;
}

static int multiCursorAddRhs(MultiCursor *pCsr, Level *pLvl){
  int i;
  int nRhs = pLvl->nRight;

  assert( pLvl->nRight>0 );
  assert( pCsr->aPtr==0 );
  pCsr->aPtr = lsmMallocZero(pCsr->pDb->pEnv, sizeof(SegmentPtr) * nRhs);
  if( !pCsr->aPtr ) return LSM_NOMEM_BKPT;
  pCsr->nPtr = nRhs;

  for(i=0; i<nRhs; i++){
    pCsr->aPtr[i].pSeg = &pLvl->aRhs[i];
    pCsr->aPtr[i].pLevel = pLvl;
  }

  return LSM_OK;
}

static void multiCursorAddOne(MultiCursor *pCsr, Level *pLvl, int *pRc){
  if( *pRc==LSM_OK ){
    int iPtr = pCsr->nPtr;
    int i;
    pCsr->aPtr[iPtr].pLevel = pLvl;
    pCsr->aPtr[iPtr].pSeg = &pLvl->lhs;
    iPtr++;
    for(i=0; i<pLvl->nRight; i++){
      pCsr->aPtr[iPtr].pLevel = pLvl;
      pCsr->aPtr[iPtr].pSeg = &pLvl->aRhs[i];
      iPtr++;
    }

    if( pLvl->nRight && pLvl->pSplitKey==0 ){
      sortedSplitkey(pCsr->pDb, pLvl, pRc);
    }
    pCsr->nPtr = iPtr;
  }
}

static int multiCursorAddAll(MultiCursor *pCsr, Snapshot *pSnap){
  Level *pLvl;
  int nPtr = 0;
  int rc = LSM_OK;

  for(pLvl=pSnap->pLevel; pLvl; pLvl=pLvl->pNext){
    /* If the LEVEL_INCOMPLETE flag is set, then this function is being
    ** called (indirectly) from within a sortedNewToplevel() call to
    ** construct pLvl. In this case ignore pLvl - this cursor is going to
    ** be used to retrieve a freelist entry from the LSM, and the partially
    ** complete level may confuse it.  */
    if( pLvl->flags & LEVEL_INCOMPLETE ) continue;
    nPtr += (1 + pLvl->nRight);
  }

  assert( pCsr->aPtr==0 );
  pCsr->aPtr = lsmMallocZeroRc(pCsr->pDb->pEnv, sizeof(SegmentPtr) * nPtr, &rc);

  for(pLvl=pSnap->pLevel; pLvl; pLvl=pLvl->pNext){
    if( (pLvl->flags & LEVEL_INCOMPLETE)==0 ){
      multiCursorAddOne(pCsr, pLvl, &rc);
    }
  }

  return rc;
}

static int multiCursorInit(MultiCursor *pCsr, Snapshot *pSnap){
  int rc;
  rc = multiCursorAddAll(pCsr, pSnap);
  if( rc==LSM_OK ){
    rc = multiCursorAddTree(pCsr, pSnap, TREE_BOTH);
  }
  pCsr->flags |= (CURSOR_IGNORE_SYSTEM | CURSOR_IGNORE_DELETE);
  return rc;
}

static MultiCursor *multiCursorNew(lsm_db *db, int *pRc){
  MultiCursor *pCsr;
  pCsr = (MultiCursor *)lsmMallocZeroRc(db->pEnv, sizeof(MultiCursor), pRc);
  if( pCsr ){
    pCsr->pNext = db->pCsr;
    db->pCsr = pCsr;
    pCsr->pDb = db;
  }
  return pCsr;
}


void lsmSortedRemap(lsm_db *pDb){
  MultiCursor *pCsr;
  for(pCsr=pDb->pCsr; pCsr; pCsr=pCsr->pNext){
    int iPtr;
    if( pCsr->pBtCsr ){
      btreeCursorLoadKey(pCsr->pBtCsr);
    }
    for(iPtr=0; iPtr<pCsr->nPtr; iPtr++){
      segmentPtrLoadCell(&pCsr->aPtr[iPtr], pCsr->aPtr[iPtr].iCell);
    }
  }
}

static void multiCursorReadSeparators(MultiCursor *pCsr){
  if( pCsr->nPtr>0 ){
    pCsr->flags |= CURSOR_READ_SEPARATORS;
  }
}

/*
** Have this cursor skip over SORTED_DELETE entries.
*/
static void multiCursorIgnoreDelete(MultiCursor *pCsr){
  if( pCsr ) pCsr->flags |= CURSOR_IGNORE_DELETE;
}

/*
** If the free-block list is not empty, then have this cursor visit a key
** with (a) the system bit set, and (b) the key "FREELIST" and (c) a value 
** blob containing the serialized free-block list.
*/
static int multiCursorVisitFreelist(MultiCursor *pCsr){
  int rc = LSM_OK;
  pCsr->flags |= CURSOR_FLUSH_FREELIST;
  pCsr->pSystemVal = lsmMallocRc(pCsr->pDb->pEnv, 4 + 8, &rc);
  return rc;
}

/*
** Allocate and return a new database cursor.
**
** This method should only be called to allocate user cursors. As it may
** recycle a cursor from lsm_db.pCsrCache.
*/
int lsmMCursorNew(
  lsm_db *pDb,                    /* Database handle */
  MultiCursor **ppCsr             /* OUT: Allocated cursor */
){
  MultiCursor *pCsr = 0;
  int rc = LSM_OK;

  if( pDb->pCsrCache ){
    int bOld;                     /* True if there is an old in-memory tree */

    /* Remove a cursor from the pCsrCache list and add it to the open list. */
    pCsr = pDb->pCsrCache;
    pDb->pCsrCache = pCsr->pNext;
    pCsr->pNext = pDb->pCsr;
    pDb->pCsr = pCsr;

    /* The cursor can almost be used as is, except that the old in-memory
    ** tree cursor may be present and not required, or required and not
    ** present. Fix this if required.  */
    bOld = (lsmTreeHasOld(pDb) && pDb->treehdr.iOldLog!=pDb->pClient->iLogOff);
    if( !bOld && pCsr->apTreeCsr[1] ){
      lsmTreeCursorDestroy(pCsr->apTreeCsr[1]);
      pCsr->apTreeCsr[1] = 0;
    }else if( bOld && !pCsr->apTreeCsr[1] ){
      rc = lsmTreeCursorNew(pDb, 1, &pCsr->apTreeCsr[1]);
    }

    pCsr->flags = (CURSOR_IGNORE_SYSTEM | CURSOR_IGNORE_DELETE);

  }else{
    pCsr = multiCursorNew(pDb, &rc);
    if( rc==LSM_OK ) rc = multiCursorInit(pCsr, pDb->pClient);
  }

  if( rc!=LSM_OK ){
    lsmMCursorClose(pCsr, 0);
    pCsr = 0;
  }
  assert( (rc==LSM_OK)==(pCsr!=0) );
  *ppCsr = pCsr;
  return rc;
}

static int multiCursorGetVal(
  MultiCursor *pCsr, 
  int iVal, 
  void **ppVal, 
  int *pnVal
){
  int rc = LSM_OK;

  *ppVal = 0;
  *pnVal = 0;

  switch( iVal ){
    case CURSOR_DATA_TREE0:
    case CURSOR_DATA_TREE1: {
      TreeCursor *pTreeCsr = pCsr->apTreeCsr[iVal-CURSOR_DATA_TREE0];
      if( lsmTreeCursorValid(pTreeCsr) ){
        lsmTreeCursorValue(pTreeCsr, ppVal, pnVal);
      }else{
        *ppVal = 0;
        *pnVal = 0;
      }
      break;
    }

    case CURSOR_DATA_SYSTEM: {
      Snapshot *pWorker = pCsr->pDb->pWorker;
      if( pWorker 
       && (pCsr->iFree % 2)==0
       && pCsr->iFree < (pWorker->freelist.nEntry*2)
      ){
        int iEntry = pWorker->freelist.nEntry - 1 - (pCsr->iFree / 2);
        u8 *aVal = &((u8 *)(pCsr->pSystemVal))[4];
        lsmPutU64(aVal, pWorker->freelist.aEntry[iEntry].iId);
        *ppVal = aVal;
        *pnVal = 8;
      }
      break;
    }

    default: {
      int iPtr = iVal-CURSOR_DATA_SEGMENT;
      if( iPtr<pCsr->nPtr ){
        SegmentPtr *pPtr = &pCsr->aPtr[iPtr];
        if( pPtr->pPg ){
          *ppVal = pPtr->pVal;
          *pnVal = pPtr->nVal;
        }
      }
    }
  }

  assert( rc==LSM_OK || (*ppVal==0 && *pnVal==0) );
  return rc;
}

static int multiCursorAdvance(MultiCursor *pCsr, int bReverse);

/*
** This function is called by worker connections to walk the part of the
** free-list stored within the LSM data structure.
*/
int lsmSortedWalkFreelist(
  lsm_db *pDb,                    /* Database handle */
  int bReverse,                   /* True to iterate from largest to smallest */
  int (*x)(void *, int, i64),     /* Callback function */
  void *pCtx                      /* First argument to pass to callback */
){
  MultiCursor *pCsr;              /* Cursor used to read db */
  int rc = LSM_OK;                /* Return Code */
  Snapshot *pSnap = 0;

  assert( pDb->pWorker );
  if( pDb->bIncrMerge ){
    rc = lsmCheckpointDeserialize(pDb, 0, pDb->pShmhdr->aSnap1, &pSnap);
    if( rc!=LSM_OK ) return rc;
  }else{
    pSnap = pDb->pWorker;
  }

  pCsr = multiCursorNew(pDb, &rc);
  if( pCsr ){
    rc = multiCursorAddAll(pCsr, pSnap);
    pCsr->flags |= CURSOR_IGNORE_DELETE;
  }
  
  if( rc==LSM_OK ){
    if( bReverse==0 ){
      rc = lsmMCursorLast(pCsr);
    }else{
      rc = lsmMCursorSeek(pCsr, 1, "", 0, LSM_SEEK_GE);
    }

    while( rc==LSM_OK && lsmMCursorValid(pCsr) && rtIsSystem(pCsr->eType) ){
      void *pKey; int nKey;
      void *pVal; int nVal;

      rc = lsmMCursorKey(pCsr, &pKey, &nKey);
      if( rc==LSM_OK ) rc = lsmMCursorValue(pCsr, &pVal, &nVal);
      if( rc==LSM_OK && (nKey!=4 || nVal!=8) ) rc = LSM_CORRUPT_BKPT;

      if( rc==LSM_OK ){
        int iBlk;
        i64 iSnap;
        iBlk = (int)(~(lsmGetU32((u8 *)pKey)));
        iSnap = (i64)lsmGetU64((u8 *)pVal);
        if( x(pCtx, iBlk, iSnap) ) break;
        rc = multiCursorAdvance(pCsr, !bReverse);
      }
    }
  }

  lsmMCursorClose(pCsr, 0);
  if( pSnap!=pDb->pWorker ){
    lsmFreeSnapshot(pDb->pEnv, pSnap);
  }

  return rc;
}

int lsmSortedLoadFreelist(
  lsm_db *pDb,                    /* Database handle (must be worker) */
  void **ppVal,                   /* OUT: Blob containing LSM free-list */
  int *pnVal                      /* OUT: Size of *ppVal blob in bytes */
){
  MultiCursor *pCsr;              /* Cursor used to retreive free-list */
  int rc = LSM_OK;                /* Return Code */

  assert( pDb->pWorker );
  assert( *ppVal==0 && *pnVal==0 );

  pCsr = multiCursorNew(pDb, &rc);
  if( pCsr ){
    rc = multiCursorAddAll(pCsr, pDb->pWorker);
    pCsr->flags |= CURSOR_IGNORE_DELETE;
  }
  
  if( rc==LSM_OK ){
    rc = lsmMCursorLast(pCsr);
    if( rc==LSM_OK 
     && rtIsWrite(pCsr->eType) && rtIsSystem(pCsr->eType)
     && pCsr->key.nData==8 
     && 0==memcmp(pCsr->key.pData, "FREELIST", 8)
    ){
      void *pVal; int nVal;         /* Value read from database */
      rc = lsmMCursorValue(pCsr, &pVal, &nVal);
      if( rc==LSM_OK ){
        *ppVal = lsmMallocRc(pDb->pEnv, nVal, &rc);
        if( *ppVal ){
          memcpy(*ppVal, pVal, nVal);
          *pnVal = nVal;
        }
      }
    }

    lsmMCursorClose(pCsr, 0);
  }

  return rc;
}

static int multiCursorAllocTree(MultiCursor *pCsr){
  int rc = LSM_OK;
  if( pCsr->aTree==0 ){
    int nByte;                    /* Bytes of space to allocate */
    int nMin;                     /* Total number of cursors being merged */

    nMin = CURSOR_DATA_SEGMENT + pCsr->nPtr + (pCsr->pBtCsr!=0);
    pCsr->nTree = 2;
    while( pCsr->nTree<nMin ){
      pCsr->nTree = pCsr->nTree*2;
    }

    nByte = sizeof(int)*pCsr->nTree*2;
    pCsr->aTree = (int *)lsmMallocZeroRc(pCsr->pDb->pEnv, nByte, &rc);
  }
  return rc;
}

static void multiCursorCacheKey(MultiCursor *pCsr, int *pRc){
  if( *pRc==LSM_OK ){
    void *pKey;
    int nKey;
    multiCursorGetKey(pCsr, pCsr->aTree[1], &pCsr->eType, &pKey, &nKey);
    *pRc = sortedBlobSet(pCsr->pDb->pEnv, &pCsr->key, pKey, nKey);
  }
}

#ifdef LSM_DEBUG_EXPENSIVE
static void assertCursorTree(MultiCursor *pCsr){
  int bRev = !!(pCsr->flags & CURSOR_PREV_OK);
  int *aSave = pCsr->aTree;
  int nSave = pCsr->nTree;
  int rc;

  pCsr->aTree = 0;
  pCsr->nTree = 0;
  rc = multiCursorAllocTree(pCsr);
  if( rc==LSM_OK ){
    int i;
    for(i=pCsr->nTree-1; i>0; i--){
      multiCursorDoCompare(pCsr, i, bRev);
    }

    assert( nSave==pCsr->nTree 
        && 0==memcmp(aSave, pCsr->aTree, sizeof(int)*nSave)
    );

    lsmFree(pCsr->pDb->pEnv, pCsr->aTree);
  }

  pCsr->aTree = aSave;
  pCsr->nTree = nSave;
}
#else
# define assertCursorTree(x)
#endif

static int mcursorLocationOk(MultiCursor *pCsr, int bDeleteOk){
  int eType = pCsr->eType;
  int iKey;
  int i;
  int rdmask;
  
  assert( pCsr->flags & (CURSOR_NEXT_OK|CURSOR_PREV_OK) );
  assertCursorTree(pCsr);

  rdmask = (pCsr->flags & CURSOR_NEXT_OK) ? LSM_END_DELETE : LSM_START_DELETE;

  /* If the cursor does not currently point to an actual database key (i.e.
  ** it points to a delete key, or the start or end of a range-delete), and
  ** the CURSOR_IGNORE_DELETE flag is set, skip past this entry.  */
  if( (pCsr->flags & CURSOR_IGNORE_DELETE) && bDeleteOk==0 ){
    if( (eType & LSM_INSERT)==0 ) return 0;
  }

  /* If the cursor points to a system key (free-list entry), and the
  ** CURSOR_IGNORE_SYSTEM flag is set, skip thie entry.  */
  if( (pCsr->flags & CURSOR_IGNORE_SYSTEM) && rtTopic(eType)!=0 ){
    return 0;
  }

#ifndef NDEBUG
  /* This block fires assert() statements to check one of the assumptions
  ** in the comment below - that if the lhs sub-cursor of a level undergoing
  ** a merge is valid, then all the rhs sub-cursors must be at EOF. 
  **
  ** Also assert that all rhs sub-cursors are either at EOF or point to
  ** a key that is not less than the level split-key.  */
  for(i=0; i<pCsr->nPtr; i++){
    SegmentPtr *pPtr = &pCsr->aPtr[i];
    Level *pLvl = pPtr->pLevel;
    if( pLvl->nRight && pPtr->pPg ){
      if( pPtr->pSeg==&pLvl->lhs ){
        int j;
        for(j=0; j<pLvl->nRight; j++) assert( pPtr[j+1].pPg==0 );
      }else{
        int res = sortedKeyCompare(pCsr->pDb->xCmp, 
            rtTopic(pPtr->eType), pPtr->pKey, pPtr->nKey,
            pLvl->iSplitTopic, pLvl->pSplitKey, pLvl->nSplitKey
        );
        assert( res>=0 );
      }
    }
  }
#endif

  /* Now check if this key has already been deleted by a range-delete. If 
  ** so, skip past it.
  **
  ** Assume, for the moment, that the tree contains no levels currently 
  ** undergoing incremental merge, and that this cursor is iterating forwards
  ** through the database keys. The cursor currently points to a key in
  ** level L. This key has already been deleted if any of the sub-cursors
  ** that point to levels newer than L (or to the in-memory tree) point to
  ** a key greater than the current key with the LSM_END_DELETE flag set.
  **
  ** Or, if the cursor is iterating backwards through data keys, if any
  ** such sub-cursor points to a key smaller than the current key with the
  ** LSM_START_DELETE flag set.
  **
  ** Why it works with levels undergoing a merge too:
  **
  ** When a cursor iterates forwards, the sub-cursors for the rhs of a 
  ** level are only activated once the lhs reaches EOF. So when iterating
  ** forwards, the keys visited are the same as if the level was completely
  ** merged.
  **
  ** If the cursor is iterating backwards, then the lhs sub-cursor is not 
  ** initialized until the last of the rhs sub-cursors has reached EOF.
  ** Additionally, if the START_DELETE flag is set on the last entry (in
  ** reverse order - so the entry with the smallest key) of a rhs sub-cursor,
  ** then a pseudo-key equal to the levels split-key with the END_DELETE
  ** flag set is visited by the sub-cursor.
  */ 
  iKey = pCsr->aTree[1];
  for(i=0; i<iKey; i++){
    int csrflags;
    multiCursorGetKey(pCsr, i, &csrflags, 0, 0);
    if( (rdmask & csrflags) ){
      const int SD_ED = (LSM_START_DELETE|LSM_END_DELETE);
      if( (csrflags & SD_ED)==SD_ED 
       || (pCsr->flags & CURSOR_IGNORE_DELETE)==0
      ){
        void *pKey; int nKey;
        multiCursorGetKey(pCsr, i, 0, &pKey, &nKey);
        if( 0==sortedKeyCompare(pCsr->pDb->xCmp,
              rtTopic(eType), pCsr->key.pData, pCsr->key.nData,
              rtTopic(csrflags), pKey, nKey
        )){
          continue;
        }
      }
      return 0;
    }
  }

  /* The current cursor position is one this cursor should visit. Return 1. */
  return 1;
}

static int multiCursorSetupTree(MultiCursor *pCsr, int bRev){
  int rc;

  rc = multiCursorAllocTree(pCsr);
  if( rc==LSM_OK ){
    int i;
    for(i=pCsr->nTree-1; i>0; i--){
      multiCursorDoCompare(pCsr, i, bRev);
    }
  }

  assertCursorTree(pCsr);
  multiCursorCacheKey(pCsr, &rc);

  if( rc==LSM_OK && mcursorLocationOk(pCsr, 0)==0 ){
    rc = multiCursorAdvance(pCsr, bRev);
  }
  return rc;
}


static int multiCursorEnd(MultiCursor *pCsr, int bLast){
  int rc = LSM_OK;
  int i;

  pCsr->flags &= ~(CURSOR_NEXT_OK | CURSOR_PREV_OK);
  pCsr->flags |= (bLast ? CURSOR_PREV_OK : CURSOR_NEXT_OK);
  pCsr->iFree = 0;

  /* Position the two in-memory tree cursors */
  for(i=0; rc==LSM_OK && i<2; i++){
    if( pCsr->apTreeCsr[i] ){
      rc = lsmTreeCursorEnd(pCsr->apTreeCsr[i], bLast);
    }
  }

  for(i=0; rc==LSM_OK && i<pCsr->nPtr; i++){
    SegmentPtr *pPtr = &pCsr->aPtr[i];
    Level *pLvl = pPtr->pLevel;
    int iRhs;
    int bHit = 0;

    if( bLast ){
      for(iRhs=0; iRhs<pLvl->nRight && rc==LSM_OK; iRhs++){
        rc = segmentPtrEnd(pCsr, &pPtr[iRhs+1], 1);
        if( pPtr[iRhs+1].pPg ) bHit = 1;
      }
      if( bHit==0 && rc==LSM_OK ){
        rc = segmentPtrEnd(pCsr, pPtr, 1);
      }else{
        segmentPtrReset(pPtr);
      }
    }else{
      int bLhs = (pPtr->pSeg==&pLvl->lhs);
      assert( pPtr->pSeg==&pLvl->lhs || pPtr->pSeg==&pLvl->aRhs[0] );

      if( bLhs ){
        rc = segmentPtrEnd(pCsr, pPtr, 0);
        if( pPtr->pKey ) bHit = 1;
      }
      for(iRhs=0; iRhs<pLvl->nRight && rc==LSM_OK; iRhs++){
        if( bHit ){
          segmentPtrReset(&pPtr[iRhs+1]);
        }else{
          rc = sortedRhsFirst(pCsr, pLvl, &pPtr[iRhs+bLhs]);
        }
      }
    }
    i += pLvl->nRight;
  }

  /* And the b-tree cursor, if applicable */
  if( rc==LSM_OK && pCsr->pBtCsr ){
    assert( bLast==0 );
    rc = btreeCursorFirst(pCsr->pBtCsr);
  }

  if( rc==LSM_OK ){
    rc = multiCursorSetupTree(pCsr, bLast);
  }
  
  return rc;
}


int mcursorSave(MultiCursor *pCsr){
  int rc = LSM_OK;
  if( pCsr->aTree ){
    int iTree = pCsr->aTree[1];
    if( iTree==CURSOR_DATA_TREE0 || iTree==CURSOR_DATA_TREE1 ){
      multiCursorCacheKey(pCsr, &rc);
    }
  }
  mcursorFreeComponents(pCsr);
  return rc;
}

int mcursorRestore(lsm_db *pDb, MultiCursor *pCsr){
  int rc;
  rc = multiCursorInit(pCsr, pDb->pClient);
  if( rc==LSM_OK && pCsr->key.pData ){
    rc = lsmMCursorSeek(pCsr, 
         rtTopic(pCsr->eType), pCsr->key.pData, pCsr->key.nData, +1
    );
  }
  return rc;
}

int lsmSaveCursors(lsm_db *pDb){
  int rc = LSM_OK;
  MultiCursor *pCsr;

  for(pCsr=pDb->pCsr; rc==LSM_OK && pCsr; pCsr=pCsr->pNext){
    rc = mcursorSave(pCsr);
  }
  return rc;
}

int lsmRestoreCursors(lsm_db *pDb){
  int rc = LSM_OK;
  MultiCursor *pCsr;

  for(pCsr=pDb->pCsr; rc==LSM_OK && pCsr; pCsr=pCsr->pNext){
    rc = mcursorRestore(pDb, pCsr);
  }
  return rc;
}

int lsmMCursorFirst(MultiCursor *pCsr){
  return multiCursorEnd(pCsr, 0);
}

int lsmMCursorLast(MultiCursor *pCsr){
  return multiCursorEnd(pCsr, 1);
}

lsm_db *lsmMCursorDb(MultiCursor *pCsr){
  return pCsr->pDb;
}

void lsmMCursorReset(MultiCursor *pCsr){
  int i;
  lsmTreeCursorReset(pCsr->apTreeCsr[0]);
  lsmTreeCursorReset(pCsr->apTreeCsr[1]);
  for(i=0; i<pCsr->nPtr; i++){
    segmentPtrReset(&pCsr->aPtr[i]);
  }
  pCsr->key.nData = 0;
}

static int treeCursorSeek(
  MultiCursor *pCsr,
  TreeCursor *pTreeCsr, 
  void *pKey, int nKey, 
  int eSeek,
  int *pbStop
){
  int rc = LSM_OK;
  if( pTreeCsr ){
    int res = 0;
    lsmTreeCursorSeek(pTreeCsr, pKey, nKey, &res);
    switch( eSeek ){
      case LSM_SEEK_EQ: {
        int eType = lsmTreeCursorFlags(pTreeCsr);
        if( (res<0 && (eType & LSM_START_DELETE))
         || (res>0 && (eType & LSM_END_DELETE))
         || (res==0 && (eType & LSM_POINT_DELETE))
        ){
          *pbStop = 1;
        }else if( res==0 && (eType & LSM_INSERT) ){
          lsm_env *pEnv = pCsr->pDb->pEnv;
          void *p; int n;         /* Key/value from tree-cursor */
          *pbStop = 1;
          pCsr->flags |= CURSOR_SEEK_EQ;
          rc = lsmTreeCursorKey(pTreeCsr, &pCsr->eType, &p, &n);
          if( rc==LSM_OK ) rc = sortedBlobSet(pEnv, &pCsr->key, p, n);
          if( rc==LSM_OK ) rc = lsmTreeCursorValue(pTreeCsr, &p, &n);
          if( rc==LSM_OK ) rc = sortedBlobSet(pEnv, &pCsr->val, p, n);
        }
        lsmTreeCursorReset(pTreeCsr);
        break;
      }
      case LSM_SEEK_GE:
        if( res<0 && lsmTreeCursorValid(pTreeCsr) ){
          lsmTreeCursorNext(pTreeCsr);
        }
        break;
      default:
        if( res>0 ){
          assert( lsmTreeCursorValid(pTreeCsr) );
          lsmTreeCursorPrev(pTreeCsr);
        }
        break;
    }
  }
  return rc;
}


/*
** Seek the cursor.
*/
int lsmMCursorSeek(
  MultiCursor *pCsr, 
  int iTopic, 
  void *pKey, int nKey, 
  int eSeek
){
  int eESeek = eSeek;             /* Effective eSeek parameter */
  int bStop = 0;                  /* Set to true to halt search operation */
  int rc = LSM_OK;                /* Return code */
  int iPtr = 0;                   /* Used to iterate through pCsr->aPtr[] */
  Pgno iPgno = 0;                 /* FC pointer value */

  assert( pCsr->apTreeCsr[0]==0 || iTopic==0 );
  assert( pCsr->apTreeCsr[1]==0 || iTopic==0 );

  if( eESeek==LSM_SEEK_LEFAST ) eESeek = LSM_SEEK_LE;

  assert( eESeek==LSM_SEEK_EQ || eESeek==LSM_SEEK_LE || eESeek==LSM_SEEK_GE );
  assert( (pCsr->flags & CURSOR_FLUSH_FREELIST)==0 );
  assert( pCsr->nPtr==0 || pCsr->aPtr[0].pLevel );

  pCsr->flags &= ~(CURSOR_NEXT_OK | CURSOR_PREV_OK | CURSOR_SEEK_EQ);
  rc = treeCursorSeek(pCsr, pCsr->apTreeCsr[0], pKey, nKey, eESeek, &bStop);
  if( rc==LSM_OK && bStop==0 ){
    rc = treeCursorSeek(pCsr, pCsr->apTreeCsr[1], pKey, nKey, eESeek, &bStop);
  }

  /* Seek all segment pointers. */
  for(iPtr=0; iPtr<pCsr->nPtr && rc==LSM_OK && bStop==0; iPtr++){
    SegmentPtr *pPtr = &pCsr->aPtr[iPtr];
    assert( pPtr->pSeg==&pPtr->pLevel->lhs );
    rc = seekInLevel(pCsr, pPtr, eESeek, iTopic, pKey, nKey, &iPgno, &bStop);
    iPtr += pPtr->pLevel->nRight;
  }

  if( eSeek!=LSM_SEEK_EQ ){
    if( rc==LSM_OK ){
      rc = multiCursorAllocTree(pCsr);
    }
    if( rc==LSM_OK ){
      int i;
      for(i=pCsr->nTree-1; i>0; i--){
        multiCursorDoCompare(pCsr, i, eESeek==LSM_SEEK_LE);
      }
      if( eSeek==LSM_SEEK_GE ) pCsr->flags |= CURSOR_NEXT_OK;
      if( eSeek==LSM_SEEK_LE ) pCsr->flags |= CURSOR_PREV_OK;
    }

    multiCursorCacheKey(pCsr, &rc);
    if( rc==LSM_OK && eSeek!=LSM_SEEK_LEFAST && 0==mcursorLocationOk(pCsr, 0) ){
      switch( eESeek ){
        case LSM_SEEK_EQ:
          lsmMCursorReset(pCsr);
          break;
        case LSM_SEEK_GE:
          rc = lsmMCursorNext(pCsr);
          break;
        default:
          rc = lsmMCursorPrev(pCsr);
          break;
      }
    }
  }

  return rc;
}

int lsmMCursorValid(MultiCursor *pCsr){
  int res = 0;
  if( pCsr->flags & CURSOR_SEEK_EQ ){
    res = 1;
  }else if( pCsr->aTree ){
    int iKey = pCsr->aTree[1];
    if( iKey==CURSOR_DATA_TREE0 || iKey==CURSOR_DATA_TREE1 ){
      res = lsmTreeCursorValid(pCsr->apTreeCsr[iKey-CURSOR_DATA_TREE0]);
    }else{
      void *pKey; 
      multiCursorGetKey(pCsr, iKey, 0, &pKey, 0);
      res = pKey!=0;
    }
  }
  return res;
}

static int mcursorAdvanceOk(
  MultiCursor *pCsr, 
  int bReverse,
  int *pRc
){
  void *pNew;                     /* Pointer to buffer containing new key */
  int nNew;                       /* Size of buffer pNew in bytes */
  int eNewType;                   /* Type of new record */

  if( *pRc ) return 1;

  /* Check the current key value. If it is not greater than (if bReverse==0)
  ** or less than (if bReverse!=0) the key currently cached in pCsr->key, 
  ** then the cursor has not yet been successfully advanced.  
  */
  multiCursorGetKey(pCsr, pCsr->aTree[1], &eNewType, &pNew, &nNew);
  if( pNew ){
    int typemask = (pCsr->flags & CURSOR_IGNORE_DELETE) ? ~(0) : LSM_SYSTEMKEY;
    int res = sortedDbKeyCompare(pCsr,
      eNewType & typemask, pNew, nNew, 
      pCsr->eType & typemask, pCsr->key.pData, pCsr->key.nData
    );

    if( (bReverse==0 && res<=0) || (bReverse!=0 && res>=0) ){
      return 0;
    }

    multiCursorCacheKey(pCsr, pRc);
    assert( pCsr->eType==eNewType );

    /* If this cursor is configured to skip deleted keys, and the current
    ** cursor points to a SORTED_DELETE entry, then the cursor has not been 
    ** successfully advanced.  
    **
    ** Similarly, if the cursor is configured to skip system keys and the
    ** current cursor points to a system key, it has not yet been advanced.
    */
    if( *pRc==LSM_OK && 0==mcursorLocationOk(pCsr, 0) ) return 0;
  }
  return 1;
}

static void flCsrAdvance(MultiCursor *pCsr){
  assert( pCsr->flags & CURSOR_FLUSH_FREELIST );
  if( pCsr->iFree % 2 ){
    pCsr->iFree++;
  }else{
    int nEntry = pCsr->pDb->pWorker->freelist.nEntry;
    FreelistEntry *aEntry = pCsr->pDb->pWorker->freelist.aEntry;

    int i = nEntry - 1 - (pCsr->iFree / 2);

    /* If the current entry is a delete and the "end-delete" key will not
    ** be attached to the next entry, increment iFree by 1 only. */
    if( aEntry[i].iId<0 ){
      while( 1 ){
        if( i==0 || aEntry[i-1].iBlk!=aEntry[i].iBlk-1 ){
          pCsr->iFree--;
          break;
        }
        if( aEntry[i-1].iId>=0 ) break;
        pCsr->iFree += 2;
        i--;
      }
    }
    pCsr->iFree += 2;
  }
}

static int multiCursorAdvance(MultiCursor *pCsr, int bReverse){
  int rc = LSM_OK;                /* Return Code */
  if( lsmMCursorValid(pCsr) ){
    do {
      int iKey = pCsr->aTree[1];

      assertCursorTree(pCsr);

      /* If this multi-cursor is advancing forwards, and the sub-cursor
      ** being advanced is the one that separator keys may be being read
      ** from, record the current absolute pointer value.  */
      if( pCsr->pPrevMergePtr ){
        if( iKey==(CURSOR_DATA_SEGMENT+pCsr->nPtr) ){
          assert( pCsr->pBtCsr );
          *pCsr->pPrevMergePtr = pCsr->pBtCsr->iPtr;
        }else if( pCsr->pBtCsr==0 && pCsr->nPtr>0
               && iKey==(CURSOR_DATA_SEGMENT+pCsr->nPtr-1) 
        ){
          SegmentPtr *pPtr = &pCsr->aPtr[iKey-CURSOR_DATA_SEGMENT];
          *pCsr->pPrevMergePtr = pPtr->iPtr+pPtr->iPgPtr;
        }
      }

      if( iKey==CURSOR_DATA_TREE0 || iKey==CURSOR_DATA_TREE1 ){
        TreeCursor *pTreeCsr = pCsr->apTreeCsr[iKey-CURSOR_DATA_TREE0];
        if( bReverse ){
          rc = lsmTreeCursorPrev(pTreeCsr);
        }else{
          rc = lsmTreeCursorNext(pTreeCsr);
        }
      }else if( iKey==CURSOR_DATA_SYSTEM ){
        assert( pCsr->flags & CURSOR_FLUSH_FREELIST );
        assert( bReverse==0 );
        flCsrAdvance(pCsr);
      }else if( iKey==(CURSOR_DATA_SEGMENT+pCsr->nPtr) ){
        assert( bReverse==0 && pCsr->pBtCsr );
        rc = btreeCursorNext(pCsr->pBtCsr);
      }else{
        rc = segmentCursorAdvance(pCsr, iKey-CURSOR_DATA_SEGMENT, bReverse);
      }
      if( rc==LSM_OK ){
        int i;
        for(i=(iKey+pCsr->nTree)/2; i>0; i=i/2){
          multiCursorDoCompare(pCsr, i, bReverse);
        }
        assertCursorTree(pCsr);
      }
    }while( mcursorAdvanceOk(pCsr, bReverse, &rc)==0 );
  }
  return rc;
}

int lsmMCursorNext(MultiCursor *pCsr){
  if( (pCsr->flags & CURSOR_NEXT_OK)==0 ) return LSM_MISUSE_BKPT;
  return multiCursorAdvance(pCsr, 0);
}

int lsmMCursorPrev(MultiCursor *pCsr){
  if( (pCsr->flags & CURSOR_PREV_OK)==0 ) return LSM_MISUSE_BKPT;
  return multiCursorAdvance(pCsr, 1);
}

int lsmMCursorKey(MultiCursor *pCsr, void **ppKey, int *pnKey){
  if( (pCsr->flags & CURSOR_SEEK_EQ) || pCsr->aTree==0 ){
    *pnKey = pCsr->key.nData;
    *ppKey = pCsr->key.pData;
  }else{
    int iKey = pCsr->aTree[1];

    if( iKey==CURSOR_DATA_TREE0 || iKey==CURSOR_DATA_TREE1 ){
      TreeCursor *pTreeCsr = pCsr->apTreeCsr[iKey-CURSOR_DATA_TREE0];
      lsmTreeCursorKey(pTreeCsr, 0, ppKey, pnKey);
    }else{
      int nKey;

#ifndef NDEBUG
      void *pKey;
      int eType;
      multiCursorGetKey(pCsr, iKey, &eType, &pKey, &nKey);
      assert( eType==pCsr->eType );
      assert( nKey==pCsr->key.nData );
      assert( memcmp(pKey, pCsr->key.pData, nKey)==0 );
#endif

      nKey = pCsr->key.nData;
      if( nKey==0 ){
        *ppKey = 0;
      }else{
        *ppKey = pCsr->key.pData;
      }
      *pnKey = nKey; 
    }
  }
  return LSM_OK;
}

/*
** Compare the current key that cursor csr points to with pKey/nKey. Set
** *piRes to the result and return LSM_OK.
*/
int lsm_csr_cmp(lsm_cursor *csr, const void *pKey, int nKey, int *piRes){
  MultiCursor *pCsr = (MultiCursor *)csr;
  void *pCsrkey; int nCsrkey;
  int rc;
  rc = lsmMCursorKey(pCsr, &pCsrkey, &nCsrkey);
  if( rc==LSM_OK ){
    int (*xCmp)(void *, int, void *, int) = pCsr->pDb->xCmp;
    *piRes = sortedKeyCompare(xCmp, 0, pCsrkey, nCsrkey, 0, (void *)pKey, nKey);
  }
  return rc;
}

int lsmMCursorValue(MultiCursor *pCsr, void **ppVal, int *pnVal){
  void *pVal;
  int nVal;
  int rc;
  if( (pCsr->flags & CURSOR_SEEK_EQ) || pCsr->aTree==0 ){
    rc = LSM_OK;
    nVal = pCsr->val.nData;
    pVal = pCsr->val.pData;
  }else{

    assert( pCsr->aTree );
    assert( mcursorLocationOk(pCsr, (pCsr->flags & CURSOR_IGNORE_DELETE)) );

    rc = multiCursorGetVal(pCsr, pCsr->aTree[1], &pVal, &nVal);
    if( pVal && rc==LSM_OK ){
      rc = sortedBlobSet(pCsr->pDb->pEnv, &pCsr->val, pVal, nVal);
      pVal = pCsr->val.pData;
    }

    if( rc!=LSM_OK ){
      pVal = 0;
      nVal = 0;
    }
  }
  *ppVal = pVal;
  *pnVal = nVal;
  return rc;
}

int lsmMCursorType(MultiCursor *pCsr, int *peType){
  assert( pCsr->aTree );
  multiCursorGetKey(pCsr, pCsr->aTree[1], peType, 0, 0);
  return LSM_OK;
}

/*
** Buffer aData[], size nData, is assumed to contain a valid b-tree 
** hierarchy page image. Return the offset in aData[] of the next free
** byte in the data area (where a new cell may be written if there is
** space).
*/
static int mergeWorkerPageOffset(u8 *aData, int nData){
  int nRec;
  int iOff;
  int nKey;
  int eType;

  nRec = lsmGetU16(&aData[SEGMENT_NRECORD_OFFSET(nData)]);
  iOff = lsmGetU16(&aData[SEGMENT_CELLPTR_OFFSET(nData, nRec-1)]);
  eType = aData[iOff++];
  assert( eType==0 
       || eType==(LSM_SYSTEMKEY|LSM_SEPARATOR) 
       || eType==(LSM_SEPARATOR)
  );

  iOff += lsmVarintGet32(&aData[iOff], &nKey);
  iOff += lsmVarintGet32(&aData[iOff], &nKey);

  return iOff + (eType ? nKey : 0);
}

/*
** Following a checkpoint operation, database pages that are part of the
** checkpointed state of the LSM are deemed read-only. This includes the
** right-most page of the b-tree hierarchy of any separators array under
** construction, and all pages between it and the b-tree root, inclusive.
** This is a problem, as when further pages are appended to the separators
** array, entries must be added to the indicated b-tree hierarchy pages.
**
** This function copies all such b-tree pages to new locations, so that
** they can be modified as required.
**
** The complication is that not all database pages are the same size - due
** to the way the file.c module works some (the first and last in each block)
** are 4 bytes smaller than the others.
*/
static int mergeWorkerMoveHierarchy(
  MergeWorker *pMW,               /* Merge worker */
  int bSep                        /* True for separators run */
){
  lsm_db *pDb = pMW->pDb;         /* Database handle */
  int rc = LSM_OK;                /* Return code */
  int i;
  Page **apHier = pMW->hier.apHier;
  int nHier = pMW->hier.nHier;

  for(i=0; rc==LSM_OK && i<nHier; i++){
    Page *pNew = 0;
    rc = lsmFsSortedAppend(pDb->pFS, pDb->pWorker, pMW->pLevel, 1, &pNew);
    assert( rc==LSM_OK );

    if( rc==LSM_OK ){
      u8 *a1; int n1;
      u8 *a2; int n2;

      a1 = fsPageData(pNew, &n1);
      a2 = fsPageData(apHier[i], &n2);

      assert( n1==n2 || n1+4==n2 );

      if( n1==n2 ){
        memcpy(a1, a2, n2);
      }else{
        int nEntry = pageGetNRec(a2, n2);
        int iEof1 = SEGMENT_EOF(n1, nEntry);
        int iEof2 = SEGMENT_EOF(n2, nEntry);

        memcpy(a1, a2, iEof2 - 4);
        memcpy(&a1[iEof1], &a2[iEof2], n2 - iEof2);
      }

      lsmFsPageRelease(apHier[i]);
      apHier[i] = pNew;

#if 0
      assert( n1==n2 || n1+4==n2 || n2+4==n1 );
      if( n1>=n2 ){
        /* If n1 (size of the new page) is equal to or greater than n2 (the
        ** size of the old page), then copy the data into the new page. If
        ** n1==n2, this could be done with a single memcpy(). However, 
        ** since sometimes n1>n2, the page content and footer must be copied 
        ** separately. */
        int nEntry = pageGetNRec(a2, n2);
        int iEof1 = SEGMENT_EOF(n1, nEntry);
        int iEof2 = SEGMENT_EOF(n2, nEntry);
        memcpy(a1, a2, iEof2);
        memcpy(&a1[iEof1], &a2[iEof2], n2 - iEof2);
        lsmFsPageRelease(apHier[i]);
        apHier[i] = pNew;
      }else{
        lsmPutU16(&a1[SEGMENT_FLAGS_OFFSET(n1)], SEGMENT_BTREE_FLAG);
        lsmPutU16(&a1[SEGMENT_NRECORD_OFFSET(n1)], 0);
        lsmPutU64(&a1[SEGMENT_POINTER_OFFSET(n1)], 0);
        i = i - 1;
        lsmFsPageRelease(pNew);
      }
#endif
    }
  }

#ifdef LSM_DEBUG
  if( rc==LSM_OK ){
    for(i=0; i<nHier; i++) assert( lsmFsPageWritable(apHier[i]) );
  }
#endif

  return rc;
}

/*
** Allocate and populate the MergeWorker.apHier[] array.
*/
static int mergeWorkerLoadHierarchy(MergeWorker *pMW){
  int rc = LSM_OK;
  Segment *pSeg;
  Hierarchy *p;
 
  pSeg = &pMW->pLevel->lhs;
  p = &pMW->hier;

  if( p->apHier==0 && pSeg->iRoot!=0 ){
    FileSystem *pFS = pMW->pDb->pFS;
    lsm_env *pEnv = pMW->pDb->pEnv;
    Page **apHier = 0;
    int nHier = 0;
    int iPg = pSeg->iRoot;

    do {
      Page *pPg = 0;
      u8 *aData;
      int nData;
      int flags;

      rc = lsmFsDbPageGet(pFS, pSeg, iPg, &pPg);
      if( rc!=LSM_OK ) break;

      aData = fsPageData(pPg, &nData);
      flags = pageGetFlags(aData, nData);
      if( flags&SEGMENT_BTREE_FLAG ){
        Page **apNew = (Page **)lsmRealloc(
            pEnv, apHier, sizeof(Page *)*(nHier+1)
        );
        if( apNew==0 ){
          rc = LSM_NOMEM_BKPT;
          break;
        }
        apHier = apNew;
        memmove(&apHier[1], &apHier[0], sizeof(Page *) * nHier);
        nHier++;

        apHier[0] = pPg;
        iPg = pageGetPtr(aData, nData);
      }else{
        lsmFsPageRelease(pPg);
        break;
      }
    }while( 1 );

    if( rc==LSM_OK ){
      u8 *aData;
      int nData;
      aData = fsPageData(apHier[0], &nData);
      pMW->aSave[0].iPgno = pageGetPtr(aData, nData);
      p->nHier = nHier;
      p->apHier = apHier;
      rc = mergeWorkerMoveHierarchy(pMW, 0);
    }else{
      int i;
      for(i=0; i<nHier; i++){
        lsmFsPageRelease(apHier[i]);
      }
      lsmFree(pEnv, apHier);
    }
  }

  return rc;
}

/*
** B-tree pages use almost the same format as regular pages. The 
** differences are:
**
**   1. The record format is (usually, see below) as follows:
**
**         + Type byte (always SORTED_SEPARATOR or SORTED_SYSTEM_SEPARATOR),
**         + Absolute pointer value (varint),
**         + Number of bytes in key (varint),
**         + Blob containing key data.
**
**   2. All pointer values are stored as absolute values (not offsets 
**      relative to the footer pointer value).
**
**   3. Each pointer that is part of a record points to a page that 
**      contains keys smaller than the records key (note: not "equal to or
**      smaller than - smaller than").
**
**   4. The pointer in the page footer of a b-tree page points to a page
**      that contains keys equal to or larger than the largest key on the
**      b-tree page.
**
** The reason for having the page footer pointer point to the right-child
** (instead of the left) is that doing things this way makes the 
** mergeWorkerMoveHierarchy() operation less complicated (since the pointers 
** that need to be updated are all stored as fixed-size integers within the 
** page footer, not varints in page records).
**
** Records may not span b-tree pages. If this function is called to add a
** record larger than (page-size / 4) bytes, then a pointer to the indexed
** array page that contains the main record is added to the b-tree instead.
** In this case the record format is:
**
**         + 0x00 byte (1 byte) 
**         + Absolute pointer value (varint),
**         + Absolute page number of page containing key (varint).
**
** See function seekInBtree() for the code that traverses b-tree pages.
*/

static int mergeWorkerBtreeWrite(
  MergeWorker *pMW,
  u8 eType,
  Pgno iPtr,
  Pgno iKeyPg,
  void *pKey,
  int nKey
){
  Hierarchy *p = &pMW->hier;
  lsm_db *pDb = pMW->pDb;         /* Database handle */
  int rc = LSM_OK;                /* Return Code */
  int iLevel;                     /* Level of b-tree hierachy to write to */
  int nData;                      /* Size of aData[] in bytes */
  u8 *aData;                      /* Page data for level iLevel */
  int iOff;                       /* Offset on b-tree page to write record to */
  int nRec;                       /* Initial number of records on b-tree page */

  /* iKeyPg should be zero for an ordinary b-tree key, or non-zero for an
  ** indirect key. The flags byte for an indirect key is 0x00.  */
  assert( (eType==0)==(iKeyPg!=0) );

  /* The MergeWorker.apHier[] array contains the right-most leaf of the b-tree
  ** hierarchy, the root node, and all nodes that lie on the path between.
  ** apHier[0] is the right-most leaf and apHier[pMW->nHier-1] is the current
  ** root page.
  **
  ** This loop searches for a node with enough space to store the key on,
  ** starting with the leaf and iterating up towards the root. When the loop
  ** exits, the key may be written to apHier[iLevel].  */
  for(iLevel=0; iLevel<=p->nHier; iLevel++){
    int nByte;                    /* Number of free bytes required */

    if( iLevel==p->nHier ){
      /* Extend the array and allocate a new root page. */
      Page **aNew;
      aNew = (Page **)lsmRealloc(
          pMW->pDb->pEnv, p->apHier, sizeof(Page *)*(p->nHier+1)
      );
      if( !aNew ){
        return LSM_NOMEM_BKPT;
      }
      p->apHier = aNew;
    }else{
      Page *pOld;
      int nFree;

      /* If the key will fit on this page, break out of the loop here.
      ** The new entry will be written to page apHier[iLevel]. */
      pOld = p->apHier[iLevel];
      assert( lsmFsPageWritable(pOld) );
      aData = fsPageData(pOld, &nData);
      if( eType==0 ){
        nByte = 2 + 1 + lsmVarintLen32(iPtr) + lsmVarintLen32(iKeyPg);
      }else{
        nByte = 2 + 1 + lsmVarintLen32(iPtr) + lsmVarintLen32(nKey) + nKey;
      }
      nRec = pageGetNRec(aData, nData);
      nFree = SEGMENT_EOF(nData, nRec) - mergeWorkerPageOffset(aData, nData);
      if( nByte<=nFree ) break;

      /* Otherwise, this page is full. Set the right-hand-child pointer
      ** to iPtr and release it.  */
      lsmPutU64(&aData[SEGMENT_POINTER_OFFSET(nData)], iPtr);
      assert( lsmFsPageNumber(pOld)==0 );
      rc = lsmFsPagePersist(pOld);
      if( rc==LSM_OK ){
        iPtr = lsmFsPageNumber(pOld);
        lsmFsPageRelease(pOld);
      }
    }

    /* Allocate a new page for apHier[iLevel]. */
    p->apHier[iLevel] = 0;
    if( rc==LSM_OK ){
      rc = lsmFsSortedAppend(
          pDb->pFS, pDb->pWorker, pMW->pLevel, 1, &p->apHier[iLevel]
      );
    }
    if( rc!=LSM_OK ) return rc;

    aData = fsPageData(p->apHier[iLevel], &nData);
    memset(aData, 0, nData);
    lsmPutU16(&aData[SEGMENT_FLAGS_OFFSET(nData)], SEGMENT_BTREE_FLAG);
    lsmPutU16(&aData[SEGMENT_NRECORD_OFFSET(nData)], 0);

    if( iLevel==p->nHier ){
      p->nHier++;
      break;
    }
  }

  /* Write the key into page apHier[iLevel]. */
  aData = fsPageData(p->apHier[iLevel], &nData);
  iOff = mergeWorkerPageOffset(aData, nData);
  nRec = pageGetNRec(aData, nData);
  lsmPutU16(&aData[SEGMENT_CELLPTR_OFFSET(nData, nRec)], iOff);
  lsmPutU16(&aData[SEGMENT_NRECORD_OFFSET(nData)], nRec+1);
  if( eType==0 ){
    aData[iOff++] = 0x00;
    iOff += lsmVarintPut32(&aData[iOff], iPtr);
    iOff += lsmVarintPut32(&aData[iOff], iKeyPg);
  }else{
    aData[iOff++] = eType;
    iOff += lsmVarintPut32(&aData[iOff], iPtr);
    iOff += lsmVarintPut32(&aData[iOff], nKey);
    memcpy(&aData[iOff], pKey, nKey);
  }

  return rc;
}

static int mergeWorkerBtreeIndirect(MergeWorker *pMW){
  int rc = LSM_OK;
  if( pMW->iIndirect ){
    Pgno iKeyPg = pMW->aSave[1].iPgno;
    rc = mergeWorkerBtreeWrite(pMW, 0, pMW->iIndirect, iKeyPg, 0, 0);
    pMW->iIndirect = 0;
  }
  return rc;
}

/*
** Append the database key (iTopic/pKey/nKey) to the b-tree under 
** construction. This key has not yet been written to a segment page.
** The pointer that will accompany the new key in the b-tree - that
** points to the completed segment page that contains keys smaller than
** (pKey/nKey) is currently stored in pMW->aSave[0].iPgno.
*/
static int mergeWorkerPushHierarchy(
  MergeWorker *pMW,               /* Merge worker object */
  int iTopic,                     /* Topic value for this key */
  void *pKey,                     /* Pointer to key buffer */
  int nKey                        /* Size of pKey buffer in bytes */
){
  int rc = LSM_OK;                /* Return Code */
  Pgno iPtr;                      /* Pointer value to accompany pKey/nKey */

  assert( pMW->aSave[0].bStore==0 );
  assert( pMW->aSave[1].bStore==0 );
  rc = mergeWorkerBtreeIndirect(pMW);

  /* Obtain the absolute pointer value to store along with the key in the
  ** page body. This pointer points to a page that contains keys that are
  ** smaller than pKey/nKey.  */
  iPtr = pMW->aSave[0].iPgno;
  assert( iPtr!=0 );

  /* Determine if the indirect format should be used. */
  if( (nKey*4 > lsmFsPageSize(pMW->pDb->pFS)) ){
    pMW->iIndirect = iPtr;
    pMW->aSave[1].bStore = 1;
  }else{
    rc = mergeWorkerBtreeWrite(
        pMW, (u8)(iTopic | LSM_SEPARATOR), iPtr, 0, pKey, nKey
    );
  }

  /* Ensure that the SortedRun.iRoot field is correct. */
  return rc;
}

static int mergeWorkerFinishHierarchy(
  MergeWorker *pMW                /* Merge worker object */
){
  int i;                          /* Used to loop through apHier[] */
  int rc = LSM_OK;                /* Return code */
  Pgno iPtr;                      /* New right-hand-child pointer value */

  iPtr = pMW->aSave[0].iPgno;
  for(i=0; i<pMW->hier.nHier && rc==LSM_OK; i++){
    Page *pPg = pMW->hier.apHier[i];
    int nData;                    /* Size of aData[] in bytes */
    u8 *aData;                    /* Page data for pPg */

    aData = fsPageData(pPg, &nData);
    lsmPutU64(&aData[SEGMENT_POINTER_OFFSET(nData)], iPtr);

    rc = lsmFsPagePersist(pPg);
    iPtr = lsmFsPageNumber(pPg);
    lsmFsPageRelease(pPg);
  }

  if( pMW->hier.nHier ){
    pMW->pLevel->lhs.iRoot = iPtr;
    lsmFree(pMW->pDb->pEnv, pMW->hier.apHier);
    pMW->hier.apHier = 0;
    pMW->hier.nHier = 0;
  }

  return rc;
}

static int mergeWorkerAddPadding(
  MergeWorker *pMW                /* Merge worker object */
){
  FileSystem *pFS = pMW->pDb->pFS;
  return lsmFsSortedPadding(pFS, pMW->pDb->pWorker, &pMW->pLevel->lhs);
}

/*
** Release all page references currently held by the merge-worker passed
** as the only argument. Unless an error has occurred, all pages have
** already been released.
*/
static void mergeWorkerReleaseAll(MergeWorker *pMW){
  int i;
  lsmFsPageRelease(pMW->pPage);
  pMW->pPage = 0;

  for(i=0; i<pMW->hier.nHier; i++){
    lsmFsPageRelease(pMW->hier.apHier[i]);
    pMW->hier.apHier[i] = 0;
  }
  lsmFree(pMW->pDb->pEnv, pMW->hier.apHier);
  pMW->hier.apHier = 0;
  pMW->hier.nHier = 0;
}

static int keyszToSkip(FileSystem *pFS, int nKey){
  int nPgsz;                /* Nominal database page size */
  nPgsz = lsmFsPageSize(pFS);
  return LSM_MIN(((nKey * 4) / nPgsz), 3);
}

/*
** Release the reference to the current output page of merge-worker *pMW
** (reference pMW->pPage). Set the page number values in aSave[] as 
** required (see comments above struct MergeWorker for details).
*/
static int mergeWorkerPersistAndRelease(MergeWorker *pMW){
  int rc;
  int i;

  assert( pMW->pPage || (pMW->aSave[0].bStore==0 && pMW->aSave[1].bStore==0) );

  /* Persist the page */
  rc = lsmFsPagePersist(pMW->pPage);

  /* If required, save the page number. */
  for(i=0; i<2; i++){
    if( pMW->aSave[i].bStore ){
      pMW->aSave[i].iPgno = lsmFsPageNumber(pMW->pPage);
      pMW->aSave[i].bStore = 0;
    }
  }

  /* Release the completed output page. */
  lsmFsPageRelease(pMW->pPage);
  pMW->pPage = 0;
  return rc;
}

/*
** Advance to the next page of an output run being populated by merge-worker
** pMW. The footer of the new page is initialized to indicate that it contains
** zero records. The flags field is cleared. The page footer pointer field
** is set to iFPtr.
**
** If successful, LSM_OK is returned. Otherwise, an error code.
*/
static int mergeWorkerNextPage(
  MergeWorker *pMW,               /* Merge worker object to append page to */
  Pgno iFPtr                      /* Pointer value for footer of new page */
){
  int rc = LSM_OK;                /* Return code */
  Page *pNext = 0;                /* New page appended to run */
  lsm_db *pDb = pMW->pDb;         /* Database handle */

  rc = lsmFsSortedAppend(pDb->pFS, pDb->pWorker, pMW->pLevel, 0, &pNext);
  assert( rc || pMW->pLevel->lhs.iFirst>0 || pMW->pDb->compress.xCompress );

  if( rc==LSM_OK ){
    u8 *aData;                    /* Data buffer belonging to page pNext */
    int nData;                    /* Size of aData[] in bytes */

    rc = mergeWorkerPersistAndRelease(pMW);

    pMW->pPage = pNext;
    pMW->pLevel->pMerge->iOutputOff = 0;
    aData = fsPageData(pNext, &nData);
    lsmPutU16(&aData[SEGMENT_NRECORD_OFFSET(nData)], 0);
    lsmPutU16(&aData[SEGMENT_FLAGS_OFFSET(nData)], 0);
    lsmPutU64(&aData[SEGMENT_POINTER_OFFSET(nData)], iFPtr);
    pMW->nWork++;
  }

  return rc;
}

/*
** Write a blob of data into an output segment being populated by a 
** merge-worker object. If argument bSep is true, write into the separators
** array. Otherwise, the main array.
**
** This function is used to write the blobs of data for keys and values.
*/
static int mergeWorkerData(
  MergeWorker *pMW,               /* Merge worker object */
  int bSep,                       /* True to write to separators run */
  int iFPtr,                      /* Footer ptr for new pages */
  u8 *aWrite,                     /* Write data from this buffer */
  int nWrite                      /* Size of aWrite[] in bytes */
){
  int rc = LSM_OK;                /* Return code */
  int nRem = nWrite;              /* Number of bytes still to write */

  while( nRem>0 ){
    Merge *pMerge = pMW->pLevel->pMerge;
    int nCopy;                    /* Number of bytes to copy */
    u8 *aData;                    /* Pointer to buffer of current output page */
    int nData;                    /* Size of aData[] in bytes */
    int nRec;                     /* Number of records on current output page */
    int iOff;                     /* Offset in aData[] to write to */

    assert( lsmFsPageWritable(pMW->pPage) );
   
    aData = fsPageData(pMW->pPage, &nData);
    nRec = pageGetNRec(aData, nData);
    iOff = pMerge->iOutputOff;
    nCopy = LSM_MIN(nRem, SEGMENT_EOF(nData, nRec) - iOff);

    memcpy(&aData[iOff], &aWrite[nWrite-nRem], nCopy);
    nRem -= nCopy;

    if( nRem>0 ){
      rc = mergeWorkerNextPage(pMW, iFPtr);
    }else{
      pMerge->iOutputOff = iOff + nCopy;
    }
  }

  return rc;
}


/*
** The MergeWorker passed as the only argument is working to merge two or
** more existing segments together (not to flush an in-memory tree). It
** has not yet written the first key to the first page of the output.
*/
static int mergeWorkerFirstPage(MergeWorker *pMW){
  int rc = LSM_OK;                /* Return code */
  Page *pPg = 0;                  /* First page of run pSeg */
  int iFPtr = 0;                  /* Pointer value read from footer of pPg */
  MultiCursor *pCsr = pMW->pCsr;

  assert( pMW->pPage==0 );

  if( pCsr->pBtCsr ){
    rc = LSM_OK;
    iFPtr = pMW->pLevel->pNext->lhs.iFirst;
  }else if( pCsr->nPtr>0 ){
    Segment *pSeg;
    pSeg = pCsr->aPtr[pCsr->nPtr-1].pSeg;
    rc = lsmFsDbPageGet(pMW->pDb->pFS, pSeg, pSeg->iFirst, &pPg);
    if( rc==LSM_OK ){
      u8 *aData;                    /* Buffer for page pPg */
      int nData;                    /* Size of aData[] in bytes */
      aData = fsPageData(pPg, &nData);
      iFPtr = pageGetPtr(aData, nData);
      lsmFsPageRelease(pPg);
    }
  }

  if( rc==LSM_OK ){
    rc = mergeWorkerNextPage(pMW, iFPtr);
    if( pCsr->pPrevMergePtr ) *pCsr->pPrevMergePtr = iFPtr;
    pMW->aSave[0].bStore = 1;
  }

  return rc;
}

static int mergeWorkerWrite(
  MergeWorker *pMW,               /* Merge worker object to write into */
  int eType,                      /* One of SORTED_SEPARATOR, WRITE or DELETE */
  void *pKey, int nKey,           /* Key value */
  void *pVal, int nVal,           /* Value value */
  int iPtr                        /* Absolute value of page pointer, or 0 */
){
  int rc = LSM_OK;                /* Return code */
  Merge *pMerge;                  /* Persistent part of level merge state */
  int nHdr;                       /* Space required for this record header */
  Page *pPg;                      /* Page to write to */
  u8 *aData;                      /* Data buffer for page pWriter->pPage */
  int nData;                      /* Size of buffer aData[] in bytes */
  int nRec;                       /* Number of records on page pPg */
  int iFPtr;                      /* Value of pointer in footer of pPg */
  int iRPtr = 0;                  /* Value of pointer written into record */
  int iOff;                       /* Current write offset within page pPg */
  Segment *pSeg;                  /* Segment being written */
  int flags = 0;                  /* If != 0, flags value for page footer */
  int bFirst = 0;                 /* True for first key of output run */

  pMerge = pMW->pLevel->pMerge;    
  pSeg = &pMW->pLevel->lhs;

  if( pSeg->iFirst==0 && pMW->pPage==0 ){
    rc = mergeWorkerFirstPage(pMW);
    bFirst = 1;
  }
  pPg = pMW->pPage;
  if( pPg ){
    aData = fsPageData(pPg, &nData);
    nRec = pageGetNRec(aData, nData);
    iFPtr = pageGetPtr(aData, nData);
    iRPtr = iPtr - iFPtr;
  }
     
  /* Figure out how much space is required by the new record. The space
  ** required is divided into two sections: the header and the body. The
  ** header consists of the intial varint fields. The body are the blobs 
  ** of data that correspond to the key and value data. The entire header 
  ** must be stored on the page. The body may overflow onto the next and
  ** subsequent pages.
  **
  ** The header space is:
  **
  **     1) record type - 1 byte.
  **     2) Page-pointer-offset - 1 varint
  **     3) Key size - 1 varint
  **     4) Value size - 1 varint (only if LSM_INSERT flag is set)
  */
  if( rc==LSM_OK ){
    nHdr = 1 + lsmVarintLen32(iRPtr) + lsmVarintLen32(nKey);
    if( rtIsWrite(eType) ) nHdr += lsmVarintLen32(nVal);

    /* If the entire header will not fit on page pPg, or if page pPg is 
    ** marked read-only, advance to the next page of the output run. */
    iOff = pMerge->iOutputOff;
    if( iOff<0 || pPg==0 || iOff+nHdr > SEGMENT_EOF(nData, nRec+1) ){
      iFPtr = *pMW->pCsr->pPrevMergePtr;
      iRPtr = iPtr - iFPtr;
      iOff = 0;
      nRec = 0;
      rc = mergeWorkerNextPage(pMW, iFPtr);
      pPg = pMW->pPage;
    }
  }

  /* If this record header will be the first on the page, and the page is 
  ** not the very first in the entire run, add a copy of the key to the
  ** b-tree hierarchy.
  */
  if( rc==LSM_OK && nRec==0 && bFirst==0 ){
    assert( pMerge->nSkip>=0 );

    if( pMerge->nSkip==0 ){
      rc = mergeWorkerPushHierarchy(pMW, rtTopic(eType), pKey, nKey);
      assert( pMW->aSave[0].bStore==0 );
      pMW->aSave[0].bStore = 1;
      pMerge->nSkip = keyszToSkip(pMW->pDb->pFS, nKey);
    }else{
      pMerge->nSkip--;
      flags = PGFTR_SKIP_THIS_FLAG;
    }

    if( pMerge->nSkip ) flags |= PGFTR_SKIP_NEXT_FLAG;
  }

  /* Update the output segment */
  if( rc==LSM_OK ){
    aData = fsPageData(pPg, &nData);

    /* Update the page footer. */
    lsmPutU16(&aData[SEGMENT_NRECORD_OFFSET(nData)], nRec+1);
    lsmPutU16(&aData[SEGMENT_CELLPTR_OFFSET(nData, nRec)], iOff);
    if( flags ) lsmPutU16(&aData[SEGMENT_FLAGS_OFFSET(nData)], flags);

    /* Write the entry header into the current page. */
    aData[iOff++] = eType;                                               /* 1 */
    iOff += lsmVarintPut32(&aData[iOff], iRPtr);                         /* 2 */
    iOff += lsmVarintPut32(&aData[iOff], nKey);                          /* 3 */
    if( rtIsWrite(eType) ) iOff += lsmVarintPut32(&aData[iOff], nVal);   /* 4 */
    pMerge->iOutputOff = iOff;

    /* Write the key and data into the segment. */
    assert( iFPtr==pageGetPtr(aData, nData) );
    rc = mergeWorkerData(pMW, 0, iFPtr+iRPtr, pKey, nKey);
    if( rc==LSM_OK && rtIsWrite(eType) ){
      if( rc==LSM_OK ){
        rc = mergeWorkerData(pMW, 0, iFPtr+iRPtr, pVal, nVal);
      }
    }
  }

  return rc;
}


/*
** Free all resources allocated by mergeWorkerInit().
*/
static void mergeWorkerShutdown(MergeWorker *pMW, int *pRc){
  int i;                          /* Iterator variable */
  int rc = *pRc;
  MultiCursor *pCsr = pMW->pCsr;

  /* Unless the merge has finished, save the cursor position in the
  ** Merge.aInput[] array. See function mergeWorkerInit() for the 
  ** code to restore a cursor position based on aInput[].  */
  if( rc==LSM_OK && pCsr && lsmMCursorValid(pCsr) ){
    Merge *pMerge = pMW->pLevel->pMerge;
    int bBtree = (pCsr->pBtCsr!=0);
    int iPtr;

    /* pMerge->nInput==0 indicates that this is a FlushTree() operation. */
    assert( pMerge->nInput==0 || pMW->pLevel->nRight>0 );
    assert( pMerge->nInput==0 || pMerge->nInput==(pCsr->nPtr+bBtree) );

    for(i=0; i<(pMerge->nInput-bBtree); i++){
      SegmentPtr *pPtr = &pCsr->aPtr[i];
      if( pPtr->pPg ){
        pMerge->aInput[i].iPg = lsmFsPageNumber(pPtr->pPg);
        pMerge->aInput[i].iCell = pPtr->iCell;
      }else{
        pMerge->aInput[i].iPg = 0;
        pMerge->aInput[i].iCell = 0;
      }
    }
    if( bBtree && pMerge->nInput ){
      assert( i==pCsr->nPtr );
      btreeCursorPosition(pCsr->pBtCsr, &pMerge->aInput[i]);
    }

    /* Store the location of the split-key */
    iPtr = pCsr->aTree[1] - CURSOR_DATA_SEGMENT;
    if( iPtr<pCsr->nPtr ){
      pMerge->splitkey = pMerge->aInput[iPtr];
    }else{
      btreeCursorSplitkey(pCsr->pBtCsr, &pMerge->splitkey);
    }
    
    pMerge->iOutputOff = -1;
  }

  lsmMCursorClose(pCsr, 0);

  /* Persist and release the output page. */
  if( rc==LSM_OK ) rc = mergeWorkerPersistAndRelease(pMW);
  if( rc==LSM_OK ) rc = mergeWorkerBtreeIndirect(pMW);
  if( rc==LSM_OK ) rc = mergeWorkerFinishHierarchy(pMW);
  if( rc==LSM_OK ) rc = mergeWorkerAddPadding(pMW);
  lsmFsFlushWaiting(pMW->pDb->pFS, &rc);
  mergeWorkerReleaseAll(pMW);

  lsmFree(pMW->pDb->pEnv, pMW->aGobble);
  pMW->aGobble = 0;
  pMW->pCsr = 0;

  *pRc = rc;
}

/*
** The cursor passed as the first argument is being used as the input for
** a merge operation. When this function is called, *piFlags contains the
** database entry flags for the current entry. The entry about to be written
** to the output.
**
** Note that this function only has to work for cursors configured to 
** iterate forwards (not backwards).
*/
static void mergeRangeDeletes(MultiCursor *pCsr, int *piVal, int *piFlags){
  int f = *piFlags;
  int iKey = pCsr->aTree[1];
  int i;

  assert( pCsr->flags & CURSOR_NEXT_OK );
  if( pCsr->flags & CURSOR_IGNORE_DELETE ){
    /* The ignore-delete flag is set when the output of the merge will form
    ** the oldest level in the database. In this case there is no point in
    ** retaining any range-delete flags.  */
    assert( (f & LSM_POINT_DELETE)==0 );
    f &= ~(LSM_START_DELETE|LSM_END_DELETE);
  }else{
    for(i=0; i<(CURSOR_DATA_SEGMENT + pCsr->nPtr); i++){
      if( i!=iKey ){
        int eType;
        void *pKey;
        int nKey;
        int res;
        multiCursorGetKey(pCsr, i, &eType, &pKey, &nKey);

        if( pKey ){
          res = sortedKeyCompare(pCsr->pDb->xCmp, 
              rtTopic(pCsr->eType), pCsr->key.pData, pCsr->key.nData,
              rtTopic(eType), pKey, nKey
          );
          assert( res<=0 );
          if( res==0 ){
            if( (f & (LSM_INSERT|LSM_POINT_DELETE))==0 ){
              if( eType & LSM_INSERT ){
                f |= LSM_INSERT;
                *piVal = i;
              }
              else if( eType & LSM_POINT_DELETE ){
                f |= LSM_POINT_DELETE;
              }
            }
            f |= (eType & (LSM_END_DELETE|LSM_START_DELETE));
          }

          if( i>iKey && (eType & LSM_END_DELETE) && res<0 ){
            if( f & (LSM_INSERT|LSM_POINT_DELETE) ){
              f |= (LSM_END_DELETE|LSM_START_DELETE);
            }else{
              f = 0;
            }
            break;
          }
        }
      }
    }

    assert( (f & LSM_INSERT)==0 || (f & LSM_POINT_DELETE)==0 );
    if( (f & LSM_START_DELETE) 
     && (f & LSM_END_DELETE) 
     && (f & LSM_POINT_DELETE )
    ){
      f = 0;
    }
  }

  *piFlags = f;
}

static int mergeWorkerStep(MergeWorker *pMW){
  lsm_db *pDb = pMW->pDb;       /* Database handle */
  MultiCursor *pCsr;            /* Cursor to read input data from */
  int rc = LSM_OK;              /* Return code */
  int eType;                    /* SORTED_SEPARATOR, WRITE or DELETE */
  void *pKey; int nKey;         /* Key */
  Pgno iPtr;
  int iVal;

  pCsr = pMW->pCsr;

  /* Pull the next record out of the source cursor. */
  lsmMCursorKey(pCsr, &pKey, &nKey);
  eType = pCsr->eType;

  /* Figure out if the output record may have a different pointer value
  ** than the previous. This is the case if the current key is identical to
  ** a key that appears in the lowest level run being merged. If so, set 
  ** iPtr to the absolute pointer value. If not, leave iPtr set to zero, 
  ** indicating that the output pointer value should be a copy of the pointer 
  ** value written with the previous key.  */
  iPtr = (pCsr->pPrevMergePtr ? *pCsr->pPrevMergePtr : 0);
  if( pCsr->pBtCsr ){
    BtreeCursor *pBtCsr = pCsr->pBtCsr;
    if( pBtCsr->pKey ){
      int res = rtTopic(pBtCsr->eType) - rtTopic(eType);
      if( res==0 ) res = pDb->xCmp(pBtCsr->pKey, pBtCsr->nKey, pKey, nKey);
      if( 0==res ) iPtr = pBtCsr->iPtr;
      assert( res>=0 );
    }
  }else if( pCsr->nPtr ){
    SegmentPtr *pPtr = &pCsr->aPtr[pCsr->nPtr-1];
    if( pPtr->pPg
     && 0==pDb->xCmp(pPtr->pKey, pPtr->nKey, pKey, nKey)
    ){
      iPtr = pPtr->iPtr+pPtr->iPgPtr;
    }
  }

  iVal = pCsr->aTree[1];
  mergeRangeDeletes(pCsr, &iVal, &eType);

  if( eType!=0 ){
    if( pMW->aGobble ){
      int iGobble = pCsr->aTree[1] - CURSOR_DATA_SEGMENT;
      if( iGobble<pCsr->nPtr && iGobble>=0 ){
        SegmentPtr *pGobble = &pCsr->aPtr[iGobble];
        if( (pGobble->flags & PGFTR_SKIP_THIS_FLAG)==0 ){
          pMW->aGobble[iGobble] = lsmFsPageNumber(pGobble->pPg);
        }
      }
    }

    /* If this is a separator key and we know that the output pointer has not
    ** changed, there is no point in writing an output record. Otherwise,
    ** proceed. */
    if( rc==LSM_OK && (rtIsSeparator(eType)==0 || iPtr!=0) ){
      /* Write the record into the main run. */
      void *pVal; int nVal;
      rc = multiCursorGetVal(pCsr, iVal, &pVal, &nVal);
      if( pVal && rc==LSM_OK ){
        assert( nVal>=0 );
        rc = sortedBlobSet(pDb->pEnv, &pCsr->val, pVal, nVal);
        pVal = pCsr->val.pData;
      }
      if( rc==LSM_OK ){
        rc = mergeWorkerWrite(pMW, eType, pKey, nKey, pVal, nVal, iPtr);
      }
    }
  }

  /* Advance the cursor to the next input record (assuming one exists). */
  assert( lsmMCursorValid(pMW->pCsr) );
  if( rc==LSM_OK ) rc = lsmMCursorNext(pMW->pCsr);

  return rc;
}

static int mergeWorkerDone(MergeWorker *pMW){
  return pMW->pCsr==0 || !lsmMCursorValid(pMW->pCsr);
}

static void sortedFreeLevel(lsm_env *pEnv, Level *p){
  if( p ){
    lsmFree(pEnv, p->pSplitKey);
    lsmFree(pEnv, p->pMerge);
    lsmFree(pEnv, p->aRhs);
    lsmFree(pEnv, p);
  }
}

static void sortedInvokeWorkHook(lsm_db *pDb){
  if( pDb->xWork ){
    pDb->xWork(pDb, pDb->pWorkCtx);
  }
}

static int sortedNewToplevel(
  lsm_db *pDb,                    /* Connection handle */
  int eTree,                      /* One of the TREE_XXX constants */
  int *pnWrite                    /* OUT: Number of database pages written */
){
  int rc = LSM_OK;                /* Return Code */
  MultiCursor *pCsr = 0;
  Level *pNext = 0;               /* The current top level */
  Level *pNew;                    /* The new level itself */
  Segment *pLinked = 0;           /* Delete separators from this segment */
  Level *pDel = 0;                /* Delete this entire level */
  int nWrite = 0;                 /* Number of database pages written */
  Freelist freelist;

  if( eTree!=TREE_NONE ){
    rc = lsmShmCacheChunks(pDb, pDb->treehdr.nChunk);
  }

  assert( pDb->bUseFreelist==0 );
  pDb->pFreelist = &freelist;
  pDb->bUseFreelist = 1;
  memset(&freelist, 0, sizeof(freelist));

  /* Allocate the new level structure to write to. */
  pNext = lsmDbSnapshotLevel(pDb->pWorker);
  pNew = (Level *)lsmMallocZeroRc(pDb->pEnv, sizeof(Level), &rc);
  if( pNew ){
    pNew->pNext = pNext;
    lsmDbSnapshotSetLevel(pDb->pWorker, pNew);
  }

  /* Create a cursor to gather the data required by the new segment. The new
  ** segment contains everything in the tree and pointers to the next segment
  ** in the database (if any).  */
  pCsr = multiCursorNew(pDb, &rc);
  if( pCsr ){
    pCsr->pDb = pDb;
    rc = multiCursorVisitFreelist(pCsr);
    if( rc==LSM_OK ){
      rc = multiCursorAddTree(pCsr, pDb->pWorker, eTree);
    }
    if( rc==LSM_OK && pNext && pNext->pMerge==0 ){
      if( (pNext->flags & LEVEL_FREELIST_ONLY) ){
        pDel = pNext;
        pCsr->aPtr = lsmMallocZeroRc(pDb->pEnv, sizeof(SegmentPtr), &rc);
        multiCursorAddOne(pCsr, pNext, &rc);
      }else if( eTree!=TREE_NONE && pNext->lhs.iRoot ){
        pLinked = &pNext->lhs;
        rc = btreeCursorNew(pDb, pLinked, &pCsr->pBtCsr);
      }
    }

    /* If this will be the only segment in the database, discard any delete
    ** markers present in the in-memory tree.  */
    if( pNext==0 ){
      multiCursorIgnoreDelete(pCsr);
    }
  }

  if( rc!=LSM_OK ){
    lsmMCursorClose(pCsr, 0);
  }else{
    Pgno iLeftPtr = 0;
    Merge merge;                  /* Merge object used to create new level */
    MergeWorker mergeworker;      /* MergeWorker object for the same purpose */

    memset(&merge, 0, sizeof(Merge));
    memset(&mergeworker, 0, sizeof(MergeWorker));

    pNew->pMerge = &merge;
    pNew->flags |= LEVEL_INCOMPLETE;
    mergeworker.pDb = pDb;
    mergeworker.pLevel = pNew;
    mergeworker.pCsr = pCsr;
    pCsr->pPrevMergePtr = &iLeftPtr;

    /* Mark the separators array for the new level as a "phantom". */
    mergeworker.bFlush = 1;

    /* Do the work to create the new merged segment on disk */
    if( rc==LSM_OK ) rc = lsmMCursorFirst(pCsr);
    while( rc==LSM_OK && mergeWorkerDone(&mergeworker)==0 ){
      rc = mergeWorkerStep(&mergeworker);
    }
    mergeWorkerShutdown(&mergeworker, &rc);
    assert( rc!=LSM_OK || mergeworker.nWork==0 || pNew->lhs.iFirst );
    if( rc==LSM_OK && pNew->lhs.iFirst ){
      rc = lsmFsSortedFinish(pDb->pFS, &pNew->lhs);
    }
    nWrite = mergeworker.nWork;
    pNew->flags &= ~LEVEL_INCOMPLETE;
    if( eTree==TREE_NONE ){
      pNew->flags |= LEVEL_FREELIST_ONLY;
    }
    pNew->pMerge = 0;
  }

  if( rc!=LSM_OK || pNew->lhs.iFirst==0 ){
    assert( rc!=LSM_OK || pDb->pWorker->freelist.nEntry==0 );
    lsmDbSnapshotSetLevel(pDb->pWorker, pNext);
    sortedFreeLevel(pDb->pEnv, pNew);
  }else{
    if( pLinked ){
      pLinked->iRoot = 0;
    }else if( pDel ){
      assert( pNew->pNext==pDel );
      pNew->pNext = pDel->pNext;
      lsmFsSortedDelete(pDb->pFS, pDb->pWorker, 1, &pDel->lhs);
      sortedFreeLevel(pDb->pEnv, pDel);
    }

#if LSM_LOG_STRUCTURE
    lsmSortedDumpStructure(pDb, pDb->pWorker, LSM_LOG_DATA, 0, "new-toplevel");
#endif

    if( freelist.nEntry ){
      Freelist *p = &pDb->pWorker->freelist;
      lsmFree(pDb->pEnv, p->aEntry);
      memcpy(p, &freelist, sizeof(freelist));
      freelist.aEntry = 0;
    }else{
      pDb->pWorker->freelist.nEntry = 0;
    }

    assertBtreeOk(pDb, &pNew->lhs);
    sortedInvokeWorkHook(pDb);
  }

  if( pnWrite ) *pnWrite = nWrite;
  pDb->pWorker->nWrite += nWrite;
  pDb->pFreelist = 0;
  pDb->bUseFreelist = 0;
  lsmFree(pDb->pEnv, freelist.aEntry);
  return rc;
}

/*
** The nMerge levels in the LSM beginning with pLevel consist of a
** left-hand-side segment only. Replace these levels with a single new
** level consisting of a new empty segment on the left-hand-side and the
** nMerge segments from the replaced levels on the right-hand-side.
**
** Also, allocate and populate a Merge object and set Level.pMerge to
** point to it.
*/
static int sortedMergeSetup(
  lsm_db *pDb,                    /* Database handle */
  Level *pLevel,                  /* First level to merge */
  int nMerge,                     /* Merge this many levels together */
  Level **ppNew                   /* New, merged, level */
){
  int rc = LSM_OK;                /* Return Code */
  Level *pNew;                    /* New Level object */
  int bUseNext = 0;               /* True to link in next separators */
  Merge *pMerge;                  /* New Merge object */
  int nByte;                      /* Bytes of space allocated at pMerge */

#ifdef LSM_DEBUG
  int iLevel;
  Level *pX = pLevel;
  for(iLevel=0; iLevel<nMerge; iLevel++){
    assert( pX->nRight==0 );
    pX = pX->pNext;
  }
#endif

  /* Allocate the new Level object */
  pNew = (Level *)lsmMallocZeroRc(pDb->pEnv, sizeof(Level), &rc);
  if( pNew ){
    pNew->aRhs = (Segment *)lsmMallocZeroRc(pDb->pEnv, 
                                        nMerge * sizeof(Segment), &rc);
  }

  /* Populate the new Level object */
  if( rc==LSM_OK ){
    Level *pNext = 0;             /* Level following pNew */
    int i;
    int bFreeOnly = 1;
    Level *pTopLevel;
    Level *p = pLevel;
    Level **pp;
    pNew->nRight = nMerge;
    pNew->iAge = pLevel->iAge+1;
    for(i=0; i<nMerge; i++){
      assert( p->nRight==0 );
      pNext = p->pNext;
      pNew->aRhs[i] = p->lhs;
      if( (p->flags & LEVEL_FREELIST_ONLY)==0 ) bFreeOnly = 0;
      sortedFreeLevel(pDb->pEnv, p);
      p = pNext;
    }

    if( bFreeOnly ) pNew->flags |= LEVEL_FREELIST_ONLY;

    /* Replace the old levels with the new. */
    pTopLevel = lsmDbSnapshotLevel(pDb->pWorker);
    pNew->pNext = p;
    for(pp=&pTopLevel; *pp!=pLevel; pp=&((*pp)->pNext));
    *pp = pNew;
    lsmDbSnapshotSetLevel(pDb->pWorker, pTopLevel);

    /* Determine whether or not the next separators will be linked in */
    if( pNext && pNext->pMerge==0 && pNext->lhs.iRoot && pNext 
     && (bFreeOnly==0 || (pNext->flags & LEVEL_FREELIST_ONLY))
    ){
      bUseNext = 1;
    }
  }

  /* Allocate the merge object */
  nByte = sizeof(Merge) + sizeof(MergeInput) * (nMerge + bUseNext);
  pMerge = (Merge *)lsmMallocZeroRc(pDb->pEnv, nByte, &rc);
  if( pMerge ){
    pMerge->aInput = (MergeInput *)&pMerge[1];
    pMerge->nInput = nMerge + bUseNext;
    pNew->pMerge = pMerge;
  }

  *ppNew = pNew;
  return rc;
}

static int mergeWorkerInit(
  lsm_db *pDb,                    /* Db connection to do merge work */
  Level *pLevel,                  /* Level to work on merging */
  MergeWorker *pMW                /* Object to initialize */
){
  int rc = LSM_OK;                /* Return code */
  Merge *pMerge = pLevel->pMerge; /* Persistent part of merge state */
  MultiCursor *pCsr = 0;          /* Cursor opened for pMW */
  Level *pNext = pLevel->pNext;   /* Next level in LSM */

  assert( pDb->pWorker );
  assert( pLevel->pMerge );
  assert( pLevel->nRight>0 );

  memset(pMW, 0, sizeof(MergeWorker));
  pMW->pDb = pDb;
  pMW->pLevel = pLevel;
  pMW->aGobble = lsmMallocZeroRc(pDb->pEnv, sizeof(Pgno) * pLevel->nRight, &rc);

  /* Create a multi-cursor to read the data to write to the new
  ** segment. The new segment contains:
  **
  **   1. Records from LHS of each of the nMerge levels being merged.
  **   2. Separators from either the last level being merged, or the
  **      separators attached to the LHS of the following level, or neither.
  **
  ** If the new level is the lowest (oldest) in the db, discard any
  ** delete keys. Key annihilation.
  */
  pCsr = multiCursorNew(pDb, &rc);
  if( pCsr ){
    pCsr->flags |= CURSOR_NEXT_OK;
    rc = multiCursorAddRhs(pCsr, pLevel);
  }
  if( rc==LSM_OK && pMerge->nInput > pLevel->nRight ){
    rc = btreeCursorNew(pDb, &pNext->lhs, &pCsr->pBtCsr);
  }else if( pNext ){
    multiCursorReadSeparators(pCsr);
  }else{
    multiCursorIgnoreDelete(pCsr);
  }

  assert( rc!=LSM_OK || pMerge->nInput==(pCsr->nPtr+(pCsr->pBtCsr!=0)) );
  pMW->pCsr = pCsr;

  /* Load the b-tree hierarchy into memory. */
  if( rc==LSM_OK ) rc = mergeWorkerLoadHierarchy(pMW);
  if( rc==LSM_OK && pMW->hier.nHier==0 ){
    pMW->aSave[0].iPgno = pLevel->lhs.iFirst;
  }

  /* Position the cursor. */
  if( rc==LSM_OK ){
    pCsr->pPrevMergePtr = &pMerge->iCurrentPtr;
    if( pLevel->lhs.iFirst==0 ){
      /* The output array is still empty. So position the cursor at the very 
      ** start of the input.  */
      rc = multiCursorEnd(pCsr, 0);
    }else{
      /* The output array is non-empty. Position the cursor based on the
      ** page/cell data saved in the Merge.aInput[] array.  */
      int i;
      for(i=0; rc==LSM_OK && i<pCsr->nPtr; i++){
        MergeInput *pInput = &pMerge->aInput[i];
        if( pInput->iPg ){
          SegmentPtr *pPtr;
          assert( pCsr->aPtr[i].pPg==0 );
          pPtr = &pCsr->aPtr[i];
          rc = segmentPtrLoadPage(pDb->pFS, pPtr, pInput->iPg);
          if( rc==LSM_OK && pPtr->nCell>0 ){
            rc = segmentPtrLoadCell(pPtr, pInput->iCell);
          }
        }
      }

      if( rc==LSM_OK && pCsr->pBtCsr ){
        int (*xCmp)(void *, int, void *, int) = pCsr->pDb->xCmp;
        assert( i==pCsr->nPtr );
        rc = btreeCursorRestore(pCsr->pBtCsr, xCmp, &pMerge->aInput[i]);
      }

      if( rc==LSM_OK ){
        rc = multiCursorSetupTree(pCsr, 0);
      }
    }
    pCsr->flags |= CURSOR_NEXT_OK;
  }

  return rc;
}

static int sortedBtreeGobble(
  lsm_db *pDb,                    /* Worker connection */
  MultiCursor *pCsr,              /* Multi-cursor being used for a merge */
  int iGobble                     /* pCsr->aPtr[] entry to operate on */
){
  int rc = LSM_OK;
  if( rtTopic(pCsr->eType)==0 ){
    Segment *pSeg = pCsr->aPtr[iGobble].pSeg;
    Pgno *aPg;
    int nPg;

    /* Seek from the root of the b-tree to the segment leaf that may contain
    ** a key equal to the one multi-cursor currently points to. Record the
    ** page number of each b-tree page and the leaf. The segment may be
    ** gobbled up to (but not including) the first of these page numbers.
    */
    assert( pSeg->iRoot>0 );
    aPg = lsmMallocZeroRc(pDb->pEnv, sizeof(Pgno)*32, &rc);
    if( rc==LSM_OK ){
      rc = seekInBtree(pCsr, pSeg, 
          rtTopic(pCsr->eType), pCsr->key.pData, pCsr->key.nData, aPg, 0
      ); 
    }

    if( rc==LSM_OK ){
      for(nPg=0; aPg[nPg]; nPg++);
      lsmFsGobble(pDb, pSeg, aPg, nPg);
    }

    lsmFree(pDb->pEnv, aPg);
  }
  return rc;
}

/*
** Argument p points to a level of age N. Return the number of levels in
** the linked list starting at p that have age=N (always at least 1).
*/
static int sortedCountLevels(Level *p){
  int iAge = p->iAge;
  int nRet = 0;
  do {
    nRet++;
    p = p->pNext;
  }while( p && p->iAge==iAge );
  return nRet;
}

static int sortedSelectLevel(lsm_db *pDb, int nMerge, Level **ppOut){
  Level *pTopLevel = lsmDbSnapshotLevel(pDb->pWorker);
  int rc = LSM_OK;
  Level *pLevel = 0;            /* Output value */
  Level *pBest = 0;             /* Best level to work on found so far */
  int nBest;                    /* Number of segments merged at pBest */
  Level *pThis = 0;             /* First in run of levels with age=iAge */
  int nThis = 0;                /* Number of levels starting at pThis */

  assert( nMerge>=1 );
  nBest = LSM_MAX(1, nMerge-1);

  /* Find the longest contiguous run of levels not currently undergoing a 
  ** merge with the same age in the structure. Or the level being merged
  ** with the largest number of right-hand segments. Work on it. */
  for(pLevel=pTopLevel; pLevel; pLevel=pLevel->pNext){
    if( pLevel->nRight==0 && pThis && pLevel->iAge==pThis->iAge ){
      nThis++;
    }else{
      if( nThis>nBest ){
        if( (pLevel->iAge!=pThis->iAge+1)
         || (pLevel->nRight==0 && sortedCountLevels(pLevel)<=pDb->nMerge)
        ){
          pBest = pThis;
          nBest = nThis;
        }
      }
      if( pLevel->nRight ){
        if( pLevel->nRight>nBest ){
          nBest = pLevel->nRight;
          pBest = pLevel;
        }
        nThis = 0;
        pThis = 0;
      }else{
        pThis = pLevel;
        nThis = 1;
      }
    }
  }
  if( nThis>nBest ){
    assert( pThis );
    pBest = pThis;
    nBest = nThis;
  }

  if( pBest==0 && nMerge==1 ){
    int nFree = 0;
    int nUsr = 0;
    for(pLevel=pTopLevel; pLevel; pLevel=pLevel->pNext){
      assert( !pLevel->nRight );
      if( pLevel->flags & LEVEL_FREELIST_ONLY ){
        nFree++;
      }else{
        nUsr++;
      }
    }
    if( nUsr>1 ){
      pBest = pTopLevel;
      nBest = nFree + nUsr;
    }
  }

  if( pBest ){
    if( pBest->nRight==0 ){
      rc = sortedMergeSetup(pDb, pBest, nBest, ppOut);
    }else{
      *ppOut = pBest;
    }
  }

  return rc;
}

static int sortedDbIsFull(lsm_db *pDb){
  Level *pTop = lsmDbSnapshotLevel(pDb->pWorker);

  if( lsmDatabaseFull(pDb) ) return 1;
  if( pTop && pTop->iAge==0
   && (pTop->nRight || sortedCountLevels(pTop)>=pDb->nMerge)
  ){
    return 1;
  }
  return 0;
}

typedef struct MoveBlockCtx MoveBlockCtx;
struct MoveBlockCtx {
  int iSeen;                      /* Previous free block on list */
  int iFrom;                      /* Total number of blocks in file */
};

static int moveBlockCb(void *pCtx, int iBlk, i64 iSnapshot){
  MoveBlockCtx *p = (MoveBlockCtx *)pCtx;
  assert( p->iFrom==0 );
  if( iBlk==(p->iSeen-1) ){
    p->iSeen = iBlk;
    return 0;
  }
  p->iFrom = p->iSeen-1;
  return 1;
}

/*
** This function is called to further compact a database for which all 
** of the content has already been merged into a single segment. If 
** possible, it moves the contents of a single block from the end of the
** file to a free-block that lies closer to the start of the file (allowing
** the file to be eventually truncated).
*/
static int sortedMoveBlock(lsm_db *pDb, int *pnWrite){
  Snapshot *p = pDb->pWorker;
  Level *pLvl = lsmDbSnapshotLevel(p);
  int iFrom;                      /* Block to move */
  int iTo;                        /* Destination to move block to */
  int rc;                         /* Return code */

  MoveBlockCtx sCtx;

  assert( pLvl->pNext==0 && pLvl->nRight==0 );
  assert( p->redirect.n<=LSM_MAX_BLOCK_REDIRECTS );

  *pnWrite = 0;

  /* Check that the redirect array is not already full. If it is, return
  ** without moving any database content.  */
  if( p->redirect.n>=LSM_MAX_BLOCK_REDIRECTS ) return LSM_OK;

  /* Find the last block of content in the database file. Do this by 
  ** traversing the free-list in reverse (descending block number) order.
  ** The first block not on the free list is the one that will be moved.
  ** Since the db consists of a single segment, there is no ambiguity as
  ** to which segment the block belongs to.  */
  sCtx.iSeen = p->nBlock+1;
  sCtx.iFrom = 0;
  rc = lsmWalkFreelist(pDb, 1, moveBlockCb, &sCtx);
  if( rc!=LSM_OK || sCtx.iFrom==0 ) return rc;
  iFrom = sCtx.iFrom;

  /* Find the first free block in the database, ignoring block 1. Block
  ** 1 is tricky as it is smaller than the other blocks.  */
  rc = lsmBlockAllocate(pDb, iFrom, &iTo);
  if( rc!=LSM_OK || iTo==0 ) return rc;
  assert( iTo!=1 && iTo<iFrom );

  rc = lsmFsMoveBlock(pDb->pFS, &pLvl->lhs, iTo, iFrom);
  if( rc==LSM_OK ){
    if( p->redirect.a==0 ){
      int nByte = sizeof(struct RedirectEntry) * LSM_MAX_BLOCK_REDIRECTS;
      p->redirect.a = lsmMallocZeroRc(pDb->pEnv, nByte, &rc);
    }
    if( rc==LSM_OK ){

      /* Check if the block just moved was already redirected. */
      int i;
      for(i=0; i<p->redirect.n; i++){
        if( p->redirect.a[i].iTo==iFrom ) break;
      }

      if( i==p->redirect.n ){
        /* Block iFrom was not already redirected. Add a new array entry. */
        memmove(&p->redirect.a[1], &p->redirect.a[0], 
            sizeof(struct RedirectEntry) * p->redirect.n
            );
        p->redirect.a[0].iFrom = iFrom;
        p->redirect.a[0].iTo = iTo;
        p->redirect.n++;
      }else{
        /* Block iFrom was already redirected. Overwrite existing entry. */
        p->redirect.a[i].iTo = iTo;
      }

      rc = lsmBlockFree(pDb, iFrom);

      *pnWrite = lsmFsBlockSize(pDb->pFS) / lsmFsPageSize(pDb->pFS);
      pLvl->lhs.pRedirect = &p->redirect;
    }
  }

#if LSM_LOG_STRUCTURE
  if( rc==LSM_OK ){
    char aBuf[64];
    sprintf(aBuf, "move-block %d/%d", p->redirect.n-1, LSM_MAX_BLOCK_REDIRECTS);
    lsmSortedDumpStructure(pDb, pDb->pWorker, LSM_LOG_DATA, 0, aBuf);
  }
#endif
  return rc;
}

/*
*/
static int mergeInsertFreelistSegments(
  lsm_db *pDb, 
  int nFree,
  MergeWorker *pMW
){
  int rc = LSM_OK;
  if( nFree>0 ){
    MultiCursor *pCsr = pMW->pCsr;
    Level *pLvl = pMW->pLevel;
    SegmentPtr *aNew1;
    Segment *aNew2;

    Level *pIter;
    Level *pNext;
    int i = 0;

    aNew1 = (SegmentPtr *)lsmMallocZeroRc(
        pDb->pEnv, sizeof(SegmentPtr) * (pCsr->nPtr+nFree), &rc
    );
    if( rc ) return rc;
    memcpy(&aNew1[nFree], pCsr->aPtr, sizeof(SegmentPtr)*pCsr->nPtr);
    pCsr->nPtr += nFree;
    lsmFree(pDb->pEnv, pCsr->aTree);
    lsmFree(pDb->pEnv, pCsr->aPtr);
    pCsr->aTree = 0;
    pCsr->aPtr = aNew1;

    aNew2 = (Segment *)lsmMallocZeroRc(
        pDb->pEnv, sizeof(Segment) * (pLvl->nRight+nFree), &rc
    );
    if( rc ) return rc;
    memcpy(&aNew2[nFree], pLvl->aRhs, sizeof(Segment)*pLvl->nRight);
    pLvl->nRight += nFree;
    lsmFree(pDb->pEnv, pLvl->aRhs);
    pLvl->aRhs = aNew2;

    for(pIter=pDb->pWorker->pLevel; rc==LSM_OK && pIter!=pLvl; pIter=pNext){
      Segment *pSeg = &pLvl->aRhs[i];
      memcpy(pSeg, &pIter->lhs, sizeof(Segment));

      pCsr->aPtr[i].pSeg = pSeg;
      pCsr->aPtr[i].pLevel = pLvl;
      rc = segmentPtrEnd(pCsr, &pCsr->aPtr[i], 0);

      pDb->pWorker->pLevel = pNext = pIter->pNext;
      sortedFreeLevel(pDb->pEnv, pIter);
      i++;
    }
    assert( i==nFree );
    assert( rc!=LSM_OK || pDb->pWorker->pLevel==pLvl );

    for(i=nFree; i<pCsr->nPtr; i++){
      pCsr->aPtr[i].pSeg = &pLvl->aRhs[i];
    }

    lsmFree(pDb->pEnv, pMW->aGobble);
    pMW->aGobble = 0;
  }
  return rc;
}

static int sortedWork(
  lsm_db *pDb,                    /* Database handle. Must be worker. */
  int nWork,                      /* Number of pages of work to do */
  int nMerge,                     /* Try to merge this many levels at once */
  int bFlush,                     /* Set if call is to make room for a flush */
  int *pnWrite                    /* OUT: Actual number of pages written */
){
  int rc = LSM_OK;                /* Return Code */
  int nRemaining = nWork;         /* Units of work to do before returning */
  Snapshot *pWorker = pDb->pWorker;

  assert( pWorker );
  if( lsmDbSnapshotLevel(pWorker)==0 ) return LSM_OK;

  while( nRemaining>0 ){
    Level *pLevel = 0;

    /* Find a level to work on. */
    rc = sortedSelectLevel(pDb, nMerge, &pLevel);
    assert( rc==LSM_OK || pLevel==0 );

    if( pLevel==0 ){
      int nDone = 0;
      Level *pTopLevel = lsmDbSnapshotLevel(pDb->pWorker);
      if( bFlush==0 && nMerge==1 && pTopLevel && pTopLevel->pNext==0 ){
        rc = sortedMoveBlock(pDb, &nDone);
      }
      nRemaining -= nDone;

      /* Could not find any work to do. Finished. */
      if( nDone==0 ) break;
    }else{
      int bSave = 0;
      Freelist freelist = {0, 0, 0};
      MergeWorker mergeworker;    /* State used to work on the level merge */

      assert( pDb->bIncrMerge==0 );
      assert( pDb->pFreelist==0 && pDb->bUseFreelist==0 );

      pDb->bIncrMerge = 1;
      rc = mergeWorkerInit(pDb, pLevel, &mergeworker);
      assert( mergeworker.nWork==0 );
      
      while( rc==LSM_OK 
          && 0==mergeWorkerDone(&mergeworker) 
          && (mergeworker.nWork<nRemaining || pDb->bUseFreelist)
      ){
        int eType = rtTopic(mergeworker.pCsr->eType);
        rc = mergeWorkerStep(&mergeworker);

        /* If the cursor now points at the first entry past the end of the
        ** user data (i.e. either to EOF or to the first free-list entry
        ** that will be added to the run), then check if it is possible to
        ** merge in any free-list entries that are either in-memory or in
        ** free-list-only blocks.  */
        if( rc==LSM_OK && nMerge==1 && eType==0
         && (rtTopic(mergeworker.pCsr->eType) || mergeWorkerDone(&mergeworker))
        ){
          int nFree = 0;          /* Number of free-list-only levels to merge */
          Level *pLvl;
          assert( pDb->pFreelist==0 && pDb->bUseFreelist==0 );

          /* Now check if all levels containing data newer than this one
          ** are single-segment free-list only levels. If so, they will be
          ** merged in now.  */
          for(pLvl=pDb->pWorker->pLevel; 
              pLvl!=mergeworker.pLevel && (pLvl->flags & LEVEL_FREELIST_ONLY); 
              pLvl=pLvl->pNext
          ){
            assert( pLvl->nRight==0 );
            nFree++;
          }
          if( pLvl==mergeworker.pLevel ){

            rc = mergeInsertFreelistSegments(pDb, nFree, &mergeworker);
            if( rc==LSM_OK ){
              rc = multiCursorVisitFreelist(mergeworker.pCsr);
            }
            if( rc==LSM_OK ){
              rc = multiCursorSetupTree(mergeworker.pCsr, 0);
              pDb->pFreelist = &freelist;
              pDb->bUseFreelist = 1;
            }
          }
        }
      }
      nRemaining -= LSM_MAX(mergeworker.nWork, 1);

      if( rc==LSM_OK ){
        /* Check if the merge operation is completely finished. If not,
        ** gobble up (declare eligible for recycling) any pages from rhs
        ** segments for which the content has been completely merged into 
        ** the lhs of the level.  */
        if( mergeWorkerDone(&mergeworker)==0 ){
          int i;
          for(i=0; i<pLevel->nRight; i++){
            SegmentPtr *pGobble = &mergeworker.pCsr->aPtr[i];
            if( pGobble->pSeg->iRoot ){
              rc = sortedBtreeGobble(pDb, mergeworker.pCsr, i);
            }else if( mergeworker.aGobble[i] ){
              lsmFsGobble(pDb, pGobble->pSeg, &mergeworker.aGobble[i], 1);
            }
          }
        }else{
          int i;
          int bEmpty;
          mergeWorkerShutdown(&mergeworker, &rc);
          bEmpty = (pLevel->lhs.iFirst==0);

          if( bEmpty==0 && rc==LSM_OK ){
            rc = lsmFsSortedFinish(pDb->pFS, &pLevel->lhs);
          }

          if( pDb->bUseFreelist ){
            Freelist *p = &pDb->pWorker->freelist;
            lsmFree(pDb->pEnv, p->aEntry);
            memcpy(p, &freelist, sizeof(freelist));
            pDb->bUseFreelist = 0;
            pDb->pFreelist = 0;
            bSave = 1;
          }

          for(i=0; i<pLevel->nRight; i++){
            lsmFsSortedDelete(pDb->pFS, pWorker, 1, &pLevel->aRhs[i]);
          }

          if( bEmpty ){
            /* If the new level is completely empty, remove it from the 
            ** database snapshot. This can only happen if all input keys were
            ** annihilated. Since keys are only annihilated if the new level
            ** is the last in the linked list (contains the most ancient of
            ** database content), this guarantees that pLevel->pNext==0.  */ 
            Level *pTop;          /* Top level of worker snapshot */
            Level **pp;           /* Read/write iterator for Level.pNext list */

            assert( pLevel->pNext==0 );

            /* Remove the level from the worker snapshot. */
            pTop = lsmDbSnapshotLevel(pWorker);
            for(pp=&pTop; *pp!=pLevel; pp=&((*pp)->pNext));
            *pp = pLevel->pNext;
            lsmDbSnapshotSetLevel(pWorker, pTop);

            /* Free the Level structure. */
            sortedFreeLevel(pDb->pEnv, pLevel);
          }else{

            /* Free the separators of the next level, if required. */
            if( pLevel->pMerge->nInput > pLevel->nRight ){
              assert( pLevel->pNext->lhs.iRoot );
              pLevel->pNext->lhs.iRoot = 0;
            }

            /* Zero the right-hand-side of pLevel */
            lsmFree(pDb->pEnv, pLevel->aRhs);
            pLevel->nRight = 0;
            pLevel->aRhs = 0;

            /* Free the Merge object */
            lsmFree(pDb->pEnv, pLevel->pMerge);
            pLevel->pMerge = 0;
          }

          if( bSave && rc==LSM_OK ){
            pDb->bIncrMerge = 0;
            rc = lsmSaveWorker(pDb, 0);
          }
        }
      }

      /* Clean up the MergeWorker object initialized above. If no error
      ** has occurred, invoke the work-hook to inform the application that
      ** the database structure has changed. */
      mergeWorkerShutdown(&mergeworker, &rc);
      pDb->bIncrMerge = 0;
      if( rc==LSM_OK ) sortedInvokeWorkHook(pDb);

#if LSM_LOG_STRUCTURE
      lsmSortedDumpStructure(pDb, pDb->pWorker, LSM_LOG_DATA, 0, "work");
#endif
      assertBtreeOk(pDb, &pLevel->lhs);
      assertRunInOrder(pDb, &pLevel->lhs);

      /* If bFlush is true and the database is no longer considered "full",
      ** break out of the loop even if nRemaining is still greater than
      ** zero. The caller has an in-memory tree to flush to disk.  */
      if( bFlush && sortedDbIsFull(pDb)==0 ) break;
    }
  }

  if( pnWrite ) *pnWrite = (nWork - nRemaining);
  pWorker->nWrite += (nWork - nRemaining);

#ifdef LSM_LOG_WORK
  lsmLogMessage(pDb, rc, "sortedWork(): %d pages", (nWork-nRemaining));
#endif
  return rc;
}

/*
** The database connection passed as the first argument must be a worker
** connection. This function checks if there exists an "old" in-memory tree
** ready to be flushed to disk. If so, true is returned. Otherwise false.
**
** If an error occurs, *pRc is set to an LSM error code before returning.
** It is assumed that *pRc is set to LSM_OK when this function is called.
*/
static int sortedTreeHasOld(lsm_db *pDb, int *pRc){
  int rc = LSM_OK;
  int bRet = 0;

  assert( pDb->pWorker );
  if( *pRc==LSM_OK ){
    if( rc==LSM_OK 
        && pDb->treehdr.iOldShmid
        && pDb->treehdr.iOldLog!=pDb->pWorker->iLogOff 
      ){
      bRet = 1;
    }else{
      bRet = 0;
    }
    *pRc = rc;
  }
  assert( *pRc==LSM_OK || bRet==0 );
  return bRet;
}

/*
** Create a new free-list only top-level segment. Return LSM_OK if successful
** or an LSM error code if some error occurs.
*/
static int sortedNewFreelistOnly(lsm_db *pDb){
  return sortedNewToplevel(pDb, TREE_NONE, 0);
}

int lsmSaveWorker(lsm_db *pDb, int bFlush){
  Snapshot *p = pDb->pWorker;
  if( p->freelist.nEntry>pDb->nMaxFreelist ){
    int rc = sortedNewFreelistOnly(pDb);
    if( rc!=LSM_OK ) return rc;
  }
  return lsmCheckpointSaveWorker(pDb, bFlush);
}

static int doLsmSingleWork(
  lsm_db *pDb, 
  int bShutdown,
  int nMerge,                     /* Minimum segments to merge together */
  int nPage,                      /* Number of pages to write to disk */
  int *pnWrite,                   /* OUT: Pages actually written to disk */
  int *pbCkpt                     /* OUT: True if an auto-checkpoint is req. */
){
  Snapshot *pWorker;              /* Worker snapshot */
  int rc = LSM_OK;                /* Return code */
  int bDirty = 0;
  int nMax = nPage;               /* Maximum pages to write to disk */
  int nRem = nPage;
  int bCkpt = 0;

  assert( nPage>0 );

  /* Open the worker 'transaction'. It will be closed before this function
  ** returns.  */
  assert( pDb->pWorker==0 );
  rc = lsmBeginWork(pDb);
  if( rc!=LSM_OK ) return rc;
  pWorker = pDb->pWorker;

  /* If this connection is doing auto-checkpoints, set nMax (and nRem) so
  ** that this call stops writing when the auto-checkpoint is due. The
  ** caller will do the checkpoint, then possibly call this function again. */
  if( bShutdown==0 && pDb->nAutockpt ){
    u32 nSync;
    u32 nUnsync;
    int nPgsz;

    lsmCheckpointSynced(pDb, 0, 0, &nSync);
    nUnsync = lsmCheckpointNWrite(pDb->pShmhdr->aSnap1, 0);
    nPgsz = lsmCheckpointPgsz(pDb->pShmhdr->aSnap1);

    nMax = LSM_MIN(nMax, (pDb->nAutockpt/nPgsz) - (int)(nUnsync-nSync));
    if( nMax<nRem ){
      bCkpt = 1;
      nRem = LSM_MAX(nMax, 0);
    }
  }

  /* If there exists in-memory data ready to be flushed to disk, attempt
  ** to flush it now.  */
  if( pDb->nTransOpen==0 ){
    rc = lsmTreeLoadHeader(pDb, 0);
  }
  if( sortedTreeHasOld(pDb, &rc) ){
    /* sortedDbIsFull() returns non-zero if either (a) there are too many
    ** levels in total in the db, or (b) there are too many levels with the
    ** the same age in the db. Either way, call sortedWork() to merge 
    ** existing segments together until this condition is cleared.  */
    if( sortedDbIsFull(pDb) ){
      int nPg = 0;
      rc = sortedWork(pDb, nRem, nMerge, 1, &nPg);
      nRem -= nPg;
      assert( rc!=LSM_OK || nRem<=0 || !sortedDbIsFull(pDb) );
      bDirty = 1;
    }

    if( rc==LSM_OK && nRem>0 ){
      int nPg = 0;
      rc = sortedNewToplevel(pDb, TREE_OLD, &nPg);
      nRem -= nPg;
      if( rc==LSM_OK ){
        if( pDb->nTransOpen>0 ){
          lsmTreeDiscardOld(pDb);
        }
        rc = lsmSaveWorker(pDb, 1);
        bDirty = 0;
      }
    }
  }

  /* If nPage is still greater than zero, do some merging. */
  if( rc==LSM_OK && nRem>0 && bShutdown==0 ){
    int nPg = 0;
    rc = sortedWork(pDb, nRem, nMerge, 0, &nPg);
    nRem -= nPg;
    if( nPg ) bDirty = 1;
  }

  /* If the in-memory part of the free-list is too large, write a new 
  ** top-level containing just the in-memory free-list entries to disk. */
  if( rc==LSM_OK && pDb->pWorker->freelist.nEntry > pDb->nMaxFreelist ){
    int nPg = 0;
    while( rc==LSM_OK && lsmDatabaseFull(pDb) ){
      rc = sortedWork(pDb, 16, nMerge, 1, &nPg);
      nRem -= nPg;
    }
    if( rc==LSM_OK ){
      rc = sortedNewFreelistOnly(pDb);
    }
    nRem -= nPg;
    if( nPg ) bDirty = 1;
  }

  if( rc==LSM_OK ){
    *pnWrite = (nMax - nRem);
    *pbCkpt = (bCkpt && nRem<=0);
    if( nMerge==1 && pDb->nAutockpt>0 && *pnWrite>0
     && pWorker->pLevel 
     && pWorker->pLevel->nRight==0 
     && pWorker->pLevel->pNext==0 
    ){
      *pbCkpt = 1;
    }
  }

  if( rc==LSM_OK && bDirty ){
    lsmFinishWork(pDb, 0, &rc);
  }else{
    int rcdummy = LSM_BUSY;
    lsmFinishWork(pDb, 0, &rcdummy);
    *pnWrite = 0;
  }
  assert( pDb->pWorker==0 );
  return rc;
}

static int doLsmWork(lsm_db *pDb, int nMerge, int nPage, int *pnWrite){
  int rc = LSM_OK;                /* Return code */
  int nWrite = 0;                 /* Number of pages written */

  assert( nMerge>=1 );

  if( nPage!=0 ){
    int bCkpt = 0;
    do {
      int nThis = 0;
      int nReq = (nPage>=0) ? (nPage-nWrite) : ((int)0x7FFFFFFF);

      bCkpt = 0;
      rc = doLsmSingleWork(pDb, 0, nMerge, nReq, &nThis, &bCkpt);
      nWrite += nThis;
      if( rc==LSM_OK && bCkpt ){
        rc = lsm_checkpoint(pDb, 0);
      }
    }while( rc==LSM_OK && bCkpt && (nWrite<nPage || nPage<0) );
  }

  if( pnWrite ){
    if( rc==LSM_OK ){
      *pnWrite = nWrite;
    }else{
      *pnWrite = 0;
    }
  }
  return rc;
}

/*
** Perform work to merge database segments together.
*/
int lsm_work(lsm_db *pDb, int nMerge, int nKB, int *pnWrite){
  int rc;                         /* Return code */
  int nPgsz;                      /* Nominal page size in bytes */
  int nPage;                      /* Equivalent of nKB in pages */
  int nWrite = 0;                 /* Number of pages written */

  /* This function may not be called if pDb has an open read or write
  ** transaction. Return LSM_MISUSE if an application attempts this.  */
  if( pDb->nTransOpen || pDb->pCsr ) return LSM_MISUSE_BKPT;
  if( nMerge<=0 ) nMerge = pDb->nMerge;

  lsmFsPurgeCache(pDb->pFS);

  /* Convert from KB to pages */
  nPgsz = lsmFsPageSize(pDb->pFS);
  if( nKB>=0 ){
    nPage = ((i64)nKB * 1024 + nPgsz - 1) / nPgsz;
  }else{
    nPage = -1;
  }

  rc = doLsmWork(pDb, nMerge, nPage, &nWrite);
  
  if( pnWrite ){
    /* Convert back from pages to KB */
    *pnWrite = (int)(((i64)nWrite * 1024 + nPgsz - 1) / nPgsz);
  }
  return rc;
}

int lsm_flush(lsm_db *db){
  int rc;

  if( db->nTransOpen>0 || db->pCsr ){
    rc = LSM_MISUSE_BKPT;
  }else{
    rc = lsmBeginWriteTrans(db);
    if( rc==LSM_OK ){
      lsmFlushTreeToDisk(db);
      lsmTreeDiscardOld(db);
      lsmTreeMakeOld(db);
      lsmTreeDiscardOld(db);
    }

    if( rc==LSM_OK ){
      rc = lsmFinishWriteTrans(db, 1);
    }else{
      lsmFinishWriteTrans(db, 0);
    }
    lsmFinishReadTrans(db);
  }

  return rc;
}

/*
** This function is called in auto-work mode to perform merging work on
** the data structure. It performs enough merging work to prevent the
** height of the tree from growing indefinitely assuming that roughly
** nUnit database pages worth of data have been written to the database
** (i.e. the in-memory tree) since the last call.
*/
int lsmSortedAutoWork(
  lsm_db *pDb,                    /* Database handle */
  int nUnit                       /* Pages of data written to in-memory tree */
){
  int rc = LSM_OK;                /* Return code */
  int nDepth = 0;                 /* Current height of tree (longest path) */
  Level *pLevel;                  /* Used to iterate through levels */
  int bRestore = 0;

  assert( pDb->pWorker==0 );
  assert( pDb->nTransOpen>0 );

  /* Determine how many units of work to do before returning. One unit of
  ** work is achieved by writing one page (~4KB) of merged data.  */
  for(pLevel=lsmDbSnapshotLevel(pDb->pClient); pLevel; pLevel=pLevel->pNext){
    /* nDepth += LSM_MAX(1, pLevel->nRight); */
    nDepth += 1;
  }
  if( lsmTreeHasOld(pDb) ){
    nDepth += 1;
    bRestore = 1;
    rc = lsmSaveCursors(pDb);
    if( rc!=LSM_OK ) return rc;
  }

  if( nDepth>0 ){
    int nRemaining;               /* Units of work to do before returning */

    nRemaining = nUnit * nDepth;
#ifdef LSM_LOG_WORK
    lsmLogMessage(pDb, rc, "lsmSortedAutoWork(): %d*%d = %d pages", 
        nUnit, nDepth, nRemaining);
#endif
    assert( nRemaining>=0 );
    rc = doLsmWork(pDb, pDb->nMerge, nRemaining, 0);
    if( rc==LSM_BUSY ) rc = LSM_OK;

    if( bRestore && pDb->pCsr ){
      lsmMCursorFreeCache(pDb);
      lsmFreeSnapshot(pDb->pEnv, pDb->pClient);
      pDb->pClient = 0;
      rc = lsmCheckpointLoad(pDb, 0);
      if( rc==LSM_OK ){
        rc = lsmCheckpointDeserialize(pDb, 0, pDb->aSnapshot, &pDb->pClient);
      }
      if( rc==LSM_OK ){
        rc = lsmRestoreCursors(pDb);
      }
    }
  }

  return rc;
}

/*
** This function is only called during system shutdown. The contents of
** any in-memory trees present (old or current) are written out to disk.
*/
int lsmFlushTreeToDisk(lsm_db *pDb){
  int rc;

  rc = lsmBeginWork(pDb);
  while( rc==LSM_OK && sortedDbIsFull(pDb) ){
    rc = sortedWork(pDb, 256, pDb->nMerge, 1, 0);
  }

  if( rc==LSM_OK ){
    rc = sortedNewToplevel(pDb, TREE_BOTH, 0);
  }

  lsmFinishWork(pDb, 1, &rc);
  return rc;
}

/*
** Return a string representation of the segment passed as the only argument.
** Space for the returned string is allocated using lsmMalloc(), and should
** be freed by the caller using lsmFree().
*/
static char *segToString(lsm_env *pEnv, Segment *pSeg, int nMin){
  int nSize = pSeg->nSize;
  Pgno iRoot = pSeg->iRoot;
  Pgno iFirst = pSeg->iFirst;
  Pgno iLast = pSeg->iLastPg;
  char *z;

  char *z1;
  char *z2;
  int nPad;

  z1 = lsmMallocPrintf(pEnv, "%d.%d", iFirst, iLast);
  if( iRoot ){
    z2 = lsmMallocPrintf(pEnv, "root=%d", iRoot);
  }else{
    z2 = lsmMallocPrintf(pEnv, "size=%d", nSize);
  }

  nPad = nMin - 2 - strlen(z1) - 1 - strlen(z2);
  nPad = LSM_MAX(0, nPad);

  if( iRoot ){
    z = lsmMallocPrintf(pEnv, "/%s %*s%s\\", z1, nPad, "", z2);
  }else{
    z = lsmMallocPrintf(pEnv, "|%s %*s%s|", z1, nPad, "", z2);
  }
  lsmFree(pEnv, z1);
  lsmFree(pEnv, z2);

  return z;
}

static int fileToString(
  lsm_db *pDb,                    /* For xMalloc() */
  char *aBuf, 
  int nBuf, 
  int nMin,
  Segment *pSeg
){
  int i = 0;
  if( pSeg ){
    char *zSeg;

    zSeg = segToString(pDb->pEnv, pSeg, nMin);
    snprintf(&aBuf[i], nBuf-i, "%s", zSeg);
    i += strlen(&aBuf[i]);
    lsmFree(pDb->pEnv, zSeg);

#ifdef LSM_LOG_FREELIST
    lsmInfoArrayStructure(pDb, 1, pSeg->iFirst, &zSeg);
    snprintf(&aBuf[i], nBuf-1, "    (%s)", zSeg);
    i += strlen(&aBuf[i]);
    lsmFree(pDb->pEnv, zSeg);
#endif
    aBuf[nBuf] = 0;
  }else{
    aBuf[0] = '\0';
  }

  return i;
}

void sortedDumpPage(lsm_db *pDb, Segment *pRun, Page *pPg, int bVals){
  Blob blob = {0, 0, 0};         /* Blob used for keys */
  LsmString s;
  int i;

  int nRec;
  int iPtr;
  int flags;
  u8 *aData;
  int nData;

  aData = fsPageData(pPg, &nData);

  nRec = pageGetNRec(aData, nData);
  iPtr = pageGetPtr(aData, nData);
  flags = pageGetFlags(aData, nData);

  lsmStringInit(&s, pDb->pEnv);
  lsmStringAppendf(&s,"nCell=%d iPtr=%d flags=%d {", nRec, iPtr, flags);
  if( flags&SEGMENT_BTREE_FLAG ) iPtr = 0;

  for(i=0; i<nRec; i++){
    Page *pRef = 0;               /* Pointer to page iRef */
    int iChar;
    u8 *aKey; int nKey = 0;       /* Key */
    u8 *aVal; int nVal = 0;       /* Value */
    int iTopic;
    u8 *aCell;
    int iPgPtr;
    int eType;

    aCell = pageGetCell(aData, nData, i);
    eType = *aCell++;
    assert( (flags & SEGMENT_BTREE_FLAG) || eType!=0 );
    aCell += lsmVarintGet32(aCell, &iPgPtr);

    if( eType==0 ){
      Pgno iRef;                  /* Page number of referenced page */
      aCell += lsmVarintGet64(aCell, &iRef);
      lsmFsDbPageGet(pDb->pFS, pRun, iRef, &pRef);
      aKey = pageGetKey(pRun, pRef, 0, &iTopic, &nKey, &blob);
    }else{
      aCell += lsmVarintGet32(aCell, &nKey);
      if( rtIsWrite(eType) ) aCell += lsmVarintGet32(aCell, &nVal);
      sortedReadData(0, pPg, (aCell-aData), nKey+nVal, (void **)&aKey, &blob);
      aVal = &aKey[nKey];
      iTopic = eType;
    }

    lsmStringAppendf(&s, "%s%2X:", (i==0?"":" "), iTopic);
    for(iChar=0; iChar<nKey; iChar++){
      lsmStringAppendf(&s, "%c", isalnum(aKey[iChar]) ? aKey[iChar] : '.');
    }
    if( nVal>0 && bVals ){
      lsmStringAppendf(&s, "##");
      for(iChar=0; iChar<nVal; iChar++){
        lsmStringAppendf(&s, "%c", isalnum(aVal[iChar]) ? aVal[iChar] : '.');
      }
    }

    lsmStringAppendf(&s, " %d", iPgPtr+iPtr);
    lsmFsPageRelease(pRef);
  }
  lsmStringAppend(&s, "}", 1);

  lsmLogMessage(pDb, LSM_OK, "      Page %d: %s", lsmFsPageNumber(pPg), s.z);
  lsmStringClear(&s);

  sortedBlobFree(&blob);
}

static void infoCellDump(
  lsm_db *pDb,                    /* Database handle */
  Segment *pSeg,                  /* Segment page belongs to */
  int bIndirect,                  /* True to follow indirect refs */
  Page *pPg,
  int iCell,
  int *peType,
  int *piPgPtr,
  u8 **paKey, int *pnKey,
  u8 **paVal, int *pnVal,
  Blob *pBlob
){
  u8 *aData; int nData;           /* Page data */
  u8 *aKey; int nKey = 0;         /* Key */
  u8 *aVal; int nVal = 0;         /* Value */
  int eType;
  int iPgPtr;
  Page *pRef = 0;                 /* Pointer to page iRef */
  u8 *aCell;

  aData = fsPageData(pPg, &nData);

  aCell = pageGetCell(aData, nData, iCell);
  eType = *aCell++;
  aCell += lsmVarintGet32(aCell, &iPgPtr);

  if( eType==0 ){
    int dummy;
    Pgno iRef;                  /* Page number of referenced page */
    aCell += lsmVarintGet64(aCell, &iRef);
    if( bIndirect ){
      lsmFsDbPageGet(pDb->pFS, pSeg, iRef, &pRef);
      pageGetKeyCopy(pDb->pEnv, pSeg, pRef, 0, &dummy, pBlob);
      aKey = (u8 *)pBlob->pData;
      nKey = pBlob->nData;
      lsmFsPageRelease(pRef);
    }else{
      aKey = (u8 *)"<indirect>";
      nKey = 11;
    }
  }else{
    aCell += lsmVarintGet32(aCell, &nKey);
    if( rtIsWrite(eType) ) aCell += lsmVarintGet32(aCell, &nVal);
    sortedReadData(pSeg, pPg, (aCell-aData), nKey+nVal, (void **)&aKey, pBlob);
    aVal = &aKey[nKey];
  }

  if( peType ) *peType = eType;
  if( piPgPtr ) *piPgPtr = iPgPtr;
  if( paKey ) *paKey = aKey;
  if( paVal ) *paVal = aVal;
  if( pnKey ) *pnKey = nKey;
  if( pnVal ) *pnVal = nVal;
}

static int infoAppendBlob(LsmString *pStr, int bHex, u8 *z, int n){
  int iChar;
  for(iChar=0; iChar<n; iChar++){
    if( bHex ){
      lsmStringAppendf(pStr, "%02X", z[iChar]);
    }else{
      lsmStringAppendf(pStr, "%c", isalnum(z[iChar]) ?z[iChar] : '.');
    }
  }
  return LSM_OK;
}

#define INFO_PAGE_DUMP_DATA     0x01
#define INFO_PAGE_DUMP_VALUES   0x02
#define INFO_PAGE_DUMP_HEX      0x04
#define INFO_PAGE_DUMP_INDIRECT 0x08

static int infoPageDump(
  lsm_db *pDb,                    /* Database handle */
  Pgno iPg,                       /* Page number of page to dump */
  int flags,
  char **pzOut                    /* OUT: lsmMalloc'd string */
){
  int rc = LSM_OK;                /* Return code */
  Page *pPg = 0;                  /* Handle for page iPg */
  int i, j;                       /* Loop counters */
  const int perLine = 16;         /* Bytes per line in the raw hex dump */
  Segment *pSeg = 0;
  Snapshot *pSnap;

  int bValues = (flags & INFO_PAGE_DUMP_VALUES);
  int bHex = (flags & INFO_PAGE_DUMP_HEX);
  int bData = (flags & INFO_PAGE_DUMP_DATA);
  int bIndirect = (flags & INFO_PAGE_DUMP_INDIRECT);

  *pzOut = 0;
  if( iPg==0 ) return LSM_ERROR;

  assert( pDb->pClient || pDb->pWorker );
  pSnap = pDb->pClient;
  if( pSnap==0 ) pSnap = pDb->pWorker;
  if( pSnap->redirect.n>0 ){
    Level *pLvl;
    int bUse = 0;
    for(pLvl=pSnap->pLevel; pLvl->pNext; pLvl=pLvl->pNext);
    pSeg = (pLvl->nRight==0 ? &pLvl->lhs : &pLvl->aRhs[pLvl->nRight-1]);
    rc = lsmFsSegmentContainsPg(pDb->pFS, pSeg, iPg, &bUse);
    if( bUse==0 ){
      pSeg = 0;
    }
  }

  /* iPg is a real page number (not subject to redirection). So it is safe 
  ** to pass a NULL in place of the segment pointer as the second argument
  ** to lsmFsDbPageGet() here.  */
  if( rc==LSM_OK ){
    rc = lsmFsDbPageGet(pDb->pFS, 0, iPg, &pPg);
  }

  if( rc==LSM_OK ){
    Blob blob = {0, 0, 0, 0};
    int nKeyWidth = 0;
    LsmString str;
    int nRec;
    int iPtr;
    int flags;
    int iCell;
    u8 *aData; int nData;         /* Page data and size thereof */

    aData = fsPageData(pPg, &nData);
    nRec = pageGetNRec(aData, nData);
    iPtr = pageGetPtr(aData, nData);
    flags = pageGetFlags(aData, nData);

    lsmStringInit(&str, pDb->pEnv);
    lsmStringAppendf(&str, "Page : %lld  (%d bytes)\n", iPg, nData);
    lsmStringAppendf(&str, "nRec : %d\n", nRec);
    lsmStringAppendf(&str, "iPtr : %d\n", iPtr);
    lsmStringAppendf(&str, "flags: %04x\n", flags);
    lsmStringAppendf(&str, "\n");

    for(iCell=0; iCell<nRec; iCell++){
      int nKey;
      infoCellDump(
          pDb, pSeg, bIndirect, pPg, iCell, 0, 0, 0, &nKey, 0, 0, &blob
      );
      if( nKey>nKeyWidth ) nKeyWidth = nKey;
    }
    if( bHex ) nKeyWidth = nKeyWidth * 2;

    for(iCell=0; iCell<nRec; iCell++){
      u8 *aKey; int nKey = 0;       /* Key */
      u8 *aVal; int nVal = 0;       /* Value */
      int iPgPtr;
      int eType;
      Pgno iAbsPtr;
      char zFlags[8];

      infoCellDump(pDb, pSeg, bIndirect, pPg, iCell, &eType, &iPgPtr,
          &aKey, &nKey, &aVal, &nVal, &blob
      );
      iAbsPtr = iPgPtr + ((flags & SEGMENT_BTREE_FLAG) ? 0 : iPtr);

      lsmFlagsToString(eType, zFlags);
      lsmStringAppendf(&str, "%s %d (%s) ", 
          zFlags, iAbsPtr, (rtTopic(eType) ? "sys" : "usr")
      );
      infoAppendBlob(&str, bHex, aKey, nKey); 
      if( nVal>0 && bValues ){
        lsmStringAppendf(&str, "%*s", nKeyWidth - (nKey*(1+bHex)), "");
        lsmStringAppendf(&str, " ");
        infoAppendBlob(&str, bHex, aVal, nVal); 
      }
      if( rtTopic(eType) ){
        int iBlk = (int)~lsmGetU32(aKey);
        lsmStringAppendf(&str, "  (block=%d", iBlk);
        if( nVal>0 ){
          i64 iSnap = lsmGetU64(aVal);
          lsmStringAppendf(&str, " snapshot=%lld", iSnap);
        }
        lsmStringAppendf(&str, ")");
      }
      lsmStringAppendf(&str, "\n");
    }

    if( bData ){
      lsmStringAppendf(&str, "\n-------------------" 
          "-------------------------------------------------------------\n");
      lsmStringAppendf(&str, "Page %d\n",
          iPg, (iPg-1)*nData, iPg*nData - 1);
      for(i=0; i<nData; i += perLine){
        lsmStringAppendf(&str, "%04x: ", i);
        for(j=0; j<perLine; j++){
          if( i+j>nData ){
            lsmStringAppendf(&str, "   ");
          }else{
            lsmStringAppendf(&str, "%02x ", aData[i+j]);
          }
        }
        lsmStringAppendf(&str, "  ");
        for(j=0; j<perLine; j++){
          if( i+j>nData ){
            lsmStringAppendf(&str, " ");
          }else{
            lsmStringAppendf(&str,"%c", isprint(aData[i+j]) ? aData[i+j] : '.');
          }
        }
        lsmStringAppendf(&str,"\n");
      }
    }

    *pzOut = str.z;
    sortedBlobFree(&blob);
    lsmFsPageRelease(pPg);
  }

  return rc;
}

int lsmInfoPageDump(
  lsm_db *pDb,                    /* Database handle */
  Pgno iPg,                       /* Page number of page to dump */
  int bHex,                       /* True to output key/value in hex form */
  char **pzOut                    /* OUT: lsmMalloc'd string */
){
  int flags = INFO_PAGE_DUMP_DATA | INFO_PAGE_DUMP_VALUES;
  if( bHex ) flags |= INFO_PAGE_DUMP_HEX;
  return infoPageDump(pDb, iPg, flags, pzOut);
}

void sortedDumpSegment(lsm_db *pDb, Segment *pRun, int bVals){
  assert( pDb->xLog );
  if( pRun && pRun->iFirst ){
    int flags = (bVals ? INFO_PAGE_DUMP_VALUES : 0);
    char *zSeg;
    Page *pPg;

    zSeg = segToString(pDb->pEnv, pRun, 0);
    lsmLogMessage(pDb, LSM_OK, "Segment: %s", zSeg);
    lsmFree(pDb->pEnv, zSeg);

    lsmFsDbPageGet(pDb->pFS, pRun, pRun->iFirst, &pPg);
    while( pPg ){
      Page *pNext;
      char *z = 0;
      infoPageDump(pDb, lsmFsPageNumber(pPg), flags, &z);
      lsmLogMessage(pDb, LSM_OK, "%s", z);
      lsmFree(pDb->pEnv, z);
#if 0
      sortedDumpPage(pDb, pRun, pPg, bVals);
#endif
      lsmFsDbPageNext(pRun, pPg, 1, &pNext);
      lsmFsPageRelease(pPg);
      pPg = pNext;
    }
  }
}

/*
** Invoke the log callback zero or more times with messages that describe
** the current database structure.
*/
void lsmSortedDumpStructure(
  lsm_db *pDb,                    /* Database handle (used for xLog callback) */
  Snapshot *pSnap,                /* Snapshot to dump */
  int bKeys,                      /* Output the keys from each segment */
  int bVals,                      /* Output the values from each segment */
  const char *zWhy                /* Caption to print near top of dump */
){
  Snapshot *pDump = pSnap;
  Level *pTopLevel;
  char *zFree = 0;

  assert( pSnap );
  pTopLevel = lsmDbSnapshotLevel(pDump);
  if( pDb->xLog && pTopLevel ){
    static int nCall = 0;
    Level *pLevel;
    int iLevel = 0;

    nCall++;
    lsmLogMessage(pDb, LSM_OK, "Database structure %d (%s)", nCall, zWhy);

#if 0
    if( nCall==1031 || nCall==1032 ) bKeys=1;
#endif

    for(pLevel=pTopLevel; pLevel; pLevel=pLevel->pNext){
      char zLeft[1024];
      char zRight[1024];
      int i = 0;

      Segment *aLeft[24];  
      Segment *aRight[24];

      int nLeft = 0;
      int nRight = 0;

      Segment *pSeg = &pLevel->lhs;
      aLeft[nLeft++] = pSeg;

      for(i=0; i<pLevel->nRight; i++){
        aRight[nRight++] = &pLevel->aRhs[i];
      }

#ifdef LSM_LOG_FREELIST
      if( nRight ){
        memmove(&aRight[1], aRight, sizeof(aRight[0])*nRight);
        aRight[0] = 0;
        nRight++;
      }
#endif

      for(i=0; i<nLeft || i<nRight; i++){
        int iPad = 0;
        char zLevel[32];
        zLeft[0] = '\0';
        zRight[0] = '\0';

        if( i<nLeft ){ 
          fileToString(pDb, zLeft, sizeof(zLeft), 24, aLeft[i]); 
        }
        if( i<nRight ){ 
          fileToString(pDb, zRight, sizeof(zRight), 24, aRight[i]); 
        }

        if( i==0 ){
          snprintf(zLevel, sizeof(zLevel), "L%d: (age=%d) (flags=%.4x)",
              iLevel, (int)pLevel->iAge, (int)pLevel->flags
          );
        }else{
          zLevel[0] = '\0';
        }

        if( nRight==0 ){
          iPad = 10;
        }

        lsmLogMessage(pDb, LSM_OK, "% 25s % *s% -35s %s", 
            zLevel, iPad, "", zLeft, zRight
        );
      }

      iLevel++;
    }

    if( bKeys ){
      for(pLevel=pTopLevel; pLevel; pLevel=pLevel->pNext){
        int i;
        sortedDumpSegment(pDb, &pLevel->lhs, bVals);
        for(i=0; i<pLevel->nRight; i++){
          sortedDumpSegment(pDb, &pLevel->aRhs[i], bVals);
        }
      }
    }
  }

  lsmInfoFreelist(pDb, &zFree);
  lsmLogMessage(pDb, LSM_OK, "Freelist: %s", zFree);
  lsmFree(pDb->pEnv, zFree);

  assert( lsmFsIntegrityCheck(pDb) );
}

void lsmSortedFreeLevel(lsm_env *pEnv, Level *pLevel){
  Level *pNext;
  Level *p;

  for(p=pLevel; p; p=pNext){
    pNext = p->pNext;
    sortedFreeLevel(pEnv, p);
  }
}

void lsmSortedSaveTreeCursors(lsm_db *pDb){
  MultiCursor *pCsr;
  for(pCsr=pDb->pCsr; pCsr; pCsr=pCsr->pNext){
    lsmTreeCursorSave(pCsr->apTreeCsr[0]);
    lsmTreeCursorSave(pCsr->apTreeCsr[1]);
  }
}

void lsmSortedExpandBtreePage(Page *pPg, int nOrig){
  u8 *aData;
  int nData;
  int nEntry;
  int iHdr;

  aData = lsmFsPageData(pPg, &nData);
  nEntry = pageGetNRec(aData, nOrig);
  iHdr = SEGMENT_EOF(nOrig, nEntry);
  memmove(&aData[iHdr + (nData-nOrig)], &aData[iHdr], nOrig-iHdr);
}

#ifdef LSM_DEBUG_EXPENSIVE
static void assertRunInOrder(lsm_db *pDb, Segment *pSeg){
  Page *pPg = 0;
  Blob blob1 = {0, 0, 0, 0};
  Blob blob2 = {0, 0, 0, 0};

  lsmFsDbPageGet(pDb->pFS, pSeg, pSeg->iFirst, &pPg);
  while( pPg ){
    u8 *aData; int nData;
    Page *pNext;

    aData = lsmFsPageData(pPg, &nData);
    if( 0==(pageGetFlags(aData, nData) & SEGMENT_BTREE_FLAG) ){
      int i;
      int nRec = pageGetNRec(aData, nData);
      for(i=0; i<nRec; i++){
        int iTopic1, iTopic2;
        pageGetKeyCopy(pDb->pEnv, pSeg, pPg, i, &iTopic1, &blob1);

        if( i==0 && blob2.nData ){
          assert( sortedKeyCompare(
                pDb->xCmp, iTopic2, blob2.pData, blob2.nData,
                iTopic1, blob1.pData, blob1.nData
          )<0 );
        }

        if( i<(nRec-1) ){
          pageGetKeyCopy(pDb->pEnv, pSeg, pPg, i+1, &iTopic2, &blob2);
          assert( sortedKeyCompare(
                pDb->xCmp, iTopic1, blob1.pData, blob1.nData,
                iTopic2, blob2.pData, blob2.nData
          )<0 );
        }
      }
    }

    lsmFsDbPageNext(pSeg, pPg, 1, &pNext);
    lsmFsPageRelease(pPg);
    pPg = pNext;
  }

  sortedBlobFree(&blob1);
  sortedBlobFree(&blob2);
}
#endif

#ifdef LSM_DEBUG_EXPENSIVE
/*
** This function is only included in the build if LSM_DEBUG_EXPENSIVE is 
** defined. Its only purpose is to evaluate various assert() statements to 
** verify that the database is well formed in certain respects.
**
** More specifically, it checks that the array pOne contains the required 
** pointers to pTwo. Array pTwo must be a main array. pOne may be either a 
** separators array or another main array. If pOne does not contain the 
** correct set of pointers, an assert() statement fails.
*/
static int assertPointersOk(
  lsm_db *pDb,                    /* Database handle */
  Segment *pOne,                  /* Segment containing pointers */
  Segment *pTwo,                  /* Segment containing pointer targets */
  int bRhs                        /* True if pTwo may have been Gobble()d */
){
  int rc = LSM_OK;                /* Error code */
  SegmentPtr ptr1;                /* Iterates through pOne */
  SegmentPtr ptr2;                /* Iterates through pTwo */
  Pgno iPrev;

  assert( pOne && pTwo );

  memset(&ptr1, 0, sizeof(ptr1));
  memset(&ptr2, 0, sizeof(ptr1));
  ptr1.pSeg = pOne;
  ptr2.pSeg = pTwo;
  segmentPtrEndPage(pDb->pFS, &ptr1, 0, &rc);
  segmentPtrEndPage(pDb->pFS, &ptr2, 0, &rc);

  /* Check that the footer pointer of the first page of pOne points to
  ** the first page of pTwo. */
  iPrev = pTwo->iFirst;
  if( ptr1.iPtr!=iPrev && !bRhs ){
    assert( 0 );
  }

  if( rc==LSM_OK && ptr1.nCell>0 ){
    rc = segmentPtrLoadCell(&ptr1, 0);
  }
      
  while( rc==LSM_OK && ptr2.pPg ){
    Pgno iThis;

    /* Advance to the next page of segment pTwo that contains at least
    ** one cell. Break out of the loop if the iterator reaches EOF.  */
    do{
      rc = segmentPtrNextPage(&ptr2, 1);
      assert( rc==LSM_OK );
    }while( rc==LSM_OK && ptr2.pPg && ptr2.nCell==0 );
    if( rc!=LSM_OK || ptr2.pPg==0 ) break;
    iThis = lsmFsPageNumber(ptr2.pPg);

    if( (ptr2.flags & (PGFTR_SKIP_THIS_FLAG|SEGMENT_BTREE_FLAG))==0 ){

      /* Load the first cell in the array pTwo page. */
      rc = segmentPtrLoadCell(&ptr2, 0);

      /* Iterate forwards through pOne, searching for a key that matches the
      ** key ptr2.pKey/nKey. This key should have a pointer to the page that
      ** ptr2 currently points to. */
      while( rc==LSM_OK ){
        int res = rtTopic(ptr1.eType) - rtTopic(ptr2.eType);
        if( res==0 ){
          res = pDb->xCmp(ptr1.pKey, ptr1.nKey, ptr2.pKey, ptr2.nKey);
        }

        if( res<0 ){
          assert( bRhs || ptr1.iPtr+ptr1.iPgPtr==iPrev );
        }else if( res>0 ){
          assert( 0 );
        }else{
          assert( ptr1.iPtr+ptr1.iPgPtr==iThis );
          iPrev = iThis;
          break;
        }

        rc = segmentPtrAdvance(0, &ptr1, 0);
        if( ptr1.pPg==0 ){
          assert( 0 );
        }
      }
    }
  }

  segmentPtrReset(&ptr1);
  segmentPtrReset(&ptr2);
  return LSM_OK;
}

/*
** This function is only included in the build if LSM_DEBUG_EXPENSIVE is 
** defined. Its only purpose is to evaluate various assert() statements to 
** verify that the database is well formed in certain respects.
**
** More specifically, it checks that the b-tree embedded in array pRun
** contains the correct keys. If not, an assert() fails.
*/
static int assertBtreeOk(
  lsm_db *pDb,
  Segment *pSeg
){
  int rc = LSM_OK;                /* Return code */
  if( pSeg->iRoot ){
    Blob blob = {0, 0, 0};        /* Buffer used to cache overflow keys */
    FileSystem *pFS = pDb->pFS;   /* File system to read from */
    Page *pPg = 0;                /* Main run page */
    BtreeCursor *pCsr = 0;        /* Btree cursor */

    rc = btreeCursorNew(pDb, pSeg, &pCsr);
    if( rc==LSM_OK ){
      rc = btreeCursorFirst(pCsr);
    }
    if( rc==LSM_OK ){
      rc = lsmFsDbPageGet(pFS, pSeg, pSeg->iFirst, &pPg);
    }

    while( rc==LSM_OK ){
      Page *pNext;
      u8 *aData;
      int nData;
      int flags;

      rc = lsmFsDbPageNext(pSeg, pPg, 1, &pNext);
      lsmFsPageRelease(pPg);
      pPg = pNext;
      if( pPg==0 ) break;
      aData = fsPageData(pPg, &nData);
      flags = pageGetFlags(aData, nData);
      if( rc==LSM_OK 
       && 0==((SEGMENT_BTREE_FLAG|PGFTR_SKIP_THIS_FLAG) & flags)
       && 0!=pageGetNRec(aData, nData)
      ){
        u8 *pKey;
        int nKey;
        int iTopic;
        pKey = pageGetKey(pSeg, pPg, 0, &iTopic, &nKey, &blob);
        assert( nKey==pCsr->nKey && 0==memcmp(pKey, pCsr->pKey, nKey) );
        assert( lsmFsPageNumber(pPg)==pCsr->iPtr );
        rc = btreeCursorNext(pCsr);
      }
    }
    assert( rc!=LSM_OK || pCsr->pKey==0 );

    if( pPg ) lsmFsPageRelease(pPg);

    btreeCursorFree(pCsr);
    sortedBlobFree(&blob);
  }

  return rc;
}
#endif /* ifdef LSM_DEBUG_EXPENSIVE */

Added ext/lsm1/lsm_str.c.













































































































































































































































































































>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
/*
** 2012-04-27
**
** The author disclaims copyright to this source code.  In place of
** a legal notice, here is a blessing:
**
**    May you do good and not evil.
**    May you find forgiveness for yourself and forgive others.
**    May you share freely, never taking more than you give.
**
*************************************************************************
**
** Dynamic string functions.
*/
#include "lsmInt.h"

/*
** Turn bulk and uninitialized memory into an LsmString object
*/
void lsmStringInit(LsmString *pStr, lsm_env *pEnv){
  memset(pStr, 0, sizeof(pStr[0]));
  pStr->pEnv = pEnv;
}

/*
** Increase the memory allocated for holding the string.  Realloc as needed.
**
** If a memory allocation error occurs, set pStr->n to -1 and free the existing
** allocation.  If a prior memory allocation has occurred, this routine is a
** no-op.
*/
int lsmStringExtend(LsmString *pStr, int nNew){
  assert( nNew>0 );
  if( pStr->n<0 ) return LSM_NOMEM;
  if( pStr->n + nNew >= pStr->nAlloc ){
    int nAlloc = pStr->n + nNew + 100;
    char *zNew = lsmRealloc(pStr->pEnv, pStr->z, nAlloc);
    if( zNew==0 ){
      lsmFree(pStr->pEnv, pStr->z);
      nAlloc = 0;
      pStr->n = -1;
      pStr->z = 0;
    }else{
      pStr->nAlloc = nAlloc;
      pStr->z = zNew;
    }
  }
  return (pStr->z ? LSM_OK : LSM_NOMEM_BKPT);
}

/*
** Clear an LsmString object, releasing any allocated memory that it holds.
** This also clears the error indication (if any).
*/
void lsmStringClear(LsmString *pStr){
  lsmFree(pStr->pEnv, pStr->z);
  lsmStringInit(pStr, pStr->pEnv);
}

/*
** Append N bytes of text to the end of an LsmString object.  If
** N is negative, append the entire string.
**
** If the string is in an error state, this routine is a no-op.
*/
int lsmStringAppend(LsmString *pStr, const char *z, int N){
  int rc;
  if( N<0 ) N = (int)strlen(z);
  rc = lsmStringExtend(pStr, N+1);
  if( pStr->nAlloc ){
    memcpy(pStr->z+pStr->n, z, N+1);
    pStr->n += N;
  }
  return rc;
}

int lsmStringBinAppend(LsmString *pStr, const u8 *a, int n){
  int rc;
  rc = lsmStringExtend(pStr, n);
  if( pStr->nAlloc ){
    memcpy(pStr->z+pStr->n, a, n);
    pStr->n += n;
  }
  return rc;
}

/*
** Append printf-formatted content to an LsmString.
*/
void lsmStringVAppendf(
  LsmString *pStr, 
  const char *zFormat, 
  va_list ap1,
  va_list ap2
){
#if (!defined(__STDC_VERSION__) || (__STDC_VERSION__<199901L)) && \
    !defined(__APPLE__)
  extern int vsnprintf(char *str, size_t size, const char *format, va_list ap)
    /* Compatibility crutch for C89 compilation mode. sqlite3_vsnprintf()
       does not work identically and causes test failures if used here.
       For the time being we are assuming that the target has vsnprintf(),
       but that is not guaranteed to be the case for pure C89 platforms.
    */;
#endif
  int nWrite;
  int nAvail;

  nAvail = pStr->nAlloc - pStr->n;
  nWrite = vsnprintf(pStr->z + pStr->n, nAvail, zFormat, ap1);

  if( nWrite>=nAvail ){
    lsmStringExtend(pStr, nWrite+1);
    if( pStr->nAlloc==0 ) return;
    nWrite = vsnprintf(pStr->z + pStr->n, nWrite+1, zFormat, ap2);
  }

  pStr->n += nWrite;
  pStr->z[pStr->n] = 0;
}

void lsmStringAppendf(LsmString *pStr, const char *zFormat, ...){
  va_list ap, ap2;
  va_start(ap, zFormat);
  va_start(ap2, zFormat);
  lsmStringVAppendf(pStr, zFormat, ap, ap2);
  va_end(ap);
  va_end(ap2);
}

int lsmStrlen(const char *zName){
  int nRet = 0;
  while( zName[nRet] ) nRet++;
  return nRet;
}

/*
** Write into memory obtained from lsm_malloc().
*/
char *lsmMallocPrintf(lsm_env *pEnv, const char *zFormat, ...){
  LsmString s;
  va_list ap, ap2;
  lsmStringInit(&s, pEnv);
  va_start(ap, zFormat);
  va_start(ap2, zFormat);
  lsmStringVAppendf(&s, zFormat, ap, ap2);
  va_end(ap);
  va_end(ap2);
  if( s.n<0 ) return 0;
  return (char *)lsmReallocOrFree(pEnv, s.z, s.n+1);
}

Added ext/lsm1/lsm_tree.c.

































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
1388
1389
1390
1391
1392
1393
1394
1395
1396
1397
1398
1399
1400
1401
1402
1403
1404
1405
1406
1407
1408
1409
1410
1411
1412
1413
1414
1415
1416
1417
1418
1419
1420
1421
1422
1423
1424
1425
1426
1427
1428
1429
1430
1431
1432
1433
1434
1435
1436
1437
1438
1439
1440
1441
1442
1443
1444
1445
1446
1447
1448
1449
1450
1451
1452
1453
1454
1455
1456
1457
1458
1459
1460
1461
1462
1463
1464
1465
1466
1467
1468
1469
1470
1471
1472
1473
1474
1475
1476
1477
1478
1479
1480
1481
1482
1483
1484
1485
1486
1487
1488
1489
1490
1491
1492
1493
1494
1495
1496
1497
1498
1499
1500
1501
1502
1503
1504
1505
1506
1507
1508
1509
1510
1511
1512
1513
1514
1515
1516
1517
1518
1519
1520
1521
1522
1523
1524
1525
1526
1527
1528
1529
1530
1531
1532
1533
1534
1535
1536
1537
1538
1539
1540
1541
1542
1543
1544
1545
1546
1547
1548
1549
1550
1551
1552
1553
1554
1555
1556
1557
1558
1559
1560
1561
1562
1563
1564
1565
1566
1567
1568
1569
1570
1571
1572
1573
1574
1575
1576
1577
1578
1579
1580
1581
1582
1583
1584
1585
1586
1587
1588
1589
1590
1591
1592
1593
1594
1595
1596
1597
1598
1599
1600
1601
1602
1603
1604
1605
1606
1607
1608
1609
1610
1611
1612
1613
1614
1615
1616
1617
1618
1619
1620
1621
1622
1623
1624
1625
1626
1627
1628
1629
1630
1631
1632
1633
1634
1635
1636
1637
1638
1639
1640
1641
1642
1643
1644
1645
1646
1647
1648
1649
1650
1651
1652
1653
1654
1655
1656
1657
1658
1659
1660
1661
1662
1663
1664
1665
1666
1667
1668
1669
1670
1671
1672
1673
1674
1675
1676
1677
1678
1679
1680
1681
1682
1683
1684
1685
1686
1687
1688
1689
1690
1691
1692
1693
1694
1695
1696
1697
1698
1699
1700
1701
1702
1703
1704
1705
1706
1707
1708
1709
1710
1711
1712
1713
1714
1715
1716
1717
1718
1719
1720
1721
1722
1723
1724
1725
1726
1727
1728
1729
1730
1731
1732
1733
1734
1735
1736
1737
1738
1739
1740
1741
1742
1743
1744
1745
1746
1747
1748
1749
1750
1751
1752
1753
1754
1755
1756
1757
1758
1759
1760
1761
1762
1763
1764
1765
1766
1767
1768
1769
1770
1771
1772
1773
1774
1775
1776
1777
1778
1779
1780
1781
1782
1783
1784
1785
1786
1787
1788
1789
1790
1791
1792
1793
1794
1795
1796
1797
1798
1799
1800
1801
1802
1803
1804
1805
1806
1807
1808
1809
1810
1811
1812
1813
1814
1815
1816
1817
1818
1819
1820
1821
1822
1823
1824
1825
1826
1827
1828
1829
1830
1831
1832
1833
1834
1835
1836
1837
1838
1839
1840
1841
1842
1843
1844
1845
1846
1847
1848
1849
1850
1851
1852
1853
1854
1855
1856
1857
1858
1859
1860
1861
1862
1863
1864
1865
1866
1867
1868
1869
1870
1871
1872
1873
1874
1875
1876
1877
1878
1879
1880
1881
1882
1883
1884
1885
1886
1887
1888
1889
1890
1891
1892
1893
1894
1895
1896
1897
1898
1899
1900
1901
1902
1903
1904
1905
1906
1907
1908
1909
1910
1911
1912
1913
1914
1915
1916
1917
1918
1919
1920
1921
1922
1923
1924
1925
1926
1927
1928
1929
1930
1931
1932
1933
1934
1935
1936
1937
1938
1939
1940
1941
1942
1943
1944
1945
1946
1947
1948
1949
1950
1951
1952
1953
1954
1955
1956
1957
1958
1959
1960
1961
1962
1963
1964
1965
1966
1967
1968
1969
1970
1971
1972
1973
1974
1975
1976
1977
1978
1979
1980
1981
1982
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
2018
2019
2020
2021
2022
2023
2024
2025
2026
2027
2028
2029
2030
2031
2032
2033
2034
2035
2036
2037
2038
2039
2040
2041
2042
2043
2044
2045
2046
2047
2048
2049
2050
2051
2052
2053
2054
2055
2056
2057
2058
2059
2060
2061
2062
2063
2064
2065
2066
2067
2068
2069
2070
2071
2072
2073
2074
2075
2076
2077
2078
2079
2080
2081
2082
2083
2084
2085
2086
2087
2088
2089
2090
2091
2092
2093
2094
2095
2096
2097
2098
2099
2100
2101
2102
2103
2104
2105
2106
2107
2108
2109
2110
2111
2112
2113
2114
2115
2116
2117
2118
2119
2120
2121
2122
2123
2124
2125
2126
2127
2128
2129
2130
2131
2132
2133
2134
2135
2136
2137
2138
2139
2140
2141
2142
2143
2144
2145
2146
2147
2148
2149
2150
2151
2152
2153
2154
2155
2156
2157
2158
2159
2160
2161
2162
2163
2164
2165
2166
2167
2168
2169
2170
2171
2172
2173
2174
2175
2176
2177
2178
2179
2180
2181
2182
2183
2184
2185
2186
2187
2188
2189
2190
2191
2192
2193
2194
2195
2196
2197
2198
2199
2200
2201
2202
2203
2204
2205
2206
2207
2208
2209
2210
2211
2212
2213
2214
2215
2216
2217
2218
2219
2220
2221
2222
2223
2224
2225
2226
2227
2228
2229
2230
2231
2232
2233
2234
2235
2236
2237
2238
2239
2240
2241
2242
2243
2244
2245
2246
2247
2248
2249
2250
2251
2252
2253
2254
2255
2256
2257
2258
2259
2260
2261
2262
2263
2264
2265
2266
2267
2268
2269
2270
2271
2272
2273
2274
2275
2276
2277
2278
2279
2280
2281
2282
2283
2284
2285
2286
2287
2288
2289
2290
2291
2292
2293
2294
2295
2296
2297
2298
2299
2300
2301
2302
2303
2304
2305
2306
2307
2308
2309
2310
2311
2312
2313
2314
2315
2316
2317
2318
2319
2320
2321
2322
2323
2324
2325
2326
2327
2328
2329
2330
2331
2332
2333
2334
2335
2336
2337
2338
2339
2340
2341
2342
2343
2344
2345
2346
2347
2348
2349
2350
2351
2352
2353
2354
2355
2356
2357
2358
2359
2360
2361
2362
2363
2364
2365
2366
2367
2368
2369
2370
2371
2372
2373
2374
2375
2376
2377
2378
2379
2380
2381
2382
2383
2384
2385
2386
2387
2388
2389
2390
2391
2392
2393
2394
2395
2396
2397
2398
2399
2400
2401
2402
2403
2404
2405
2406
2407
2408
2409
2410
2411
2412
2413
2414
2415
2416
2417
2418
2419
2420
2421
2422
2423
2424
2425
2426
2427
2428
2429
2430
2431
2432
2433
2434
2435
2436
2437
2438
2439
2440
2441
2442
2443
2444
2445
2446
2447
2448
2449
2450
2451
2452
2453
2454
2455
2456
2457
2458
2459
2460
2461
2462
2463
2464
/*
** 2011-08-18
**
** The author disclaims copyright to this source code.  In place of
** a legal notice, here is a blessing:
**
**    May you do good and not evil.
**    May you find forgiveness for yourself and forgive others.
**    May you share freely, never taking more than you give.
**
*************************************************************************
**
** This file contains the implementation of an in-memory tree structure.
**
** Technically the tree is a B-tree of order 4 (in the Knuth sense - each 
** node may have up to 4 children). Keys are stored within B-tree nodes by
** reference. This may be slightly slower than a conventional red-black
** tree, but it is simpler. It is also an easier structure to modify to 
** create a version that supports nested transaction rollback.
**
** This tree does not currently support a delete operation. One is not 
** required. When LSM deletes a key from a database, it inserts a DELETE
** marker into the data structure. As a result, although the value associated
** with a key stored in the in-memory tree structure may be modified, no
** keys are ever removed. 
*/

/*
** MVCC NOTES
**
**   The in-memory tree structure supports SQLite-style MVCC. This means
**   that while one client is writing to the tree structure, other clients
**   may still be querying an older snapshot of the tree.
**
**   One way to implement this is to use an append-only b-tree. In this 
**   case instead of modifying nodes in-place, a copy of the node is made
**   and the required modifications made to the copy. The parent of the
**   node is then modified (to update the pointer so that it points to
**   the new copy), which causes a copy of the parent to be made, and so on.
**   This means that each time the tree is written to a new root node is
**   created. A snapshot is identified by the root node that it uses.
**
**   The problem with the above is that each time the tree is written to,
**   a copy of the node structure modified and all of its ancestor nodes
**   is made. This may prove excessive with large tree structures.
**
**   To reduce this overhead, the data structure used for a tree node is
**   designed so that it may be edited in place exactly once without 
**   affecting existing users. In other words, the node structure is capable
**   of storing two separate versions of the node at the same time.
**   When a node is to be edited, if the node structure already contains 
**   two versions, a copy is made as in the append-only approach. Or, if
**   it only contains a single version, it is edited in place.
**
**   This reduces the overhead so that, roughly, one new node structure
**   must be allocated for each write (on top of those allocations that 
**   would have been required by a non-MVCC tree). Logic: Assume that at 
**   any time, 50% of nodes in the tree already contain 2 versions. When
**   a new entry is written to a node, there is a 50% chance that a copy
**   of the node will be required. And a 25% chance that a copy of its 
**   parent is required. And so on.
**
** ROLLBACK
**
**   The in-memory tree also supports transaction and sub-transaction 
**   rollback. In order to rollback to point in time X, the following is
**   necessary:
**
**     1. All memory allocated since X must be freed, and 
**     2. All "v2" data adding to nodes that existed at X should be zeroed.
**     3. The root node must be restored to its X value.
**
**   The Mempool object used to allocate memory for the tree supports 
**   operation (1) - see the lsmPoolMark() and lsmPoolRevert() functions.
**
**   To support (2), all nodes that have v2 data are part of a singly linked 
**   list, sorted by the age of the v2 data (nodes that have had data added 
**   most recently are at the end of the list). So to zero all v2 data added
**   since X, the linked list is traversed from the first node added following
**   X onwards.
**
*/

#ifndef _LSM_INT_H
# include "lsmInt.h"
#endif

#include <string.h>

#define MAX_DEPTH 32

typedef struct TreeKey TreeKey;
typedef struct TreeNode TreeNode;
typedef struct TreeLeaf TreeLeaf;
typedef struct NodeVersion NodeVersion;

struct TreeOld {
  u32 iShmid;                     /* Last shared-memory chunk in use by old */
  u32 iRoot;                      /* Offset of root node in shm file */
  u32 nHeight;                    /* Height of tree structure */
};

#if 0
/*
** assert() that a TreeKey.flags value is sane. Usage:
**
**   assert( lsmAssertFlagsOk(pTreeKey->flags) );
*/
static int lsmAssertFlagsOk(u8 keyflags){
  /* At least one flag must be set. Otherwise, what is this key doing? */
  assert( keyflags!=0 );

  /* The POINT_DELETE and INSERT flags cannot both be set. */
  assert( (keyflags & LSM_POINT_DELETE)==0 || (keyflags & LSM_INSERT)==0 );

  /* If both the START_DELETE and END_DELETE flags are set, then the INSERT
  ** flag must also be set. In other words - the three DELETE flags cannot
  ** all be set */
  assert( (keyflags & LSM_END_DELETE)==0 
       || (keyflags & LSM_START_DELETE)==0 
       || (keyflags & LSM_POINT_DELETE)==0 
  );

  return 1;
}
#endif
static int assert_delete_ranges_match(lsm_db *);
static int treeCountEntries(lsm_db *db);

/*
** Container for a key-value pair. Within the *-shm file, each key/value
** pair is stored in a single allocation (which may not actually be 
** contiguous in memory). Layout is the TreeKey structure, followed by
** the nKey bytes of key blob, followed by the nValue bytes of value blob
** (if nValue is non-negative).
*/
struct TreeKey {
  int nKey;                       /* Size of pKey in bytes */
  int nValue;                     /* Size of pValue. Or negative. */
  u8 flags;                       /* Various LSM_XXX flags */
};

#define TKV_KEY(p) ((void *)&(p)[1])
#define TKV_VAL(p) ((void *)(((u8 *)&(p)[1]) + (p)->nKey))

/*
** A single tree node. A node structure may contain up to 3 key/value
** pairs. Internal (non-leaf) nodes have up to 4 children.
**
** TODO: Update the format of this to be more compact. Get it working
** first though...
*/
struct TreeNode {
  u32 aiKeyPtr[3];                /* Array of pointers to TreeKey objects */

  /* The following fields are present for interior nodes only, not leaves. */
  u32 aiChildPtr[4];              /* Array of pointers to child nodes */

  /* The extra child pointer slot. */
  u32 iV2;                        /* Transaction number of v2 */
  u8 iV2Child;                    /* apChild[] entry replaced by pV2Ptr */
  u32 iV2Ptr;                     /* Substitute pointer */
};

struct TreeLeaf {
  u32 aiKeyPtr[3];                /* Array of pointers to TreeKey objects */
};

typedef struct TreeBlob TreeBlob;
struct TreeBlob {
  int n;
  u8 *a;
};

/*
** Cursor for searching a tree structure.
**
** If a cursor does not point to any element (a.k.a. EOF), then the
** TreeCursor.iNode variable is set to a negative value. Otherwise, the
** cursor currently points to key aiCell[iNode] on node apTreeNode[iNode].
**
** Entries in the apTreeNode[] and aiCell[] arrays contain the node and
** index of the TreeNode.apChild[] pointer followed to descend to the 
** current element. Hence apTreeNode[0] always contains the root node of
** the tree.
*/
struct TreeCursor {
  lsm_db *pDb;                    /* Database handle for this cursor */
  TreeRoot *pRoot;                /* Root node and height of tree to access */
  int iNode;                      /* Cursor points at apTreeNode[iNode] */
  TreeNode *apTreeNode[MAX_DEPTH];/* Current position in tree */
  u8 aiCell[MAX_DEPTH];           /* Current position in tree */
  TreeKey *pSave;                 /* Saved key */
  TreeBlob blob;                  /* Dynamic storage for a key */
};

/*
** A value guaranteed to be larger than the largest possible transaction
** id (TreeHeader.iTransId).
*/
#define WORKING_VERSION (1<<30)

static int tblobGrow(lsm_db *pDb, TreeBlob *p, int n, int *pRc){
  if( n>p->n ){
    lsmFree(pDb->pEnv, p->a);
    p->a = lsmMallocRc(pDb->pEnv, n, pRc);
    p->n = n;
  }
  return (p->a==0);
}
static void tblobFree(lsm_db *pDb, TreeBlob *p){
  lsmFree(pDb->pEnv, p->a);
}


/***********************************************************************
** Start of IntArray methods.  */
/*
** Append value iVal to the contents of IntArray *p. Return LSM_OK if 
** successful, or LSM_NOMEM if an OOM condition is encountered.
*/
static int intArrayAppend(lsm_env *pEnv, IntArray *p, u32 iVal){
  assert( p->nArray<=p->nAlloc );
  if( p->nArray>=p->nAlloc ){
    u32 *aNew;
    int nNew = p->nArray ? p->nArray*2 : 128;
    aNew = lsmRealloc(pEnv, p->aArray, nNew*sizeof(u32));
    if( !aNew ) return LSM_NOMEM_BKPT;
    p->aArray = aNew;
    p->nAlloc = nNew;
  }

  p->aArray[p->nArray++] = iVal;
  return LSM_OK;
}

/*
** Zero the IntArray object.
*/
static void intArrayFree(lsm_env *pEnv, IntArray *p){
  p->nArray = 0;
}

/*
** Return the number of entries currently in the int-array object.
*/
static int intArraySize(IntArray *p){
  return p->nArray;
}

/*
** Return a copy of the iIdx'th entry in the int-array.
*/
static u32 intArrayEntry(IntArray *p, int iIdx){
  return p->aArray[iIdx];
}

/*
** Truncate the int-array so that all but the first nVal values are 
** discarded.
*/
static void intArrayTruncate(IntArray *p, int nVal){
  p->nArray = nVal;
}
/* End of IntArray methods.
***********************************************************************/

static int treeKeycmp(void *p1, int n1, void *p2, int n2){
  int res;
  res = memcmp(p1, p2, LSM_MIN(n1, n2));
  if( res==0 ) res = (n1-n2);
  return res;
}

/*
** The pointer passed as the first argument points to an interior node,
** not a leaf. This function returns the offset of the iCell'th child
** sub-tree of the node.
*/
static u32 getChildPtr(TreeNode *p, int iVersion, int iCell){
  assert( iCell>=0 && iCell<=array_size(p->aiChildPtr) );
  if( p->iV2 && p->iV2<=iVersion && iCell==p->iV2Child ) return p->iV2Ptr;
  return p->aiChildPtr[iCell];
}

/*
** Given an offset within the *-shm file, return the associated chunk number.
*/
static int treeOffsetToChunk(u32 iOff){
  assert( LSM_SHM_CHUNK_SIZE==(1<<15) );
  return (int)(iOff>>15);
}

#define treeShmptrUnsafe(pDb, iPtr) \
(&((u8*)((pDb)->apShm[(iPtr)>>15]))[(iPtr) & (LSM_SHM_CHUNK_SIZE-1)])

/*
** Return a pointer to the mapped memory location associated with *-shm 
** file offset iPtr.
*/
static void *treeShmptr(lsm_db *pDb, u32 iPtr){

  assert( (iPtr>>15)<pDb->nShm );
  assert( pDb->apShm[iPtr>>15] );

  return iPtr ? treeShmptrUnsafe(pDb, iPtr) : 0;
}

static ShmChunk * treeShmChunk(lsm_db *pDb, int iChunk){
  return (ShmChunk *)(pDb->apShm[iChunk]);
}

static ShmChunk * treeShmChunkRc(lsm_db *pDb, int iChunk, int *pRc){
  assert( *pRc==LSM_OK );
  if( iChunk<pDb->nShm || LSM_OK==(*pRc = lsmShmCacheChunks(pDb, iChunk+1)) ){
    return (ShmChunk *)(pDb->apShm[iChunk]);
  }
  return 0;
}


#ifndef NDEBUG
static void assertIsWorkingChild(
  lsm_db *db, 
  TreeNode *pNode, 
  TreeNode *pParent, 
  int iCell
){
  TreeNode *p;
  u32 iPtr = getChildPtr(pParent, WORKING_VERSION, iCell);
  p = treeShmptr(db, iPtr);
  assert( p==pNode );
}
#else
# define assertIsWorkingChild(w,x,y,z)
#endif

/* Values for the third argument to treeShmkey(). */
#define TKV_LOADKEY  1
#define TKV_LOADVAL  2

static TreeKey *treeShmkey(
  lsm_db *pDb,                    /* Database handle */
  u32 iPtr,                       /* Shmptr to TreeKey struct */
  int eLoad,                      /* Either zero or a TREEKEY_LOADXXX value */
  TreeBlob *pBlob,                /* Used if dynamic memory is required */
  int *pRc                        /* IN/OUT: Error code */
){
  TreeKey *pRet;

  assert( eLoad==TKV_LOADKEY || eLoad==TKV_LOADVAL );
  pRet = (TreeKey *)treeShmptr(pDb, iPtr);
  if( pRet ){
    int nReq;                     /* Bytes of space required at pRet */
    int nAvail;                   /* Bytes of space available at pRet */

    nReq = sizeof(TreeKey) + pRet->nKey;
    if( eLoad==TKV_LOADVAL && pRet->nValue>0 ){
      nReq += pRet->nValue;
    }
    assert( LSM_SHM_CHUNK_SIZE==(1<<15) );
    nAvail = LSM_SHM_CHUNK_SIZE - (iPtr & (LSM_SHM_CHUNK_SIZE-1));

    if( nAvail<nReq ){
      if( tblobGrow(pDb, pBlob, nReq, pRc)==0 ){
        int nLoad = 0;
        while( *pRc==LSM_OK ){
          ShmChunk *pChunk;
          void *p = treeShmptr(pDb, iPtr);
          int n = LSM_MIN(nAvail, nReq-nLoad);

          memcpy(&pBlob->a[nLoad], p, n);
          nLoad += n;
          if( nLoad==nReq ) break;

          pChunk = treeShmChunk(pDb, treeOffsetToChunk(iPtr));
          assert( pChunk );
          iPtr = (pChunk->iNext * LSM_SHM_CHUNK_SIZE) + LSM_SHM_CHUNK_HDR;
          nAvail = LSM_SHM_CHUNK_SIZE - LSM_SHM_CHUNK_HDR;
        }
      }
      pRet = (TreeKey *)(pBlob->a);
    }
  }

  return pRet;
}

#if defined(LSM_DEBUG) && defined(LSM_EXPENSIVE_ASSERT)
void assert_leaf_looks_ok(TreeNode *pNode){
  assert( pNode->apKey[1] );
}

void assert_node_looks_ok(TreeNode *pNode, int nHeight){
  if( pNode ){
    assert( pNode->apKey[1] );
    if( nHeight>1 ){
      int i;
      assert( getChildPtr(pNode, WORKING_VERSION, 1) );
      assert( getChildPtr(pNode, WORKING_VERSION, 2) );
      for(i=0; i<4; i++){
        assert_node_looks_ok(getChildPtr(pNode, WORKING_VERSION, i), nHeight-1);
      }
    }
  }
}

/*
** Run various assert() statements to check that the working-version of the
** tree is correct in the following respects:
**
**   * todo...
*/
void assert_tree_looks_ok(int rc, Tree *pTree){
}
#else
# define assert_tree_looks_ok(x,y)
#endif

void lsmFlagsToString(int flags, char *zFlags){

  zFlags[0] = (flags & LSM_END_DELETE)   ? ']' : '.';

  /* Only one of LSM_POINT_DELETE, LSM_INSERT and LSM_SEPARATOR should ever
  ** be set. If this is not true, write a '?' to the output.  */
  switch( flags & (LSM_POINT_DELETE|LSM_INSERT|LSM_SEPARATOR) ){
    case 0:                zFlags[1] = '.'; break;
    case LSM_POINT_DELETE: zFlags[1] = '-'; break;
    case LSM_INSERT:       zFlags[1] = '+'; break;
    case LSM_SEPARATOR:    zFlags[1] = '^'; break;
    default:               zFlags[1] = '?'; break;
  }

  zFlags[2] = (flags & LSM_SYSTEMKEY)    ? '*' : '.';
  zFlags[3] = (flags & LSM_START_DELETE) ? '[' : '.';
  zFlags[4] = '\0';
}

#ifdef LSM_DEBUG

/*
** Pointer pBlob points to a buffer containing a blob of binary data
** nBlob bytes long. Append the contents of this blob to *pStr, with
** each octet represented by a 2-digit hexadecimal number. For example,
** if the input blob is three bytes in size and contains {0x01, 0x44, 0xFF},
** then "0144ff" is appended to *pStr.
*/
static void lsmAppendStrBlob(LsmString *pStr, void *pBlob, int nBlob){
  int i;
  lsmStringExtend(pStr, nBlob*2);
  if( pStr->nAlloc==0 ) return;
  for(i=0; i<nBlob; i++){
    u8 c = ((u8*)pBlob)[i];
    if( c>='a' && c<='z' ){
      pStr->z[pStr->n++] = c;
    }else if( c!=0 || nBlob==1 || i!=(nBlob-1) ){
      pStr->z[pStr->n++] = "0123456789abcdef"[(c>>4)&0xf];
      pStr->z[pStr->n++] = "0123456789abcdef"[c&0xf];
    }
  }
  pStr->z[pStr->n] = 0;
}

#if 0  /* NOT USED */
/*
** Append nIndent space (0x20) characters to string *pStr.
*/
static void lsmAppendIndent(LsmString *pStr, int nIndent){
  int i;
  lsmStringExtend(pStr, nIndent);
  for(i=0; i<nIndent; i++) lsmStringAppend(pStr, " ", 1);
}
#endif

static void strAppendFlags(LsmString *pStr, u8 flags){
  char zFlags[8];

  lsmFlagsToString(flags, zFlags);
  zFlags[4] = ':';

  lsmStringAppend(pStr, zFlags, 5);
}

void dump_node_contents(
  lsm_db *pDb,
  u32 iNode,                      /* Print out the contents of this node */
  char *zPath,                    /* Path from root to this node */
  int nPath,                      /* Number of bytes in zPath */
  int nHeight                     /* Height: (0==leaf) (1==parent-of-leaf) */
){
  const char *zSpace = "                                           ";
  int i;
  int rc = LSM_OK;
  LsmString s;
  TreeNode *pNode;
  TreeBlob b = {0, 0};

  pNode = (TreeNode *)treeShmptr(pDb, iNode);

  if( nHeight==0 ){
    /* Append the nIndent bytes of space to string s. */
    lsmStringInit(&s, pDb->pEnv);

    /* Append each key to string s. */
    for(i=0; i<3; i++){
      u32 iPtr = pNode->aiKeyPtr[i];
      if( iPtr ){
        TreeKey *pKey = treeShmkey(pDb, pNode->aiKeyPtr[i],TKV_LOADKEY, &b,&rc);
        strAppendFlags(&s, pKey->flags);
        lsmAppendStrBlob(&s, TKV_KEY(pKey), pKey->nKey);
        lsmStringAppend(&s, "     ", -1);
      }
    }

    printf("% 6d %.*sleaf%.*s: %s\n", 
        iNode, nPath, zPath, 20-nPath-4, zSpace, s.z
    );
    lsmStringClear(&s);
  }else{
    for(i=0; i<4 && nHeight>0; i++){
      u32 iPtr = getChildPtr(pNode, pDb->treehdr.root.iTransId, i);
      zPath[nPath] = i+'0';
      zPath[nPath+1] = '/';

      if( iPtr ){
        dump_node_contents(pDb, iPtr, zPath, nPath+2, nHeight-1);
      }
      if( i!=3 && pNode->aiKeyPtr[i] ){
        TreeKey *pKey = treeShmkey(pDb, pNode->aiKeyPtr[i], TKV_LOADKEY,&b,&rc);
        lsmStringInit(&s, pDb->pEnv);
        strAppendFlags(&s, pKey->flags);
        lsmAppendStrBlob(&s, TKV_KEY(pKey), pKey->nKey);
        printf("% 6d %.*s%.*s: %s\n", 
            iNode, nPath+1, zPath, 20-nPath-1, zSpace, s.z);
        lsmStringClear(&s);
      }
    }
  }

  tblobFree(pDb, &b);
}

void dump_tree_contents(lsm_db *pDb, const char *zCaption){
  char zPath[64];
  TreeRoot *p = &pDb->treehdr.root;
  printf("\n%s\n", zCaption);
  zPath[0] = '/';
  if( p->iRoot ){
    dump_node_contents(pDb, p->iRoot, zPath, 1, p->nHeight-1);
  }
  fflush(stdout);
}

#endif

/*
** Initialize a cursor object, the space for which has already been
** allocated.
*/
static void treeCursorInit(lsm_db *pDb, int bOld, TreeCursor *pCsr){
  memset(pCsr, 0, sizeof(TreeCursor));
  pCsr->pDb = pDb;
  if( bOld ){
    pCsr->pRoot = &pDb->treehdr.oldroot;
  }else{
    pCsr->pRoot = &pDb->treehdr.root;
  }
  pCsr->iNode = -1;
}

/*
** Return a pointer to the mapping of the TreeKey object that the cursor
** is pointing to. 
*/
static TreeKey *csrGetKey(TreeCursor *pCsr, TreeBlob *pBlob, int *pRc){
  TreeKey *pRet;
  lsm_db *pDb = pCsr->pDb;
  u32 iPtr = pCsr->apTreeNode[pCsr->iNode]->aiKeyPtr[pCsr->aiCell[pCsr->iNode]];

  assert( iPtr );
  pRet = (TreeKey*)treeShmptrUnsafe(pDb, iPtr);
  if( !(pRet->flags & LSM_CONTIGUOUS) ){
    pRet = treeShmkey(pDb, iPtr, TKV_LOADVAL, pBlob, pRc);
  }

  return pRet;
}

/*
** Save the current position of tree cursor pCsr.
*/
int lsmTreeCursorSave(TreeCursor *pCsr){
  int rc = LSM_OK;
  if( pCsr && pCsr->pSave==0 ){
    int iNode = pCsr->iNode;
    if( iNode>=0 ){
      pCsr->pSave = csrGetKey(pCsr, &pCsr->blob, &rc);
    }
    pCsr->iNode = -1;
  }
  return rc;
}

/*
** Restore the position of a saved tree cursor.
*/
static int treeCursorRestore(TreeCursor *pCsr, int *pRes){
  int rc = LSM_OK;
  if( pCsr->pSave ){
    TreeKey *pKey = pCsr->pSave;
    pCsr->pSave = 0;
    if( pRes ){
      rc = lsmTreeCursorSeek(pCsr, TKV_KEY(pKey), pKey->nKey, pRes);
    }
  }
  return rc;
}

/*
** Allocate nByte bytes of space within the *-shm file. If successful, 
** return LSM_OK and set *piPtr to the offset within the file at which
** the allocated space is located.
*/
static u32 treeShmalloc(lsm_db *pDb, int bAlign, int nByte, int *pRc){
  u32 iRet = 0;
  if( *pRc==LSM_OK ){
    const static int CHUNK_SIZE = LSM_SHM_CHUNK_SIZE;
    const static int CHUNK_HDR = LSM_SHM_CHUNK_HDR;
    u32 iWrite;                   /* Current write offset */
    u32 iEof;                     /* End of current chunk */
    int iChunk;                   /* Current chunk */

    assert( nByte <= (CHUNK_SIZE-CHUNK_HDR) );

    /* Check if there is enough space on the current chunk to fit the
    ** new allocation. If not, link in a new chunk and put the new
    ** allocation at the start of it.  */
    iWrite = pDb->treehdr.iWrite;
    if( bAlign ){
      iWrite = (iWrite + 3) & ~0x0003;
      assert( (iWrite % 4)==0 );
    }

    assert( iWrite );
    iChunk = treeOffsetToChunk(iWrite-1);
    iEof = (iChunk+1) * CHUNK_SIZE;
    assert( iEof>=iWrite && (iEof-iWrite)<CHUNK_SIZE );
    if( (iWrite+nByte)>iEof ){
      ShmChunk *pHdr;           /* Header of chunk just finished (iChunk) */
      ShmChunk *pFirst;         /* Header of chunk treehdr.iFirst */
      ShmChunk *pNext;          /* Header of new chunk */
      int iNext = 0;            /* Next chunk */
      int rc = LSM_OK;

      pFirst = treeShmChunk(pDb, pDb->treehdr.iFirst);

      assert( shm_sequence_ge(pDb->treehdr.iUsedShmid, pFirst->iShmid) );
      assert( (pDb->treehdr.iNextShmid+1-pDb->treehdr.nChunk)==pFirst->iShmid );

      /* Check if the chunk at the start of the linked list is still in
      ** use. If not, reuse it. If so, allocate a new chunk by appending
      ** to the *-shm file.  */
      if( pDb->treehdr.iUsedShmid!=pFirst->iShmid ){
        int bInUse;
        rc = lsmTreeInUse(pDb, pFirst->iShmid, &bInUse);
        if( rc!=LSM_OK ){
          *pRc = rc;
          return 0;
        }
        if( bInUse==0 ){
          iNext = pDb->treehdr.iFirst;
          pDb->treehdr.iFirst = pFirst->iNext;
          assert( pDb->treehdr.iFirst );
        }
      }
      if( iNext==0 ) iNext = pDb->treehdr.nChunk++;

      /* Set the header values for the new chunk */
      pNext = treeShmChunkRc(pDb, iNext, &rc);
      if( pNext ){
        pNext->iNext = 0;
        pNext->iShmid = (pDb->treehdr.iNextShmid++);
      }else{
        *pRc = rc;
        return 0;
      }

      /* Set the header values for the chunk just finished */
      pHdr = (ShmChunk *)treeShmptr(pDb, iChunk*CHUNK_SIZE);
      pHdr->iNext = iNext;

      /* Advance to the next chunk */
      iWrite = iNext * CHUNK_SIZE + CHUNK_HDR;
    }

    /* Allocate space at iWrite. */
    iRet = iWrite;
    pDb->treehdr.iWrite = iWrite + nByte;
    pDb->treehdr.root.nByte += nByte;
  }
  return iRet;
}

/*
** Allocate and zero nByte bytes of space within the *-shm file.
*/
static void *treeShmallocZero(lsm_db *pDb, int nByte, u32 *piPtr, int *pRc){
  u32 iPtr;
  void *p;
  iPtr = treeShmalloc(pDb, 1, nByte, pRc);
  p = treeShmptr(pDb, iPtr);
  if( p ){
    assert( *pRc==LSM_OK );
    memset(p, 0, nByte);
    *piPtr = iPtr;
  }
  return p;
}

static TreeNode *newTreeNode(lsm_db *pDb, u32 *piPtr, int *pRc){
  return treeShmallocZero(pDb, sizeof(TreeNode), piPtr, pRc);
}

static TreeLeaf *newTreeLeaf(lsm_db *pDb, u32 *piPtr, int *pRc){
  return treeShmallocZero(pDb, sizeof(TreeLeaf), piPtr, pRc);
}

static TreeKey *newTreeKey(
  lsm_db *pDb, 
  u32 *piPtr, 
  void *pKey, int nKey,           /* Key data */
  void *pVal, int nVal,           /* Value data (or nVal<0 for delete) */
  int *pRc
){
  TreeKey *p;
  u32 iPtr;
  u32 iEnd;
  int nRem;
  u8 *a;
  int n;

  /* Allocate space for the TreeKey structure itself */
  *piPtr = iPtr = treeShmalloc(pDb, 1, sizeof(TreeKey), pRc);
  p = treeShmptr(pDb, iPtr);
  if( *pRc ) return 0;
  p->nKey = nKey;
  p->nValue = nVal;

  /* Allocate and populate the space required for the key and value. */
  n = nRem = nKey;
  a = (u8 *)pKey;
  while( a ){
    while( nRem>0 ){
      u8 *aAlloc;
      int nAlloc;
      u32 iWrite;

      iWrite = (pDb->treehdr.iWrite & (LSM_SHM_CHUNK_SIZE-1));
      iWrite = LSM_MAX(iWrite, LSM_SHM_CHUNK_HDR);
      nAlloc = LSM_MIN((LSM_SHM_CHUNK_SIZE-iWrite), nRem);

      aAlloc = treeShmptr(pDb, treeShmalloc(pDb, 0, nAlloc, pRc));
      if( aAlloc==0 ) break;
      memcpy(aAlloc, &a[n-nRem], nAlloc);
      nRem -= nAlloc;
    }
    a = pVal;
    n = nRem = nVal;
    pVal = 0;
  }

  iEnd = iPtr + sizeof(TreeKey) + nKey + LSM_MAX(0, nVal);
  if( (iPtr & ~(LSM_SHM_CHUNK_SIZE-1))!=(iEnd & ~(LSM_SHM_CHUNK_SIZE-1)) ){
    p->flags = 0;
  }else{
    p->flags = LSM_CONTIGUOUS;
  }

  if( *pRc ) return 0;
#if 0
  printf("store: %d %s\n", (int)iPtr, (char *)pKey);
#endif
  return p;
}

static TreeNode *copyTreeNode(
  lsm_db *pDb, 
  TreeNode *pOld, 
  u32 *piNew, 
  int *pRc
){
  TreeNode *pNew;

  pNew = newTreeNode(pDb, piNew, pRc);
  if( pNew ){
    memcpy(pNew->aiKeyPtr, pOld->aiKeyPtr, sizeof(pNew->aiKeyPtr));
    memcpy(pNew->aiChildPtr, pOld->aiChildPtr, sizeof(pNew->aiChildPtr));
    if( pOld->iV2 ) pNew->aiChildPtr[pOld->iV2Child] = pOld->iV2Ptr;
  }
  return pNew;
}

static TreeNode *copyTreeLeaf(
  lsm_db *pDb, 
  TreeLeaf *pOld, 
  u32 *piNew, 
  int *pRc
){
  TreeLeaf *pNew;
  pNew = newTreeLeaf(pDb, piNew, pRc);
  if( pNew ){
    memcpy(pNew, pOld, sizeof(TreeLeaf));
  }
  return (TreeNode *)pNew;
}

/*
** The tree cursor passed as the second argument currently points to an 
** internal node (not a leaf). Specifically, to a sub-tree pointer. This
** function replaces the sub-tree that the cursor currently points to
** with sub-tree pNew.
**
** The sub-tree may be replaced either by writing the "v2 data" on the
** internal node, or by allocating a new TreeNode structure and then 
** calling this function on the parent of the internal node.
*/
static int treeUpdatePtr(lsm_db *pDb, TreeCursor *pCsr, u32 iNew){
  int rc = LSM_OK;
  if( pCsr->iNode<0 ){
    /* iNew is the new root node */
    pDb->treehdr.root.iRoot = iNew;
  }else{
    /* If this node already has version 2 content, allocate a copy and
    ** update the copy with the new pointer value. Otherwise, store the
    ** new pointer as v2 data within the current node structure.  */

    TreeNode *p;                  /* The node to be modified */
    int iChildPtr;                /* apChild[] entry to modify */

    p = pCsr->apTreeNode[pCsr->iNode];
    iChildPtr = pCsr->aiCell[pCsr->iNode];

    if( p->iV2 ){
      /* The "allocate new TreeNode" option */
      u32 iCopy;
      TreeNode *pCopy;
      pCopy = copyTreeNode(pDb, p, &iCopy, &rc);
      if( pCopy ){
        assert( rc==LSM_OK );
        pCopy->aiChildPtr[iChildPtr] = iNew;
        pCsr->iNode--;
        rc = treeUpdatePtr(pDb, pCsr, iCopy);
      }
    }else{
      /* The "v2 data" option */
      u32 iPtr;
      assert( pDb->treehdr.root.iTransId>0 );

      if( pCsr->iNode ){
        iPtr = getChildPtr(
            pCsr->apTreeNode[pCsr->iNode-1], 
            pDb->treehdr.root.iTransId, pCsr->aiCell[pCsr->iNode-1]
        );
      }else{
        iPtr = pDb->treehdr.root.iRoot;
      }
      rc = intArrayAppend(pDb->pEnv, &pDb->rollback, iPtr);

      if( rc==LSM_OK ){
        p->iV2 = pDb->treehdr.root.iTransId;
        p->iV2Child = (u8)iChildPtr;
        p->iV2Ptr = iNew;
      }
    }
  }

  return rc;
}

/*
** Cursor pCsr points at a node that is part of pTree. This function
** inserts a new key and optionally child node pointer into that node.
**
** The position into which the new key and pointer are inserted is
** determined by the iSlot parameter. The new key will be inserted to
** the left of the key currently stored in apKey[iSlot]. Or, if iSlot is
** greater than the index of the rightmost key in the node.
**
** Pointer pLeftPtr points to a child tree that contains keys that are
** smaller than pTreeKey.
*/
static int treeInsert(
  lsm_db *pDb,                    /* Database handle */
  TreeCursor *pCsr,               /* Cursor indicating path to insert at */
  u32 iLeftPtr,                   /* Left child pointer */
  u32 iTreeKey,                   /* Location of key to insert */
  u32 iRightPtr,                  /* Right child pointer */
  int iSlot                       /* Position to insert key into */
){
  int rc = LSM_OK;
  TreeNode *pNode = pCsr->apTreeNode[pCsr->iNode];

  /* Check if the node is currently full. If so, split pNode in two and
  ** call this function recursively to add a key to the parent. Otherwise, 
  ** insert the new key directly into pNode.  */
  assert( pNode->aiKeyPtr[1] );
  if( pNode->aiKeyPtr[0] && pNode->aiKeyPtr[2] ){
    u32 iLeft; TreeNode *pLeft;   /* New left-hand sibling node */
    u32 iRight; TreeNode *pRight; /* New right-hand sibling node */

    pLeft = newTreeNode(pDb, &iLeft, &rc);
    pRight = newTreeNode(pDb, &iRight, &rc);
    if( rc ) return rc;

    pLeft->aiChildPtr[1] = getChildPtr(pNode, WORKING_VERSION, 0);
    pLeft->aiKeyPtr[1] = pNode->aiKeyPtr[0];
    pLeft->aiChildPtr[2] = getChildPtr(pNode, WORKING_VERSION, 1);

    pRight->aiChildPtr[1] = getChildPtr(pNode, WORKING_VERSION, 2);
    pRight->aiKeyPtr[1] = pNode->aiKeyPtr[2];
    pRight->aiChildPtr[2] = getChildPtr(pNode, WORKING_VERSION, 3);

    if( pCsr->iNode==0 ){
      /* pNode is the root of the tree. Grow the tree by one level. */
      u32 iRoot; TreeNode *pRoot; /* New root node */

      pRoot = newTreeNode(pDb, &iRoot, &rc);
      pRoot->aiKeyPtr[1] = pNode->aiKeyPtr[1];
      pRoot->aiChildPtr[1] = iLeft;
      pRoot->aiChildPtr[2] = iRight;

      pDb->treehdr.root.iRoot = iRoot;
      pDb->treehdr.root.nHeight++;
    }else{

      pCsr->iNode--;
      rc = treeInsert(pDb, pCsr, 
          iLeft, pNode->aiKeyPtr[1], iRight, pCsr->aiCell[pCsr->iNode]
      );
    }

    assert( pLeft->iV2==0 );
    assert( pRight->iV2==0 );
    switch( iSlot ){
      case 0:
        pLeft->aiKeyPtr[0] = iTreeKey;
        pLeft->aiChildPtr[0] = iLeftPtr;
        if( iRightPtr ) pLeft->aiChildPtr[1] = iRightPtr;
        break;
      case 1:
        pLeft->aiChildPtr[3] = (iRightPtr ? iRightPtr : pLeft->aiChildPtr[2]);
        pLeft->aiKeyPtr[2] = iTreeKey;
        pLeft->aiChildPtr[2] = iLeftPtr;
        break;
      case 2:
        pRight->aiKeyPtr[0] = iTreeKey;
        pRight->aiChildPtr[0] = iLeftPtr;
        if( iRightPtr ) pRight->aiChildPtr[1] = iRightPtr;
        break;
      case 3:
        pRight->aiChildPtr[3] = (iRightPtr ? iRightPtr : pRight->aiChildPtr[2]);
        pRight->aiKeyPtr[2] = iTreeKey;
        pRight->aiChildPtr[2] = iLeftPtr;
        break;
    }

  }else{
    TreeNode *pNew;
    u32 *piKey;
    u32 *piChild;
    u32 iStore = 0;
    u32 iNew = 0;
    int i;

    /* Allocate a new version of node pNode. */
    pNew = newTreeNode(pDb, &iNew, &rc);
    if( rc ) return rc;

    piKey = pNew->aiKeyPtr;
    piChild = pNew->aiChildPtr;

    for(i=0; i<iSlot; i++){
      if( pNode->aiKeyPtr[i] ){
        *(piKey++) = pNode->aiKeyPtr[i];
        *(piChild++) = getChildPtr(pNode, WORKING_VERSION, i);
      }
    }

    *piKey++ = iTreeKey;
    *piChild++ = iLeftPtr;

    iStore = iRightPtr;
    for(i=iSlot; i<3; i++){
      if( pNode->aiKeyPtr[i] ){
        *(piKey++) = pNode->aiKeyPtr[i];
        *(piChild++) = iStore ? iStore : getChildPtr(pNode, WORKING_VERSION, i);
        iStore = 0;
      }
    }

    if( iStore ){
      *piChild = iStore;
    }else{
      *piChild = getChildPtr(pNode, WORKING_VERSION, 
          (pNode->aiKeyPtr[2] ? 3 : 2)
      );
    }
    pCsr->iNode--;
    rc = treeUpdatePtr(pDb, pCsr, iNew);
  }

  return rc;
}

static int treeInsertLeaf(
  lsm_db *pDb,                    /* Database handle */
  TreeCursor *pCsr,               /* Cursor structure */
  u32 iTreeKey,                   /* Key pointer to insert */
  int iSlot                       /* Insert key to the left of this */
){
  int rc = LSM_OK;                /* Return code */
  TreeNode *pLeaf = pCsr->apTreeNode[pCsr->iNode];
  TreeLeaf *pNew;
  u32 iNew;

  assert( iSlot>=0 && iSlot<=4 );
  assert( pCsr->iNode>0 );
  assert( pLeaf->aiKeyPtr[1] );

  pCsr->iNode--;

  pNew = newTreeLeaf(pDb, &iNew, &rc);
  if( pNew ){
    if( pLeaf->aiKeyPtr[0] && pLeaf->aiKeyPtr[2] ){
      /* The leaf is full. Split it in two. */
      TreeLeaf *pRight;
      u32 iRight;
      pRight = newTreeLeaf(pDb, &iRight, &rc);
      if( pRight ){
        assert( rc==LSM_OK );
        pNew->aiKeyPtr[1] = pLeaf->aiKeyPtr[0];
        pRight->aiKeyPtr[1] = pLeaf->aiKeyPtr[2];
        switch( iSlot ){
          case 0: pNew->aiKeyPtr[0] = iTreeKey; break;
          case 1: pNew->aiKeyPtr[2] = iTreeKey; break;
          case 2: pRight->aiKeyPtr[0] = iTreeKey; break;
          case 3: pRight->aiKeyPtr[2] = iTreeKey; break;
        }

        rc = treeInsert(pDb, pCsr, iNew, pLeaf->aiKeyPtr[1], iRight, 
            pCsr->aiCell[pCsr->iNode]
        );
      }
    }else{
      int iOut = 0;
      int i;
      for(i=0; i<4; i++){
        if( i==iSlot ) pNew->aiKeyPtr[iOut++] = iTreeKey;
        if( i<3 && pLeaf->aiKeyPtr[i] ){
          pNew->aiKeyPtr[iOut++] = pLeaf->aiKeyPtr[i];
        }
      }
      rc = treeUpdatePtr(pDb, pCsr, iNew);
    }
  }

  return rc;
}

void lsmTreeMakeOld(lsm_db *pDb){

  /* A write transaction must be open. Otherwise the code below that
  ** assumes (pDb->pClient->iLogOff) is current may malfunction. 
  **
  ** Update: currently this assert fails due to lsm_flush(), which does
  ** not set nTransOpen.
  */
  assert( /* pDb->nTransOpen>0 && */ pDb->iReader>=0 );

  if( pDb->treehdr.iOldShmid==0 ){
    pDb->treehdr.iOldLog = (pDb->treehdr.log.aRegion[2].iEnd << 1);
    pDb->treehdr.iOldLog |= (~(pDb->pClient->iLogOff) & (i64)0x0001);

    pDb->treehdr.oldcksum0 = pDb->treehdr.log.cksum0;
    pDb->treehdr.oldcksum1 = pDb->treehdr.log.cksum1;
    pDb->treehdr.iOldShmid = pDb->treehdr.iNextShmid-1;
    memcpy(&pDb->treehdr.oldroot, &pDb->treehdr.root, sizeof(TreeRoot));

    pDb->treehdr.root.iTransId = 1;
    pDb->treehdr.root.iRoot = 0;
    pDb->treehdr.root.nHeight = 0;
    pDb->treehdr.root.nByte = 0;
  }
}

void lsmTreeDiscardOld(lsm_db *pDb){
  assert( lsmShmAssertLock(pDb, LSM_LOCK_WRITER, LSM_LOCK_EXCL) 
       || lsmShmAssertLock(pDb, LSM_LOCK_DMS2, LSM_LOCK_EXCL) 
  );
  pDb->treehdr.iUsedShmid = pDb->treehdr.iOldShmid;
  pDb->treehdr.iOldShmid = 0;
}

int lsmTreeHasOld(lsm_db *pDb){
  return pDb->treehdr.iOldShmid!=0;
}

/*
** This function is called during recovery to initialize the 
** tree header. Only the database connections private copy of the tree-header
** is initialized here - it will be copied into shared memory if log file
** recovery is successful.
*/
int lsmTreeInit(lsm_db *pDb){
  ShmChunk *pOne;
  int rc = LSM_OK;

  memset(&pDb->treehdr, 0, sizeof(TreeHeader));
  pDb->treehdr.root.iTransId = 1;
  pDb->treehdr.iFirst = 1;
  pDb->treehdr.nChunk = 2;
  pDb->treehdr.iWrite = LSM_SHM_CHUNK_SIZE + LSM_SHM_CHUNK_HDR;
  pDb->treehdr.iNextShmid = 2;
  pDb->treehdr.iUsedShmid = 1;

  pOne = treeShmChunkRc(pDb, 1, &rc);
  if( pOne ){
    pOne->iNext = 0;
    pOne->iShmid = 1;
  }
  return rc;
}

static void treeHeaderChecksum(
  TreeHeader *pHdr, 
  u32 *aCksum
){
  u32 cksum1 = 0x12345678;
  u32 cksum2 = 0x9ABCDEF0;
  u32 *a = (u32 *)pHdr;
  int i;

  assert( (offsetof(TreeHeader, aCksum) + sizeof(u32)*2)==sizeof(TreeHeader) );
  assert( (sizeof(TreeHeader) % (sizeof(u32)*2))==0 );

  for(i=0; i<(offsetof(TreeHeader, aCksum) / sizeof(u32)); i+=2){
    cksum1 += a[i];
    cksum2 += (cksum1 + a[i+1]);
  }
  aCksum[0] = cksum1;
  aCksum[1] = cksum2;
}

/*
** Return true if the checksum stored in TreeHeader object *pHdr is 
** consistent with the contents of its other fields.
*/
static int treeHeaderChecksumOk(TreeHeader *pHdr){
  u32 aCksum[2];
  treeHeaderChecksum(pHdr, aCksum);
  return (0==memcmp(aCksum, pHdr->aCksum, sizeof(aCksum)));
}

/*
** This type is used by functions lsmTreeRepair() and treeSortByShmid() to
** make relinking the linked list of shared-memory chunks easier.
*/
typedef struct ShmChunkLoc ShmChunkLoc;
struct ShmChunkLoc {
  ShmChunk *pShm;
  u32 iLoc;
};

/*
** This function checks that the linked list of shared memory chunks 
** that starts at chunk db->treehdr.iFirst:
**
**   1) Includes all chunks in the shared-memory region, and
**   2) Links them together in order of ascending shm-id.
**
** If no error occurs and the conditions above are met, LSM_OK is returned.
**
** If either of the conditions are untrue, LSM_CORRUPT is returned. Or, if
** an error is encountered before the checks are completed, another LSM error
** code (i.e. LSM_IOERR or LSM_NOMEM) may be returned.
*/
static int treeCheckLinkedList(lsm_db *db){
  int rc = LSM_OK;
  int nVisit = 0;
  ShmChunk *p;

  p = treeShmChunkRc(db, db->treehdr.iFirst, &rc);
  while( rc==LSM_OK && p ){
    if( p->iNext ){
      if( p->iNext>=db->treehdr.nChunk ){
        rc = LSM_CORRUPT_BKPT;
      }else{
        ShmChunk *pNext = treeShmChunkRc(db, p->iNext, &rc);
        if( rc==LSM_OK ){
          if( pNext->iShmid!=p->iShmid+1 ){
            rc = LSM_CORRUPT_BKPT;
          }
          p = pNext;
        }
      }
    }else{
      p = 0;
    }
    nVisit++;
  }

  if( rc==LSM_OK && nVisit!=db->treehdr.nChunk-1 ){
    rc = LSM_CORRUPT_BKPT;
  }
  return rc;
}

/*
** Iterate through the current in-memory tree. If there are any v2-pointers
** with transaction ids larger than db->treehdr.iTransId, zero them.
*/
static int treeRepairPtrs(lsm_db *db){
  int rc = LSM_OK;

  if( db->treehdr.root.nHeight>1 ){
    TreeCursor csr;               /* Cursor used to iterate through tree */
    u32 iTransId = db->treehdr.root.iTransId;

    /* Initialize the cursor structure. Also decrement the nHeight variable
    ** in the tree-header. This will prevent the cursor from visiting any
    ** leaf nodes.  */
    db->treehdr.root.nHeight--;
    treeCursorInit(db, 0, &csr);

    rc = lsmTreeCursorEnd(&csr, 0);
    while( rc==LSM_OK && lsmTreeCursorValid(&csr) ){
      TreeNode *pNode = csr.apTreeNode[csr.iNode];
      if( pNode->iV2>iTransId ){
        pNode->iV2Child = 0;
        pNode->iV2Ptr = 0;
        pNode->iV2 = 0;
      }
      rc = lsmTreeCursorNext(&csr);
    }
    tblobFree(csr.pDb, &csr.blob);

    db->treehdr.root.nHeight++;
  }

  return rc;
}

static int treeRepairList(lsm_db *db){
  int rc = LSM_OK;
  int i;
  ShmChunk *p;
  ShmChunk *pMin = 0;
  u32 iMin = 0;

  /* Iterate through all shm chunks. Find the smallest shm-id present in
  ** the shared-memory region. */
  for(i=1; rc==LSM_OK && i<db->treehdr.nChunk; i++){
    p = treeShmChunkRc(db, i, &rc);
    if( p && (pMin==0 || shm_sequence_ge(pMin->iShmid, p->iShmid)) ){
      pMin = p;
      iMin = i;
    }
  }

  /* Fix the shm-id values on any chunks with a shm-id greater than or 
  ** equal to treehdr.iNextShmid. Then do a merge-sort of all chunks to 
  ** fix the ShmChunk.iNext pointers.
  */
  if( rc==LSM_OK ){
    int nSort;
    int nByte;
    u32 iPrevShmid;
    ShmChunkLoc *aSort;

    /* Allocate space for a merge sort. */
    nSort = 1;
    while( nSort < (db->treehdr.nChunk-1) ) nSort = nSort * 2;
    nByte = sizeof(ShmChunkLoc) * nSort * 2;
    aSort = lsmMallocZeroRc(db->pEnv, nByte, &rc);
    iPrevShmid = pMin->iShmid;

    /* Fix all shm-ids, if required. */
    if( rc==LSM_OK ){
      iPrevShmid = pMin->iShmid-1;
      for(i=1; i<db->treehdr.nChunk; i++){
        p = treeShmChunk(db, i);
        aSort[i-1].pShm = p;
        aSort[i-1].iLoc = i;
        if( i!=db->treehdr.iFirst ){
          if( shm_sequence_ge(p->iShmid, db->treehdr.iNextShmid) ){
            p->iShmid = iPrevShmid--;
          }
        }
      }
      if( iMin!=db->treehdr.iFirst ){
        p = treeShmChunk(db, db->treehdr.iFirst);
        p->iShmid = iPrevShmid;
      }
    }

    if( rc==LSM_OK ){
      ShmChunkLoc *aSpace = &aSort[nSort];
      for(i=0; i<nSort; i++){
        if( aSort[i].pShm ){
          assert( shm_sequence_ge(aSort[i].pShm->iShmid, iPrevShmid) );
          assert( aSpace[aSort[i].pShm->iShmid - iPrevShmid].pShm==0 );
          aSpace[aSort[i].pShm->iShmid - iPrevShmid] = aSort[i];
        }
      }

      if( aSpace[nSort-1].pShm ) aSpace[nSort-1].pShm->iNext = 0;
      for(i=0; i<nSort-1; i++){
        if( aSpace[i].pShm ){
          aSpace[i].pShm->iNext = aSpace[i+1].iLoc;
        }
      }

      rc = treeCheckLinkedList(db);
      lsmFree(db->pEnv, aSort);
    }
  }

  return rc;
}

/*
** This function is called as part of opening a write-transaction if the
** writer-flag is already set - indicating that the previous writer 
** failed before ending its transaction.
*/
int lsmTreeRepair(lsm_db *db){
  int rc = LSM_OK;
  TreeHeader hdr;
  ShmHeader *pHdr = db->pShmhdr;

  /* Ensure that the two tree-headers are consistent. Copy one over the other
  ** if necessary. Prefer the data from a tree-header for which the checksum
  ** computes. Or, if they both compute, prefer tree-header-1.  */
  if( memcmp(&pHdr->hdr1, &pHdr->hdr2, sizeof(TreeHeader)) ){
    if( treeHeaderChecksumOk(&pHdr->hdr1) ){
      memcpy(&pHdr->hdr2, &pHdr->hdr1, sizeof(TreeHeader));
    }else{
      memcpy(&pHdr->hdr1, &pHdr->hdr2, sizeof(TreeHeader));
    }
  }

  /* Save the connections current copy of the tree-header. It will be 
  ** restored before returning.  */
  memcpy(&hdr, &db->treehdr, sizeof(TreeHeader));

  /* Walk the tree. Zero any v2 pointers with a transaction-id greater than
  ** the transaction-id currently in the tree-headers.  */
  rc = treeRepairPtrs(db);

  /* Repair the linked list of shared-memory chunks. */
  if( rc==LSM_OK ){
    rc = treeRepairList(db);
  }

  memcpy(&db->treehdr, &hdr, sizeof(TreeHeader));
  return rc;
}

static void treeOverwriteKey(lsm_db *db, TreeCursor *pCsr, u32 iKey, int *pRc){
  if( *pRc==LSM_OK ){
    TreeRoot *p = &db->treehdr.root;
    TreeNode *pNew;
    u32 iNew;
    TreeNode *pNode = pCsr->apTreeNode[pCsr->iNode];
    int iCell = pCsr->aiCell[pCsr->iNode];

    /* Create a copy of this node */
    if( (pCsr->iNode>0 && pCsr->iNode==(p->nHeight-1)) ){
      pNew = copyTreeLeaf(db, (TreeLeaf *)pNode, &iNew, pRc);
    }else{
      pNew = copyTreeNode(db, pNode, &iNew, pRc);
    }

    if( pNew ){
      /* Modify the value in the new version */
      pNew->aiKeyPtr[iCell] = iKey;

      /* Change the pointer in the parent (if any) to point at the new 
       ** TreeNode */
      pCsr->iNode--;
      treeUpdatePtr(db, pCsr, iNew);
    }
  }
}

static int treeNextIsEndDelete(lsm_db *db, TreeCursor *pCsr){
  int iNode = pCsr->iNode;
  int iCell = pCsr->aiCell[iNode]+1;

  /* Cursor currently points to a leaf node. */
  assert( pCsr->iNode==(db->treehdr.root.nHeight-1) );

  while( iNode>=0 ){
    TreeNode *pNode = pCsr->apTreeNode[iNode];
    if( iCell<3 && pNode->aiKeyPtr[iCell] ){
      int rc = LSM_OK;
      TreeKey *pKey = treeShmptr(db, pNode->aiKeyPtr[iCell]);
      assert( rc==LSM_OK );
      return ((pKey->flags & LSM_END_DELETE) ? 1 : 0);
    }
    iNode--;
    iCell = pCsr->aiCell[iNode];
  }

  return 0;
}

static int treePrevIsStartDelete(lsm_db *db, TreeCursor *pCsr){
  int iNode = pCsr->iNode;

  /* Cursor currently points to a leaf node. */
  assert( pCsr->iNode==(db->treehdr.root.nHeight-1) );

  while( iNode>=0 ){
    TreeNode *pNode = pCsr->apTreeNode[iNode];
    int iCell = pCsr->aiCell[iNode]-1;
    if( iCell>=0 && pNode->aiKeyPtr[iCell] ){
      int rc = LSM_OK;
      TreeKey *pKey = treeShmptr(db, pNode->aiKeyPtr[iCell]);
      assert( rc==LSM_OK );
      return ((pKey->flags & LSM_START_DELETE) ? 1 : 0);
    }
    iNode--;
  }

  return 0;
}


static int treeInsertEntry(
  lsm_db *pDb,                    /* Database handle */
  int flags,                      /* Flags associated with entry */
  void *pKey,                     /* Pointer to key data */
  int nKey,                       /* Size of key data in bytes */
  void *pVal,                     /* Pointer to value data (or NULL) */
  int nVal                        /* Bytes in value data (or -ve for delete) */
){
  int rc = LSM_OK;                /* Return Code */
  TreeKey *pTreeKey;              /* New key-value being inserted */
  u32 iTreeKey;
  TreeRoot *p = &pDb->treehdr.root;
  TreeCursor csr;                 /* Cursor to seek to pKey/nKey */
  int res;                        /* Result of seek operation on csr */

  assert( nVal>=0 || pVal==0 );
  assert_tree_looks_ok(LSM_OK, pTree);
  assert( flags==LSM_INSERT       || flags==LSM_POINT_DELETE 
       || flags==LSM_START_DELETE || flags==LSM_END_DELETE 
  );
  assert( (flags & LSM_CONTIGUOUS)==0 );
#if 0
  dump_tree_contents(pDb, "before");
#endif

  if( p->iRoot ){
    TreeKey *pRes;                /* Key at end of seek operation */
    treeCursorInit(pDb, 0, &csr);

    /* Seek to the leaf (or internal node) that the new key belongs on */
    rc = lsmTreeCursorSeek(&csr, pKey, nKey, &res);
    pRes = csrGetKey(&csr, &csr.blob, &rc);
    if( rc!=LSM_OK ) return rc;

    if( flags==LSM_START_DELETE ){
      /* When inserting a start-delete-range entry, if the key that
      ** occurs immediately before the new entry is already a START_DELETE,
      ** then the new entry is not required.  */
      if( (res<=0 && (pRes->flags & LSM_START_DELETE))
       || (res>0  && treePrevIsStartDelete(pDb, &csr))
      ){ 
        goto insert_entry_out;
      }
    }else if( flags==LSM_END_DELETE ){
      /* When inserting an start-delete-range entry, if the key that
      ** occurs immediately after the new entry is already an END_DELETE,
      ** then the new entry is not required.  */
      if( (res<0  && treeNextIsEndDelete(pDb, &csr))
       || (res>=0 && (pRes->flags & LSM_END_DELETE))
      ){
        goto insert_entry_out;
      }
    }

    if( res==0 && (flags & (LSM_END_DELETE|LSM_START_DELETE)) ){
      if( pRes->flags & LSM_INSERT ){
        nVal = pRes->nValue;
        pVal = TKV_VAL(pRes);
      }
      flags = flags | pRes->flags;
    }

    if( flags & (LSM_INSERT|LSM_POINT_DELETE) ){
      if( (res<0 && (pRes->flags & LSM_START_DELETE))
       || (res>0 && (pRes->flags & LSM_END_DELETE)) 
      ){
        flags = flags | (LSM_END_DELETE|LSM_START_DELETE);
      }else if( res==0 ){
        flags = flags | (pRes->flags & (LSM_END_DELETE|LSM_START_DELETE));
      }
    }
  }else{
    memset(&csr, 0, sizeof(TreeCursor));
  }

  /* Allocate and populate a new key-value pair structure */
  pTreeKey = newTreeKey(pDb, &iTreeKey, pKey, nKey, pVal, nVal, &rc);
  if( rc!=LSM_OK ) return rc;
  assert( pTreeKey->flags==0 || pTreeKey->flags==LSM_CONTIGUOUS );
  pTreeKey->flags |= flags;

  if( p->iRoot==0 ){
    /* The tree is completely empty. Add a new root node and install
    ** (pKey/nKey) as the middle entry. Even though it is a leaf at the
    ** moment, use newTreeNode() to allocate the node (i.e. allocate enough
    ** space for the fields used by interior nodes). This is because the
    ** treeInsert() routine may convert this node to an interior node. */
    TreeNode *pRoot = newTreeNode(pDb, &p->iRoot, &rc);
    if( rc==LSM_OK ){
      assert( p->nHeight==0 );
      pRoot->aiKeyPtr[1] = iTreeKey;
      p->nHeight = 1;
    }
  }else{
    if( res==0 ){
      /* The search found a match within the tree. */
      treeOverwriteKey(pDb, &csr, iTreeKey, &rc);
    }else{
      /* The cursor now points to the leaf node into which the new entry should
      ** be inserted. There may or may not be a free slot within the leaf for
      ** the new key-value pair. 
      **
      ** iSlot is set to the index of the key within pLeaf that the new key
      ** should be inserted to the left of (or to a value 1 greater than the
      ** index of the rightmost key if the new key is larger than all keys
      ** currently stored in the node).
      */
      int iSlot = csr.aiCell[csr.iNode] + (res<0);
      if( csr.iNode==0 ){
        rc = treeInsert(pDb, &csr, 0, iTreeKey, 0, iSlot);
      }else{
        rc = treeInsertLeaf(pDb, &csr, iTreeKey, iSlot);
      }
    }
  }

#if 0
  dump_tree_contents(pDb, "after");
#endif
 insert_entry_out:
  tblobFree(pDb, &csr.blob);
  assert_tree_looks_ok(rc, pTree);
  return rc;
}

/*
** Insert a new entry into the in-memory tree.
**
** If the value of the 5th parameter, nVal, is negative, then a delete-marker
** is inserted into the tree. In this case the value pointer, pVal, must be
** NULL.
*/
int lsmTreeInsert(
  lsm_db *pDb,                    /* Database handle */
  void *pKey,                     /* Pointer to key data */
  int nKey,                       /* Size of key data in bytes */
  void *pVal,                     /* Pointer to value data (or NULL) */
  int nVal                        /* Bytes in value data (or -ve for delete) */
){
  int flags;
  if( nVal<0 ){
    flags = LSM_POINT_DELETE;
  }else{
    flags = LSM_INSERT;
  }

  return treeInsertEntry(pDb, flags, pKey, nKey, pVal, nVal);
}

static int treeDeleteEntry(lsm_db *db, TreeCursor *pCsr, u32 iNewptr){
  TreeRoot *p = &db->treehdr.root;
  TreeNode *pNode = pCsr->apTreeNode[pCsr->iNode];
  int iSlot = pCsr->aiCell[pCsr->iNode];
  int bLeaf;
  int rc = LSM_OK;

  assert( pNode->aiKeyPtr[1] );
  assert( pNode->aiKeyPtr[iSlot] );
  assert( iSlot==0 || iSlot==1 || iSlot==2 );
  assert( (pCsr->iNode==(db->treehdr.root.nHeight-1))==(iNewptr==0) );

  bLeaf = (pCsr->iNode==(p->nHeight-1) && p->nHeight>1);
  
  if( pNode->aiKeyPtr[0] || pNode->aiKeyPtr[2] ){
    /* There are currently at least 2 keys on this node. So just create
    ** a new copy of the node with one of the keys removed. If the node
    ** happens to be the root node of the tree, allocate an entire 
    ** TreeNode structure instead of just a TreeLeaf.  */
    TreeNode *pNew;
    u32 iNew;

    if( bLeaf ){
      pNew = (TreeNode *)newTreeLeaf(db, &iNew, &rc);
    }else{
      pNew = newTreeNode(db, &iNew, &rc);
    }
    if( pNew ){
      int i;
      int iOut = 1;
      for(i=0; i<4; i++){
        if( i==iSlot ){
          i++;
          if( bLeaf==0 ) pNew->aiChildPtr[iOut] = iNewptr;
          if( i<3 ) pNew->aiKeyPtr[iOut] = pNode->aiKeyPtr[i];
          iOut++;
        }else if( bLeaf || p->nHeight==1 ){
          if( i<3 && pNode->aiKeyPtr[i] ){
            pNew->aiKeyPtr[iOut++] = pNode->aiKeyPtr[i];
          }
        }else{
          if( getChildPtr(pNode, WORKING_VERSION, i) ){
            pNew->aiChildPtr[iOut] = getChildPtr(pNode, WORKING_VERSION, i);
            if( i<3 ) pNew->aiKeyPtr[iOut] = pNode->aiKeyPtr[i];
            iOut++;
          }
        }
      }
      assert( iOut<=4 );
      assert( bLeaf || pNew->aiChildPtr[0]==0 );
      pCsr->iNode--;
      rc = treeUpdatePtr(db, pCsr, iNew);
    }

  }else if( pCsr->iNode==0 ){
    /* Removing the only key in the root node. iNewptr is the new root. */
    assert( iSlot==1 );
    db->treehdr.root.iRoot = iNewptr;
    db->treehdr.root.nHeight--;

  }else{
    /* There is only one key on this node and the node is not the root
    ** node. Find a peer for this node. Then redistribute the contents of
    ** the peer and the parent cell between the parent and either one or
    ** two new nodes.  */
    TreeNode *pParent;            /* Parent tree node */
    int iPSlot;
    u32 iPeer;                    /* Pointer to peer leaf node */
    int iDir;
    TreeNode *pPeer;              /* The peer leaf node */
    TreeNode *pNew1; u32 iNew1;   /* First new leaf node */

    assert( iSlot==1 );

    pParent = pCsr->apTreeNode[pCsr->iNode-1];
    iPSlot = pCsr->aiCell[pCsr->iNode-1];

    if( iPSlot>0 && getChildPtr(pParent, WORKING_VERSION, iPSlot-1) ){
      iDir = -1;
    }else{
      iDir = +1;
    }
    iPeer = getChildPtr(pParent, WORKING_VERSION, iPSlot+iDir);
    pPeer = (TreeNode *)treeShmptr(db, iPeer);
    assertIsWorkingChild(db, pNode, pParent, iPSlot);

    /* Allocate the first new leaf node. This is always required. */
    if( bLeaf ){
      pNew1 = (TreeNode *)newTreeLeaf(db, &iNew1, &rc);
    }else{
      pNew1 = (TreeNode *)newTreeNode(db, &iNew1, &rc);
    }

    if( pPeer->aiKeyPtr[0] && pPeer->aiKeyPtr[2] ){
      /* Peer node is completely full. This means that two new leaf nodes
      ** and a new parent node are required. */

      TreeNode *pNew2; u32 iNew2; /* Second new leaf node */
      TreeNode *pNewP; u32 iNewP; /* New parent node */

      if( bLeaf ){
        pNew2 = (TreeNode *)newTreeLeaf(db, &iNew2, &rc);
      }else{
        pNew2 = (TreeNode *)newTreeNode(db, &iNew2, &rc);
      }
      pNewP = copyTreeNode(db, pParent, &iNewP, &rc);

      if( iDir==-1 ){
        pNew1->aiKeyPtr[1] = pPeer->aiKeyPtr[0];
        if( bLeaf==0 ){
          pNew1->aiChildPtr[1] = getChildPtr(pPeer, WORKING_VERSION, 0);
          pNew1->aiChildPtr[2] = getChildPtr(pPeer, WORKING_VERSION, 1);
        }

        pNewP->aiChildPtr[iPSlot-1] = iNew1;
        pNewP->aiKeyPtr[iPSlot-1] = pPeer->aiKeyPtr[1];
        pNewP->aiChildPtr[iPSlot] = iNew2;

        pNew2->aiKeyPtr[0] = pPeer->aiKeyPtr[2];
        pNew2->aiKeyPtr[1] = pParent->aiKeyPtr[iPSlot-1];
        if( bLeaf==0 ){
          pNew2->aiChildPtr[0] = getChildPtr(pPeer, WORKING_VERSION, 2);
          pNew2->aiChildPtr[1] = getChildPtr(pPeer, WORKING_VERSION, 3);
          pNew2->aiChildPtr[2] = iNewptr;
        }
      }else{
        pNew1->aiKeyPtr[1] = pParent->aiKeyPtr[iPSlot];
        if( bLeaf==0 ){
          pNew1->aiChildPtr[1] = iNewptr;
          pNew1->aiChildPtr[2] = getChildPtr(pPeer, WORKING_VERSION, 0);
        }

        pNewP->aiChildPtr[iPSlot] = iNew1;
        pNewP->aiKeyPtr[iPSlot] = pPeer->aiKeyPtr[0];
        pNewP->aiChildPtr[iPSlot+1] = iNew2;

        pNew2->aiKeyPtr[0] = pPeer->aiKeyPtr[1];
        pNew2->aiKeyPtr[1] = pPeer->aiKeyPtr[2];
        if( bLeaf==0 ){
          pNew2->aiChildPtr[0] = getChildPtr(pPeer, WORKING_VERSION, 1);
          pNew2->aiChildPtr[1] = getChildPtr(pPeer, WORKING_VERSION, 2);
          pNew2->aiChildPtr[2] = getChildPtr(pPeer, WORKING_VERSION, 3);
        }
      }
      assert( pCsr->iNode>=1 );
      pCsr->iNode -= 2;
      if( rc==LSM_OK ){
        assert( pNew1->aiKeyPtr[1] && pNew2->aiKeyPtr[1] );
        rc = treeUpdatePtr(db, pCsr, iNewP);
      }
    }else{
      int iKOut = 0;
      int iPOut = 0;
      int i;

      pCsr->iNode--;

      if( iDir==1 ){
        pNew1->aiKeyPtr[iKOut++] = pParent->aiKeyPtr[iPSlot];
        if( bLeaf==0 ) pNew1->aiChildPtr[iPOut++] = iNewptr;
      }
      for(i=0; i<3; i++){
        if( pPeer->aiKeyPtr[i] ){
          pNew1->aiKeyPtr[iKOut++] = pPeer->aiKeyPtr[i];
        }
      }
      if( bLeaf==0 ){
        for(i=0; i<4; i++){
          if( getChildPtr(pPeer, WORKING_VERSION, i) ){
            pNew1->aiChildPtr[iPOut++] = getChildPtr(pPeer, WORKING_VERSION, i);
          }
        }
      }
      if( iDir==-1 ){
        iPSlot--;
        pNew1->aiKeyPtr[iKOut++] = pParent->aiKeyPtr[iPSlot];
        if( bLeaf==0 ) pNew1->aiChildPtr[iPOut++] = iNewptr;
        pCsr->aiCell[pCsr->iNode] = iPSlot;
      }

      rc = treeDeleteEntry(db, pCsr, iNew1);
    }
  }

  return rc;
}

/*
** Delete a range of keys from the tree structure (i.e. the lsm_delete_range()
** function, not lsm_delete()).
**
** This is a two step process: 
**
**     1) Remove all entries currently stored in the tree that have keys
**        that fall into the deleted range.
**
**        TODO: There are surely good ways to optimize this step - removing 
**        a range of keys from a b-tree. But for now, this function removes
**        them one at a time using the usual approach.
**
**     2) Unless the largest key smaller than or equal to (pKey1/nKey1) is
**        already marked as START_DELETE, insert a START_DELETE key. 
**        Similarly, unless the smallest key greater than or equal to
**        (pKey2/nKey2) is already START_END, insert a START_END key.
*/
int lsmTreeDelete(
  lsm_db *db,
  void *pKey1, int nKey1,         /* Start of range */
  void *pKey2, int nKey2          /* End of range */
){
  int rc = LSM_OK;
  int bDone = 0;
  TreeRoot *p = &db->treehdr.root;
  TreeBlob blob = {0, 0};

  /* The range must be sensible - that (key1 < key2). */
  assert( treeKeycmp(pKey1, nKey1, pKey2, nKey2)<0 );
  assert( assert_delete_ranges_match(db) );

#if 0
  static int nCall = 0;
  printf("\n");
  nCall++;
  printf("%d delete %s .. %s\n", nCall, (char *)pKey1, (char *)pKey2);
  dump_tree_contents(db, "before delete");
#endif

  /* Step 1. This loop runs until the tree contains no keys within the
  ** range being deleted. Or until an error occurs. */
  while( bDone==0 && rc==LSM_OK ){
    int res;
    TreeCursor csr;               /* Cursor to seek to first key in range */
    void *pDel; int nDel;         /* Key to (possibly) delete this iteration */
#ifndef NDEBUG
    int nEntry = treeCountEntries(db);
#endif

    /* Seek the cursor to the first entry in the tree greater than pKey1. */
    treeCursorInit(db, 0, &csr);
    lsmTreeCursorSeek(&csr, pKey1, nKey1, &res);
    if( res<=0 && lsmTreeCursorValid(&csr) ) lsmTreeCursorNext(&csr);

    /* If there is no such entry, or if it is greater than pKey2, then the
    ** tree now contains no keys in the range being deleted. In this case
    ** break out of the loop.  */
    bDone = 1;
    if( lsmTreeCursorValid(&csr) ){
      lsmTreeCursorKey(&csr, 0, &pDel, &nDel);
      if( treeKeycmp(pDel, nDel, pKey2, nKey2)<0 ) bDone = 0;
    }

    if( bDone==0 ){
      if( csr.iNode==(p->nHeight-1) ){
        /* The element to delete already lies on a leaf node */
        rc = treeDeleteEntry(db, &csr, 0);
      }else{
        /* 1. Overwrite the current key with a copy of the next key in the 
        **    tree (key N).
        **
        ** 2. Seek to key N (cursor will stop at the internal node copy of
        **    N). Move to the next key (original copy of N). Delete
        **    this entry. 
        */
        u32 iKey;
        TreeKey *pKey;
        int iNode = csr.iNode;
        lsmTreeCursorNext(&csr);
        assert( csr.iNode==(p->nHeight-1) );

        iKey = csr.apTreeNode[csr.iNode]->aiKeyPtr[csr.aiCell[csr.iNode]];
        lsmTreeCursorPrev(&csr);

        treeOverwriteKey(db, &csr, iKey, &rc);
        pKey = treeShmkey(db, iKey, TKV_LOADKEY, &blob, &rc);
        if( pKey ){
          rc = lsmTreeCursorSeek(&csr, TKV_KEY(pKey), pKey->nKey, &res);
        }
        if( rc==LSM_OK ){
          assert( res==0 && csr.iNode==iNode );
          rc = lsmTreeCursorNext(&csr);
          if( rc==LSM_OK ){
            rc = treeDeleteEntry(db, &csr, 0);
          }
        }
      }
    }

    /* Clean up any memory allocated by the cursor. */
    tblobFree(db, &csr.blob);
#if 0
    dump_tree_contents(db, "ddd delete");
#endif
    assert( bDone || treeCountEntries(db)==(nEntry-1) );
  }

#if 0
  dump_tree_contents(db, "during delete");
#endif

  /* Now insert the START_DELETE and END_DELETE keys. */
  if( rc==LSM_OK ){
    rc = treeInsertEntry(db, LSM_START_DELETE, pKey1, nKey1, 0, -1);
  }
#if 0
  dump_tree_contents(db, "during delete 2");
#endif
  if( rc==LSM_OK ){
    rc = treeInsertEntry(db, LSM_END_DELETE, pKey2, nKey2, 0, -1);
  }

#if 0
  dump_tree_contents(db, "after delete");
#endif

  tblobFree(db, &blob);
  assert( assert_delete_ranges_match(db) );
  return rc;
}

/*
** Return, in bytes, the amount of memory currently used by the tree 
** structure.
*/
int lsmTreeSize(lsm_db *pDb){
  return pDb->treehdr.root.nByte;
}

/*
** Open a cursor on the in-memory tree pTree.
*/
int lsmTreeCursorNew(lsm_db *pDb, int bOld, TreeCursor **ppCsr){
  TreeCursor *pCsr;
  *ppCsr = pCsr = lsmMalloc(pDb->pEnv, sizeof(TreeCursor));
  if( pCsr ){
    treeCursorInit(pDb, bOld, pCsr);
    return LSM_OK;
  }
  return LSM_NOMEM_BKPT;
}

/*
** Close an in-memory tree cursor.
*/
void lsmTreeCursorDestroy(TreeCursor *pCsr){
  if( pCsr ){
    tblobFree(pCsr->pDb, &pCsr->blob);
    lsmFree(pCsr->pDb->pEnv, pCsr);
  }
}

void lsmTreeCursorReset(TreeCursor *pCsr){
  if( pCsr ){
    pCsr->iNode = -1;
    pCsr->pSave = 0;
  }
}

#ifndef NDEBUG
static int treeCsrCompare(TreeCursor *pCsr, void *pKey, int nKey){
  TreeKey *p;
  int cmp = 0;
  int rc = LSM_OK;
  assert( pCsr->iNode>=0 );
  p = csrGetKey(pCsr, &pCsr->blob, &rc);
  if( p ){
    cmp = treeKeycmp(TKV_KEY(p), p->nKey, pKey, nKey);
  }
  return cmp;
}
#endif


/*
** Attempt to seek the cursor passed as the first argument to key (pKey/nKey)
** in the tree structure. If an exact match for the key is found, leave the
** cursor pointing to it and set *pRes to zero before returning. If an
** exact match cannot be found, do one of the following:
**
**   * Leave the cursor pointing to the smallest element in the tree that 
**     is larger than the key and set *pRes to +1, or
**
**   * Leave the cursor pointing to the largest element in the tree that 
**     is smaller than the key and set *pRes to -1, or
**
**   * If the tree is empty, leave the cursor at EOF and set *pRes to -1.
*/
int lsmTreeCursorSeek(TreeCursor *pCsr, void *pKey, int nKey, int *pRes){
  int rc = LSM_OK;                /* Return code */
  lsm_db *pDb = pCsr->pDb;
  TreeRoot *pRoot = pCsr->pRoot;
  u32 iNodePtr;                   /* Location of current node in search */

  /* Discard any saved position data */
  treeCursorRestore(pCsr, 0);

  iNodePtr = pRoot->iRoot;
  if( iNodePtr==0 ){
    /* Either an error occurred or the tree is completely empty. */
    assert( rc!=LSM_OK || pRoot->iRoot==0 );
    *pRes = -1;
    pCsr->iNode = -1;
  }else{
    TreeBlob b = {0, 0};
    int res = 0;                  /* Result of comparison function */
    int iNode = -1;
    while( iNodePtr ){
      TreeNode *pNode;            /* Node at location iNodePtr */
      int iTest;                  /* Index of second key to test (0 or 2) */
      u32 iTreeKey;
      TreeKey *pTreeKey;          /* Key to compare against */

      pNode = (TreeNode *)treeShmptrUnsafe(pDb, iNodePtr);
      iNode++;
      pCsr->apTreeNode[iNode] = pNode;

      /* Compare (pKey/nKey) with the key in the middle slot of B-tree node
      ** pNode. The middle slot is never empty. If the comparison is a match,
      ** then the search is finished. Break out of the loop. */
      pTreeKey = (TreeKey*)treeShmptrUnsafe(pDb, pNode->aiKeyPtr[1]);
      if( !(pTreeKey->flags & LSM_CONTIGUOUS) ){
        pTreeKey = treeShmkey(pDb, pNode->aiKeyPtr[1], TKV_LOADKEY, &b, &rc);
        if( rc!=LSM_OK ) break;
      }
      res = treeKeycmp((void *)&pTreeKey[1], pTreeKey->nKey, pKey, nKey);
      if( res==0 ){
        pCsr->aiCell[iNode] = 1;
        break;
      }

      /* Based on the results of the previous comparison, compare (pKey/nKey)
      ** to either the left or right key of the B-tree node, if such a key
      ** exists. */
      iTest = (res>0 ? 0 : 2);
      iTreeKey = pNode->aiKeyPtr[iTest];
      if( iTreeKey ){
        pTreeKey = (TreeKey*)treeShmptrUnsafe(pDb, iTreeKey);
        if( !(pTreeKey->flags & LSM_CONTIGUOUS) ){
          pTreeKey = treeShmkey(pDb, iTreeKey, TKV_LOADKEY, &b, &rc);
          if( rc ) break;
        }
        res = treeKeycmp((void *)&pTreeKey[1], pTreeKey->nKey, pKey, nKey);
        if( res==0 ){
          pCsr->aiCell[iNode] = iTest;
          break;
        }
      }else{
        iTest = 1;
      }

      if( iNode<(pRoot->nHeight-1) ){
        iNodePtr = getChildPtr(pNode, pRoot->iTransId, iTest + (res<0));
      }else{
        iNodePtr = 0;
      }
      pCsr->aiCell[iNode] = iTest + (iNodePtr && (res<0));
    }

    *pRes = res;
    pCsr->iNode = iNode;
    tblobFree(pDb, &b);
  }

  /* assert() that *pRes has been set properly */
#ifndef NDEBUG
  if( rc==LSM_OK && lsmTreeCursorValid(pCsr) ){
    int cmp = treeCsrCompare(pCsr, pKey, nKey);
    assert( *pRes==cmp || (*pRes ^ cmp)>0 );
  }
#endif

  return rc;
}

int lsmTreeCursorNext(TreeCursor *pCsr){
#ifndef NDEBUG
  TreeKey *pK1;
  TreeBlob key1 = {0, 0};
#endif
  lsm_db *pDb = pCsr->pDb;
  TreeRoot *pRoot = pCsr->pRoot;
  const int iLeaf = pRoot->nHeight-1;
  int iCell; 
  int rc = LSM_OK; 
  TreeNode *pNode; 

  /* Restore the cursor position, if required */
  int iRestore = 0;
  treeCursorRestore(pCsr, &iRestore);
  if( iRestore>0 ) return LSM_OK;

  /* Save a pointer to the current key. This is used in an assert() at the
  ** end of this function - to check that the 'next' key really is larger
  ** than the current key. */
#ifndef NDEBUG
  pK1 = csrGetKey(pCsr, &key1, &rc);
  if( rc!=LSM_OK ) return rc;
#endif

  assert( lsmTreeCursorValid(pCsr) );
  assert( pCsr->aiCell[pCsr->iNode]<3 );

  pNode = pCsr->apTreeNode[pCsr->iNode];
  iCell = ++pCsr->aiCell[pCsr->iNode];

  /* If the current node is not a leaf, and the current cell has sub-tree
  ** associated with it, descend to the left-most key on the left-most
  ** leaf of the sub-tree.  */
  if( pCsr->iNode<iLeaf && getChildPtr(pNode, pRoot->iTransId, iCell) ){
    do {
      u32 iNodePtr;
      pCsr->iNode++;
      iNodePtr = getChildPtr(pNode, pRoot->iTransId, iCell);
      pNode = (TreeNode *)treeShmptr(pDb, iNodePtr);
      pCsr->apTreeNode[pCsr->iNode] = pNode;
      iCell = pCsr->aiCell[pCsr->iNode] = (pNode->aiKeyPtr[0]==0);
    }while( pCsr->iNode < iLeaf );
  }

  /* Otherwise, the next key is found by following pointer up the tree 
  ** until there is a key immediately to the right of the pointer followed 
  ** to reach the sub-tree containing the current key. */
  else if( iCell>=3 || pNode->aiKeyPtr[iCell]==0 ){
    while( (--pCsr->iNode)>=0 ){
      iCell = pCsr->aiCell[pCsr->iNode];
      if( iCell<3 && pCsr->apTreeNode[pCsr->iNode]->aiKeyPtr[iCell] ) break;
    }
  }

#ifndef NDEBUG
  if( pCsr->iNode>=0 ){
    TreeKey *pK2 = csrGetKey(pCsr, &pCsr->blob, &rc);
    assert( rc||treeKeycmp(TKV_KEY(pK2),pK2->nKey,TKV_KEY(pK1),pK1->nKey)>=0 );
  }
  tblobFree(pDb, &key1);
#endif

  return rc;
}

int lsmTreeCursorPrev(TreeCursor *pCsr){
#ifndef NDEBUG
  TreeKey *pK1;
  TreeBlob key1 = {0, 0};
#endif
  lsm_db *pDb = pCsr->pDb;
  TreeRoot *pRoot = pCsr->pRoot;
  const int iLeaf = pRoot->nHeight-1;
  int iCell; 
  int rc = LSM_OK; 
  TreeNode *pNode; 

  /* Restore the cursor position, if required */
  int iRestore = 0;
  treeCursorRestore(pCsr, &iRestore);
  if( iRestore<0 ) return LSM_OK;

  /* Save a pointer to the current key. This is used in an assert() at the
  ** end of this function - to check that the 'next' key really is smaller
  ** than the current key. */
#ifndef NDEBUG
  pK1 = csrGetKey(pCsr, &key1, &rc);
  if( rc!=LSM_OK ) return rc;
#endif

  assert( lsmTreeCursorValid(pCsr) );
  pNode = pCsr->apTreeNode[pCsr->iNode];
  iCell = pCsr->aiCell[pCsr->iNode];
  assert( iCell>=0 && iCell<3 );

  /* If the current node is not a leaf, and the current cell has sub-tree
  ** associated with it, descend to the right-most key on the right-most
  ** leaf of the sub-tree.  */
  if( pCsr->iNode<iLeaf && getChildPtr(pNode, pRoot->iTransId, iCell) ){
    do {
      u32 iNodePtr;
      pCsr->iNode++;
      iNodePtr = getChildPtr(pNode, pRoot->iTransId, iCell);
      pNode = (TreeNode *)treeShmptr(pDb, iNodePtr);
      if( rc!=LSM_OK ) break;
      pCsr->apTreeNode[pCsr->iNode] = pNode;
      iCell = 1 + (pNode->aiKeyPtr[2]!=0) + (pCsr->iNode < iLeaf);
      pCsr->aiCell[pCsr->iNode] = iCell;
    }while( pCsr->iNode < iLeaf );
  }

  /* Otherwise, the next key is found by following pointer up the tree until
  ** there is a key immediately to the left of the pointer followed to reach
  ** the sub-tree containing the current key. */
  else{
    do {
      iCell = pCsr->aiCell[pCsr->iNode]-1;
      if( iCell>=0 && pCsr->apTreeNode[pCsr->iNode]->aiKeyPtr[iCell] ) break;
    }while( (--pCsr->iNode)>=0 );
    pCsr->aiCell[pCsr->iNode] = iCell;
  }

#ifndef NDEBUG
  if( pCsr->iNode>=0 ){
    TreeKey *pK2 = csrGetKey(pCsr, &pCsr->blob, &rc);
    assert( rc || treeKeycmp(TKV_KEY(pK2),pK2->nKey,TKV_KEY(pK1),pK1->nKey)<0 );
  }
  tblobFree(pDb, &key1);
#endif

  return rc;
}

/*
** Move the cursor to the first (bLast==0) or last (bLast!=0) entry in the
** in-memory tree.
*/
int lsmTreeCursorEnd(TreeCursor *pCsr, int bLast){
  lsm_db *pDb = pCsr->pDb;
  TreeRoot *pRoot = pCsr->pRoot;
  int rc = LSM_OK;

  u32 iNodePtr;
  pCsr->iNode = -1;

  /* Discard any saved position data */
  treeCursorRestore(pCsr, 0);

  iNodePtr = pRoot->iRoot;
  while( iNodePtr ){
    int iCell;
    TreeNode *pNode;

    pNode = (TreeNode *)treeShmptr(pDb, iNodePtr);
    if( rc ) break;

    if( bLast ){
      iCell = ((pNode->aiKeyPtr[2]==0) ? 2 : 3);
    }else{
      iCell = ((pNode->aiKeyPtr[0]==0) ? 1 : 0);
    }
    pCsr->iNode++;
    pCsr->apTreeNode[pCsr->iNode] = pNode;

    if( pCsr->iNode<pRoot->nHeight-1 ){
      iNodePtr = getChildPtr(pNode, pRoot->iTransId, iCell);
    }else{
      iNodePtr = 0;
    }
    pCsr->aiCell[pCsr->iNode] = iCell - (iNodePtr==0 && bLast);
  }

  return rc;
}

int lsmTreeCursorFlags(TreeCursor *pCsr){
  int flags = 0;
  if( pCsr && pCsr->iNode>=0 ){
    int rc = LSM_OK;
    TreeKey *pKey = (TreeKey *)treeShmptrUnsafe(pCsr->pDb,
        pCsr->apTreeNode[pCsr->iNode]->aiKeyPtr[pCsr->aiCell[pCsr->iNode]]
    );
    assert( rc==LSM_OK );
    flags = (pKey->flags & ~LSM_CONTIGUOUS);
  }
  return flags;
}

int lsmTreeCursorKey(TreeCursor *pCsr, int *pFlags, void **ppKey, int *pnKey){
  TreeKey *pTreeKey;
  int rc = LSM_OK;

  assert( lsmTreeCursorValid(pCsr) );

  pTreeKey = pCsr->pSave;
  if( !pTreeKey ){
    pTreeKey = csrGetKey(pCsr, &pCsr->blob, &rc);
  }
  if( rc==LSM_OK ){
    *pnKey = pTreeKey->nKey;
    if( pFlags ) *pFlags = pTreeKey->flags;
    *ppKey = (void *)&pTreeKey[1];
  }

  return rc;
}

int lsmTreeCursorValue(TreeCursor *pCsr, void **ppVal, int *pnVal){
  int res = 0;
  int rc;

  rc = treeCursorRestore(pCsr, &res);
  if( res==0 ){
    TreeKey *pTreeKey = csrGetKey(pCsr, &pCsr->blob, &rc);
    if( rc==LSM_OK ){
      if( pTreeKey->flags & LSM_INSERT ){
        *pnVal = pTreeKey->nValue;
        *ppVal = TKV_VAL(pTreeKey);
      }else{
        *ppVal = 0;
        *pnVal = -1;
      }
    }
  }else{
    *ppVal = 0;
    *pnVal = 0;
  }

  return rc;
}

/*
** Return true if the cursor currently points to a valid entry. 
*/
int lsmTreeCursorValid(TreeCursor *pCsr){
  return (pCsr && (pCsr->pSave || pCsr->iNode>=0));
}

/*
** Store a mark in *pMark. Later on, a call to lsmTreeRollback() with a
** pointer to the same TreeMark structure may be used to roll the tree
** contents back to their current state.
*/
void lsmTreeMark(lsm_db *pDb, TreeMark *pMark){
  pMark->iRoot = pDb->treehdr.root.iRoot;
  pMark->nHeight = pDb->treehdr.root.nHeight;
  pMark->iWrite = pDb->treehdr.iWrite;
  pMark->nChunk = pDb->treehdr.nChunk;
  pMark->iNextShmid = pDb->treehdr.iNextShmid;
  pMark->iRollback = intArraySize(&pDb->rollback);
}

/*
** Roll back to mark pMark. Structure *pMark should have been previously
** populated by a call to lsmTreeMark().
*/
void lsmTreeRollback(lsm_db *pDb, TreeMark *pMark){
  int iIdx;
  int nIdx;
  u32 iNext;
  ShmChunk *pChunk;
  u32 iChunk;
  u32 iShmid;

  /* Revert all required v2 pointers. */
  nIdx = intArraySize(&pDb->rollback);
  for(iIdx = pMark->iRollback; iIdx<nIdx; iIdx++){
    TreeNode *pNode;
    pNode = treeShmptr(pDb, intArrayEntry(&pDb->rollback, iIdx));
    assert( pNode );
    pNode->iV2 = 0;
    pNode->iV2Child = 0;
    pNode->iV2Ptr = 0;
  }
  intArrayTruncate(&pDb->rollback, pMark->iRollback);

  /* Restore the free-chunk list. */
  assert( pMark->iWrite!=0 );
  iChunk = treeOffsetToChunk(pMark->iWrite-1);
  pChunk = treeShmChunk(pDb, iChunk);
  iNext = pChunk->iNext;
  pChunk->iNext = 0;

  pChunk = treeShmChunk(pDb, pDb->treehdr.iFirst);
  iShmid = pChunk->iShmid-1;

  while( iNext ){
    u32 iFree = iNext;            /* Current chunk being rollback-freed */
    ShmChunk *pFree;              /* Pointer to chunk iFree */

    pFree = treeShmChunk(pDb, iFree);
    iNext = pFree->iNext;

    if( iFree<pMark->nChunk ){
      pFree->iNext = pDb->treehdr.iFirst;
      pFree->iShmid = iShmid--;
      pDb->treehdr.iFirst = iFree;
    }
  }

  /* Restore the tree-header fields */
  pDb->treehdr.root.iRoot = pMark->iRoot;
  pDb->treehdr.root.nHeight = pMark->nHeight;
  pDb->treehdr.iWrite = pMark->iWrite;
  pDb->treehdr.nChunk = pMark->nChunk;
  pDb->treehdr.iNextShmid = pMark->iNextShmid;
}

/*
** Load the in-memory tree header from shared-memory into pDb->treehdr.
** If the header cannot be loaded, return LSM_PROTOCOL.
**
** If the header is successfully loaded and parameter piRead is not NULL,
** is is set to 1 if the header was loaded from ShmHeader.hdr1, or 2 if
** the header was loaded from ShmHeader.hdr2.
*/
int lsmTreeLoadHeader(lsm_db *pDb, int *piRead){
  int nRem = LSM_ATTEMPTS_BEFORE_PROTOCOL;
  while( (nRem--)>0 ){
    ShmHeader *pShm = pDb->pShmhdr;

    memcpy(&pDb->treehdr, &pShm->hdr1, sizeof(TreeHeader));
    if( treeHeaderChecksumOk(&pDb->treehdr) ){
      if( piRead ) *piRead = 1;
      return LSM_OK;
    }
    memcpy(&pDb->treehdr, &pShm->hdr2, sizeof(TreeHeader));
    if( treeHeaderChecksumOk(&pDb->treehdr) ){
      if( piRead ) *piRead = 2;
      return LSM_OK;
    }

    lsmShmBarrier(pDb);
  }
  return LSM_PROTOCOL_BKPT;
}

int lsmTreeLoadHeaderOk(lsm_db *pDb, int iRead){
  TreeHeader *p = (iRead==1) ? &pDb->pShmhdr->hdr1 : &pDb->pShmhdr->hdr2;
  assert( iRead==1 || iRead==2 );
  return (0==memcmp(pDb->treehdr.aCksum, p->aCksum, sizeof(u32)*2));
}

/*
** This function is called to conclude a transaction. If argument bCommit
** is true, the transaction is committed. Otherwise it is rolled back.
*/
int lsmTreeEndTransaction(lsm_db *pDb, int bCommit){
  ShmHeader *pShm = pDb->pShmhdr;

  treeHeaderChecksum(&pDb->treehdr, pDb->treehdr.aCksum);
  memcpy(&pShm->hdr2, &pDb->treehdr, sizeof(TreeHeader));
  lsmShmBarrier(pDb);
  memcpy(&pShm->hdr1, &pDb->treehdr, sizeof(TreeHeader));
  pShm->bWriter = 0;
  intArrayFree(pDb->pEnv, &pDb->rollback);

  return LSM_OK;
}

#ifndef NDEBUG
static int assert_delete_ranges_match(lsm_db *db){
  int prev = 0;
  TreeBlob blob = {0, 0};
  TreeCursor csr;               /* Cursor used to iterate through tree */
  int rc;

  treeCursorInit(db, 0, &csr);
  for( rc = lsmTreeCursorEnd(&csr, 0);
       rc==LSM_OK && lsmTreeCursorValid(&csr);
       rc = lsmTreeCursorNext(&csr)
  ){
    TreeKey *pKey = csrGetKey(&csr, &blob, &rc);
    if( rc!=LSM_OK ) break;
    assert( ((prev&LSM_START_DELETE)==0)==((pKey->flags&LSM_END_DELETE)==0) );
    prev = pKey->flags;
  }

  tblobFree(csr.pDb, &csr.blob);
  tblobFree(csr.pDb, &blob);

  return 1;
}

static int treeCountEntries(lsm_db *db){
  TreeCursor csr;               /* Cursor used to iterate through tree */
  int rc;
  int nEntry = 0;

  treeCursorInit(db, 0, &csr);
  for( rc = lsmTreeCursorEnd(&csr, 0);
       rc==LSM_OK && lsmTreeCursorValid(&csr);
       rc = lsmTreeCursorNext(&csr)
  ){
    nEntry++;
  }

  tblobFree(csr.pDb, &csr.blob);

  return nEntry;
}
#endif

Added ext/lsm1/lsm_unix.c.





















































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
/*
** 2011-12-03
**
** The author disclaims copyright to this source code.  In place of
** a legal notice, here is a blessing:
**
**    May you do good and not evil.
**    May you find forgiveness for yourself and forgive others.
**    May you share freely, never taking more than you give.
**
*************************************************************************
**
** Unix-specific run-time environment implementation for LSM.
*/

#ifndef _WIN32

#if defined(__GNUC__) || defined(__TINYC__)
/* workaround for ftruncate() visibility on gcc. */
# ifndef _XOPEN_SOURCE
#  define _XOPEN_SOURCE 500
# endif
#endif

#include <unistd.h>
#include <sys/types.h>

#include <sys/stat.h>
#include <fcntl.h>
#include <assert.h>
#include <string.h>

#include <stdlib.h>
#include <stdarg.h>
#include <stdio.h>
#include <ctype.h>

#include <unistd.h>
#include <errno.h>

#include <sys/mman.h>
#include "lsmInt.h"

/* There is no fdatasync() call on Android */
#ifdef __ANDROID__
# define fdatasync(x) fsync(x)
#endif

/*
** An open file is an instance of the following object
*/
typedef struct PosixFile PosixFile;
struct PosixFile {
  lsm_env *pEnv;                  /* The run-time environment */
  const char *zName;              /* Full path to file */
  int fd;                         /* The open file descriptor */
  int shmfd;                      /* Shared memory file-descriptor */
  void *pMap;                     /* Pointer to mapping of file fd */
  off_t nMap;                     /* Size of mapping at pMap in bytes */
  int nShm;                       /* Number of entries in array apShm[] */
  void **apShm;                   /* Array of 32K shared memory segments */
};

static char *posixShmFile(PosixFile *p){
  char *zShm;
  int nName = strlen(p->zName);
  zShm = (char *)lsmMalloc(p->pEnv, nName+4+1);
  if( zShm ){
    memcpy(zShm, p->zName, nName);
    memcpy(&zShm[nName], "-shm", 5);
  }
  return zShm;
}

static int lsmPosixOsOpen(
  lsm_env *pEnv,
  const char *zFile,
  int flags,
  lsm_file **ppFile
){
  int rc = LSM_OK;
  PosixFile *p;

  p = lsm_malloc(pEnv, sizeof(PosixFile));
  if( p==0 ){
    rc = LSM_NOMEM;
  }else{
    int bReadonly = (flags & LSM_OPEN_READONLY);
    int oflags = (bReadonly ? O_RDONLY : (O_RDWR|O_CREAT));
    memset(p, 0, sizeof(PosixFile));
    p->zName = zFile;
    p->pEnv = pEnv;
    p->fd = open(zFile, oflags, 0644);
    if( p->fd<0 ){
      lsm_free(pEnv, p);
      p = 0;
      if( errno==ENOENT ){
        rc = lsmErrorBkpt(LSM_IOERR_NOENT);
      }else{
        rc = LSM_IOERR_BKPT;
      }
    }
  }

  *ppFile = (lsm_file *)p;
  return rc;
}

static int lsmPosixOsWrite(
  lsm_file *pFile,                /* File to write to */
  lsm_i64 iOff,                   /* Offset to write to */
  void *pData,                    /* Write data from this buffer */
  int nData                       /* Bytes of data to write */
){
  int rc = LSM_OK;
  PosixFile *p = (PosixFile *)pFile;
  off_t offset;

  offset = lseek(p->fd, (off_t)iOff, SEEK_SET);
  if( offset!=iOff ){
    rc = LSM_IOERR_BKPT;
  }else{
    ssize_t prc = write(p->fd, pData, (size_t)nData);
    if( prc<0 ) rc = LSM_IOERR_BKPT;
  }

  return rc;
}

static int lsmPosixOsTruncate(
  lsm_file *pFile,                /* File to write to */
  lsm_i64 nSize                   /* Size to truncate file to */
){
  PosixFile *p = (PosixFile *)pFile;
  int rc = LSM_OK;                /* Return code */
  int prc;                        /* Posix Return Code */
  struct stat sStat;              /* Result of fstat() invocation */
  
  prc = fstat(p->fd, &sStat);
  if( prc==0 && sStat.st_size>nSize ){
    prc = ftruncate(p->fd, (off_t)nSize);
  }
  if( prc<0 ) rc = LSM_IOERR_BKPT;

  return rc;
}

static int lsmPosixOsRead(
  lsm_file *pFile,                /* File to read from */
  lsm_i64 iOff,                   /* Offset to read from */
  void *pData,                    /* Read data into this buffer */
  int nData                       /* Bytes of data to read */
){
  int rc = LSM_OK;
  PosixFile *p = (PosixFile *)pFile;
  off_t offset;

  offset = lseek(p->fd, (off_t)iOff, SEEK_SET);
  if( offset!=iOff ){
    rc = LSM_IOERR_BKPT;
  }else{
    ssize_t prc = read(p->fd, pData, (size_t)nData);
    if( prc<0 ){ 
      rc = LSM_IOERR_BKPT;
    }else if( prc<nData ){
      memset(&((u8 *)pData)[prc], 0, nData - prc);
    }

  }

  return rc;
}

static int lsmPosixOsSync(lsm_file *pFile){
  int rc = LSM_OK;

#ifndef LSM_NO_SYNC
  PosixFile *p = (PosixFile *)pFile;
  int prc = 0;

  if( p->pMap ){
    prc = msync(p->pMap, p->nMap, MS_SYNC);
  }
  if( prc==0 ) prc = fdatasync(p->fd);
  if( prc<0 ) rc = LSM_IOERR_BKPT;
#else
  (void)pFile;
#endif

  return rc;
}

static int lsmPosixOsSectorSize(lsm_file *pFile){
  return 512;
}

static int lsmPosixOsRemap(
  lsm_file *pFile, 
  lsm_i64 iMin, 
  void **ppOut,
  lsm_i64 *pnOut
){
  off_t iSz;
  int prc;
  PosixFile *p = (PosixFile *)pFile;
  struct stat buf;

  /* If the file is between 0 and 2MB in size, extend it in chunks of 256K.
  ** Thereafter, in chunks of 1MB at a time.  */
  const int aIncrSz[] = {256*1024, 1024*1024};
  int nIncrSz = aIncrSz[iMin>(2*1024*1024)];

  if( p->pMap ){
    munmap(p->pMap, p->nMap);
    *ppOut = p->pMap = 0;
    *pnOut = p->nMap = 0;
  }

  if( iMin>=0 ){
    memset(&buf, 0, sizeof(buf));
    prc = fstat(p->fd, &buf);
    if( prc!=0 ) return LSM_IOERR_BKPT;
    iSz = buf.st_size;
    if( iSz<iMin ){
      iSz = ((iMin + nIncrSz-1) / nIncrSz) * nIncrSz;
      prc = ftruncate(p->fd, iSz);
      if( prc!=0 ) return LSM_IOERR_BKPT;
    }

    p->pMap = mmap(0, iSz, PROT_READ|PROT_WRITE, MAP_SHARED, p->fd, 0);
    p->nMap = iSz;
  }

  *ppOut = p->pMap;
  *pnOut = p->nMap;
  return LSM_OK;
}

static int lsmPosixOsFullpath(
  lsm_env *pEnv,
  const char *zName,
  char *zOut,
  int *pnOut
){
  int nBuf = *pnOut;
  int nReq;

  if( zName[0]!='/' ){
    char *z;
    char *zTmp;
    int nTmp = 512;
    zTmp = lsmMalloc(pEnv, nTmp);
    while( zTmp ){
      z = getcwd(zTmp, nTmp);
      if( z || errno!=ERANGE ) break;
      nTmp = nTmp*2;
      zTmp = lsmReallocOrFree(pEnv, zTmp, nTmp);
    }
    if( zTmp==0 ) return LSM_NOMEM_BKPT;
    if( z==0 ) return LSM_IOERR_BKPT;
    assert( z==zTmp );

    nTmp = strlen(zTmp);
    nReq = nTmp + 1 + strlen(zName) + 1;
    if( nReq<=nBuf ){
      memcpy(zOut, zTmp, nTmp);
      zOut[nTmp] = '/';
      memcpy(&zOut[nTmp+1], zName, strlen(zName)+1);
    }
    lsmFree(pEnv, zTmp);
  }else{
    nReq = strlen(zName)+1;
    if( nReq<=nBuf ){
      memcpy(zOut, zName, strlen(zName)+1);
    }
  }

  *pnOut = nReq;
  return LSM_OK;
}

static int lsmPosixOsFileid(
  lsm_file *pFile, 
  void *pBuf,
  int *pnBuf
){
  int prc;
  int nBuf;
  int nReq;
  PosixFile *p = (PosixFile *)pFile;
  struct stat buf;

  nBuf = *pnBuf;
  nReq = (sizeof(buf.st_dev) + sizeof(buf.st_ino));
  *pnBuf = nReq;
  if( nReq>nBuf ) return LSM_OK;

  memset(&buf, 0, sizeof(buf));
  prc = fstat(p->fd, &buf);
  if( prc!=0 ) return LSM_IOERR_BKPT;

  memcpy(pBuf, &buf.st_dev, sizeof(buf.st_dev));
  memcpy(&(((u8 *)pBuf)[sizeof(buf.st_dev)]), &buf.st_ino, sizeof(buf.st_ino));
  return LSM_OK;
}

static int lsmPosixOsUnlink(lsm_env *pEnv, const char *zFile){
  int prc = unlink(zFile);
  return prc ? LSM_IOERR_BKPT : LSM_OK;
}

int lsmPosixOsLock(lsm_file *pFile, int iLock, int eType){
  int rc = LSM_OK;
  PosixFile *p = (PosixFile *)pFile;
  static const short aType[3] = { F_UNLCK, F_RDLCK, F_WRLCK };
  struct flock lock;

  assert( aType[LSM_LOCK_UNLOCK]==F_UNLCK );
  assert( aType[LSM_LOCK_SHARED]==F_RDLCK );
  assert( aType[LSM_LOCK_EXCL]==F_WRLCK );
  assert( eType>=0 && eType<array_size(aType) );
  assert( iLock>0 && iLock<=32 );

  memset(&lock, 0, sizeof(lock));
  lock.l_whence = SEEK_SET;
  lock.l_len = 1;
  lock.l_type = aType[eType];
  lock.l_start = (4096-iLock);

  if( fcntl(p->fd, F_SETLK, &lock) ){
    int e = errno;
    if( e==EACCES || e==EAGAIN ){
      rc = LSM_BUSY;
    }else{
      rc = LSM_IOERR_BKPT;
    }
  }

  return rc;
}

int lsmPosixOsTestLock(lsm_file *pFile, int iLock, int nLock, int eType){
  int rc = LSM_OK;
  PosixFile *p = (PosixFile *)pFile;
  static const short aType[3] = { 0, F_RDLCK, F_WRLCK };
  struct flock lock;

  assert( eType==LSM_LOCK_SHARED || eType==LSM_LOCK_EXCL );
  assert( aType[LSM_LOCK_SHARED]==F_RDLCK );
  assert( aType[LSM_LOCK_EXCL]==F_WRLCK );
  assert( eType>=0 && eType<array_size(aType) );
  assert( iLock>0 && iLock<=32 );

  memset(&lock, 0, sizeof(lock));
  lock.l_whence = SEEK_SET;
  lock.l_len = nLock;
  lock.l_type = aType[eType];
  lock.l_start = (4096-iLock);

  if( fcntl(p->fd, F_GETLK, &lock) ){
    rc = LSM_IOERR_BKPT;
  }else if( lock.l_type!=F_UNLCK ){
    rc = LSM_BUSY;
  }

  return rc;
}

int lsmPosixOsShmMap(lsm_file *pFile, int iChunk, int sz, void **ppShm){
  PosixFile *p = (PosixFile *)pFile;

  *ppShm = 0;
  assert( sz==LSM_SHM_CHUNK_SIZE );
  if( iChunk>=p->nShm ){
    int i;
    void **apNew;
    int nNew = iChunk+1;
    off_t nReq = nNew * LSM_SHM_CHUNK_SIZE;
    struct stat sStat;

    /* If the shared-memory file has not been opened, open it now. */
    if( p->shmfd<=0 ){
      char *zShm = posixShmFile(p);
      if( !zShm ) return LSM_NOMEM_BKPT;
      p->shmfd = open(zShm, O_RDWR|O_CREAT, 0644);
      lsmFree(p->pEnv, zShm);
      if( p->shmfd<0 ){ 
        return LSM_IOERR_BKPT;
      }
    }

    /* If the shared-memory file is not large enough to contain the 
    ** requested chunk, cause it to grow.  */
    if( fstat(p->shmfd, &sStat) ){
      return LSM_IOERR_BKPT;
    }
    if( sStat.st_size<nReq ){
      if( ftruncate(p->shmfd, nReq) ){
        return LSM_IOERR_BKPT;
      }
    }

    apNew = (void **)lsmRealloc(p->pEnv, p->apShm, sizeof(void *) * nNew);
    if( !apNew ) return LSM_NOMEM_BKPT;
    for(i=p->nShm; i<nNew; i++){
      apNew[i] = 0;
    }
    p->apShm = apNew;
    p->nShm = nNew;
  }

  if( p->apShm[iChunk]==0 ){
    p->apShm[iChunk] = mmap(0, LSM_SHM_CHUNK_SIZE, 
        PROT_READ|PROT_WRITE, MAP_SHARED, p->shmfd, iChunk*LSM_SHM_CHUNK_SIZE
    );
    if( p->apShm[iChunk]==0 ) return LSM_IOERR_BKPT;
  }

  *ppShm = p->apShm[iChunk];
  return LSM_OK;
}

void lsmPosixOsShmBarrier(void){
}

int lsmPosixOsShmUnmap(lsm_file *pFile, int bDelete){
  PosixFile *p = (PosixFile *)pFile;
  if( p->shmfd>0 ){
    int i;
    for(i=0; i<p->nShm; i++){
      if( p->apShm[i] ){
        munmap(p->apShm[i], LSM_SHM_CHUNK_SIZE);
        p->apShm[i] = 0;
      }
    }
    close(p->shmfd);
    p->shmfd = 0;
    if( bDelete ){
      char *zShm = posixShmFile(p);
      if( zShm ) unlink(zShm);
      lsmFree(p->pEnv, zShm);
    }
  }
  return LSM_OK;
}


static int lsmPosixOsClose(lsm_file *pFile){
   PosixFile *p = (PosixFile *)pFile;
   lsmPosixOsShmUnmap(pFile, 0);
   if( p->pMap ) munmap(p->pMap, p->nMap);
   close(p->fd);
   lsm_free(p->pEnv, p->apShm);
   lsm_free(p->pEnv, p);
   return LSM_OK;
}

static int lsmPosixOsSleep(lsm_env *pEnv, int us){
#if 0
  /* Apparently on Android usleep() returns void */
  if( usleep(us) ) return LSM_IOERR;
#endif
  usleep(us);
  return LSM_OK;
}

/****************************************************************************
** Memory allocation routines.
*/
#define BLOCK_HDR_SIZE ROUND8( sizeof(size_t) )

static void *lsmPosixOsMalloc(lsm_env *pEnv, size_t N){
  unsigned char * m;
  N += BLOCK_HDR_SIZE;
  m = (unsigned char *)malloc(N);
  *((size_t*)m) = N;
  return m + BLOCK_HDR_SIZE;
}

static void lsmPosixOsFree(lsm_env *pEnv, void *p){
  if(p){
    free( ((unsigned char *)p) - BLOCK_HDR_SIZE );
  }
}

static void *lsmPosixOsRealloc(lsm_env *pEnv, void *p, size_t N){
  unsigned char * m = (unsigned char *)p;
  if(1>N){
    lsmPosixOsFree( pEnv, p );
    return NULL;
  }else if(NULL==p){
    return lsmPosixOsMalloc(pEnv, N);
  }else{
    void * re = NULL;
    m -= BLOCK_HDR_SIZE;
#if 0 /* arguable: don't shrink */
    size_t * sz = (size_t*)m;
    if(*sz >= (size_t)N){
      return p;
    }
#endif
    re = realloc( m, N + BLOCK_HDR_SIZE );
    if(re){
      m = (unsigned char *)re;
      *((size_t*)m) = N;
      return m + BLOCK_HDR_SIZE;
    }else{
      return NULL;
    }
  }
}

static size_t lsmPosixOsMSize(lsm_env *pEnv, void *p){
  unsigned char * m = (unsigned char *)p;
  return *((size_t*)(m-BLOCK_HDR_SIZE));
}
#undef BLOCK_HDR_SIZE


#ifdef LSM_MUTEX_PTHREADS 
/*************************************************************************
** Mutex methods for pthreads based systems.  If LSM_MUTEX_PTHREADS is
** missing then a no-op implementation of mutexes found in lsm_mutex.c
** will be used instead.
*/
#include <pthread.h>

typedef struct PthreadMutex PthreadMutex;
struct PthreadMutex {
  lsm_env *pEnv;
  pthread_mutex_t mutex;
#ifdef LSM_DEBUG
  pthread_t owner;
#endif
};

#ifdef LSM_DEBUG
# define LSM_PTHREAD_STATIC_MUTEX { 0, PTHREAD_MUTEX_INITIALIZER, 0 }
#else
# define LSM_PTHREAD_STATIC_MUTEX { 0, PTHREAD_MUTEX_INITIALIZER }
#endif

static int lsmPosixOsMutexStatic(
  lsm_env *pEnv,
  int iMutex,
  lsm_mutex **ppStatic
){
  static PthreadMutex sMutex[2] = {
    LSM_PTHREAD_STATIC_MUTEX,
    LSM_PTHREAD_STATIC_MUTEX
  };

  assert( iMutex==LSM_MUTEX_GLOBAL || iMutex==LSM_MUTEX_HEAP );
  assert( LSM_MUTEX_GLOBAL==1 && LSM_MUTEX_HEAP==2 );

  *ppStatic = (lsm_mutex *)&sMutex[iMutex-1];
  return LSM_OK;
}

static int lsmPosixOsMutexNew(lsm_env *pEnv, lsm_mutex **ppNew){
  PthreadMutex *pMutex;           /* Pointer to new mutex */
  pthread_mutexattr_t attr;       /* Attributes object */

  pMutex = (PthreadMutex *)lsmMallocZero(pEnv, sizeof(PthreadMutex));
  if( !pMutex ) return LSM_NOMEM_BKPT;

  pMutex->pEnv = pEnv;
  pthread_mutexattr_init(&attr);
  pthread_mutexattr_settype(&attr, PTHREAD_MUTEX_RECURSIVE);
  pthread_mutex_init(&pMutex->mutex, &attr);
  pthread_mutexattr_destroy(&attr);

  *ppNew = (lsm_mutex *)pMutex;
  return LSM_OK;
}

static void lsmPosixOsMutexDel(lsm_mutex *p){
  PthreadMutex *pMutex = (PthreadMutex *)p;
  pthread_mutex_destroy(&pMutex->mutex);
  lsmFree(pMutex->pEnv, pMutex);
}

static void lsmPosixOsMutexEnter(lsm_mutex *p){
  PthreadMutex *pMutex = (PthreadMutex *)p;
  pthread_mutex_lock(&pMutex->mutex);

#ifdef LSM_DEBUG
  assert( !pthread_equal(pMutex->owner, pthread_self()) );
  pMutex->owner = pthread_self();
  assert( pthread_equal(pMutex->owner, pthread_self()) );
#endif
}

static int lsmPosixOsMutexTry(lsm_mutex *p){
  int ret;
  PthreadMutex *pMutex = (PthreadMutex *)p;
  ret = pthread_mutex_trylock(&pMutex->mutex);
#ifdef LSM_DEBUG
  if( ret==0 ){
    assert( !pthread_equal(pMutex->owner, pthread_self()) );
    pMutex->owner = pthread_self();
    assert( pthread_equal(pMutex->owner, pthread_self()) );
  }
#endif
  return ret;
}

static void lsmPosixOsMutexLeave(lsm_mutex *p){
  PthreadMutex *pMutex = (PthreadMutex *)p;
#ifdef LSM_DEBUG
  assert( pthread_equal(pMutex->owner, pthread_self()) );
  pMutex->owner = 0;
  assert( !pthread_equal(pMutex->owner, pthread_self()) );
#endif
  pthread_mutex_unlock(&pMutex->mutex);
}

#ifdef LSM_DEBUG
static int lsmPosixOsMutexHeld(lsm_mutex *p){
  PthreadMutex *pMutex = (PthreadMutex *)p;
  return pMutex ? pthread_equal(pMutex->owner, pthread_self()) : 1;
}
static int lsmPosixOsMutexNotHeld(lsm_mutex *p){
  PthreadMutex *pMutex = (PthreadMutex *)p;
  return pMutex ? !pthread_equal(pMutex->owner, pthread_self()) : 1;
}
#endif
/*
** End of pthreads mutex implementation.
*************************************************************************/
#else
/*************************************************************************
** Noop mutex implementation
*/
typedef struct NoopMutex NoopMutex;
struct NoopMutex {
  lsm_env *pEnv;                  /* Environment handle (for xFree()) */
  int bHeld;                      /* True if mutex is held */
  int bStatic;                    /* True for a static mutex */
};
static NoopMutex aStaticNoopMutex[2] = {
  {0, 0, 1},
  {0, 0, 1},
};

static int lsmPosixOsMutexStatic(
  lsm_env *pEnv,
  int iMutex,
  lsm_mutex **ppStatic
){
  assert( iMutex>=1 && iMutex<=(int)array_size(aStaticNoopMutex) );
  *ppStatic = (lsm_mutex *)&aStaticNoopMutex[iMutex-1];
  return LSM_OK;
}
static int lsmPosixOsMutexNew(lsm_env *pEnv, lsm_mutex **ppNew){
  NoopMutex *p;
  p = (NoopMutex *)lsmMallocZero(pEnv, sizeof(NoopMutex));
  if( p ) p->pEnv = pEnv;
  *ppNew = (lsm_mutex *)p;
  return (p ? LSM_OK : LSM_NOMEM_BKPT);
}
static void lsmPosixOsMutexDel(lsm_mutex *pMutex)  { 
  NoopMutex *p = (NoopMutex *)pMutex;
  assert( p->bStatic==0 && p->pEnv );
  lsmFree(p->pEnv, p);
}
static void lsmPosixOsMutexEnter(lsm_mutex *pMutex){ 
  NoopMutex *p = (NoopMutex *)pMutex;
  assert( p->bHeld==0 );
  p->bHeld = 1;
}
static int lsmPosixOsMutexTry(lsm_mutex *pMutex){
  NoopMutex *p = (NoopMutex *)pMutex;
  assert( p->bHeld==0 );
  p->bHeld = 1;
  return 0;
}
static void lsmPosixOsMutexLeave(lsm_mutex *pMutex){ 
  NoopMutex *p = (NoopMutex *)pMutex;
  assert( p->bHeld==1 );
  p->bHeld = 0;
}
#ifdef LSM_DEBUG
static int lsmPosixOsMutexHeld(lsm_mutex *pMutex){ 
  NoopMutex *p = (NoopMutex *)pMutex;
  return p ? p->bHeld : 1;
}
static int lsmPosixOsMutexNotHeld(lsm_mutex *pMutex){ 
  NoopMutex *p = (NoopMutex *)pMutex;
  return p ? !p->bHeld : 1;
}
#endif
/***************************************************************************/
#endif /* else LSM_MUTEX_NONE */

/* Without LSM_DEBUG, the MutexHeld tests are never called */
#ifndef LSM_DEBUG
# define lsmPosixOsMutexHeld    0
# define lsmPosixOsMutexNotHeld 0
#endif

lsm_env *lsm_default_env(void){
  static lsm_env posix_env = {
    sizeof(lsm_env),         /* nByte */
    1,                       /* iVersion */
    /***** file i/o ******************/
    0,                       /* pVfsCtx */
    lsmPosixOsFullpath,      /* xFullpath */
    lsmPosixOsOpen,          /* xOpen */
    lsmPosixOsRead,          /* xRead */
    lsmPosixOsWrite,         /* xWrite */
    lsmPosixOsTruncate,      /* xTruncate */
    lsmPosixOsSync,          /* xSync */
    lsmPosixOsSectorSize,    /* xSectorSize */
    lsmPosixOsRemap,         /* xRemap */
    lsmPosixOsFileid,        /* xFileid */
    lsmPosixOsClose,         /* xClose */
    lsmPosixOsUnlink,        /* xUnlink */
    lsmPosixOsLock,          /* xLock */
    lsmPosixOsTestLock,      /* xTestLock */
    lsmPosixOsShmMap,        /* xShmMap */
    lsmPosixOsShmBarrier,    /* xShmBarrier */
    lsmPosixOsShmUnmap,      /* xShmUnmap */
    /***** memory allocation *********/
    0,                       /* pMemCtx */
    lsmPosixOsMalloc,        /* xMalloc */
    lsmPosixOsRealloc,       /* xRealloc */
    lsmPosixOsFree,          /* xFree */
    lsmPosixOsMSize,         /* xSize */
    /***** mutexes *********************/
    0,                       /* pMutexCtx */
    lsmPosixOsMutexStatic,   /* xMutexStatic */
    lsmPosixOsMutexNew,      /* xMutexNew */
    lsmPosixOsMutexDel,      /* xMutexDel */
    lsmPosixOsMutexEnter,    /* xMutexEnter */
    lsmPosixOsMutexTry,      /* xMutexTry */
    lsmPosixOsMutexLeave,    /* xMutexLeave */
    lsmPosixOsMutexHeld,     /* xMutexHeld */
    lsmPosixOsMutexNotHeld,  /* xMutexNotHeld */
    /***** other *********************/
    lsmPosixOsSleep,         /* xSleep */
  };
  return &posix_env;
}

#endif

Added ext/lsm1/lsm_varint.c.









































































































































































































































































































































































































>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196

/*
** 2012-02-08
**
** The author disclaims copyright to this source code.  In place of
** a legal notice, here is a blessing:
**
**    May you do good and not evil.
**    May you find forgiveness for yourself and forgive others.
**    May you share freely, never taking more than you give.
**
*************************************************************************
**
** SQLite4-compatible varint implementation.
*/
#include "lsmInt.h"

/*************************************************************************
** The following is a copy of the varint.c module from SQLite 4.
*/

/*
** Decode the varint in z[].  Write the integer value into *pResult and
** return the number of bytes in the varint.
*/
static int lsmSqlite4GetVarint64(const unsigned char *z, u64 *pResult){
  unsigned int x;
  if( z[0]<=240 ){
    *pResult = z[0];
    return 1;
  }
  if( z[0]<=248 ){
    *pResult = (z[0]-241)*256 + z[1] + 240;
    return 2;
  }
  if( z[0]==249 ){
    *pResult = 2288 + 256*z[1] + z[2];
    return 3;
  }
  if( z[0]==250 ){
    *pResult = (z[1]<<16) + (z[2]<<8) + z[3];
    return 4;
  }
  x = (z[1]<<24) + (z[2]<<16) + (z[3]<<8) + z[4];
  if( z[0]==251 ){
    *pResult = x;
    return 5;
  }
  if( z[0]==252 ){
    *pResult = (((u64)x)<<8) + z[5];
    return 6;
  }
  if( z[0]==253 ){
    *pResult = (((u64)x)<<16) + (z[5]<<8) + z[6];
    return 7;
  }
  if( z[0]==254 ){
    *pResult = (((u64)x)<<24) + (z[5]<<16) + (z[6]<<8) + z[7];
    return 8;
  }
  *pResult = (((u64)x)<<32) +
               (0xffffffff & ((z[5]<<24) + (z[6]<<16) + (z[7]<<8) + z[8]));
  return 9;
}

/*
** Write a 32-bit unsigned integer as 4 big-endian bytes.
*/
static void lsmVarintWrite32(unsigned char *z, unsigned int y){
  z[0] = (unsigned char)(y>>24);
  z[1] = (unsigned char)(y>>16);
  z[2] = (unsigned char)(y>>8);
  z[3] = (unsigned char)(y);
}

/*
** Write a varint into z[].  The buffer z[] must be at least 9 characters
** long to accommodate the largest possible varint.  Return the number of
** bytes of z[] used.
*/
static int lsmSqlite4PutVarint64(unsigned char *z, u64 x){
  unsigned int w, y;
  if( x<=240 ){
    z[0] = (unsigned char)x;
    return 1;
  }
  if( x<=2287 ){
    y = (unsigned int)(x - 240);
    z[0] = (unsigned char)(y/256 + 241);
    z[1] = (unsigned char)(y%256);
    return 2;
  }
  if( x<=67823 ){
    y = (unsigned int)(x - 2288);
    z[0] = 249;
    z[1] = (unsigned char)(y/256);
    z[2] = (unsigned char)(y%256);
    return 3;
  }
  y = (unsigned int)x;
  w = (unsigned int)(x>>32);
  if( w==0 ){
    if( y<=16777215 ){
      z[0] = 250;
      z[1] = (unsigned char)(y>>16);
      z[2] = (unsigned char)(y>>8);
      z[3] = (unsigned char)(y);
      return 4;
    }
    z[0] = 251;
    lsmVarintWrite32(z+1, y);
    return 5;
  }
  if( w<=255 ){
    z[0] = 252;
    z[1] = (unsigned char)w;
    lsmVarintWrite32(z+2, y);
    return 6;
  }
  if( w<=32767 ){
    z[0] = 253;
    z[1] = (unsigned char)(w>>8);
    z[2] = (unsigned char)w;
    lsmVarintWrite32(z+3, y);
    return 7;
  }
  if( w<=16777215 ){
    z[0] = 254;
    z[1] = (unsigned char)(w>>16);
    z[2] = (unsigned char)(w>>8);
    z[3] = (unsigned char)w;
    lsmVarintWrite32(z+4, y);
    return 8;
  }
  z[0] = 255;
  lsmVarintWrite32(z+1, w);
  lsmVarintWrite32(z+5, y);
  return 9;
}

/*
** End of SQLite 4 code.
*************************************************************************/

int lsmVarintPut64(u8 *aData, i64 iVal){
  return lsmSqlite4PutVarint64(aData, (u64)iVal);
}

int lsmVarintGet64(const u8 *aData, i64 *piVal){
  return lsmSqlite4GetVarint64(aData, (u64 *)piVal);
}

int lsmVarintPut32(u8 *aData, int iVal){
  return lsmSqlite4PutVarint64(aData, (u64)iVal);
}

int lsmVarintGet32(u8 *z, int *piVal){
  u64 i;
  int ret;

  if( z[0]<=240 ){
    *piVal = z[0];
    return 1;
  }
  if( z[0]<=248 ){
    *piVal = (z[0]-241)*256 + z[1] + 240;
    return 2;
  }
  if( z[0]==249 ){
    *piVal = 2288 + 256*z[1] + z[2];
    return 3;
  }
  if( z[0]==250 ){
    *piVal = (z[1]<<16) + (z[2]<<8) + z[3];
    return 4;
  }

  ret = lsmSqlite4GetVarint64(z, &i);
  *piVal = i;
  return ret;
}

int lsmVarintLen32(int n){
  u8 aData[9];
  return lsmVarintPut32(aData, n);
}

/*
** The argument is the first byte of a varint. This function returns the
** total number of bytes in the entire varint (including the first byte).
*/
int lsmVarintSize(u8 c){
  if( c<241 ) return 1;
  if( c<249 ) return 2;
  return (int)(c - 246);
}

Added ext/lsm1/lsm_vtab.c.







































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
/*
** 2015-11-16
**
** The author disclaims copyright to this source code.  In place of
** a legal notice, here is a blessing:
**
**    May you do good and not evil.
**    May you find forgiveness for yourself and forgive others.
**    May you share freely, never taking more than you give.
**
*************************************************************************
**
** This file implements a simple virtual table wrapper around the LSM
** storage engine from SQLite4.
*/
#include "sqlite3ext.h"
SQLITE_EXTENSION_INIT1
#include "lsm.h"
#include <assert.h>
#include <string.h>

/* Forward declaration of subclasses of virtual table objects */
typedef struct lsm1_vtab lsm1_vtab;
typedef struct lsm1_cursor lsm1_cursor;

/* Primitive types */
typedef unsigned char u8;

/* An open connection to an LSM table */
struct lsm1_vtab {
  sqlite3_vtab base;          /* Base class - must be first */
  lsm_db *pDb;                /* Open connection to the LSM table */
};


/* lsm1_cursor is a subclass of sqlite3_vtab_cursor which will
** serve as the underlying representation of a cursor that scans
** over rows of the result
*/
struct lsm1_cursor {
  sqlite3_vtab_cursor base;  /* Base class - must be first */
  lsm_cursor *pLsmCur;       /* The LSM cursor */
  u8 isDesc;                 /* 0: scan forward.  1: scan reverse */
  u8 atEof;                  /* True if the scan is complete */
  u8 bUnique;                /* True if no more than one row of output */
};

/*
** The lsm1Connect() method is invoked to create a new
** lsm1_vtab that describes the virtual table.
*/
static int lsm1Connect(
  sqlite3 *db,
  void *pAux,
  int argc, const char *const*argv,
  sqlite3_vtab **ppVtab,
  char **pzErr
){
  lsm1_vtab *pNew;
  int rc;

  if( argc!=4 || argv[3]==0 || argv[3][0]==0 ){
    *pzErr = sqlite3_mprintf("filename argument missing");
    return SQLITE_ERROR;
  }
  *ppVtab = sqlite3_malloc( sizeof(*pNew) );
  pNew = (lsm1_vtab*)*ppVtab;
  if( pNew==0 ){
    return SQLITE_NOMEM;
  }
  memset(pNew, 0, sizeof(*pNew));
  rc = lsm_new(0, &pNew->pDb);
  if( rc ){
    *pzErr = sqlite3_mprintf("lsm_new failed with error code %d",  rc);
    rc = SQLITE_ERROR;
    goto connect_failed;
  }
  rc = lsm_open(pNew->pDb, argv[3]);
  if( rc ){
    *pzErr = sqlite3_mprintf("lsm_open failed with %d", rc);
    rc = SQLITE_ERROR;
    goto connect_failed;
  }

/* Column numbers */
#define LSM1_COLUMN_KEY         0
#define LSM1_COLUMN_BLOBKEY     1
#define LSM1_COLUMN_VALUE       2
#define LSM1_COLUMN_BLOBVALUE   3
#define LSM1_COLUMN_COMMAND     4

  rc = sqlite3_declare_vtab(db,
     "CREATE TABLE x("
     "  key,"              /* The primary key.  Any non-NULL */
     "  blobkey,"          /* Pure BLOB primary key */
     "  value,"            /* The value associated with key.  Any non-NULL */
     "  blobvalue,"        /* Pure BLOB value */
     "  command hidden"    /* Insert here for control operations */
     ");"
  );
connect_failed:
  if( rc!=SQLITE_OK ){
    if( pNew ){
      if( pNew->pDb ) lsm_close(pNew->pDb);
      sqlite3_free(pNew);
    }
    *ppVtab = 0;
  }
  return rc;
}

/*
** This method is the destructor for lsm1_cursor objects.
*/
static int lsm1Disconnect(sqlite3_vtab *pVtab){
  lsm1_vtab *p = (lsm1_vtab*)pVtab;
  lsm_close(p->pDb);
  sqlite3_free(p);
  return SQLITE_OK;
}

/*
** Constructor for a new lsm1_cursor object.
*/
static int lsm1Open(sqlite3_vtab *pVtab, sqlite3_vtab_cursor **ppCursor){
  lsm1_vtab *p = (lsm1_vtab*)pVtab;
  lsm1_cursor *pCur;
  int rc;
  pCur = sqlite3_malloc( sizeof(*pCur) );
  if( pCur==0 ) return SQLITE_NOMEM;
  memset(pCur, 0, sizeof(*pCur));
  *ppCursor = &pCur->base;
  rc = lsm_csr_open(p->pDb, &pCur->pLsmCur);
  if( rc==LSM_OK ){
    rc = SQLITE_OK;
  }else{
    sqlite3_free(pCur);
    *ppCursor = 0;
    rc = SQLITE_ERROR;
  }
  return rc;
}

/*
** Destructor for a lsm1_cursor.
*/
static int lsm1Close(sqlite3_vtab_cursor *cur){
  lsm1_cursor *pCur = (lsm1_cursor*)cur;
  lsm_csr_close(pCur->pLsmCur);
  sqlite3_free(pCur);
  return SQLITE_OK;
}


/*
** Advance a lsm1_cursor to its next row of output.
*/
static int lsm1Next(sqlite3_vtab_cursor *cur){
  lsm1_cursor *pCur = (lsm1_cursor*)cur;
  int rc;
  if( pCur->bUnique ){
    pCur->atEof = 1;
  }else{
    if( pCur->isDesc ){
      rc = lsm_csr_prev(pCur->pLsmCur);
    }else{
      rc = lsm_csr_next(pCur->pLsmCur);
    }
    if( rc==LSM_OK && lsm_csr_valid(pCur->pLsmCur)==0 ){
      pCur->atEof = 1;
    }
  }
  return rc==LSM_OK ? SQLITE_OK : SQLITE_ERROR;
}

/*
** Return TRUE if the cursor has been moved off of the last
** row of output.
*/
static int lsm1Eof(sqlite3_vtab_cursor *cur){
  lsm1_cursor *pCur = (lsm1_cursor*)cur;
  return pCur->atEof;
}

/*
** Rowids are not supported by the underlying virtual table.  So always
** return 0 for the rowid.
*/
static int lsm1Rowid(sqlite3_vtab_cursor *cur, sqlite_int64 *pRowid){
  *pRowid = 0;
  return SQLITE_OK;
}

/*
** Type prefixes on LSM keys
*/
#define LSM1_TYPE_NEGATIVE   0
#define LSM1_TYPE_POSITIVE   1
#define LSM1_TYPE_TEXT       2
#define LSM1_TYPE_BLOB       3

/*
** Write a 32-bit unsigned integer as 4 big-endian bytes.
*/
static void varintWrite32(unsigned char *z, unsigned int y){
  z[0] = (unsigned char)(y>>24);
  z[1] = (unsigned char)(y>>16);
  z[2] = (unsigned char)(y>>8);
  z[3] = (unsigned char)(y);
}

/*
** Write a varint into z[].  The buffer z[] must be at least 9 characters
** long to accommodate the largest possible varint.  Return the number of
** bytes of z[] used.
*/
static int lsm1PutVarint64(unsigned char *z, sqlite3_uint64 x){
  unsigned int w, y;
  if( x<=240 ){
    z[0] = (unsigned char)x;
    return 1;
  }
  if( x<=2287 ){
    y = (unsigned int)(x - 240);
    z[0] = (unsigned char)(y/256 + 241);
    z[1] = (unsigned char)(y%256);
    return 2;
  }
  if( x<=67823 ){
    y = (unsigned int)(x - 2288);
    z[0] = 249;
    z[1] = (unsigned char)(y/256);
    z[2] = (unsigned char)(y%256);
    return 3;
  }
  y = (unsigned int)x;
  w = (unsigned int)(x>>32);
  if( w==0 ){
    if( y<=16777215 ){
      z[0] = 250;
      z[1] = (unsigned char)(y>>16);
      z[2] = (unsigned char)(y>>8);
      z[3] = (unsigned char)(y);
      return 4;
    }
    z[0] = 251;
    varintWrite32(z+1, y);
    return 5;
  }
  if( w<=255 ){
    z[0] = 252;
    z[1] = (unsigned char)w;
    varintWrite32(z+2, y);
    return 6;
  }
  if( w<=65535 ){
    z[0] = 253;
    z[1] = (unsigned char)(w>>8);
    z[2] = (unsigned char)w;
    varintWrite32(z+3, y);
    return 7;
  }
  if( w<=16777215 ){
    z[0] = 254;
    z[1] = (unsigned char)(w>>16);
    z[2] = (unsigned char)(w>>8);
    z[3] = (unsigned char)w;
    varintWrite32(z+4, y);
    return 8;
  }
  z[0] = 255;
  varintWrite32(z+1, w);
  varintWrite32(z+5, y);
  return 9;
}

/*
** Decode the varint in the first n bytes z[].  Write the integer value
** into *pResult and return the number of bytes in the varint.
**
** If the decode fails because there are not enough bytes in z[] then
** return 0;
*/
static int lsm1GetVarint64(
  const unsigned char *z,
  int n,
  sqlite3_uint64 *pResult
){
  unsigned int x;
  if( n<1 ) return 0;
  if( z[0]<=240 ){
    *pResult = z[0];
    return 1;
  }
  if( z[0]<=248 ){
    if( n<2 ) return 0;
    *pResult = (z[0]-241)*256 + z[1] + 240;
    return 2;
  }
  if( n<z[0]-246 ) return 0;
  if( z[0]==249 ){
    *pResult = 2288 + 256*z[1] + z[2];
    return 3;
  }
  if( z[0]==250 ){
    *pResult = (z[1]<<16) + (z[2]<<8) + z[3];
    return 4;
  }
  x = (z[1]<<24) + (z[2]<<16) + (z[3]<<8) + z[4];
  if( z[0]==251 ){
    *pResult = x;
    return 5;
  }
  if( z[0]==252 ){
    *pResult = (((sqlite3_uint64)x)<<8) + z[5];
    return 6;
  }
  if( z[0]==253 ){
    *pResult = (((sqlite3_uint64)x)<<16) + (z[5]<<8) + z[6];
    return 7;
  }
  if( z[0]==254 ){
    *pResult = (((sqlite3_uint64)x)<<24) + (z[5]<<16) + (z[6]<<8) + z[7];
    return 8;
  }
  *pResult = (((sqlite3_uint64)x)<<32) +
               (0xffffffff & ((z[5]<<24) + (z[6]<<16) + (z[7]<<8) + z[8]));
  return 9;
}

/*
** Generate a key encoding for pValue such that all keys compare in
** lexicographical order.  Return an SQLite error code or SQLITE_OK.
**
** The key encoding is *pnKey bytes in length written into *ppKey.
** Space to hold the key is taken from pSpace if sufficient, or else
** from sqlite3_malloc().  The caller is responsible for freeing malloced
** space.
*/
static int lsm1EncodeKey(
  sqlite3_value *pValue,     /* Value to be encoded */
  unsigned char **ppKey,     /* Write the encoding here */
  int *pnKey,                /* Write the size of the encoding here */
  unsigned char *pSpace,     /* Use this space if it is large enough */
  int nSpace                 /* Size of pSpace[] */
){
  int eType = sqlite3_value_type(pValue);
  *ppKey = 0;
  *pnKey = 0;
  assert( nSpace>=32 );
  switch( eType ){
    default: {
      return SQLITE_ERROR;  /* We cannot handle NULL keys */
    }
    case SQLITE_BLOB:
    case SQLITE_TEXT: {
      int nVal = sqlite3_value_bytes(pValue);
      const void *pVal;
      if( eType==SQLITE_BLOB ){
        eType = LSM1_TYPE_BLOB;
        pVal = sqlite3_value_blob(pValue);
      }else{
        eType = LSM1_TYPE_TEXT;
        pVal = (const void*)sqlite3_value_text(pValue);
        if( pVal==0 ) return SQLITE_NOMEM;
      }
      if( nVal+1>nSpace ){
        pSpace = sqlite3_malloc( nVal+1 );
        if( pSpace==0 ) return SQLITE_NOMEM;
      }
      pSpace[0] = eType;
      memcpy(&pSpace[1], pVal, nVal);
      *ppKey = pSpace;
      *pnKey = nVal+1;
      break;
    }
    case SQLITE_INTEGER: {
      sqlite3_int64 iVal = sqlite3_value_int64(pValue);
      sqlite3_uint64 uVal;
      if( iVal<0 ){
        if( iVal==0xffffffffffffffffLL ) return SQLITE_ERROR;
        uVal = *(sqlite3_uint64*)&iVal;
        eType = LSM1_TYPE_NEGATIVE;
      }else{
        uVal = iVal;
        eType = LSM1_TYPE_POSITIVE;
      }
      pSpace[0] = eType;
      *ppKey = pSpace;
      *pnKey = 1 + lsm1PutVarint64(&pSpace[1], uVal);
    }
  }
  return SQLITE_OK;
}

/*
** Return values of columns for the row at which the lsm1_cursor
** is currently pointing.
*/
static int lsm1Column(
  sqlite3_vtab_cursor *cur,   /* The cursor */
  sqlite3_context *ctx,       /* First argument to sqlite3_result_...() */
  int i                       /* Which column to return */
){
  lsm1_cursor *pCur = (lsm1_cursor*)cur;
  switch( i ){
    case LSM1_COLUMN_BLOBKEY: {
      const void *pVal;
      int nVal;
      if( lsm_csr_key(pCur->pLsmCur, &pVal, &nVal)==LSM_OK ){
        sqlite3_result_blob(ctx, pVal, nVal, SQLITE_TRANSIENT);
      }
      break;
    }
    case LSM1_COLUMN_KEY: {
      const unsigned char *pVal;
      int nVal;
      if( lsm_csr_key(pCur->pLsmCur, (const void**)&pVal, &nVal)==LSM_OK
       && nVal>=1
      ){
        if( pVal[0]==LSM1_TYPE_BLOB ){
          sqlite3_result_blob(ctx, (const void*)&pVal[1],nVal-1,
                              SQLITE_TRANSIENT);
        }else if( pVal[0]==LSM1_TYPE_TEXT ){
          sqlite3_result_text(ctx, (const char*)&pVal[1],nVal-1,
                              SQLITE_TRANSIENT);
        }else if( nVal>=2 && nVal<=10 &&
           (pVal[0]==LSM1_TYPE_POSITIVE || pVal[0]==LSM1_TYPE_NEGATIVE)
        ){
          sqlite3_int64 iVal;
          lsm1GetVarint64(pVal+1, nVal-1, (sqlite3_uint64*)&iVal);
          sqlite3_result_int64(ctx, iVal);
        }         
      }
      break;
    }
    case LSM1_COLUMN_BLOBVALUE: {
      const void *pVal;
      int nVal;
      if( lsm_csr_value(pCur->pLsmCur, (const void**)&pVal, &nVal)==LSM_OK ){
        sqlite3_result_blob(ctx, pVal, nVal, SQLITE_TRANSIENT);
      }
      break;
    }
    case LSM1_COLUMN_VALUE: {
      const unsigned char *aVal;
      int nVal;
      if( lsm_csr_value(pCur->pLsmCur, (const void**)&aVal, &nVal)==LSM_OK
          && nVal>=1
      ){
        switch( aVal[0] ){
          case SQLITE_FLOAT:
          case SQLITE_INTEGER: {
            sqlite3_uint64 x = 0;
            int j;
            for(j=1; j<nVal; j++){
              x = (x<<8) | aVal[j];
            }
            if( aVal[0]==SQLITE_INTEGER ){
              sqlite3_result_int64(ctx, *(sqlite3_int64*)&x);
            }else{
              sqlite3_result_double(ctx, *(double*)&x);
            }
            break;
          }
          case SQLITE_TEXT: {
            sqlite3_result_text(ctx, (char*)&aVal[1], nVal-1, SQLITE_TRANSIENT);
            break;
          }
          case SQLITE_BLOB: {
            sqlite3_result_blob(ctx, &aVal[1], nVal-1, SQLITE_TRANSIENT);
            break;
          }
        }
      }
      break;
    }
    default: {
      break;
    }
  }
  return SQLITE_OK;
}

/* Move to the first row to return.
*/
static int lsm1Filter(
  sqlite3_vtab_cursor *pVtabCursor, 
  int idxNum, const char *idxStr,
  int argc, sqlite3_value **argv
){
  lsm1_cursor *pCur = (lsm1_cursor *)pVtabCursor;
  int rc = LSM_OK;
  pCur->atEof = 1;
  if( idxNum==1 ){
    assert( argc==1 );
    pCur->isDesc = 0;
    pCur->bUnique = 1;
    if( sqlite3_value_type(argv[0])==SQLITE_BLOB ){
      const void *pVal = sqlite3_value_blob(argv[0]);
      int nVal = sqlite3_value_bytes(argv[0]);
      rc = lsm_csr_seek(pCur->pLsmCur, pVal, nVal, LSM_SEEK_EQ);
    }
  }else{
    rc = lsm_csr_first(pCur->pLsmCur);
    pCur->isDesc = 0;
    pCur->bUnique = 0;
  }
  if( rc==LSM_OK && lsm_csr_valid(pCur->pLsmCur)!=0 ){
    pCur->atEof = 0;
  }
  return rc==LSM_OK ? SQLITE_OK : SQLITE_ERROR;
}

/*
** Only comparisons against the key are allowed.  The idxNum defines
** which comparisons are available:
**
**     0        Full table scan only
**   bit 1      key==?1  single argument for ?1
**   bit 2      key>?1
**   bit 3      key>=?1
**   bit 4      key<?N   (N==1 if bits 2,3 clear, or 2 if bits2,3 set)
**   bit 5      key<=?N  (N==1 if bits 2,3 clear, or 2 if bits2,3 set)
**   bit 6      Use blobkey instead of key
**
** To put it another way:
**
**     0        Full table scan.
**     1        key==?1
**     2        key>?1
**     4        key>=?1
**     8        key<?1
**     10       key>?1 AND key<?2
**     12       key>=?1 AND key<?2
**     16       key<=?1
**     18       key>?1 AND key<=?2
**     20       key>=?1 AND key<=?2
**     33..52   Use blobkey in place of key...
*/
static int lsm1BestIndex(
  sqlite3_vtab *tab,
  sqlite3_index_info *pIdxInfo
){
  int i;                 /* Loop over constraints */
  int idxNum = 0;        /* The query plan bitmask */
  int nArg = 0;          /* Number of arguments to xFilter */
  int eqIdx = -1;        /* Index of the key== constraint, or -1 if none */

  const struct sqlite3_index_constraint *pConstraint;
  pConstraint = pIdxInfo->aConstraint;
  for(i=0; i<pIdxInfo->nConstraint && idxNum<16; i++, pConstraint++){
    if( pConstraint->usable==0 ) continue;
    if( pConstraint->iColumn!=LSM1_COLUMN_KEY ) continue;
    if( pConstraint->op!=SQLITE_INDEX_CONSTRAINT_EQ ) continue;
    switch( pConstraint->op ){
      case SQLITE_INDEX_CONSTRAINT_EQ: {
        eqIdx = i;
        idxNum = 1;
        break;
      }
    }
  }
  if( eqIdx>=0 ){
    pIdxInfo->aConstraintUsage[eqIdx].argvIndex = ++nArg;
    pIdxInfo->aConstraintUsage[eqIdx].omit = 1;
  }
  if( idxNum==1 ){
    pIdxInfo->estimatedCost = (double)1;
    pIdxInfo->estimatedRows = 1;
    pIdxInfo->orderByConsumed = 1;
  }else{
    /* Full table scan */
    pIdxInfo->estimatedCost = (double)2147483647;
    pIdxInfo->estimatedRows = 2147483647;
  }
  pIdxInfo->idxNum = idxNum;
  return SQLITE_OK;
}

/*
** The xUpdate method is normally used for INSERT, REPLACE, UPDATE, and
** DELETE.  But this virtual table only supports INSERT and REPLACE.
** DELETE is accomplished by inserting a record with a value of NULL.
** UPDATE is achieved by using REPLACE.
*/
int lsm1Update(
  sqlite3_vtab *pVTab,
  int argc,
  sqlite3_value **argv,
  sqlite_int64 *pRowid
){
  lsm1_vtab *p = (lsm1_vtab*)pVTab;
  const void *pKey;
  int nKey;
  int eType;
  int rc;
  sqlite3_value *pValue;
  const unsigned char *pVal;
  unsigned char *pData;
  int nVal;
  unsigned char pSpace[100];

  if( argc==1 ){
    pVTab->zErrMsg = sqlite3_mprintf("cannot DELETE");
    return SQLITE_ERROR;
  }
  if( sqlite3_value_type(argv[0])!=SQLITE_NULL ){
    pVTab->zErrMsg = sqlite3_mprintf("cannot UPDATE");
    return SQLITE_ERROR;
  }

  /* "INSERT INTO tab(command) VALUES('....')" is used to implement
  ** special commands.
  */
  if( sqlite3_value_type(argv[2+LSM1_COLUMN_COMMAND])!=SQLITE_NULL ){
    return SQLITE_OK;
  }
  if( sqlite3_value_type(argv[2+LSM1_COLUMN_BLOBKEY])==SQLITE_BLOB ){
    /* Use the blob key exactly as supplied */
    pKey = sqlite3_value_blob(argv[2+LSM1_COLUMN_BLOBKEY]);
    nKey = sqlite3_value_bytes(argv[2+LSM1_COLUMN_BLOBKEY]);
  }else{
    /* Use a key encoding that sorts in lexicographical order */
    rc = lsm1EncodeKey(argv[2+LSM1_COLUMN_KEY],
                       (unsigned char**)&pKey,&nKey,
                       pSpace,sizeof(pSpace));
    if( rc ) return rc;
  }
  if( sqlite3_value_type(argv[2+LSM1_COLUMN_BLOBVALUE])==SQLITE_BLOB ){
    pVal = sqlite3_value_blob(argv[2+LSM1_COLUMN_BLOBVALUE]);
    nVal = sqlite3_value_bytes(argv[2+LSM1_COLUMN_BLOBVALUE]);
    rc = lsm_insert(p->pDb, pKey, nKey, pVal, nVal);
  }else{
    pValue = argv[2+LSM1_COLUMN_VALUE];
    eType = sqlite3_value_type(pValue);
    switch( eType ){
      case SQLITE_NULL: {
        rc = lsm_delete(p->pDb, pKey, nKey);
        break;
      }
      case SQLITE_BLOB:
      case SQLITE_TEXT: {
        if( eType==SQLITE_TEXT ){
          pVal = sqlite3_value_text(pValue);
        }else{
          pVal = (unsigned char*)sqlite3_value_blob(pValue);
        }
        nVal = sqlite3_value_bytes(pValue);
        pData = sqlite3_malloc( nVal+1 );
        if( pData==0 ){
          rc = SQLITE_NOMEM;
        }else{
          pData[0] = eType;
          memcpy(&pData[1], pVal, nVal);
          rc = lsm_insert(p->pDb, pKey, nKey, pData, nVal+1);
          sqlite3_free(pData);
        }
        break;
      }
      case SQLITE_INTEGER:
      case SQLITE_FLOAT: {
        sqlite3_uint64 x;
        unsigned char aVal[9];
        int i;
        if( eType==SQLITE_INTEGER ){
          *(sqlite3_int64*)&x = sqlite3_value_int64(pValue);
        }else{
          *(double*)&x = sqlite3_value_double(pValue);
        }
        for(i=8; x>0 && i>=1; i--){
          aVal[i] = x & 0xff;
          x >>= 8;
        }
        aVal[i] = eType;
        rc = lsm_insert(p->pDb, pKey, nKey, &aVal[i], 9-i);
        break;
      }
    }
  }
  if( pKey!=(const void*)pSpace ) sqlite3_free((void*)pKey);
  return rc==LSM_OK ? SQLITE_OK : SQLITE_ERROR;
}      

/* Begin a transaction
*/
static int lsm1Begin(sqlite3_vtab *pVtab){
  lsm1_vtab *p = (lsm1_vtab*)pVtab;
  int rc = lsm_begin(p->pDb, 1);
  return rc==LSM_OK ? SQLITE_OK : SQLITE_ERROR;
}

/* Phase 1 of a transaction commit.
*/
static int lsm1Sync(sqlite3_vtab *pVtab){
  return SQLITE_OK;
}

/* Commit a transaction
*/
static int lsm1Commit(sqlite3_vtab *pVtab){
  lsm1_vtab *p = (lsm1_vtab*)pVtab;
  int rc = lsm_commit(p->pDb, 0);
  return rc==LSM_OK ? SQLITE_OK : SQLITE_ERROR;
}

/* Rollback a transaction
*/
static int lsm1Rollback(sqlite3_vtab *pVtab){
  lsm1_vtab *p = (lsm1_vtab*)pVtab;
  int rc = lsm_rollback(p->pDb, 0);
  return rc==LSM_OK ? SQLITE_OK : SQLITE_ERROR;
}

/*
** This following structure defines all the methods for the 
** generate_lsm1 virtual table.
*/
static sqlite3_module lsm1Module = {
  0,                       /* iVersion */
  lsm1Connect,             /* xCreate */
  lsm1Connect,             /* xConnect */
  lsm1BestIndex,           /* xBestIndex */
  lsm1Disconnect,          /* xDisconnect */
  lsm1Disconnect,          /* xDestroy */
  lsm1Open,                /* xOpen - open a cursor */
  lsm1Close,               /* xClose - close a cursor */
  lsm1Filter,              /* xFilter - configure scan constraints */
  lsm1Next,                /* xNext - advance a cursor */
  lsm1Eof,                 /* xEof - check for end of scan */
  lsm1Column,              /* xColumn - read data */
  lsm1Rowid,               /* xRowid - read data */
  lsm1Update,              /* xUpdate */
  lsm1Begin,               /* xBegin */
  lsm1Sync,                /* xSync */
  lsm1Commit,              /* xCommit */
  lsm1Rollback,            /* xRollback */
  0,                       /* xFindMethod */
  0,                       /* xRename */
};


#ifdef _WIN32
__declspec(dllexport)
#endif
int sqlite3_lsm_init(
  sqlite3 *db, 
  char **pzErrMsg, 
  const sqlite3_api_routines *pApi
){
  int rc = SQLITE_OK;
  SQLITE_EXTENSION_INIT2(pApi);
  rc = sqlite3_create_module(db, "lsm1", &lsm1Module, 0);
  return rc;
}

Added ext/lsm1/lsm_win32.c.









































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
/*
** 2011-12-03
**
** The author disclaims copyright to this source code.  In place of
** a legal notice, here is a blessing:
**
**    May you do good and not evil.
**    May you find forgiveness for yourself and forgive others.
**    May you share freely, never taking more than you give.
**
*************************************************************************
**
** Win32-specific run-time environment implementation for LSM.
*/

#ifdef _WIN32

#include <assert.h>
#include <string.h>

#include <stdlib.h>
#include <stdarg.h>
#include <stdio.h>
#include <ctype.h>

#include "windows.h"

#include "lsmInt.h"

/*
** An open file is an instance of the following object
*/
typedef struct Win32File Win32File;
struct Win32File {
  lsm_env *pEnv;                  /* The run-time environment */
  const char *zName;              /* Full path to file */

  HANDLE hFile;                   /* Open file handle */
  HANDLE hShmFile;                /* File handle for *-shm file */

  HANDLE hMap;                    /* File handle for mapping */
  LPVOID pMap;                    /* Pointer to mapping of file fd */
  size_t nMap;                    /* Size of mapping at pMap in bytes */
  int nShm;                       /* Number of entries in ahShm[]/apShm[] */
  LPHANDLE ahShm;                 /* Array of handles for shared mappings */
  LPVOID *apShm;                  /* Array of 32K shared memory segments */
};

static char *win32ShmFile(Win32File *p){
  char *zShm;
  int nName = strlen(p->zName);
  zShm = (char *)lsmMallocZero(p->pEnv, nName+4+1);
  if( zShm ){
    memcpy(zShm, p->zName, nName);
    memcpy(&zShm[nName], "-shm", 5);
  }
  return zShm;
}

static int win32Sleep(int us){
  Sleep((us + 999) / 1000);
  return LSM_OK;
}

/*
** The number of times that an I/O operation will be retried following a
** locking error - probably caused by antivirus software.  Also the initial
** delay before the first retry.  The delay increases linearly with each
** retry.
*/
#ifndef LSM_WIN32_IOERR_RETRY
# define LSM_WIN32_IOERR_RETRY 10
#endif
#ifndef LSM_WIN32_IOERR_RETRY_DELAY
# define LSM_WIN32_IOERR_RETRY_DELAY 25000
#endif
static int win32IoerrRetry = LSM_WIN32_IOERR_RETRY;
static int win32IoerrRetryDelay = LSM_WIN32_IOERR_RETRY_DELAY;

/*
** The "win32IoerrCanRetry1" macro is used to determine if a particular
** I/O error code obtained via GetLastError() is eligible to be retried.
** It must accept the error code DWORD as its only argument and should
** return non-zero if the error code is transient in nature and the
** operation responsible for generating the original error might succeed
** upon being retried.  The argument to this macro should be a variable.
**
** Additionally, a macro named "win32IoerrCanRetry2" may be defined.  If
** it is defined, it will be consulted only when the macro
** "win32IoerrCanRetry1" returns zero.  The "win32IoerrCanRetry2" macro
** is completely optional and may be used to include additional error
** codes in the set that should result in the failing I/O operation being
** retried by the caller.  If defined, the "win32IoerrCanRetry2" macro
** must exhibit external semantics identical to those of the
** "win32IoerrCanRetry1" macro.
*/
#if !defined(win32IoerrCanRetry1)
#define win32IoerrCanRetry1(a) (((a)==ERROR_ACCESS_DENIED)        || \
                                ((a)==ERROR_SHARING_VIOLATION)    || \
                                ((a)==ERROR_LOCK_VIOLATION)       || \
                                ((a)==ERROR_DEV_NOT_EXIST)        || \
                                ((a)==ERROR_NETNAME_DELETED)      || \
                                ((a)==ERROR_SEM_TIMEOUT)          || \
                                ((a)==ERROR_NETWORK_UNREACHABLE))
#endif

/*
** If an I/O error occurs, invoke this routine to see if it should be
** retried.  Return TRUE to retry.  Return FALSE to give up with an
** error.
*/
static int win32RetryIoerr(
  lsm_env *pEnv,
  int *pnRetry
){
  DWORD lastErrno;
  if( *pnRetry>=win32IoerrRetry ){
    return 0;
  }
  lastErrno = GetLastError();
  if( win32IoerrCanRetry1(lastErrno) ){
    win32Sleep(win32IoerrRetryDelay*(1+*pnRetry));
    ++*pnRetry;
    return 1;
  }
#if defined(win32IoerrCanRetry2)
  else if( win32IoerrCanRetry2(lastErrno) ){
    win32Sleep(win32IoerrRetryDelay*(1+*pnRetry));
    ++*pnRetry;
    return 1;
  }
#endif
  return 0;
}

/*
** Convert a UTF-8 string to Microsoft Unicode.
**
** Space to hold the returned string is obtained from lsmMalloc().
*/
static LPWSTR win32Utf8ToUnicode(lsm_env *pEnv, const char *zText){
  int nChar;
  LPWSTR zWideText;

  nChar = MultiByteToWideChar(CP_UTF8, 0, zText, -1, NULL, 0);
  if( nChar==0 ){
    return 0;
  }
  zWideText = lsmMallocZero(pEnv, nChar * sizeof(WCHAR));
  if( zWideText==0 ){
    return 0;
  }
  nChar = MultiByteToWideChar(CP_UTF8, 0, zText, -1, zWideText, nChar);
  if( nChar==0 ){
    lsmFree(pEnv, zWideText);
    zWideText = 0;
  }
  return zWideText;
}

/*
** Convert a Microsoft Unicode string to UTF-8.
**
** Space to hold the returned string is obtained from lsmMalloc().
*/
static char *win32UnicodeToUtf8(lsm_env *pEnv, LPCWSTR zWideText){
  int nByte;
  char *zText;

  nByte = WideCharToMultiByte(CP_UTF8, 0, zWideText, -1, 0, 0, 0, 0);
  if( nByte == 0 ){
    return 0;
  }
  zText = lsmMallocZero(pEnv, nByte);
  if( zText==0 ){
    return 0;
  }
  nByte = WideCharToMultiByte(CP_UTF8, 0, zWideText, -1, zText, nByte, 0, 0);
  if( nByte == 0 ){
    lsmFree(pEnv, zText);
    zText = 0;
  }
  return zText;
}

#if !defined(win32IsNotFound)
#define win32IsNotFound(a) (((a)==ERROR_FILE_NOT_FOUND)  || \
                            ((a)==ERROR_PATH_NOT_FOUND))
#endif

static int win32Open(
  lsm_env *pEnv,
  const char *zFile,
  int flags,
  LPHANDLE phFile
){
  int rc;
  LPWSTR zConverted;

  zConverted = win32Utf8ToUnicode(pEnv, zFile);
  if( zConverted==0 ){
    rc = LSM_NOMEM_BKPT;
  }else{
    int bReadonly = (flags & LSM_OPEN_READONLY);
    DWORD dwDesiredAccess;
    DWORD dwShareMode = FILE_SHARE_READ | FILE_SHARE_WRITE;
    DWORD dwCreationDisposition;
    DWORD dwFlagsAndAttributes = FILE_ATTRIBUTE_NORMAL;
    HANDLE hFile;
    int nRetry = 0;
    if( bReadonly ){
      dwDesiredAccess = GENERIC_READ;
      dwCreationDisposition = OPEN_EXISTING;
    }else{
      dwDesiredAccess = GENERIC_READ | GENERIC_WRITE;
      dwCreationDisposition = OPEN_ALWAYS;
    }
    while( (hFile = CreateFileW((LPCWSTR)zConverted,
                                dwDesiredAccess,
                                dwShareMode, NULL,
                                dwCreationDisposition,
                                dwFlagsAndAttributes,
                                NULL))==INVALID_HANDLE_VALUE &&
                                win32RetryIoerr(pEnv, &nRetry) ){
      /* Noop */
    }
    lsmFree(pEnv, zConverted);
    if( hFile!=INVALID_HANDLE_VALUE ){
      *phFile = hFile;
      rc = LSM_OK;
    }else{
      if( win32IsNotFound(GetLastError()) ){
        rc = lsmErrorBkpt(LSM_IOERR_NOENT);
      }else{
        rc = LSM_IOERR_BKPT;
      }
    }
  }
  return rc;
}

static int lsmWin32OsOpen(
  lsm_env *pEnv,
  const char *zFile,
  int flags,
  lsm_file **ppFile
){
  int rc = LSM_OK;
  Win32File *pWin32File;

  pWin32File = lsmMallocZero(pEnv, sizeof(Win32File));
  if( pWin32File==0 ){
    rc = LSM_NOMEM_BKPT;
  }else{
    HANDLE hFile;

    rc = win32Open(pEnv, zFile, flags, &hFile);
    if( rc==LSM_OK ){
      pWin32File->pEnv = pEnv;
      pWin32File->zName = zFile;
      pWin32File->hFile = hFile;
    }else{
      lsmFree(pEnv, pWin32File);
      pWin32File = 0;
    }
  }
  *ppFile = (lsm_file *)pWin32File;
  return rc;
}

static int lsmWin32OsWrite(
  lsm_file *pFile, /* File to write to */
  lsm_i64 iOff,    /* Offset to write to */
  void *pData,     /* Write data from this buffer */
  int nData        /* Bytes of data to write */
){
  Win32File *pWin32File = (Win32File *)pFile;
  OVERLAPPED overlapped;  /* The offset for WriteFile. */
  u8 *aRem = (u8 *)pData; /* Data yet to be written */
  int nRem = nData;       /* Number of bytes yet to be written */
  int nRetry = 0;         /* Number of retrys */

  memset(&overlapped, 0, sizeof(OVERLAPPED));
  overlapped.Offset = (LONG)(iOff & 0xffffffff);
  overlapped.OffsetHigh = (LONG)((iOff>>32) & 0x7fffffff);
  while( nRem>0 ){
    DWORD nWrite = 0; /* Bytes written using WriteFile */
    if( !WriteFile(pWin32File->hFile, aRem, nRem, &nWrite, &overlapped) ){
      if( win32RetryIoerr(pWin32File->pEnv, &nRetry) ) continue;
      break;
    }
    assert( nWrite==0 || nWrite<=(DWORD)nRem );
    if( nWrite==0 || nWrite>(DWORD)nRem ){
      break;
    }
    iOff += nWrite;
    overlapped.Offset = (LONG)(iOff & 0xffffffff);
    overlapped.OffsetHigh = (LONG)((iOff>>32) & 0x7fffffff);
    aRem += nWrite;
    nRem -= nWrite;
  }
  if( nRem!=0 ) return LSM_IOERR_BKPT;
  return LSM_OK;
}

static int win32Truncate(
  HANDLE hFile,
  lsm_i64 nSize
){
  LARGE_INTEGER offset;
  offset.QuadPart = nSize;
  if( !SetFilePointerEx(hFile, offset, 0, FILE_BEGIN) ){
    return LSM_IOERR_BKPT;
  }
  if (!SetEndOfFile(hFile) ){
    return LSM_IOERR_BKPT;
  }
  return LSM_OK;
}

static int lsmWin32OsTruncate(
  lsm_file *pFile, /* File to write to */
  lsm_i64 nSize    /* Size to truncate file to */
){
  Win32File *pWin32File = (Win32File *)pFile;
  return win32Truncate(pWin32File->hFile, nSize);
}

static int lsmWin32OsRead(
  lsm_file *pFile, /* File to read from */
  lsm_i64 iOff,    /* Offset to read from */
  void *pData,     /* Read data into this buffer */
  int nData        /* Bytes of data to read */
){
  Win32File *pWin32File = (Win32File *)pFile;
  OVERLAPPED overlapped; /* The offset for ReadFile */
  DWORD nRead = 0;       /* Bytes read using ReadFile */
  int nRetry = 0;        /* Number of retrys */

  memset(&overlapped, 0, sizeof(OVERLAPPED));
  overlapped.Offset = (LONG)(iOff & 0xffffffff);
  overlapped.OffsetHigh = (LONG)((iOff>>32) & 0x7fffffff);
  while( !ReadFile(pWin32File->hFile, pData, nData, &nRead, &overlapped) &&
         GetLastError()!=ERROR_HANDLE_EOF ){
    if( win32RetryIoerr(pWin32File->pEnv, &nRetry) ) continue;
    return LSM_IOERR_BKPT;
  }
  if( nRead<(DWORD)nData ){
    /* Unread parts of the buffer must be zero-filled */
    memset(&((char*)pData)[nRead], 0, nData - nRead);
  }
  return LSM_OK;
}

static int lsmWin32OsSync(lsm_file *pFile){
  int rc = LSM_OK;

#ifndef LSM_NO_SYNC
  Win32File *pWin32File = (Win32File *)pFile;

  if( pWin32File->pMap!=NULL ){
    if( !FlushViewOfFile(pWin32File->pMap, 0) ){
      rc = LSM_IOERR_BKPT;
    }
  }
  if( rc==LSM_OK && !FlushFileBuffers(pWin32File->hFile) ){
    rc = LSM_IOERR_BKPT;
  }
#else
#endif

  return rc;
}

static int lsmWin32OsSectorSize(lsm_file *pFile){
  return 512;
}

static int lsmWin32OsRemap(
  lsm_file *pFile,
  lsm_i64 iMin,
  void **ppOut,
  lsm_i64 *pnOut
){
  Win32File *pWin32File = (Win32File *)pFile;

  /* If the file is between 0 and 2MB in size, extend it in chunks of 256K.
  ** Thereafter, in chunks of 1MB at a time.  */
  const int aIncrSz[] = {256*1024, 1024*1024};
  int nIncrSz = aIncrSz[iMin>(2*1024*1024)];

  if( pWin32File->pMap!=NULL ){
    UnmapViewOfFile(pWin32File->pMap);
    *ppOut = pWin32File->pMap = NULL;
    *pnOut = pWin32File->nMap = 0;
  }
  if( pWin32File->hMap!=NULL ){
    CloseHandle(pWin32File->hMap);
    pWin32File->hMap = NULL;
  }
  if( iMin>=0 ){
    LARGE_INTEGER fileSize;
    DWORD dwSizeHigh;
    DWORD dwSizeLow;
    HANDLE hMap;
    LPVOID pMap;
    memset(&fileSize, 0, sizeof(LARGE_INTEGER));
    if( !GetFileSizeEx(pWin32File->hFile, &fileSize) ){
      return LSM_IOERR_BKPT;
    }
    assert( fileSize.QuadPart>=0 );
    if( fileSize.QuadPart<iMin ){
      int rc;
      fileSize.QuadPart = ((iMin + nIncrSz-1) / nIncrSz) * nIncrSz;
      rc = lsmWin32OsTruncate(pFile, fileSize.QuadPart);
      if( rc!=LSM_OK ){
        return rc;
      }
    }
    dwSizeLow = (DWORD)(fileSize.QuadPart & 0xFFFFFFFF);
    dwSizeHigh = (DWORD)((fileSize.QuadPart & 0x7FFFFFFFFFFFFFFF) >> 32);
    hMap = CreateFileMappingW(pWin32File->hFile, NULL, PAGE_READWRITE,
                              dwSizeHigh, dwSizeLow, NULL);
    if( hMap==NULL ){
      return LSM_IOERR_BKPT;
    }
    pWin32File->hMap = hMap;
    assert( fileSize.QuadPart<=0xFFFFFFFF );
    pMap = MapViewOfFile(hMap, FILE_MAP_WRITE | FILE_MAP_READ, 0, 0,
                         (SIZE_T)fileSize.QuadPart);
    if( pMap==NULL ){
      return LSM_IOERR_BKPT;
    }
    pWin32File->pMap = pMap;
    pWin32File->nMap = (SIZE_T)fileSize.QuadPart;
  }
  *ppOut = pWin32File->pMap;
  *pnOut = pWin32File->nMap;
  return LSM_OK;
}

static BOOL win32IsDriveLetterAndColon(
  const char *zPathname
){
  return ( isalpha(zPathname[0]) && zPathname[1]==':' );
}

static int lsmWin32OsFullpath(
  lsm_env *pEnv,
  const char *zName,
  char *zOut,
  int *pnOut
){
  DWORD nByte;
  void *zConverted;
  LPWSTR zTempWide;
  char *zTempUtf8;

  if( zName[0]=='/' && win32IsDriveLetterAndColon(zName+1) ){
    zName++;
  }
  zConverted = win32Utf8ToUnicode(pEnv, zName);
  if( zConverted==0 ){
    return LSM_NOMEM_BKPT;
  }
  nByte = GetFullPathNameW((LPCWSTR)zConverted, 0, 0, 0);
  if( nByte==0 ){
    lsmFree(pEnv, zConverted);
    return LSM_IOERR_BKPT;
  }
  nByte += 3;
  zTempWide = lsmMallocZero(pEnv, nByte * sizeof(zTempWide[0]));
  if( zTempWide==0 ){
    lsmFree(pEnv, zConverted);
    return LSM_NOMEM_BKPT;
  }
  nByte = GetFullPathNameW((LPCWSTR)zConverted, nByte, zTempWide, 0);
  if( nByte==0 ){
    lsmFree(pEnv, zConverted);
    lsmFree(pEnv, zTempWide);
    return LSM_IOERR_BKPT;
  }
  lsmFree(pEnv, zConverted);
  zTempUtf8 = win32UnicodeToUtf8(pEnv, zTempWide);
  lsmFree(pEnv, zTempWide);
  if( zTempUtf8 ){
    int nOut = *pnOut;
    int nLen = strlen(zTempUtf8) + 1;
    if( nLen>=nOut ){
      lsmFree(pEnv, zTempUtf8);
      return LSM_IOERR_BKPT;
    }
    snprintf(zOut, nOut, "%s", zTempUtf8);
    lsmFree(pEnv, zTempUtf8);
    *pnOut = nLen;
    return LSM_OK;
  }else{
    return LSM_NOMEM_BKPT;
  }
}

static int lsmWin32OsFileid(
  lsm_file *pFile,
  void *pBuf,
  int *pnBuf
){
  int nBuf;
  int nReq;
  u8 *pBuf2 = (u8 *)pBuf;
  Win32File *pWin32File = (Win32File *)pFile;
  BY_HANDLE_FILE_INFORMATION fileInfo;

  nBuf = *pnBuf;
  nReq = (sizeof(fileInfo.dwVolumeSerialNumber) +
          sizeof(fileInfo.nFileIndexHigh) +
          sizeof(fileInfo.nFileIndexLow));
  *pnBuf = nReq;
  if( nReq>nBuf ) return LSM_OK;
  memset(&fileInfo, 0, sizeof(BY_HANDLE_FILE_INFORMATION));
  if( !GetFileInformationByHandle(pWin32File->hFile, &fileInfo) ){
    return LSM_IOERR_BKPT;
  }
  nReq = sizeof(fileInfo.dwVolumeSerialNumber);
  memcpy(pBuf2, &fileInfo.dwVolumeSerialNumber, nReq);
  pBuf2 += nReq;
  nReq = sizeof(fileInfo.nFileIndexHigh);
  memcpy(pBuf, &fileInfo.nFileIndexHigh, nReq);
  pBuf2 += nReq;
  nReq = sizeof(fileInfo.nFileIndexLow);
  memcpy(pBuf2, &fileInfo.nFileIndexLow, nReq);
  return LSM_OK;
}

static int win32Delete(
  lsm_env *pEnv,
  const char *zFile
){
  int rc;
  LPWSTR zConverted;

  zConverted = win32Utf8ToUnicode(pEnv, zFile);
  if( zConverted==0 ){
    rc = LSM_NOMEM_BKPT;
  }else{
    int nRetry = 0;
    DWORD attr;
    DWORD lastErrno;

    do {
      attr = GetFileAttributesW(zConverted);
      if ( attr==INVALID_FILE_ATTRIBUTES ){
        rc = LSM_IOERR_BKPT;
        break;
      }
      if ( attr&FILE_ATTRIBUTE_DIRECTORY ){
        rc = LSM_IOERR_BKPT; /* Files only. */
        break;
      }
      if ( DeleteFileW(zConverted) ){
        rc = LSM_OK; /* Deleted OK. */
        break;
      }
      if ( !win32RetryIoerr(pEnv, &nRetry) ){
        rc = LSM_IOERR_BKPT; /* No more retries. */
        break;
      }
    }while( 1 );
  }
  lsmFree(pEnv, zConverted);
  return rc;
}

static int lsmWin32OsUnlink(lsm_env *pEnv, const char *zFile){
  return win32Delete(pEnv, zFile);
}

int lsmWin32OsLock(lsm_file *pFile, int iLock, int eType){
  Win32File *pWin32File = (Win32File *)pFile;
  OVERLAPPED ovlp;

  assert( LSM_LOCK_UNLOCK==0 );
  assert( LSM_LOCK_SHARED==1 );
  assert( LSM_LOCK_EXCL==2 );
  assert( eType>=LSM_LOCK_UNLOCK && eType<=LSM_LOCK_EXCL );
  assert( iLock>0 && iLock<=32 );

  memset(&ovlp, 0, sizeof(OVERLAPPED));
  ovlp.Offset = (4096-iLock);
  if( eType>LSM_LOCK_UNLOCK ){
    DWORD flags = LOCKFILE_FAIL_IMMEDIATELY;
    if( eType>=LSM_LOCK_EXCL ) flags |= LOCKFILE_EXCLUSIVE_LOCK;
    if( !LockFileEx(pWin32File->hFile, flags, 0, 1, 0, &ovlp) ){
      if( GetLastError()==ERROR_IO_PENDING ){
        return LSM_BUSY;
      }else{
        return LSM_IOERR_BKPT;
      }
    }
  }else{
    if( !UnlockFileEx(pWin32File->hFile, 0, 1, 0, &ovlp) ){
      return LSM_IOERR_BKPT;
    }
  }
  return LSM_OK;
}

int lsmWin32OsTestLock(lsm_file *pFile, int iLock, int nLock, int eType){
  Win32File *pWin32File = (Win32File *)pFile;
  DWORD flags = LOCKFILE_FAIL_IMMEDIATELY;
  OVERLAPPED ovlp;

  assert( LSM_LOCK_UNLOCK==0 );
  assert( LSM_LOCK_SHARED==1 );
  assert( LSM_LOCK_EXCL==2 );
  assert( eType==LSM_LOCK_SHARED || eType==LSM_LOCK_EXCL );
  assert( nLock>=0 );
  assert( iLock>0 && iLock<=32 );

  if( eType>=LSM_LOCK_EXCL ) flags |= LOCKFILE_EXCLUSIVE_LOCK;
  memset(&ovlp, 0, sizeof(OVERLAPPED));
  ovlp.Offset = (4096-iLock);
  if( !LockFileEx(pWin32File->hFile, flags, 0, (DWORD)nLock, 0, &ovlp) ){
    if( GetLastError()==ERROR_IO_PENDING ){
      return LSM_BUSY;
    }else{
      return LSM_IOERR_BKPT;
    }
  }
  UnlockFileEx(pWin32File->hFile, 0, (DWORD)nLock, 0, &ovlp);
  return LSM_OK;
}

int lsmWin32OsShmMap(lsm_file *pFile, int iChunk, int sz, void **ppShm){
  int rc;
  Win32File *pWin32File = (Win32File *)pFile;

  *ppShm = NULL;
  assert( sz>=0 );
  assert( sz==LSM_SHM_CHUNK_SIZE );
  if( iChunk>=pWin32File->nShm ){
    int i;
    void **apNew;
    int nNew = iChunk+1;
    lsm_i64 nReq = nNew * sz;
    LARGE_INTEGER fileSize;

    /* If the shared-memory file has not been opened, open it now. */
    if( pWin32File->hShmFile==NULL ){
      char *zShm = win32ShmFile(pWin32File);
      if( !zShm ) return LSM_NOMEM_BKPT;
      rc = win32Open(pWin32File->pEnv, zShm, 0, &pWin32File->hShmFile);
      lsmFree(pWin32File->pEnv, zShm);
      if( rc!=LSM_OK ){
        return rc;
      }
    }

    /* If the shared-memory file is not large enough to contain the
    ** requested chunk, cause it to grow.  */
    memset(&fileSize, 0, sizeof(LARGE_INTEGER));
    if( !GetFileSizeEx(pWin32File->hShmFile, &fileSize) ){
      return LSM_IOERR_BKPT;
    }
    assert( fileSize.QuadPart>=0 );
    if( fileSize.QuadPart<nReq ){
      rc = win32Truncate(pWin32File->hShmFile, nReq);
      if( rc!=LSM_OK ){
        return rc;
      }
    }

    apNew = (void **)lsmRealloc(pWin32File->pEnv, pWin32File->apShm,
                                sizeof(LPVOID) * nNew);
    if( !apNew ) return LSM_NOMEM_BKPT;
    for(i=pWin32File->nShm; i<nNew; i++){
      apNew[i] = NULL;
    }
    pWin32File->apShm = apNew;
    pWin32File->nShm = nNew;
  }

  if( pWin32File->apShm[iChunk]==NULL ){
    HANDLE hMap;
    LPVOID pMap;
    hMap = CreateFileMappingW(pWin32File->hShmFile, NULL, PAGE_READWRITE, 0,
                              (DWORD)sz, NULL);
    if( hMap==NULL ){
      return LSM_IOERR_BKPT;
    }
    pWin32File->ahShm[iChunk] = hMap;
    pMap = MapViewOfFile(hMap, FILE_MAP_WRITE | FILE_MAP_READ, 0, 0,
                         (SIZE_T)sz);
    if( pMap==NULL ){
      return LSM_IOERR_BKPT;
    }
    pWin32File->apShm[iChunk] = pMap;
    pWin32File->nMap = (SIZE_T)sz;
  }
  *ppShm = pWin32File->apShm[iChunk];
  return LSM_OK;
}

void lsmWin32OsShmBarrier(void){
  MemoryBarrier();
}

int lsmWin32OsShmUnmap(lsm_file *pFile, int bDelete){
  Win32File *pWin32File = (Win32File *)pFile;

  if( pWin32File->hShmFile!=NULL ){
    int i;
    for(i=0; i<pWin32File->nShm; i++){
      if( pWin32File->apShm[i]!=NULL ){
        UnmapViewOfFile(pWin32File->apShm[i]);
        pWin32File->apShm[i] = NULL;
      }
      if( pWin32File->ahShm[i]!=NULL ){
        CloseHandle(pWin32File->ahShm[i]);
        pWin32File->ahShm[i] = NULL;
      }
    }
    CloseHandle(pWin32File->hShmFile);
    pWin32File->hShmFile = 0;
    if( bDelete ){
      char *zShm = win32ShmFile(pWin32File);
      if( zShm ){ win32Delete(pWin32File->pEnv, zShm); }
      lsmFree(pWin32File->pEnv, zShm);
    }
  }
  return LSM_OK;
}

#define MX_CLOSE_ATTEMPT 3
static int lsmWin32OsClose(lsm_file *pFile){
  int rc;
  int nRetry = 0;
  Win32File *pWin32File = (Win32File *)pFile;
  lsmWin32OsShmUnmap(pFile, 0);
  if( pWin32File->pMap!=NULL ){
    UnmapViewOfFile(pWin32File->pMap);
    pWin32File->pMap = NULL;
    pWin32File->nMap = 0;
  }
  if( pWin32File->hMap!=NULL ){
    CloseHandle(pWin32File->hMap);
    pWin32File->hMap = NULL;
  }
  do{
    if( pWin32File->hFile==NULL ){
      rc = LSM_IOERR_BKPT;
      break;
    }
    rc = CloseHandle(pWin32File->hFile);
    if( rc ){
      pWin32File->hFile = NULL;
      rc = LSM_OK;
      break;
    }
    if( ++nRetry>=MX_CLOSE_ATTEMPT ){
      rc = LSM_IOERR_BKPT;
      break;
    }
  }while( 1 );
  lsmFree(pWin32File->pEnv, pWin32File->apShm);
  lsmFree(pWin32File->pEnv, pWin32File);
  return rc;
}

static int lsmWin32OsSleep(lsm_env *pEnv, int us){
  unused_parameter(pEnv);
  return win32Sleep(us);
}

/****************************************************************************
** Memory allocation routines.
*/

static void *lsmWin32OsMalloc(lsm_env *pEnv, size_t N){
  return HeapAlloc(GetProcessHeap(), 0, (SIZE_T)N);
}

static void lsmWin32OsFree(lsm_env *pEnv, void *p){
  if( p ){
    HeapFree(GetProcessHeap(), 0, p);
  }
}

static void *lsmWin32OsRealloc(lsm_env *pEnv, void *p, size_t N){
  unsigned char *m = (unsigned char *)p;
  if( 1>N ){
    lsmWin32OsFree(pEnv, p);
    return NULL;
  }else if( NULL==p ){
    return lsmWin32OsMalloc(pEnv, N);
  }else{
#if 0 /* arguable: don't shrink */
    SIZE_T sz = HeapSize(GetProcessHeap(), 0, m);
    if( sz>=(SIZE_T)N ){
      return p;
    }
#endif
    return HeapReAlloc(GetProcessHeap(), 0, m, N);
  }
}

static size_t lsmWin32OsMSize(lsm_env *pEnv, void *p){
  return (size_t)HeapSize(GetProcessHeap(), 0, p);
}


#ifdef LSM_MUTEX_WIN32
/*************************************************************************
** Mutex methods for Win32 based systems.  If LSM_MUTEX_WIN32 is
** missing then a no-op implementation of mutexes found below will be
** used instead.
*/
#include "windows.h"

typedef struct Win32Mutex Win32Mutex;
struct Win32Mutex {
  lsm_env *pEnv;
  CRITICAL_SECTION mutex;
#ifdef LSM_DEBUG
  DWORD owner;
#endif
};

#ifndef WIN32_MUTEX_INITIALIZER
# define WIN32_MUTEX_INITIALIZER { 0 }
#endif

#ifdef LSM_DEBUG
# define LSM_WIN32_STATIC_MUTEX { 0, WIN32_MUTEX_INITIALIZER, 0 }
#else
# define LSM_WIN32_STATIC_MUTEX { 0, WIN32_MUTEX_INITIALIZER }
#endif

static int lsmWin32OsMutexStatic(
  lsm_env *pEnv,
  int iMutex,
  lsm_mutex **ppStatic
){
  static volatile LONG initialized = 0;
  static Win32Mutex sMutex[2] = {
    LSM_WIN32_STATIC_MUTEX,
    LSM_WIN32_STATIC_MUTEX
  };

  assert( iMutex==LSM_MUTEX_GLOBAL || iMutex==LSM_MUTEX_HEAP );
  assert( LSM_MUTEX_GLOBAL==1 && LSM_MUTEX_HEAP==2 );

  if( InterlockedCompareExchange(&initialized, 1, 0)==0 ){
    int i;
    for(i=0; i<array_size(sMutex); i++){
      InitializeCriticalSection(&sMutex[i].mutex);
    }
  }
  *ppStatic = (lsm_mutex *)&sMutex[iMutex-1];
  return LSM_OK;
}

static int lsmWin32OsMutexNew(lsm_env *pEnv, lsm_mutex **ppNew){
  Win32Mutex *pMutex;           /* Pointer to new mutex */

  pMutex = (Win32Mutex *)lsmMallocZero(pEnv, sizeof(Win32Mutex));
  if( !pMutex ) return LSM_NOMEM_BKPT;

  pMutex->pEnv = pEnv;
  InitializeCriticalSection(&pMutex->mutex);

  *ppNew = (lsm_mutex *)pMutex;
  return LSM_OK;
}

static void lsmWin32OsMutexDel(lsm_mutex *p){
  Win32Mutex *pMutex = (Win32Mutex *)p;
  DeleteCriticalSection(&pMutex->mutex);
  lsmFree(pMutex->pEnv, pMutex);
}

static void lsmWin32OsMutexEnter(lsm_mutex *p){
  Win32Mutex *pMutex = (Win32Mutex *)p;
  EnterCriticalSection(&pMutex->mutex);

#ifdef LSM_DEBUG
  assert( pMutex->owner!=GetCurrentThreadId() );
  pMutex->owner = GetCurrentThreadId();
  assert( pMutex->owner==GetCurrentThreadId() );
#endif
}

static int lsmWin32OsMutexTry(lsm_mutex *p){
  BOOL bRet;
  Win32Mutex *pMutex = (Win32Mutex *)p;
  bRet = TryEnterCriticalSection(&pMutex->mutex);
#ifdef LSM_DEBUG
  if( bRet ){
    assert( pMutex->owner!=GetCurrentThreadId() );
    pMutex->owner = GetCurrentThreadId();
    assert( pMutex->owner==GetCurrentThreadId() );
  }
#endif
  return !bRet;
}

static void lsmWin32OsMutexLeave(lsm_mutex *p){
  Win32Mutex *pMutex = (Win32Mutex *)p;
#ifdef LSM_DEBUG
  assert( pMutex->owner==GetCurrentThreadId() );
  pMutex->owner = 0;
  assert( pMutex->owner!=GetCurrentThreadId() );
#endif
  LeaveCriticalSection(&pMutex->mutex);
}

#ifdef LSM_DEBUG
static int lsmWin32OsMutexHeld(lsm_mutex *p){
  Win32Mutex *pMutex = (Win32Mutex *)p;
  return pMutex ? pMutex->owner==GetCurrentThreadId() : 1;
}
static int lsmWin32OsMutexNotHeld(lsm_mutex *p){
  Win32Mutex *pMutex = (Win32Mutex *)p;
  return pMutex ? pMutex->owner!=GetCurrentThreadId() : 1;
}
#endif
/*
** End of pthreads mutex implementation.
*************************************************************************/
#else
/*************************************************************************
** Noop mutex implementation
*/
typedef struct NoopMutex NoopMutex;
struct NoopMutex {
  lsm_env *pEnv;                  /* Environment handle (for xFree()) */
  int bHeld;                      /* True if mutex is held */
  int bStatic;                    /* True for a static mutex */
};
static NoopMutex aStaticNoopMutex[2] = {
  {0, 0, 1},
  {0, 0, 1},
};

static int lsmWin32OsMutexStatic(
  lsm_env *pEnv,
  int iMutex,
  lsm_mutex **ppStatic
){
  assert( iMutex>=1 && iMutex<=(int)array_size(aStaticNoopMutex) );
  *ppStatic = (lsm_mutex *)&aStaticNoopMutex[iMutex-1];
  return LSM_OK;
}
static int lsmWin32OsMutexNew(lsm_env *pEnv, lsm_mutex **ppNew){
  NoopMutex *p;
  p = (NoopMutex *)lsmMallocZero(pEnv, sizeof(NoopMutex));
  if( p ) p->pEnv = pEnv;
  *ppNew = (lsm_mutex *)p;
  return (p ? LSM_OK : LSM_NOMEM_BKPT);
}
static void lsmWin32OsMutexDel(lsm_mutex *pMutex)  {
  NoopMutex *p = (NoopMutex *)pMutex;
  assert( p->bStatic==0 && p->pEnv );
  lsmFree(p->pEnv, p);
}
static void lsmWin32OsMutexEnter(lsm_mutex *pMutex){
  NoopMutex *p = (NoopMutex *)pMutex;
  assert( p->bHeld==0 );
  p->bHeld = 1;
}
static int lsmWin32OsMutexTry(lsm_mutex *pMutex){
  NoopMutex *p = (NoopMutex *)pMutex;
  assert( p->bHeld==0 );
  p->bHeld = 1;
  return 0;
}
static void lsmWin32OsMutexLeave(lsm_mutex *pMutex){
  NoopMutex *p = (NoopMutex *)pMutex;
  assert( p->bHeld==1 );
  p->bHeld = 0;
}
#ifdef LSM_DEBUG
static int lsmWin32OsMutexHeld(lsm_mutex *pMutex){
  NoopMutex *p = (NoopMutex *)pMutex;
  return p ? p->bHeld : 1;
}
static int lsmWin32OsMutexNotHeld(lsm_mutex *pMutex){
  NoopMutex *p = (NoopMutex *)pMutex;
  return p ? !p->bHeld : 1;
}
#endif
/***************************************************************************/
#endif /* else LSM_MUTEX_NONE */

/* Without LSM_DEBUG, the MutexHeld tests are never called */
#ifndef LSM_DEBUG
# define lsmWin32OsMutexHeld    0
# define lsmWin32OsMutexNotHeld 0
#endif

lsm_env *lsm_default_env(void){
  static lsm_env win32_env = {
    sizeof(lsm_env),         /* nByte */
    1,                       /* iVersion */
    /***** file i/o ******************/
    0,                       /* pVfsCtx */
    lsmWin32OsFullpath,      /* xFullpath */
    lsmWin32OsOpen,          /* xOpen */
    lsmWin32OsRead,          /* xRead */
    lsmWin32OsWrite,         /* xWrite */
    lsmWin32OsTruncate,      /* xTruncate */
    lsmWin32OsSync,          /* xSync */
    lsmWin32OsSectorSize,    /* xSectorSize */
    lsmWin32OsRemap,         /* xRemap */
    lsmWin32OsFileid,        /* xFileid */
    lsmWin32OsClose,         /* xClose */
    lsmWin32OsUnlink,        /* xUnlink */
    lsmWin32OsLock,          /* xLock */
    lsmWin32OsTestLock,      /* xTestLock */
    lsmWin32OsShmMap,        /* xShmMap */
    lsmWin32OsShmBarrier,    /* xShmBarrier */
    lsmWin32OsShmUnmap,      /* xShmUnmap */
    /***** memory allocation *********/
    0,                       /* pMemCtx */
    lsmWin32OsMalloc,        /* xMalloc */
    lsmWin32OsRealloc,       /* xRealloc */
    lsmWin32OsFree,          /* xFree */
    lsmWin32OsMSize,         /* xSize */
    /***** mutexes *********************/
    0,                       /* pMutexCtx */
    lsmWin32OsMutexStatic,   /* xMutexStatic */
    lsmWin32OsMutexNew,      /* xMutexNew */
    lsmWin32OsMutexDel,      /* xMutexDel */
    lsmWin32OsMutexEnter,    /* xMutexEnter */
    lsmWin32OsMutexTry,      /* xMutexTry */
    lsmWin32OsMutexLeave,    /* xMutexLeave */
    lsmWin32OsMutexHeld,     /* xMutexHeld */
    lsmWin32OsMutexNotHeld,  /* xMutexNotHeld */
    /***** other *********************/
    lsmWin32OsSleep,         /* xSleep */
  };
  return &win32_env;
}

#endif