SQLite

#define UNIXFILE_RDONLY      0x02     /* Connection is read only */
#define UNIXFILE_PERSIST_WAL 0x04     /* Persistent WAL mode */
#ifndef SQLITE_DISABLE_DIRSYNC
# define UNIXFILE_DIRSYNC    0x08     /* Directory sync needed */
#else
# define UNIXFILE_DIRSYNC    0x00
#endif
#define UNIXFILE_PSOW        0x10     /* SQLITE_IOCAP_POWERSAFE_OVERWRITE */

/*
** Include code that is common to all os_*.c files
*/
#include "os_common.h"

/*

      SimulateIOErrorBenign(0);
      return rc;
    }
    case SQLITE_FCNTL_PERSIST_WAL: {
      unixModeBit(pFile, UNIXFILE_PERSIST_WAL, (int*)pArg);
      return SQLITE_OK;
    }
    case SQLITE_FCNTL_POWERSAFE_OVERWRITE: {
      unixModeBit(pFile, UNIXFILE_PSOW, (int*)pArg);
      return SQLITE_OK;
    }
    case SQLITE_FCNTL_VFSNAME: {
      *(char**)pArg = sqlite3_mprintf("%s", pFile->pVfs->zName);
      return SQLITE_OK;
    }
#ifndef NDEBUG

  }
  return p->szSector*512;
}

/*
** Return the device characteristics for the file.
**
** This VFS is set up to return SQLITE_IOCAP_POWERSAFE_OVERWRITE by default.
** However, that choice is contraversial since technically the underlying
** file system does not always provide powersafe overwrites.  (In other
** words, after a power-loss event, parts of the file that were never
** written might end up being altered.)  However, non-PSOW behavior is very,
** very rare.  And asserting PSOW makes a large reduction in the amount
** of required I/O for journaling, since a lot of padding is eliminated.
**  Hence, while POWERSAFE_OVERWRITE is on by default, there is a file-control
** available to turn it off and URI query parameter available to turn it off.
*/
static int unixDeviceCharacteristics(sqlite3_file *id){
  unixFile *p = (unixFile*)id;
  if( p->ctrlFlags & UNIXFILE_PSOW ){
    return SQLITE_IOCAP_POWERSAFE_OVERWRITE;
  }else{
    return 0;
  }
}

#ifndef SQLITE_OMIT_WAL


/*
** Object used to represent an shared memory buffer.  

  assert( zFilename!=0 || noLock );

  OSTRACE(("OPEN    %-3d %s\n", h, zFilename));
  pNew->h = h;
  pNew->pVfs = pVfs;
  pNew->zPath = zFilename;
  pNew->ctrlFlags = 0;
  if( sqlite3_uri_boolean(zFilename, "psow", SQLITE_POWERSAFE_OVERWRITE) ){
    pNew->ctrlFlags |= UNIXFILE_PSOW;
  }
  if( memcmp(pVfs->zName,"unix-excl",10)==0 ){
    pNew->ctrlFlags |= UNIXFILE_EXCL;
  }
  if( isReadOnly ){
    pNew->ctrlFlags |= UNIXFILE_RDONLY;
  }

#endif
};

/*
** Allowed values for winFile.ctrlFlags
*/
#define WINFILE_PERSIST_WAL     0x04   /* Persistent WAL mode */
#define WINFILE_PSOW            0x10   /* SQLITE_IOCAP_POWERSAFE_OVERWRITE */

/*
 * If compiled with SQLITE_WIN32_MALLOC on Windows, we will use the
 * various Win32 API heap functions instead of our own.
 */
#ifdef SQLITE_WIN32_MALLOC
/*

      }
      return SQLITE_OK;
    }
    case SQLITE_FCNTL_PERSIST_WAL: {
      winModeBit(pFile, WINFILE_PERSIST_WAL, (int*)pArg);
      return SQLITE_OK;
    }
    case SQLITE_FCNTL_POWERSAFE_OVERWRITE: {
      winModeBit(pFile, WINFILE_PSOW, (int*)pArg);
      return SQLITE_OK;
    }
    case SQLITE_FCNTL_VFSNAME: {
      *(char**)pArg = sqlite3_mprintf("win32");
      return SQLITE_OK;
    }
    case SQLITE_FCNTL_SYNC_OMITTED: {

/*
** Return a vector of device characteristics.
*/
static int winDeviceCharacteristics(sqlite3_file *id){
  winFile *p = (winFile*)id;
  return SQLITE_IOCAP_UNDELETABLE_WHEN_OPEN |
         ((p->ctrlFlags & WINFILE_PSOW)?SQLITE_IOCAP_POWERSAFE_OVERWRITE:0);
}

#ifndef SQLITE_OMIT_WAL

/* 
** Windows will only let you create file view mappings
** on allocation size granularity boundaries.

  memset(pFile, 0, sizeof(*pFile));
  pFile->pMethod = &winIoMethod;
  pFile->h = h;
  pFile->lastErrno = NO_ERROR;
  pFile->pVfs = pVfs;
  pFile->pShm = 0;
  pFile->zPath = zName;
  if( sqlite3_uri_boolean(zName, "psow", SQLITE_POWERSAFE_OVERWRITE) ){
    pFile->ctrlFlags |= WINFILE_PSOW;
  }
  pFile->sectorSize = getSectorSize(pVfs, zUtf8Name);

#if SQLITE_OS_WINCE
  if( isReadWrite && eType==SQLITE_OPEN_MAIN_DB
       && !winceCreateLock(zName, pFile)
  ){

** For temporary files the effective sector size is always 512 bytes.
**
** Otherwise, for non-temporary files, the effective sector size is
** the value returned by the xSectorSize() method rounded up to 32 if
** it is less than 32, or rounded down to MAX_SECTOR_SIZE if it
** is greater than MAX_SECTOR_SIZE.
**
** If the file has the SQLITE_IOCAP_POWERSAFE_OVERWRITE property, then set
** the effective sector size to its minimum value (512).  The purpose of
** pPager->sectorSize is to define the "blast radius" of bytes that
** might change if a crash occurs while writing to a single byte in
** that range.  But with POWERSAFE_OVERWRITE, the blast radius is zero
** (that is what POWERSAFE_OVERWRITE means), so we minimize the sector
** size.  For backwards compatibility of the rollback journal file format,
** we cannot reduce the effective sector size below 512.

*/
static void setSectorSize(Pager *pPager){
  assert( isOpen(pPager->fd) || pPager->tempFile );

  if( pPager->tempFile
   || (sqlite3OsDeviceCharacteristics(pPager->fd) & 
              SQLITE_IOCAP_POWERSAFE_OVERWRITE)!=0
  ){
    /* Sector size doesn't matter for temporary files. Also, the file
    ** may not have been opened yet, in which case the OsSectorSize()
    ** call will segfault. */
    pPager->sectorSize = 512;
  }else{
    pPager->sectorSize = sqlite3OsSectorSize(pPager->fd);

** mean that writes of blocks that are nnn bytes in size and
** are aligned to an address which is an integer multiple of
** nnn are atomic.  The SQLITE_IOCAP_SAFE_APPEND value means
** that when data is appended to a file, the data is appended
** first then the size of the file is extended, never the other
** way around.  The SQLITE_IOCAP_SEQUENTIAL property means that
** information is written to disk in the same order as calls
** to xWrite().  The SQLITE_IOCAP_POWERSAFE_OVERWRITE property means that
** after reboot following a crash or power loss, the value of
** each byte in a file is a value that was actually written
** into that byte at some point.  In other words, a crash will
** not cause unwritten bytes of the file to change nor introduce 
** randomness into a file nor zero out parts of the file, and any byte of
** a file that are never written will not change values due to
** writes to nearby bytes.
*/
#define SQLITE_IOCAP_ATOMIC                 0x00000001
#define SQLITE_IOCAP_ATOMIC512              0x00000002
#define SQLITE_IOCAP_ATOMIC1K               0x00000004
#define SQLITE_IOCAP_ATOMIC2K               0x00000008
#define SQLITE_IOCAP_ATOMIC4K               0x00000010
#define SQLITE_IOCAP_ATOMIC8K               0x00000020
#define SQLITE_IOCAP_ATOMIC16K              0x00000040
#define SQLITE_IOCAP_ATOMIC32K              0x00000080
#define SQLITE_IOCAP_ATOMIC64K              0x00000100
#define SQLITE_IOCAP_SAFE_APPEND            0x00000200
#define SQLITE_IOCAP_SEQUENTIAL             0x00000400
#define SQLITE_IOCAP_UNDELETABLE_WHEN_OPEN  0x00000800
#define SQLITE_IOCAP_POWERSAFE_OVERWRITE    0x00001000

/*
** CAPI3REF: File Locking Levels
**
** SQLite uses one of these integer values as the second
** argument to calls it makes to the xLock() and xUnlock() methods
** of an [sqlite3_io_methods] object.

** to read the database file, as the WAL and shared memory files must exist
** in order for the database to be readable.  The fourth parameter to
** [sqlite3_file_control()] for this opcode should be a pointer to an integer.
** That integer is 0 to disable persistent WAL mode or 1 to enable persistent
** WAL mode.  If the integer is -1, then it is overwritten with the current
** WAL persistence setting.
**
** ^The [SQLITE_FCNTL_POWERSAFE_OVERWRITE] opcode is used to set or query the
** persistent "powersafe-overwrite" or "PSOW" setting.  The PSOW setting
** determines the [SQLITE_IOCAP_POWERSAFE_OVERWRITE] bit of the
** xDeviceCharacteristics methods. The fourth parameter to
** [sqlite3_file_control()] for this opcode should be a pointer to an integer.
** That integer is 0 to disable zero-damage mode or 1 to enable zero-damage
** mode.  If the integer is -1, then it is overwritten with the current
** zero-damage mode setting.
**
** ^The [SQLITE_FCNTL_OVERWRITE] opcode is invoked by SQLite after opening
** a write transaction to indicate that, unless it is rolled back for some

** that the fourth parameter of [sqlite3_file_control()] points to.
** The caller is responsible for freeing the memory when done.  As with
** all file-control actions, there is no guarantee that this will actually
** do anything.  Callers should initialize the char* variable to a NULL
** pointer in case this file-control is not implemented.  This file-control
** is intended for diagnostic use only.
*/
#define SQLITE_FCNTL_LOCKSTATE               1
#define SQLITE_GET_LOCKPROXYFILE             2
#define SQLITE_SET_LOCKPROXYFILE             3
#define SQLITE_LAST_ERRNO                    4
#define SQLITE_FCNTL_SIZE_HINT               5
#define SQLITE_FCNTL_CHUNK_SIZE              6
#define SQLITE_FCNTL_FILE_POINTER            7
#define SQLITE_FCNTL_SYNC_OMITTED            8
#define SQLITE_FCNTL_WIN32_AV_RETRY          9
#define SQLITE_FCNTL_PERSIST_WAL            10
#define SQLITE_FCNTL_OVERWRITE              11
#define SQLITE_FCNTL_VFSNAME                12
#define SQLITE_FCNTL_POWERSAFE_OVERWRITE    13

/*
** CAPI3REF: Mutex Handle
**
** The mutex module within SQLite defines [sqlite3_mutex] to be an
** abstract type for a mutex object.  The SQLite core never looks
** at the internal representation of an [sqlite3_mutex].  It only

#if defined(THREADSAFE)
# define SQLITE_THREADSAFE THREADSAFE
#else
# define SQLITE_THREADSAFE 1 /* IMP: R-07272-22309 */
#endif
#endif

/*
** Powersafe overwrite is on by default.  But can be turned off using
** the -DSQLITE_POWERSAFE_OVERWRITE=0 command-line option.
*/
#ifndef SQLITE_POWERSAFE_OVERWRITE
# define SQLITE_POWERSAFE_OVERWRITE 1
#endif

/*
** The SQLITE_DEFAULT_MEMSTATUS macro must be defined as either 0 or 1.
** It determines whether or not the features related to 
** SQLITE_CONFIG_MEMSTATUS are available by default or not. This value can
** be overridden at runtime using the sqlite3_config() API.
*/
#if !defined(SQLITE_DEFAULT_MEMSTATUS)

  rc = sqlite3_file_control(db, NULL, SQLITE_FCNTL_PERSIST_WAL, (void*)&bPersist);
  sqlite3_snprintf(sizeof(z), z, "%d %d", rc, bPersist);
  Tcl_AppendResult(interp, z, (char*)0);
  return TCL_OK;  
}

/*
** tclcmd:   file_control_powersafe_overwrite DB PSOW-FLAG
**
** This TCL command runs the sqlite3_file_control interface with
** the SQLITE_FCNTL_POWERSAFE_OVERWRITE opcode.
*/
static int file_control_powersafe_overwrite(
  ClientData clientData, /* Pointer to sqlite3_enable_XXX function */
  Tcl_Interp *interp,    /* The TCL interpreter that invoked this command */
  int objc,              /* Number of arguments */
  Tcl_Obj *CONST objv[]  /* Command arguments */
){
  sqlite3 *db;
  int rc;
  int b;
  char z[100];

  if( objc!=3 ){
    Tcl_AppendResult(interp, "wrong # args: should be \"",
        Tcl_GetStringFromObj(objv[0], 0), " DB FLAG", 0);
    return TCL_ERROR;
  }
  if( getDbPointer(interp, Tcl_GetString(objv[1]), &db) ){
    return TCL_ERROR;
  }
  if( Tcl_GetIntFromObj(interp, objv[2], &b) ) return TCL_ERROR;
  rc = sqlite3_file_control(db,NULL,SQLITE_FCNTL_POWERSAFE_OVERWRITE,(void*)&b);
  sqlite3_snprintf(sizeof(z), z, "%d %d", rc, b);
  Tcl_AppendResult(interp, z, (char*)0);
  return TCL_OK;  
}


/*
** tclcmd:   file_control_vfsname DB ?AUXDB?

     { "file_control_test",          file_control_test,   0   },
     { "file_control_lasterrno_test", file_control_lasterrno_test,  0   },
     { "file_control_lockproxy_test", file_control_lockproxy_test,  0   },
     { "file_control_chunksize_test", file_control_chunksize_test,  0   },
     { "file_control_sizehint_test",  file_control_sizehint_test,   0   },
     { "file_control_win32_av_retry", file_control_win32_av_retry,  0   },
     { "file_control_persist_wal",    file_control_persist_wal,     0   },
     { "file_control_powersafe_overwrite",file_control_powersafe_overwrite,0},
     { "file_control_vfsname",        file_control_vfsname,         0   },
     { "sqlite3_vfs_list",           vfs_list,     0   },
     { "sqlite3_create_function_v2", test_create_function_v2, 0 },

     /* Functions from os.h */
#ifndef SQLITE_OMIT_UTF16
     { "add_test_collate",        test_collate, 0            },

  int *piDeviceChar,
  int *piSectorSize
){
  struct DeviceFlag {
    char *zName;
    int iValue;
  } aFlag[] = {
    { "atomic",              SQLITE_IOCAP_ATOMIC                },
    { "atomic512",           SQLITE_IOCAP_ATOMIC512             },
    { "atomic1k",            SQLITE_IOCAP_ATOMIC1K              },
    { "atomic2k",            SQLITE_IOCAP_ATOMIC2K              },
    { "atomic4k",            SQLITE_IOCAP_ATOMIC4K              },
    { "atomic8k",            SQLITE_IOCAP_ATOMIC8K              },
    { "atomic16k",           SQLITE_IOCAP_ATOMIC16K             },
    { "atomic32k",           SQLITE_IOCAP_ATOMIC32K             },
    { "atomic64k",           SQLITE_IOCAP_ATOMIC64K             },
    { "sequential",          SQLITE_IOCAP_SEQUENTIAL            },
    { "safe_append",         SQLITE_IOCAP_SAFE_APPEND           },
    { "powersafe_overwrite", SQLITE_IOCAP_POWERSAFE_OVERWRITE   },
    { 0, 0 }
  };

  int i;
  int iDc = 0;
  int iSectorSize = 0;
  int setSectorsize = 0;

    case CMD_DEVCHAR: {
      struct DeviceFlag {
        char *zName;
        int iValue;
      } aFlag[] = {
        { "default",               -1 },
        { "atomic",                SQLITE_IOCAP_ATOMIC                },
        { "atomic512",             SQLITE_IOCAP_ATOMIC512             },
        { "atomic1k",              SQLITE_IOCAP_ATOMIC1K              },
        { "atomic2k",              SQLITE_IOCAP_ATOMIC2K              },
        { "atomic4k",              SQLITE_IOCAP_ATOMIC4K              },
        { "atomic8k",              SQLITE_IOCAP_ATOMIC8K              },
        { "atomic16k",             SQLITE_IOCAP_ATOMIC16K             },
        { "atomic32k",             SQLITE_IOCAP_ATOMIC32K             },
        { "atomic64k",             SQLITE_IOCAP_ATOMIC64K             },
        { "sequential",            SQLITE_IOCAP_SEQUENTIAL            },
        { "safe_append",           SQLITE_IOCAP_SAFE_APPEND           },
        { "undeletable_when_open", SQLITE_IOCAP_UNDELETABLE_WHEN_OPEN },
        { "powersafe_overwrite",   SQLITE_IOCAP_POWERSAFE_OVERWRITE   },
        { 0, 0 }
      };
      Tcl_Obj *pRet;
      int iFlag;

      if( objc>3 ){
        Tcl_WrongNumArgs(interp, 2, objv, "?ATTR-LIST?");

  if( rc!=SQLITE_OK ){
    walIndexClose(pRet, 0);
    sqlite3OsClose(pRet->pWalFd);
    sqlite3_free(pRet);
  }else{
    int iDC = sqlite3OsDeviceCharacteristics(pRet->pWalFd);
    if( iDC & SQLITE_IOCAP_SEQUENTIAL ){ pRet->syncHeader = 0; }
    if( iDC & SQLITE_IOCAP_POWERSAFE_OVERWRITE ){
      pRet->padToSectorBoundary = 0;
    }
    *ppWal = pRet;
    WALTRACE(("WAL%d: opened\n", pRet));
  }
  return rc;
}

/*

  /* If this is the end of a transaction, then we might need to pad
  ** the transaction and/or sync the WAL file.
  **
  ** Padding and syncing only occur if this set of frames complete a
  ** transaction and if PRAGMA synchronous=FULL.  If synchronous==NORMAL
  ** or synchonous==OFF, then no padding or syncing are needed.
  **
  ** If SQLITE_IOCAP_POWERSAFE_OVERWRITE is defined, then padding is not
  ** needed and only the sync is done.  If padding is needed, then the
  ** final frame is repeated (with its commit mark) until the next sector
  ** boundary is crossed.  Only the part of the WAL prior to the last
  ** sector boundary is synced; the part of the last frame that extends
  ** past the sector boundary is written after the sync.
  */
  if( isCommit && (sync_flags & WAL_SYNC_TRANSACTIONS)!=0 ){
    if( pWal->padToSectorBoundary ){
      int sectorSize = sqlite3OsSectorSize(pWal->pWalFd);

  string range [string repeat "${a_string_counter}." $n] 1 $n
}

# Create a [testvfs] and install it as the default VFS. Set the device
# characteristics flags to "SAFE_DELETE".
#
testvfs tvfs -default 1
tvfs devchar {undeletable_when_open powersafe_overwrite}

# Set up a hook so that each time a journal file is opened, closed or
# deleted, the method name ("xOpen", "xClose" or "xDelete") and the final
# segment of the journal file-name (i.e. "test.db-journal") are appended to
# global list variable $::oplog.
#
tvfs filter {xOpen xClose xDelete}

# 2011 December 21
#
# 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 tests of the SQLITE_IOCAP_POWERSAFE_OVERWRITE property
# and the SQLITE_FCNTL_POWERSAFE_OVERWRITE file-control for manipulating it.
#
# The name of this file comes from the fact that we used to call the
# POWERSAFE_OVERWRITE property ZERO_DAMAGE.
#

set testdir [file dirname $argv0]
source $testdir/tester.tcl
set testprefix wal5

# POWERSAFE_OVERWRITE defaults to true
#
do_test zerodamage-1.0 {
  file_control_powersafe_overwrite db -1
} {0 1}

# Check the ability to turn zero-damage on and off.
#
do_test zerodamage-1.1 {
  file_control_powersafe_overwrite db 0
  file_control_powersafe_overwrite db -1
} {0 0}
do_test zerodamage-1.2 {
  file_control_powersafe_overwrite db 1
  file_control_powersafe_overwrite db -1
} {0 1}

# Run a transaction with zero-damage on, a small page size and a much larger
# sectorsize.  Verify that the maximum journal size is small - that the
# rollback journal is not being padded.
#
do_test zerodamage-2.0 {
  db close
  testvfs tv -default 1
  tv sectorsize 8192
  sqlite3 db file:test.db?psow=TRUE -uri 1
  unset -nocomplain ::max_journal_size
  set ::max_journal_size 0
  proc xDeleteCallback {method file args} {
    set sz [file size $file]
    if {$sz>$::max_journal_size} {set ::max_journal_size $sz}
  }
  tv filter xDelete

    INSERT INTO t1 SELECT value, randomblob(100) FROM nums
                    WHERE value BETWEEN 1 AND 400;
  }
  set ::max_journal_size 0
  db eval {
    UPDATE t1 SET y=randomblob(50) WHERE x=123;
  }
  concat [file_control_powersafe_overwrite db -1] [set ::max_journal_size]
} {0 1 2576}

# Repeat the previous step with zero-damage turned off.  This time the
# maximum rollback journal size should be much larger.
#
do_test zerodamage-2.1 {
  set ::max_journal_size 0
  db close
  sqlite3 db file:test.db?psow=FALSE -uri 1
  db eval {
    UPDATE t1 SET y=randomblob(50) WHERE x=124;
  }
  concat [file_control_powersafe_overwrite db -1] [set ::max_journal_size]
} {0 0 24704}

# Run a WAL-mode transaction with POWERSAFE_OVERWRITE on to verify that the
# WAL file does not get too big.
#
do_test zerodamage-3.0 {
  db eval {
     PRAGMA journal_mode=WAL;
  }
  db close
  sqlite3 db file:test.db?psow=TRUE -uri 1
  db eval {
     UPDATE t1 SET y=randomblob(50) WHERE x=124;
  }
  file size test.db-wal
} {1080}

# Repeat the previous with POWERSAFE_OVERWRITE off.  Verify that the WAL file
# is padded.
#
do_test zerodamage-3.1 {
  db close
  sqlite3 db file:test.db?psow=FALSE -uri 1
  db eval {
     UPDATE t1 SET y=randomblob(50) WHERE x=124;
  }
  file size test.db-wal
} {8416}