#include "sqliteInt.h"
#include "os.h"
#if OS_UNIX              /* This file is used on unix only */

/*
** These #defines should enable >2GB file support on Posix if the
** underlying operating system supports it.  If the OS lacks
** large file support, or if the OS is windows, these should be no-ops.
**
** Large file support can be disabled using the -DSQLITE_DISABLE_LFS switch
** on the compiler command line.  This is necessary if you are compiling
** on a recent machine (ex: RedHat 7.2) but you want your code to work
** on an older machine (ex: RedHat 6.0).  If you compile on RedHat 7.2
** without this option, LFS is enable.  But LFS does not exist in the kernel
** in RedHat 6.0, so the code won't work.  Hence, for maximum binary
** portability you should omit LFS.
**
** Similar is true for MacOS.  LFS is only supported on MacOS 9 and later.
*/
#ifndef SQLITE_DISABLE_LFS
# define _LARGE_FILE       1
# ifndef _FILE_OFFSET_BITS
#   define _FILE_OFFSET_BITS 64
# endif
# define _LARGEFILE_SOURCE 1
................................................................................
#include <fcntl.h>
#include <unistd.h>
#include <time.h>
#include <sys/time.h>
#include <errno.h>

/*
** Macros used to determine whether or not to use threads.  The
** SQLITE_UNIX_THREADS macro is defined if we are synchronizing for
** Posix threads and SQLITE_W32_THREADS is defined if we are
** synchronizing using Win32 threads.
*/
#if defined(THREADSAFE) && THREADSAFE
# include <pthread.h>
# define SQLITE_UNIX_THREADS 1
#endif

/*
................................................................................
  struct lockInfo *pLock;   /* Info about locks on this inode */
  int h;                    /* The file descriptor */
  unsigned char locktype;   /* The type of lock held on this fd */
  unsigned char isOpen;     /* True if needs to be closed */
  unsigned char fullSync;   /* Use F_FULLSYNC if available */
  int dirfd;                /* File descriptor for the directory */
#ifdef SQLITE_UNIX_THREADS
  pthread_t tid;            /* The thread authorized to use this OsFile */
#endif
};

/*
** Provide the ability to override some OS-layer functions during
** testing.  This is used to simulate OS crashes to verify that 
** commits are atomic even in the event of an OS crash.
................................................................................
/*
** Include code that is common to all os_*.c files
*/
#include "os_common.h"

/*
** Do not include any of the File I/O interface procedures if the
** SQLITE_OMIT_DISKIO macro is defined (indicating that there database
** will be in-memory only)
*/
#ifndef SQLITE_OMIT_DISKIO


/*
** Define various macros that are missing from some systems.
................................................................................
** Set or check the OsFile.tid field.  This field is set when an OsFile
** is first opened.  All subsequent uses of the OsFile verify that the
** same thread is operating on the OsFile.  Some operating systems do
** not allow locks to be overridden by other threads and that restriction
** means that sqlite3* database handles cannot be moved from one thread
** to another.  This logic makes sure a user does not try to do that
** by mistake.







*/
#if defined(SQLITE_UNIX_THREADS)
# define SET_THREADID(X)   (X)->tid = pthread_self()
# define CHECK_THREADID(X) (threadsOverrideEachOthersLocks==0 && \
                            !pthread_equal((X)->tid, pthread_self()))
#else
# define SET_THREADID(X)
................................................................................

/*
** An instance of the following structure serves as the key used
** to locate a particular lockInfo structure given its inode.
**
** If threads cannot override each others locks, then we set the
** lockKey.tid field to the thread ID.  If threads can override
** each others locks then tid is always set to zero.  tid is also
** set to zero if we compile without threading support.
*/
struct lockKey {
  dev_t dev;       /* Device number */
  ino_t ino;       /* Inode number */
#ifdef SQLITE_UNIX_THREADS
  pthread_t tid;   /* Thread ID or zero if threads can override each other */
#endif
................................................................................
  int nRef;             /* Number of pointers to this structure */
  int nLock;            /* Number of outstanding locks */
  int nPending;         /* Number of pending close() operations */
  int *aPending;        /* Malloced space holding fd's awaiting a close() */
};

/* 
** These hash table maps inodes and process IDs into lockInfo and openCnt

** structures.  Access to these hash tables must be protected by a mutex.
*/
static Hash lockHash = { SQLITE_HASH_BINARY, 0, 0, 0, 0, 0 };
static Hash openHash = { SQLITE_HASH_BINARY, 0, 0, 0, 0, 0 };


#ifdef SQLITE_UNIX_THREADS
/*
** This variable records whether or not threads can override each others
** locks.
**
**    0:  No.  Threads cannot override each others locks.
**    1:  Yes.  Threads can override each others locks.
**   -1:  We don't know yet.




*/
#ifdef SQLITE_TEST
int threadsOverrideEachOthersLocks = -1;
#else
static int threadsOverrideEachOthersLocks = -1;
#endif

................................................................................
#ifdef SQLITE_LOCK_TRACE
/*
** Print out information about all locking operations.
**
** This routine is used for troubleshooting locks on multithreaded
** platforms.  Enable by compiling with the -DSQLITE_LOCK_TRACE
** command-line option on the compiler.  This code is normally
** turnned off.
*/
static int lockTrace(int fd, int op, struct flock *p){
  char *zOpName, *zType;
  int s;
  int savedErrno;
  if( op==F_GETLK ){
    zOpName = "GETLK";
................................................................................
**
** A unixFile is only owned by a thread on systems where one thread is
** unable to override locks created by a different thread.  RedHat9 is
** an example of such a system.
**
** Ownership transfer is only allowed if the unixFile is currently unlocked.
** If the unixFile is locked and an ownership is wrong, then return
** SQLITE_MISUSE.  Otherwise return SQLITE_OK.
*/
#ifdef SQLITE_UNIX_THREADS
static int transferOwnership(unixFile *pFile){
  int rc;
  pthread_t hSelf;
  if( threadsOverrideEachOthersLocks ){
    /* Ownership transfers not needed on this system */
................................................................................
  rc = findLockInfo(pFile->h, &pFile->pLock, 0);
  TRACE5("LOCK    %d is now %s(%s,%d)\n", pFile->h,
     locktypeName(pFile->locktype),
     locktypeName(pFile->pLock->locktype), pFile->pLock->cnt);
  return rc;
}
#else

# define transferOwnership(X) SQLITE_OK
#endif

/*
** Delete the named file
*/
int sqlite3UnixDelete(const char *zFilename){
................................................................................
  return SQLITE_OK;
}

/*
** If the following global variable points to a string which is the
** name of a directory, then that directory will be used to store
** temporary files.


*/
char *sqlite3_temp_directory = 0;

/*
** Create a temporary file name in zBuf.  zBuf must be big enough to
** hold at least SQLITE_TEMPNAME_SIZE characters.
*/
................................................................................

/*
** Lower the locking level on file descriptor pFile to locktype.  locktype
** must be either NO_LOCK or SHARED_LOCK.
**
** If the locking level of the file descriptor is already at or below
** the requested locking level, this routine is a no-op.
**
** It is not possible for this routine to fail if the second argument
** is NO_LOCK.  If the second argument is SHARED_LOCK, this routine
** might return SQLITE_IOERR instead of SQLITE_OK.
*/
static int unixUnlock(OsFile *id, int locktype){
  struct lockInfo *pLock;
  struct flock lock;
  int rc = SQLITE_OK;
  unixFile *pFile = (unixFile*)id;

................................................................................
  TRACE7("UNLOCK  %d %d was %d(%d,%d) pid=%d\n", pFile->h, locktype,
      pFile->locktype, pFile->pLock->locktype, pFile->pLock->cnt, getpid());

  assert( locktype<=SHARED_LOCK );
  if( pFile->locktype<=locktype ){
    return SQLITE_OK;
  }
  if( CHECK_THREADID(pFile) ) return SQLITE_MISUSE;


  sqlite3OsEnterMutex();
  pLock = pFile->pLock;
  assert( pLock->cnt!=0 );
  if( pFile->locktype>SHARED_LOCK ){
    assert( pLock->locktype==pFile->locktype );
    if( locktype==SHARED_LOCK ){
      lock.l_type = F_RDLCK;
................................................................................
  ** prefer that the randomness be increased by making use of the
  ** uninitialized space in zBuf - but valgrind errors tend to worry
  ** some users.  Rather than argue, it seems easier just to initialize
  ** the whole array and silence valgrind, even if that means less randomness
  ** in the random seed.
  **
  ** When testing, initializing zBuf[] to zero is all we do.  That means
  ** that we always use the same random number sequence.* This makes the
  ** tests repeatable.
  */
  memset(zBuf, 0, 256);
#if !defined(SQLITE_TEST)
  {
    int pid, fd;
    fd = open("/dev/urandom", O_RDONLY);
................................................................................
  }
#endif
  return SQLITE_OK;
}

/*
** Sleep for a little while.  Return the amount of time slept.

*/
int sqlite3UnixSleep(int ms){
#if defined(HAVE_USLEEP) && HAVE_USLEEP
  usleep(ms*1000);
  return ms;
#else
  sleep((ms+999)/1000);
................................................................................
/*
** The following pair of routine implement mutual exclusion for
** multi-threaded processes.  Only a single thread is allowed to
** executed code that is surrounded by EnterMutex() and LeaveMutex().
**
** SQLite uses only a single Mutex.  There is not much critical
** code and what little there is executes quickly and without blocking.


*/
void sqlite3UnixEnterMutex(){
#ifdef SQLITE_UNIX_THREADS
  pthread_mutex_lock(&mutex);
#endif
  assert( !inMutex );
  inMutex = 1;
................................................................................
# endif
#else
# define TSD_COUNTER(N)  /* no-op */
#endif


/*
** If called with allocateFlag>1, then return a pointer to thread
** specific data for the current thread.  Allocate and zero the
** thread-specific data if it does not already exist necessary.
**
** If called with allocateFlag==0, then check the current thread
** specific data.  Return it if it exists.  If it does not exist,
** then return NULL.
**
** If called with allocateFlag<0, check to see if the thread specific
** data is allocated and is all zero.  If it is then deallocate it.

#include "sqliteInt.h"
#include "os.h"
#if OS_UNIX              /* This file is used on unix only */

/*
** These #defines should enable >2GB file support on Posix if the
** underlying operating system supports it.  If the OS lacks
** large file support, these should be no-ops.
**
** Large file support can be disabled using the -DSQLITE_DISABLE_LFS switch
** on the compiler command line.  This is necessary if you are compiling
** on a recent machine (ex: RedHat 7.2) but you want your code to work
** on an older machine (ex: RedHat 6.0).  If you compile on RedHat 7.2
** without this option, LFS is enable.  But LFS does not exist in the kernel
** in RedHat 6.0, so the code won't work.  Hence, for maximum binary
** portability you should omit LFS.


*/
#ifndef SQLITE_DISABLE_LFS
# define _LARGE_FILE       1
# ifndef _FILE_OFFSET_BITS
#   define _FILE_OFFSET_BITS 64
# endif
# define _LARGEFILE_SOURCE 1
................................................................................
#include <fcntl.h>
#include <unistd.h>
#include <time.h>
#include <sys/time.h>
#include <errno.h>

/*
** If we are to be thread-safe, include the pthreads header and define
** the SQLITE_UNIX_THREADS macro.


*/
#if defined(THREADSAFE) && THREADSAFE
# include <pthread.h>
# define SQLITE_UNIX_THREADS 1
#endif

/*
................................................................................
  struct lockInfo *pLock;   /* Info about locks on this inode */
  int h;                    /* The file descriptor */
  unsigned char locktype;   /* The type of lock held on this fd */
  unsigned char isOpen;     /* True if needs to be closed */
  unsigned char fullSync;   /* Use F_FULLSYNC if available */
  int dirfd;                /* File descriptor for the directory */
#ifdef SQLITE_UNIX_THREADS
  pthread_t tid;            /* The thread that "owns" this OsFile */
#endif
};

/*
** Provide the ability to override some OS-layer functions during
** testing.  This is used to simulate OS crashes to verify that 
** commits are atomic even in the event of an OS crash.
................................................................................
/*
** Include code that is common to all os_*.c files
*/
#include "os_common.h"

/*
** Do not include any of the File I/O interface procedures if the
** SQLITE_OMIT_DISKIO macro is defined (indicating that the database
** will be in-memory only)
*/
#ifndef SQLITE_OMIT_DISKIO


/*
** Define various macros that are missing from some systems.
................................................................................
** Set or check the OsFile.tid field.  This field is set when an OsFile
** is first opened.  All subsequent uses of the OsFile verify that the
** same thread is operating on the OsFile.  Some operating systems do
** not allow locks to be overridden by other threads and that restriction
** means that sqlite3* database handles cannot be moved from one thread
** to another.  This logic makes sure a user does not try to do that
** by mistake.
**
** Version 3.3.1 (2006-01-15):  OsFiles can be moved from one thread to
** another as long as we are running on a system that supports threads
** overriding each others locks (which now the most common behavior)
** or if no locks are held.  But the OsFile.pLock field needs to be
** recomputed because its key includes the thread-id.  See the 
** transferOwnership() function below for additional information
*/
#if defined(SQLITE_UNIX_THREADS)
# define SET_THREADID(X)   (X)->tid = pthread_self()
# define CHECK_THREADID(X) (threadsOverrideEachOthersLocks==0 && \
                            !pthread_equal((X)->tid, pthread_self()))
#else
# define SET_THREADID(X)
................................................................................

/*
** An instance of the following structure serves as the key used
** to locate a particular lockInfo structure given its inode.
**
** If threads cannot override each others locks, then we set the
** lockKey.tid field to the thread ID.  If threads can override
** each others locks then tid is always set to zero.  tid is omitted
** if we compile without threading support.
*/
struct lockKey {
  dev_t dev;       /* Device number */
  ino_t ino;       /* Inode number */
#ifdef SQLITE_UNIX_THREADS
  pthread_t tid;   /* Thread ID or zero if threads can override each other */
#endif
................................................................................
  int nRef;             /* Number of pointers to this structure */
  int nLock;            /* Number of outstanding locks */
  int nPending;         /* Number of pending close() operations */
  int *aPending;        /* Malloced space holding fd's awaiting a close() */
};

/* 
** These hash tables map inodes and file descriptors (really, lockKey and
** openKey structures) into lockInfo and openCnt structures.  Access to 
** these hash tables must be protected by a mutex.
*/
static Hash lockHash = { SQLITE_HASH_BINARY, 0, 0, 0, 0, 0 };
static Hash openHash = { SQLITE_HASH_BINARY, 0, 0, 0, 0, 0 };


#ifdef SQLITE_UNIX_THREADS
/*
** This variable records whether or not threads can override each others
** locks.
**
**    0:  No.  Threads cannot override each others locks.
**    1:  Yes.  Threads can override each others locks.
**   -1:  We don't know yet.
**
** This variable normally has file scope only.  But during testing, we make
** it a global so that the test code can change its value in order to verify
** that the right stuff happens in either case.
*/
#ifdef SQLITE_TEST
int threadsOverrideEachOthersLocks = -1;
#else
static int threadsOverrideEachOthersLocks = -1;
#endif

................................................................................
#ifdef SQLITE_LOCK_TRACE
/*
** Print out information about all locking operations.
**
** This routine is used for troubleshooting locks on multithreaded
** platforms.  Enable by compiling with the -DSQLITE_LOCK_TRACE
** command-line option on the compiler.  This code is normally
** turned off.
*/
static int lockTrace(int fd, int op, struct flock *p){
  char *zOpName, *zType;
  int s;
  int savedErrno;
  if( op==F_GETLK ){
    zOpName = "GETLK";
................................................................................
**
** A unixFile is only owned by a thread on systems where one thread is
** unable to override locks created by a different thread.  RedHat9 is
** an example of such a system.
**
** Ownership transfer is only allowed if the unixFile is currently unlocked.
** If the unixFile is locked and an ownership is wrong, then return
** SQLITE_MISUSE.  SQLITE_OK is returned if everything works.
*/
#ifdef SQLITE_UNIX_THREADS
static int transferOwnership(unixFile *pFile){
  int rc;
  pthread_t hSelf;
  if( threadsOverrideEachOthersLocks ){
    /* Ownership transfers not needed on this system */
................................................................................
  rc = findLockInfo(pFile->h, &pFile->pLock, 0);
  TRACE5("LOCK    %d is now %s(%s,%d)\n", pFile->h,
     locktypeName(pFile->locktype),
     locktypeName(pFile->pLock->locktype), pFile->pLock->cnt);
  return rc;
}
#else
  /* On single-threaded builds, ownership transfer is a no-op */
# define transferOwnership(X) SQLITE_OK
#endif

/*
** Delete the named file
*/
int sqlite3UnixDelete(const char *zFilename){
................................................................................
  return SQLITE_OK;
}

/*
** If the following global variable points to a string which is the
** name of a directory, then that directory will be used to store
** temporary files.
**
** See also the "PRAGMA temp_store_directory" SQL command.
*/
char *sqlite3_temp_directory = 0;

/*
** Create a temporary file name in zBuf.  zBuf must be big enough to
** hold at least SQLITE_TEMPNAME_SIZE characters.
*/
................................................................................

/*
** Lower the locking level on file descriptor pFile to locktype.  locktype
** must be either NO_LOCK or SHARED_LOCK.
**
** If the locking level of the file descriptor is already at or below
** the requested locking level, this routine is a no-op.




*/
static int unixUnlock(OsFile *id, int locktype){
  struct lockInfo *pLock;
  struct flock lock;
  int rc = SQLITE_OK;
  unixFile *pFile = (unixFile*)id;

................................................................................
  TRACE7("UNLOCK  %d %d was %d(%d,%d) pid=%d\n", pFile->h, locktype,
      pFile->locktype, pFile->pLock->locktype, pFile->pLock->cnt, getpid());

  assert( locktype<=SHARED_LOCK );
  if( pFile->locktype<=locktype ){
    return SQLITE_OK;
  }
  if( CHECK_THREADID(pFile) ){
    return SQLITE_MISUSE;
  }
  sqlite3OsEnterMutex();
  pLock = pFile->pLock;
  assert( pLock->cnt!=0 );
  if( pFile->locktype>SHARED_LOCK ){
    assert( pLock->locktype==pFile->locktype );
    if( locktype==SHARED_LOCK ){
      lock.l_type = F_RDLCK;
................................................................................
  ** prefer that the randomness be increased by making use of the
  ** uninitialized space in zBuf - but valgrind errors tend to worry
  ** some users.  Rather than argue, it seems easier just to initialize
  ** the whole array and silence valgrind, even if that means less randomness
  ** in the random seed.
  **
  ** When testing, initializing zBuf[] to zero is all we do.  That means
  ** that we always use the same random number sequence.  This makes the
  ** tests repeatable.
  */
  memset(zBuf, 0, 256);
#if !defined(SQLITE_TEST)
  {
    int pid, fd;
    fd = open("/dev/urandom", O_RDONLY);
................................................................................
  }
#endif
  return SQLITE_OK;
}

/*
** Sleep for a little while.  Return the amount of time slept.
** The argument is the number of milliseconds we want to sleep.
*/
int sqlite3UnixSleep(int ms){
#if defined(HAVE_USLEEP) && HAVE_USLEEP
  usleep(ms*1000);
  return ms;
#else
  sleep((ms+999)/1000);
................................................................................
/*
** The following pair of routine implement mutual exclusion for
** multi-threaded processes.  Only a single thread is allowed to
** executed code that is surrounded by EnterMutex() and LeaveMutex().
**
** SQLite uses only a single Mutex.  There is not much critical
** code and what little there is executes quickly and without blocking.
**
** This mutex is not recursive.
*/
void sqlite3UnixEnterMutex(){
#ifdef SQLITE_UNIX_THREADS
  pthread_mutex_lock(&mutex);
#endif
  assert( !inMutex );
  inMutex = 1;
................................................................................
# endif
#else
# define TSD_COUNTER(N)  /* no-op */
#endif


/*
** If called with allocateFlag>0, then return a pointer to thread
** specific data for the current thread.  Allocate and zero the
** thread-specific data if it does not already exist.
**
** If called with allocateFlag==0, then check the current thread
** specific data.  Return it if it exists.  If it does not exist,
** then return NULL.
**
** If called with allocateFlag<0, check to see if the thread specific
** data is allocated and is all zero.  If it is then deallocate it.