os_unix.c

sqlite的最新源码 This ZIP archive contains preprocessed C code for the SQLite library as individual sour

  struct lockInfo **ppLock,    /* Return the lockInfo structure here */
  struct openCnt **ppOpen      /* Return the openCnt structure here */
){
  int rc;
  struct lockKey key1;
  struct openKey key2;
  struct stat statbuf;
  struct lockInfo *pLock;
  struct openCnt *pOpen;
  rc = fstat(fd, &statbuf);
  if( rc!=0 ){
#ifdef EOVERFLOW
    if( errno==EOVERFLOW ) return SQLITE_NOLFS;
#endif
    return SQLITE_IOERR;
  }

  /* On OS X on an msdos filesystem, the inode number is reported
  ** incorrectly for zero-size files.  See ticket #3260.  To work
  ** around this problem (we consider it a bug in OS X, not SQLite)
  ** we always increase the file size to 1 by writing a single byte
  ** prior to accessing the inode number.  The one byte written is
  ** an ASCII 'S' character which also happens to be the first byte
  ** in the header of every SQLite database.  In this way, if there
  ** is a race condition such that another thread has already populated
  ** the first page of the database, no damage is done.
  */
  if( statbuf.st_size==0 ){
    write(fd, "S", 1);
    rc = fstat(fd, &statbuf);
    if( rc!=0 ){
      return SQLITE_IOERR;
    }
  }

  memset(&key1, 0, sizeof(key1));
  key1.dev = statbuf.st_dev;
#if IS_VXWORKS
  key1.rnam = rnam;
#else
  key1.ino = statbuf.st_ino;
#endif
#if SQLITE_THREADSAFE
  if( threadsOverrideEachOthersLocks<0 ){
    testThreadLockingBehavior(fd);
  }
  key1.tid = threadsOverrideEachOthersLocks ? 0 : pthread_self();
#endif
  memset(&key2, 0, sizeof(key2));
  key2.dev = statbuf.st_dev;
#if IS_VXWORKS
  key2.rnam = rnam;
#else
  key2.ino = statbuf.st_ino;
#endif
  pLock = lockList;
  while( pLock && memcmp(&key1, &pLock->key, sizeof(key1)) ){
    pLock = pLock->pNext;
  }
  if( pLock==0 ){
    pLock = sqlite3_malloc( sizeof(*pLock) );
    if( pLock==0 ){
      rc = SQLITE_NOMEM;
      goto exit_findlockinfo;
    }
    pLock->key = key1;
    pLock->nRef = 1;
    pLock->cnt = 0;
    pLock->locktype = 0;
    pLock->pNext = lockList;
    pLock->pPrev = 0;
    if( lockList ) lockList->pPrev = pLock;
    lockList = pLock;
  }else{
    pLock->nRef++;
  }
  *ppLock = pLock;
  if( ppOpen!=0 ){
    pOpen = openList;
    while( pOpen && memcmp(&key2, &pOpen->key, sizeof(key2)) ){
      pOpen = pOpen->pNext;
    }
    if( pOpen==0 ){
      pOpen = sqlite3_malloc( sizeof(*pOpen) );
      if( pOpen==0 ){
        releaseLockInfo(pLock);
        rc = SQLITE_NOMEM;
        goto exit_findlockinfo;
      }
      pOpen->key = key2;
      pOpen->nRef = 1;
      pOpen->nLock = 0;
      pOpen->nPending = 0;
      pOpen->aPending = 0;
      pOpen->pNext = openList;
      pOpen->pPrev = 0;
      if( openList ) openList->pPrev = pOpen;
      openList = pOpen;
#if IS_VXWORKS
      pOpen->pSem = NULL;
      pOpen->aSemName[0] = '\0';
#endif
    }else{
      pOpen->nRef++;
    }
    *ppOpen = pOpen;
  }

exit_findlockinfo:
  return rc;
}

#ifdef SQLITE_DEBUG
/*
** Helper function for printing out trace information from debugging
** binaries. This returns the string represetation of the supplied
** integer lock-type.
*/
static const char *locktypeName(int locktype){
  switch( locktype ){
  case NO_LOCK: return "NONE";
  case SHARED_LOCK: return "SHARED";
  case RESERVED_LOCK: return "RESERVED";
  case PENDING_LOCK: return "PENDING";
  case EXCLUSIVE_LOCK: return "EXCLUSIVE";
  }
  return "ERROR";
}
#endif

/*
** If we are currently in a different thread than the thread that the
** unixFile argument belongs to, then transfer ownership of the unixFile
** over to the current thread.
**
** 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.
*/
#if SQLITE_THREADSAFE
static int transferOwnership(unixFile *pFile){
  int rc;
  pthread_t hSelf;
  if( threadsOverrideEachOthersLocks ){
    /* Ownership transfers not needed on this system */
    return SQLITE_OK;
  }
  hSelf = pthread_self();
  if( pthread_equal(pFile->tid, hSelf) ){
    /* We are still in the same thread */
    OSTRACE1("No-transfer, same thread\n");
    return SQLITE_OK;
  }
  if( pFile->locktype!=NO_LOCK ){
    /* We cannot change ownership while we are holding a lock! */
    return SQLITE_MISUSE;
  }
  OSTRACE4("Transfer ownership of %d from %d to %d\n",
            pFile->h, pFile->tid, hSelf);
  pFile->tid = hSelf;
  if (pFile->pLock != NULL) {
    releaseLockInfo(pFile->pLock);
#if IS_VXWORKS
    rc = findLockInfo(pFile->h, pFile->zRealpath, &pFile->pLock, 0);
#else
    rc = findLockInfo(pFile->h, &pFile->pLock, 0);
#endif
    OSTRACE5("LOCK    %d is now %s(%s,%d)\n", pFile->h,
           locktypeName(pFile->locktype),
           locktypeName(pFile->pLock->locktype), pFile->pLock->cnt);
    return rc;
  } else {
    return SQLITE_OK;
  }
}
#else
  /* On single-threaded builds, ownership transfer is a no-op */
# define transferOwnership(X) SQLITE_OK
#endif

/*
** Seek to the offset passed as the second argument, then read cnt 
** bytes into pBuf. Return the number of bytes actually read.
**
** NB:  If you define USE_PREAD or USE_PREAD64, then it might also
** be necessary to define _XOPEN_SOURCE to be 500.  This varies from
** one system to another.  Since SQLite does not define USE_PREAD
** any any form by default, we will not attempt to define _XOPEN_SOURCE.
** See tickets #2741 and #2681.
*/
static int seekAndRead(unixFile *id, sqlite3_int64 offset, void *pBuf, int cnt){
  int got;
  i64 newOffset;
  TIMER_START;
#if defined(USE_PREAD)
  got = pread(id->h, pBuf, cnt, offset);
  SimulateIOError( got = -1 );
#elif defined(USE_PREAD64)
  got = pread64(id->h, pBuf, cnt, offset);
  SimulateIOError( got = -1 );
#else
  newOffset = lseek(id->h, offset, SEEK_SET);
  SimulateIOError( newOffset-- );
  if( newOffset!=offset ){
    return -1;
  }
  got = read(id->h, pBuf, cnt);
#endif
  TIMER_END;
  OSTRACE5("READ    %-3d %5d %7lld %llu\n", id->h, got, offset, TIMER_ELAPSED);
  return got;
}

/*
** Read data from a file into a buffer.  Return SQLITE_OK if all
** bytes were read successfully and SQLITE_IOERR if anything goes
** wrong.
*/
static int unixRead(
  sqlite3_file *id, 
  void *pBuf, 
  int amt,
  sqlite3_int64 offset
){
  int got;
  assert( id );
  got = seekAndRead((unixFile*)id, offset, pBuf, amt);
  if( got==amt ){
    return SQLITE_OK;
  }else if( got<0 ){
    return SQLITE_IOERR_READ;
  }else{
    /* Unread parts of the buffer must be zero-filled */
    memset(&((char*)pBuf)[got], 0, amt-got);
    return SQLITE_IOERR_SHORT_READ;
  }
}

/*
** Seek to the offset in id->offset then read cnt bytes into pBuf.
** Return the number of bytes actually read.  Update the offset.
*/
static int seekAndWrite(unixFile *id, i64 offset, const void *pBuf, int cnt){
  int got;
  i64 newOffset;
  TIMER_START;
#if defined(USE_PREAD)
  got = pwrite(id->h, pBuf, cnt, offset);
#elif defined(USE_PREAD64)
  got = pwrite64(id->h, pBuf, cnt, offset);
#else
  newOffset = lseek(id->h, offset, SEEK_SET);
  if( newOffset!=offset ){
    return -1;
  }
  got = write(id->h, pBuf, cnt);
#endif
  TIMER_END;
  OSTRACE5("WRITE   %-3d %5d %7lld %llu\n", id->h, got, offset, TIMER_ELAPSED);
  return got;
}


/*
** Write data from a buffer into a file.  Return SQLITE_OK on success
** or some other error code on failure.
*/
static int unixWrite(
  sqlite3_file *id, 
  const void *pBuf, 
  int amt,
  sqlite3_int64 offset 
){
  int wrote = 0;
  assert( id );
  assert( amt>0 );
  while( amt>0 && (wrote = seekAndWrite((unixFile*)id, offset, pBuf, amt))>0 ){
    amt -= wrote;
    offset += wrote;
    pBuf = &((char*)pBuf)[wrote];
  }
  SimulateIOError(( wrote=(-1), amt=1 ));
  SimulateDiskfullError(( wrote=0, amt=1 ));
  if( amt>0 ){
    if( wrote<0 ){
      return SQLITE_IOERR_WRITE;
    }else{
      return SQLITE_FULL;
    }
  }
  return SQLITE_OK;
}

#ifdef SQLITE_TEST
/*
** Count the number of fullsyncs and normal syncs.  This is used to test
** that syncs and fullsyncs are occuring at the right times.
*/
int sqlite3_sync_count = 0;
int sqlite3_fullsync_count = 0;
#endif

/*
** Use the fdatasync() API only if the HAVE_FDATASYNC macro is defined.
** Otherwise use fsync() in its place.
*/
#ifndef HAVE_FDATASYNC
# define fdatasync fsync
#endif

/*
** Define HAVE_FULLFSYNC to 0 or 1 depending on whether or not
** the F_FULLFSYNC macro is defined.  F_FULLFSYNC is currently
** only available on Mac OS X.  But that could change.
*/
#ifdef F_FULLFSYNC
# define HAVE_FULLFSYNC 1
#else
# define HAVE_FULLFSYNC 0
#endif


/*
** The fsync() system call does not work as advertised on many
** unix systems.  The following procedure is an attempt to make
** it work better.
**
** The SQLITE_NO_SYNC macro disables all fsync()s.  This is useful
** for testing when we want to run through the test suite quickly.
** You are strongly advised *not* to deploy with SQLITE_NO_SYNC
** enabled, however, since with SQLITE_NO_SYNC enabled, an OS crash
** or power failure will likely corrupt the database file.
*/
static int full_fsync(int fd, int fullSync, int dataOnly){
  int rc;

  /* The following "ifdef/elif/else/" block has the same structure as
  ** the one below. It is replicated here solely to avoid cluttering 
  ** up the real code with the UNUSED_PARAMETER() macros.
  */
#ifdef SQLITE_NO_SYNC
  UNUSED_PARAMETER(fd);
  UNUSED_PARAMETER(fullSync);
  UNUSED_PARAMETER(dataOnly);
#elif HAVE_FULLFSYNC
  UNUSED_PARAMETER(dataOnly);
#else
  UNUSED_PARAMETER(fullSync);
#endif

  /* Record the number of times that we do a normal fsync() and 
  ** FULLSYNC.  This is used during testing to verify that this procedure
  ** gets called with the correct arguments.
  */
#ifdef SQLITE_TEST
  if( fullSync ) sqlite3_fullsync_count++;
  sqlite3_sync_count++;
#endif

  /* If we compiled with the SQLITE_NO_SYNC flag, then syncing is a
  ** no-op
  */
#ifdef SQLITE_NO_SYNC
  rc = SQLITE_OK;
#elif HAVE_FULLFSYNC
  if( fullSync ){
    rc = fcntl(fd, F_FULLFSYNC, 0);
  }else{
    rc = 1;
  }
  /* If the FULLFSYNC failed, fall back to attempting an fsync().
   * It shouldn't be possible for fullfsync to fail on the local 
   * file system (on OSX), so failure indicates that FULLFSYNC
   * isn't supported for this file system. So, attempt an fsync 
   * and (for now) ignore the overhead of a superfluous fcntl call.  
   * It'd be better to detect fullfsync support once and avoid 
   * the fcntl call every time sync is called.
   */
  if( rc ) rc = fsync(fd);

#else 
  if( dataOnly ){
    rc = fdatasync(fd);
    if( IS_VXWORKS && rc==-1 && errno==ENOTSUP ){
      rc = fsync(fd);
    }
  }else{
    rc = fsync(fd);
  }
#endif /* ifdef SQLITE_NO_SYNC elif HAVE_FULLFSYNC */

  if( IS_VXWORKS && rc!= -1 ){
    rc = 0;
  }
  return rc;
}

/*
** Make sure all writes to a particular file are committed to disk.
**
** If dataOnly==0 then both the file itself and its metadata (file
** size, access time, etc) are synced.  If dataOnly!=0 then only the
** file data is synced.
**
** Under Unix, also make sure that the directory entry for the file
** has been created by fsync-ing the directory that contains the file.
** If we do not do this and we encounter a power failure, the directory
** entry for the journal might not exist after we reboot.  The next
** SQLite to access the file will not know that the journal exists (because
** the directory entry for the journal was never created) and the transaction
** will not roll back - possibly leading to database corruption.
*/
static int unixSync(sqlite3_file *id, int flags){
  int rc;
  unixFile *pFile = (unixFile*)id;

  int isDataOnly = (flags&SQLITE_SYNC_DATAONLY);
  int isFullsync = (flags&0x0F)==SQLITE_SYNC_FULL;

  /* Check that one of SQLITE_SYNC_NORMAL or FULL was passed */
  assert((flags&0x0F)==SQLITE_SYNC_NORMAL
      || (flags&0x0F)==SQLITE_SYNC_FULL
  );