os.c

sqlite数据库源码

    if( aNew==0 ){
      /* If a malloc fails, just leak the file descriptor */
    }else{
      pOpen->aPending = aNew;
      pOpen->aPending[pOpen->nPending-1] = id->fd;
    }
  }else{
    /* There are no outstanding locks so we can close the file immediately */
    close(id->fd);
  }
  releaseLockInfo(id->pLock);
  releaseOpenCnt(id->pOpen);
  sqliteOsLeaveMutex();
  TRACE2("CLOSE   %-3d\n", id->fd);
  OpenCounter(-1);
  return SQLITE_OK;
#endif
#if OS_WIN
  CloseHandle(id->h);
  OpenCounter(-1);
  return SQLITE_OK;
#endif
#if OS_MAC
  if( id->refNumRF!=-1 )
    FSClose(id->refNumRF);
# ifdef _LARGE_FILE
  FSCloseFork(id->refNum);
# else
  FSClose(id->refNum);
# endif
  if( id->delOnClose ){
    unlink(id->pathToDel);
    sqliteFree(id->pathToDel);
  }
  OpenCounter(-1);
  return SQLITE_OK;
#endif
}

/*
** Read data from a file into a buffer.  Return SQLITE_OK if all
** bytes were read successfully and SQLITE_IOERR if anything goes
** wrong.
*/
int sqliteOsRead(OsFile *id, void *pBuf, int amt){
#if OS_UNIX
  int got;
  SimulateIOError(SQLITE_IOERR);
  TIMER_START;
  got = read(id->fd, pBuf, amt);
  TIMER_END;
  TRACE4("READ    %-3d %7d %d\n", id->fd, last_page, elapse);
  SEEK(0);
  /* if( got<0 ) got = 0; */
  if( got==amt ){
    return SQLITE_OK;
  }else{
    return SQLITE_IOERR;
  }
#endif
#if OS_WIN
  DWORD got;
  SimulateIOError(SQLITE_IOERR);
  TRACE2("READ %d\n", last_page);
  if( !ReadFile(id->h, pBuf, amt, &got, 0) ){
    got = 0;
  }
  if( got==(DWORD)amt ){
    return SQLITE_OK;
  }else{
    return SQLITE_IOERR;
  }
#endif
#if OS_MAC
  int got;
  SimulateIOError(SQLITE_IOERR);
  TRACE2("READ %d\n", last_page);
# ifdef _LARGE_FILE
  FSReadFork(id->refNum, fsAtMark, 0, (ByteCount)amt, pBuf, (ByteCount*)&got);
# else
  got = amt;
  FSRead(id->refNum, &got, pBuf);
# endif
  if( got==amt ){
    return SQLITE_OK;
  }else{
    return SQLITE_IOERR;
  }
#endif
}

/*
** Write data from a buffer into a file.  Return SQLITE_OK on success
** or some other error code on failure.
*/
int sqliteOsWrite(OsFile *id, const void *pBuf, int amt){
#if OS_UNIX
  int wrote = 0;
  SimulateIOError(SQLITE_IOERR);
  TIMER_START;
  while( amt>0 && (wrote = write(id->fd, pBuf, amt))>0 ){
    amt -= wrote;
    pBuf = &((char*)pBuf)[wrote];
  }
  TIMER_END;
  TRACE4("WRITE   %-3d %7d %d\n", id->fd, last_page, elapse);
  SEEK(0);
  if( amt>0 ){
    return SQLITE_FULL;
  }
  return SQLITE_OK;
#endif
#if OS_WIN
  int rc;
  DWORD wrote;
  SimulateIOError(SQLITE_IOERR);
  TRACE2("WRITE %d\n", last_page);
  while( amt>0 && (rc = WriteFile(id->h, pBuf, amt, &wrote, 0))!=0 && wrote>0 ){
    amt -= wrote;
    pBuf = &((char*)pBuf)[wrote];
  }
  if( !rc || amt>(int)wrote ){
    return SQLITE_FULL;
  }
  return SQLITE_OK;
#endif
#if OS_MAC
  OSErr oserr;
  int wrote = 0;
  SimulateIOError(SQLITE_IOERR);
  TRACE2("WRITE %d\n", last_page);
  while( amt>0 ){
# ifdef _LARGE_FILE
    oserr = FSWriteFork(id->refNum, fsAtMark, 0,
                        (ByteCount)amt, pBuf, (ByteCount*)&wrote);
# else
    wrote = amt;
    oserr = FSWrite(id->refNum, &wrote, pBuf);
# endif
    if( wrote == 0 || oserr != noErr)
      break;
    amt -= wrote;
    pBuf = &((char*)pBuf)[wrote];
  }
  if( oserr != noErr || amt>wrote ){
    return SQLITE_FULL;
  }
  return SQLITE_OK;
#endif
}

/*
** Move the read/write pointer in a file.
*/
int sqliteOsSeek(OsFile *id, off_t offset){
  SEEK(offset/1024 + 1);
#if OS_UNIX
  lseek(id->fd, offset, SEEK_SET);
  return SQLITE_OK;
#endif
#if OS_WIN
  {
    LONG upperBits = offset>>32;
    LONG lowerBits = offset & 0xffffffff;
    DWORD rc;
    rc = SetFilePointer(id->h, lowerBits, &upperBits, FILE_BEGIN);
    /* TRACE3("SEEK rc=0x%x upper=0x%x\n", rc, upperBits); */
  }
  return SQLITE_OK;
#endif
#if OS_MAC
  {
    off_t curSize;
    if( sqliteOsFileSize(id, &curSize) != SQLITE_OK ){
      return SQLITE_IOERR;
    }
    if( offset >= curSize ){
      if( sqliteOsTruncate(id, offset+1) != SQLITE_OK ){
        return SQLITE_IOERR;
      }
    }
# ifdef _LARGE_FILE
    if( FSSetForkPosition(id->refNum, fsFromStart, offset) != noErr ){
# else
    if( SetFPos(id->refNum, fsFromStart, offset) != noErr ){
# endif
      return SQLITE_IOERR;
    }else{
      return SQLITE_OK;
    }
  }
#endif
}

/*
** Make sure all writes to a particular file are committed to disk.
**
** 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.
*/
int sqliteOsSync(OsFile *id){
#if OS_UNIX
  SimulateIOError(SQLITE_IOERR);
  TRACE2("SYNC    %-3d\n", id->fd);
  if( fsync(id->fd) ){
    return SQLITE_IOERR;
  }else{
    if( id->dirfd>=0 ){
      TRACE2("DIRSYNC %-3d\n", id->dirfd);
      fsync(id->dirfd);
      close(id->dirfd);  /* Only need to sync once, so close the directory */
      id->dirfd = -1;    /* when we are done. */
    }
    return SQLITE_OK;
  }
#endif
#if OS_WIN
  if( FlushFileBuffers(id->h) ){
    return SQLITE_OK;
  }else{
    return SQLITE_IOERR;
  }
#endif
#if OS_MAC
# ifdef _LARGE_FILE
  if( FSFlushFork(id->refNum) != noErr ){
# else
  ParamBlockRec params;
  memset(&params, 0, sizeof(ParamBlockRec));
  params.ioParam.ioRefNum = id->refNum;
  if( PBFlushFileSync(&params) != noErr ){
# endif
    return SQLITE_IOERR;
  }else{
    return SQLITE_OK;
  }
#endif
}

/*
** Truncate an open file to a specified size
*/
int sqliteOsTruncate(OsFile *id, off_t nByte){
  SimulateIOError(SQLITE_IOERR);
#if OS_UNIX
  return ftruncate(id->fd, nByte)==0 ? SQLITE_OK : SQLITE_IOERR;
#endif
#if OS_WIN
  {
    LONG upperBits = nByte>>32;
    SetFilePointer(id->h, nByte, &upperBits, FILE_BEGIN);
    SetEndOfFile(id->h);
  }
  return SQLITE_OK;
#endif
#if OS_MAC
# ifdef _LARGE_FILE
  if( FSSetForkSize(id->refNum, fsFromStart, nByte) != noErr){
# else
  if( SetEOF(id->refNum, nByte) != noErr ){
# endif
    return SQLITE_IOERR;
  }else{
    return SQLITE_OK;
  }
#endif
}

/*
** Determine the current size of a file in bytes
*/
int sqliteOsFileSize(OsFile *id, off_t *pSize){
#if OS_UNIX
  struct stat buf;
  SimulateIOError(SQLITE_IOERR);
  if( fstat(id->fd, &buf)!=0 ){
    return SQLITE_IOERR;
  }
  *pSize = buf.st_size;
  return SQLITE_OK;
#endif
#if OS_WIN
  DWORD upperBits, lowerBits;
  SimulateIOError(SQLITE_IOERR);
  lowerBits = GetFileSize(id->h, &upperBits);
  *pSize = (((off_t)upperBits)<<32) + lowerBits;
  return SQLITE_OK;
#endif
#if OS_MAC
# ifdef _LARGE_FILE
  if( FSGetForkSize(id->refNum, pSize) != noErr){
# else
  if( GetEOF(id->refNum, pSize) != noErr ){
# endif
    return SQLITE_IOERR;
  }else{
    return SQLITE_OK;
  }
#endif
}

#if OS_WIN
/*
** Return true (non-zero) if we are running under WinNT, Win2K or WinXP.
** Return false (zero) for Win95, Win98, or WinME.
**
** Here is an interesting observation:  Win95, Win98, and WinME lack
** the LockFileEx() API.  But we can still statically link against that
** API as long as we don't call it win running Win95/98/ME.  A call to
** this routine is used to determine if the host is Win95/98/ME or
** WinNT/2K/XP so that we will know whether or not we can safely call
** the LockFileEx() API.
*/
int isNT(void){
  static int osType = 0;   /* 0=unknown 1=win95 2=winNT */
  if( osType==0 ){
    OSVERSIONINFO sInfo;
    sInfo.dwOSVersionInfoSize = sizeof(sInfo);
    GetVersionEx(&sInfo);
    osType = sInfo.dwPlatformId==VER_PLATFORM_WIN32_NT ? 2 : 1;
  }
  return osType==2;
}
#endif

/*
** Windows file locking notes:  [similar issues apply to MacOS]
**
** We cannot use LockFileEx() or UnlockFileEx() on Win95/98/ME because
** those functions are not available.  So we use only LockFile() and
** UnlockFile().
**
** LockFile() prevents not just writing but also reading by other processes.
** (This is a design error on the part of Windows, but there is nothing
** we can do about that.)  So the region used for locking is at the
** end of the file where it is unlikely to ever interfere with an
** actual read attempt.
**
** A database read lock is obtained by locking a single randomly-chosen 
** byte out of a specific range of bytes. The lock byte is obtained at 
** random so two separate readers can probably access the file at the 
** same time, unless they are unlucky and choose the same lock byte.
** A database write lock is obtained by locking all bytes in the range.
** There can only be one writer.
**
** A lock is obtained on the first byte of the lock range before acquiring
** either a read lock or a write lock.  This prevents two processes from
** attempting to get a lock at a same time.  The semantics of 
** sqliteOsReadLock() require that if there is already a write lock, that
** lock is converted into a read lock atomically.  The lock on the first
** byte allows us to drop the old write lock and get the read lock without
** another process jumping into the middle and messing us up.  The same
** argument applies to sqliteOsWriteLock().
**
** On WinNT/2K/XP systems, LockFileEx() and UnlockFileEx() are available,
** which means we can use reader/writer locks.  When reader writer locks
** are used, the lock is placed on the same range of bytes that is used
** for probabilistic locking in Win95/98/ME.  Hence, the locking scheme
** will support two or more Win95 readers or two or more WinNT readers.
** But a single Win95 reader will lock out all WinNT readers and a single
** WinNT reader will lock out all other Win95 readers.
**
** Note: On MacOS we use the resource fork for locking.
**
** The following #defines specify the range of bytes used for locking.
** N_LOCKBYTE is the number of bytes available for doing the locking.
** The first byte used to hold the lock while the lock is changing does
** not count toward this number.  FIRST_LOCKBYTE is the address of
** the first byte in the range of bytes used for locking.
*/
#define N_LOCKBYTE       10239
#if OS_MAC
# define FIRST_LOCKBYTE   (0x000fffff - N_LOCKBYTE)
#else
# define FIRST_LOCKBYTE   (0xffffffff - N_LOCKBYTE)
#endif

/*
** Change the status of the lock on the file "id" to be a readlock.
** If the file was write locked, then this reduces the lock to a read.
** If the file was read locked, then this acquires a new read lock.
**
** Return SQLITE_OK on success and SQLITE_BUSY on failure.  If this
** library was compiled with large file support (LFS) but LFS is not
** available on the host, then an SQLITE_NOLFS is returned.
*/
int sqliteOsReadLock(OsFile *id){
#if OS_UNIX
  int rc;
  sqliteOsEnterMutex();
  if( id->pLock->cnt>0 ){
    if( !id->locked ){
      id->pLock->cnt++;
      id->locked = 1;
      id->pOpen->nLock++;
    }
    rc = SQLITE_OK;
  }else if( id->locked || id->pLock->cnt==0 ){
    struct flock lock;
    int s;
    lock.l_type = F_RDLCK;
    lock.l_whence = SEEK_SET;
    lock.l_start = lock.l_len = 0L;
    s = fcntl(id->fd, F_SETLK, &lock);
    if( s!=0 ){
      rc = (errno==EINVAL) ? SQLITE_NOLFS : SQLITE_BUSY;
    }else{
      rc = SQLITE_OK;
      if( !id->locked ){
        id->pOpen->nLock++;
        id->locked = 1;
      }
      id->pLock->cnt = 1;
    }
  }else{
    rc = SQLITE_BUSY;
  }
  sqliteOsLeaveMutex();
  return rc;
#endif
#if OS_WIN
  int rc;
  if( id->locked>0 ){
    rc = SQLITE_OK;
  }else{
    int lk;
    int res;
    int cnt = 100;
    sqliteRandomness(sizeof(lk), &lk);
    lk = (lk & 0x7fffffff)%N_LOCKBYTE + 1;
    while( cnt-->0 && (res = LockFile(id->h, FIRST_LOCKBYTE, 0, 1, 0))==0 ){
      Sleep(1);
    }
    if( res ){
      UnlockFile(id->h, FIRST_LOCKBYTE+1, 0, N_LOCKBYTE, 0);
      if( isNT() ){
        OVERLAPPED ovlp;
        ovlp.Offset = FIRST_LOCKBYTE+1;
        ovlp.OffsetHigh = 0;
        ovlp.hEvent = 0;
        res = LockFileEx(id->h, LOCKFILE_FAIL_IMMEDIATELY, 
                          0, N_LOCKBYTE, 0, &ovlp);
      }else{
        res = LockFile(id->h, FIRST_LOCKBYTE+lk, 0, 1, 0);
      }
      UnlockFile(id->h, FIRST_LOCKBYTE, 0, 1, 0);
    }
    if( res ){
      id->locked = lk;
      rc = SQLITE_OK;
    }else{
      rc = SQLITE_BUSY;
    }
  }
  return rc;
#endif
#if OS_MAC