os_unix.c

sqlite database for embed linux

    h = open(zFilename, O_RDONLY|O_LARGEFILE|O_BINARY);
    if( h<0 ){
      return SQLITE_CANTOPEN; 
    }
    *pReadonly = 1;
  }else{
    *pReadonly = 0;
  }
  return allocateUnixFile(h, pId, zFilename, 0);
}


/*
** Attempt to open a new file for exclusive access by this process.
** The file will be opened for both reading and writing.  To avoid
** a potential security problem, we do not allow the file to have
** previously existed.  Nor do we allow the file to be a symbolic
** link.
**
** If delFlag is true, then make arrangements to automatically delete
** the file when it is closed.
**
** On success, write the file handle into *id and return SQLITE_OK.
**
** On failure, return SQLITE_CANTOPEN.
*/
int sqlite3UnixOpenExclusive(const char *zFilename, OsFile **pId, int delFlag){
  int h;

  CRASH_TEST_OVERRIDE(sqlite3CrashOpenExclusive, zFilename, pId, delFlag);
  assert( 0==*pId );
  h = open(zFilename,
                O_RDWR|O_CREAT|O_EXCL|O_NOFOLLOW|O_LARGEFILE|O_BINARY,
                delFlag ? 0600 : SQLITE_DEFAULT_FILE_PERMISSIONS);
  if( h<0 ){
    return SQLITE_CANTOPEN;
  }
  return allocateUnixFile(h, pId, zFilename, delFlag);
}

/*
** Attempt to open a new file for read-only access.
**
** On success, write the file handle into *id and return SQLITE_OK.
**
** On failure, return SQLITE_CANTOPEN.
*/
int sqlite3UnixOpenReadOnly(const char *zFilename, OsFile **pId){
  int h;
  
  CRASH_TEST_OVERRIDE(sqlite3CrashOpenReadOnly, zFilename, pId, 0);
  assert( 0==*pId );
  h = open(zFilename, O_RDONLY|O_LARGEFILE|O_BINARY);
  if( h<0 ){
    return SQLITE_CANTOPEN;
  }
  return allocateUnixFile(h, pId, zFilename, 0);
}

/*
** Attempt to open a file descriptor for the directory that contains a
** file.  This file descriptor can be used to fsync() the directory
** in order to make sure the creation of a new file is actually written
** to disk.
**
** This routine is only meaningful for Unix.  It is a no-op under
** windows since windows does not support hard links.
**
** If FULL_FSYNC is enabled, this function is not longer useful, 
** a FULL_FSYNC sync applies to all pending disk operations.
**
** On success, a handle for a previously open file at *id is
** updated with the new directory file descriptor and SQLITE_OK is
** returned.
**
** On failure, the function returns SQLITE_CANTOPEN and leaves
** *id unchanged.
*/
static int unixOpenDirectory(
  OsFile *id,
  const char *zDirname
){
  unixFile *pFile = (unixFile*)id;
  assert( pFile!=0 );
  SET_THREADID(pFile);
  assert( pFile->dirfd<0 );
  pFile->dirfd = open(zDirname, O_RDONLY|O_BINARY, 0);
  if( pFile->dirfd<0 ){
    return SQLITE_CANTOPEN; 
  }
  OSTRACE3("OPENDIR %-3d %s\n", pFile->dirfd, zDirname);
  return SQLITE_OK;
}

/*
** Create a temporary file name in zBuf.  zBuf must be big enough to
** hold at least SQLITE_TEMPNAME_SIZE characters.
*/
int sqlite3UnixTempFileName(char *zBuf){
  static const char *azDirs[] = {
     0,
     "/var/tmp",
     "/usr/tmp",
     "/tmp",
     ".",
  };
  static const unsigned char zChars[] =
    "abcdefghijklmnopqrstuvwxyz"
    "ABCDEFGHIJKLMNOPQRSTUVWXYZ"
    "0123456789";
  int i, j;
  struct stat buf;
  const char *zDir = ".";
  azDirs[0] = sqlite3_temp_directory;
  for(i=0; i<sizeof(azDirs)/sizeof(azDirs[0]); i++){
    if( azDirs[i]==0 ) continue;
    if( stat(azDirs[i], &buf) ) continue;
    if( !S_ISDIR(buf.st_mode) ) continue;
    if( access(azDirs[i], 07) ) continue;
    zDir = azDirs[i];
    break;
  }
  do{
    sprintf(zBuf, "%s/"TEMP_FILE_PREFIX, zDir);
    j = strlen(zBuf);
    sqlite3Randomness(15, &zBuf[j]);
    for(i=0; i<15; i++, j++){
      zBuf[j] = (char)zChars[ ((unsigned char)zBuf[j])%(sizeof(zChars)-1) ];
    }
    zBuf[j] = 0;
  }while( access(zBuf,0)==0 );
  return SQLITE_OK; 
}

/*
** Check that a given pathname is a directory and is writable 
**
*/
int sqlite3UnixIsDirWritable(char *zBuf){
#ifndef SQLITE_OMIT_PAGER_PRAGMAS
  struct stat buf;
  if( zBuf==0 ) return 0;
  if( zBuf[0]==0 ) return 0;
  if( stat(zBuf, &buf) ) return 0;
  if( !S_ISDIR(buf.st_mode) ) return 0;
  if( access(zBuf, 07) ) return 0;
#endif /* SQLITE_OMIT_PAGER_PRAGMAS */
  return 1;
}

/*
** 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 seekAndRead(unixFile *id, void *pBuf, int cnt){
  int got;
  i64 newOffset;
  TIMER_START;
#if defined(USE_PREAD)
  got = pread(id->h, pBuf, cnt, id->offset);
  SimulateIOError( got = -1 );
#elif defined(USE_PREAD64)
  got = pread64(id->h, pBuf, cnt, id->offset);
  SimulateIOError( got = -1 );
#else
  newOffset = lseek(id->h, id->offset, SEEK_SET);
  SimulateIOError( newOffset-- );
  if( newOffset!=id->offset ){
    return -1;
  }
  got = read(id->h, pBuf, cnt);
#endif
  TIMER_END;
  OSTRACE5("READ    %-3d %5d %7lld %d\n", id->h, got, id->offset, TIMER_ELAPSED);
  if( got>0 ){
    id->offset += got;
  }
  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(OsFile *id, void *pBuf, int amt){
  int got;
  assert( id );
  got = seekAndRead((unixFile*)id, pBuf, amt);
  if( got==amt ){
    return SQLITE_OK;
  }else if( got<0 ){
    return SQLITE_IOERR_READ;
  }else{
    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, const void *pBuf, int cnt){
  int got;
  i64 newOffset;
  TIMER_START;
#if defined(USE_PREAD)
  got = pwrite(id->h, pBuf, cnt, id->offset);
#elif defined(USE_PREAD64)
  got = pwrite64(id->h, pBuf, cnt, id->offset);
#else
  newOffset = lseek(id->h, id->offset, SEEK_SET);
  if( newOffset!=id->offset ){
    return -1;
  }
  got = write(id->h, pBuf, cnt);
#endif
  TIMER_END;
  OSTRACE5("WRITE   %-3d %5d %7lld %d\n", id->h, got, id->offset, TIMER_ELAPSED);
  if( got>0 ){
    id->offset += got;
  }
  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(OsFile *id, const void *pBuf, int amt){
  int wrote = 0;
  assert( id );
  assert( amt>0 );
  while( amt>0 && (wrote = seekAndWrite((unixFile*)id, pBuf, amt))>0 ){
    amt -= 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;
}

/*
** Move the read/write pointer in a file.
*/
static int unixSeek(OsFile *id, i64 offset){
  assert( id );
#ifdef SQLITE_TEST
  if( offset ) SimulateDiskfullError(return SQLITE_FULL);
#endif
  ((unixFile*)id)->offset = offset;
  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;

  /* 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;
#else

#if 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);
  }else{
    rc = fsync(fd);
  }
#endif /* HAVE_FULLFSYNC */
#endif /* defined(SQLITE_NO_SYNC) */

  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(OsFile *id, int dataOnly){
  int rc;
  unixFile *pFile = (unixFile*)id;
  assert( pFile );
  OSTRACE2("SYNC    %-3d\n", pFile->h);
  rc = full_fsync(pFile->h, pFile->fullSync, dataOnly);
  SimulateIOError( rc=1 );
  if( rc ){
    return SQLITE_IOERR_FSYNC;
  }
  if( pFile->dirfd>=0 ){
    OSTRACE4("DIRSYNC %-3d (have_fullfsync=%d fullsync=%d)\n", pFile->dirfd,
            HAVE_FULLFSYNC, pFile->fullSync);
#ifndef SQLITE_DISABLE_DIRSYNC
    /* The directory sync is only attempted if full_fsync is
    ** turned off or unavailable.  If a full_fsync occurred above,
    ** then the directory sync is superfluous.
    */
    if( (!HAVE_FULLFSYNC || !pFile->fullSync) && full_fsync(pFile->dirfd,0,0) ){
       /*
       ** We have received multiple reports of fsync() returning
       ** errors when applied to directories on certain file systems.
       ** A failed directory sync is not a big deal.  So it seems
       ** better to ignore the error.  Ticket #1657
       */
       /* return SQLITE_IOERR; */
    }
#endif
    close(pFile->dirfd);  /* Only need to sync once, so close the directory */
    pFile->dirfd = -1;    /* when we are done. */
  }
  return SQLITE_OK;
}

/*
** Sync the directory zDirname. This is a no-op on operating systems other
** than UNIX.
**
** This is used to make sure the master journal file has truely been deleted
** before making changes to individual journals on a multi-database commit.
** The F_FULLFSYNC option is not needed here.
*/
int sqlite3UnixSyncDirectory(const char *zDirname){
#ifdef SQLITE_DISABLE_DIRSYNC
  return SQLITE_OK;
#else
  int fd;
  int r;
  fd = open(zDirname, O_RDONLY|O_BINARY, 0);
  OSTRACE3("DIRSYNC %-3d (%s)\n", fd, zDirname);
  if( fd<0 ){
    return SQLITE_CANTOPEN; 
  }
  r = fsync(fd);