os_unix.c

sqlite最新源码

  int fd;                /* File to be locked */
  struct flock lock;     /* The locking operation */
  int result;            /* Result of the locking operation */
};

#if SQLITE_THREADSAFE && defined(__linux__)
/*
** This function is used as the main routine for a thread launched by
** testThreadLockingBehavior(). It tests whether the shared-lock obtained
** by the main thread in testThreadLockingBehavior() conflicts with a
** hypothetical write-lock obtained by this thread on the same file.
**
** The write-lock is not actually acquired, as this is not possible if 
** the file is open in read-only mode (see ticket #3472).
*/ 
static void *threadLockingTest(void *pArg){
  struct threadTestData *pData = (struct threadTestData*)pArg;
  pData->result = fcntl(pData->fd, F_GETLK, &pData->lock);
  return pArg;
}
#endif /* SQLITE_THREADSAFE && defined(__linux__) */


#if SQLITE_THREADSAFE && defined(__linux__)
/*
** This procedure attempts to determine whether or not threads
** can override each others locks then sets the 
** threadsOverrideEachOthersLocks variable appropriately.
*/
static void testThreadLockingBehavior(int fd_orig){
  int fd;
  int rc;
  struct threadTestData d;
  struct flock l;
  pthread_t t;

  fd = dup(fd_orig);
  if( fd<0 ) return;
  memset(&l, 0, sizeof(l));
  l.l_type = F_RDLCK;
  l.l_len = 1;
  l.l_start = 0;
  l.l_whence = SEEK_SET;
  rc = fcntl(fd_orig, F_SETLK, &l);
  if( rc!=0 ) return;
  memset(&d, 0, sizeof(d));
  d.fd = fd;
  d.lock = l;
  d.lock.l_type = F_WRLCK;
  pthread_create(&t, 0, threadLockingTest, &d);
  pthread_join(t, 0);
  close(fd);
  if( d.result!=0 ) return;
  threadsOverrideEachOthersLocks = (d.lock.l_type==F_UNLCK);
}
#endif /* SQLITE_THERADSAFE && defined(__linux__) */

/*
** Release a unixLockInfo structure previously allocated by findLockInfo().
*/
static void releaseLockInfo(struct unixLockInfo *pLock){
  if( pLock ){
    pLock->nRef--;
    if( pLock->nRef==0 ){
      if( pLock->pPrev ){
        assert( pLock->pPrev->pNext==pLock );
        pLock->pPrev->pNext = pLock->pNext;
      }else{
        assert( lockList==pLock );
        lockList = pLock->pNext;
      }
      if( pLock->pNext ){
        assert( pLock->pNext->pPrev==pLock );
        pLock->pNext->pPrev = pLock->pPrev;
      }
      sqlite3_free(pLock);
    }
  }
}

/*
** Release a unixOpenCnt structure previously allocated by findLockInfo().
*/
static void releaseOpenCnt(struct unixOpenCnt *pOpen){
  if( pOpen ){
    pOpen->nRef--;
    if( pOpen->nRef==0 ){
      if( pOpen->pPrev ){
        assert( pOpen->pPrev->pNext==pOpen );
        pOpen->pPrev->pNext = pOpen->pNext;
      }else{
        assert( openList==pOpen );
        openList = pOpen->pNext;
      }
      if( pOpen->pNext ){
        assert( pOpen->pNext->pPrev==pOpen );
        pOpen->pNext->pPrev = pOpen->pPrev;
      }
      sqlite3_free(pOpen->aPending);
      sqlite3_free(pOpen);
    }
  }
}

/*
** Given a file descriptor, locate unixLockInfo and unixOpenCnt structures that
** describes that file descriptor.  Create new ones if necessary.  The
** return values might be uninitialized if an error occurs.
**
** Return an appropriate error code.
*/
static int findLockInfo(
  unixFile *pFile,               /* Unix file with file desc used in the key */
  struct unixLockInfo **ppLock,  /* Return the unixLockInfo structure here */
  struct unixOpenCnt **ppOpen    /* Return the unixOpenCnt structure here */
){
  int rc;                        /* System call return code */
  int fd;                        /* The file descriptor for pFile */
  struct unixLockKey lockKey;    /* Lookup key for the unixLockInfo structure */
  struct unixFileId fileId;      /* Lookup key for the unixOpenCnt struct */
  struct stat statbuf;           /* Low-level file information */
  struct unixLockInfo *pLock;    /* Candidate unixLockInfo object */
  struct unixOpenCnt *pOpen;     /* Candidate unixOpenCnt object */

  /* Get low-level information about the file that we can used to
  ** create a unique name for the file.
  */
  fd = pFile->h;
  rc = fstat(fd, &statbuf);
  if( rc!=0 ){
    pFile->lastErrno = errno;
#ifdef EOVERFLOW
    if( pFile->lastErrno==EOVERFLOW ) return SQLITE_NOLFS;
#endif
    return SQLITE_IOERR;
  }

#ifdef __APPLE__
  /* 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 ){
    rc = write(fd, "S", 1);
    if( rc!=1 ){
      return SQLITE_IOERR;
    }
    rc = fstat(fd, &statbuf);
    if( rc!=0 ){
      pFile->lastErrno = errno;
      return SQLITE_IOERR;
    }
  }
#endif

  memset(&lockKey, 0, sizeof(lockKey));
  lockKey.fid.dev = statbuf.st_dev;
#if OS_VXWORKS
  lockKey.fid.pId = pFile->pId;
#else
  lockKey.fid.ino = statbuf.st_ino;
#endif
#if SQLITE_THREADSAFE && defined(__linux__)
  if( threadsOverrideEachOthersLocks<0 ){
    testThreadLockingBehavior(fd);
  }
  lockKey.tid = threadsOverrideEachOthersLocks ? 0 : pthread_self();
#endif
  fileId = lockKey.fid;
  if( ppLock!=0 ){
    pLock = lockList;
    while( pLock && memcmp(&lockKey, &pLock->lockKey, sizeof(lockKey)) ){
      pLock = pLock->pNext;
    }
    if( pLock==0 ){
      pLock = sqlite3_malloc( sizeof(*pLock) );
      if( pLock==0 ){
        rc = SQLITE_NOMEM;
        goto exit_findlockinfo;
      }
      pLock->lockKey = lockKey;
      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(&fileId, &pOpen->fileId, sizeof(fileId)) ){
      pOpen = pOpen->pNext;
    }
    if( pOpen==0 ){
      pOpen = sqlite3_malloc( sizeof(*pOpen) );
      if( pOpen==0 ){
        releaseLockInfo(pLock);
        rc = SQLITE_NOMEM;
        goto exit_findlockinfo;
      }
      pOpen->fileId = fileId;
      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 OS_VXWORKS
      pOpen->pSem = NULL;
      pOpen->aSemName[0] = '\0';
#endif
    }else{
      pOpen->nRef++;
    }
    *ppOpen = pOpen;
  }

exit_findlockinfo:
  return rc;
}

/*
** 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 that use LinuxThreads.
**
** 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 && defined(__linux__)
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);
    rc = findLockInfo(pFile, &pFile->pLock, 0);
    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  /* if not SQLITE_THREADSAFE */
  /* On single-threaded builds, ownership transfer is a no-op */
# define transferOwnership(X) SQLITE_OK
#endif /* SQLITE_THREADSAFE */


/*
** This routine checks if there is a RESERVED lock held on the specified
** file by this or any other process. If such a lock is held, set *pResOut
** to a non-zero value otherwise *pResOut is set to zero.  The return value
** is set to SQLITE_OK unless an I/O error occurs during lock checking.
*/
static int unixCheckReservedLock(sqlite3_file *id, int *pResOut){
  int rc = SQLITE_OK;
  int reserved = 0;
  unixFile *pFile = (unixFile*)id;

  SimulateIOError( return SQLITE_IOERR_CHECKRESERVEDLOCK; );

  assert( pFile );
  unixEnterMutex(); /* Because pFile->pLock is shared across threads */

  /* Check if a thread in this process holds such a lock */
  if( pFile->pLock->locktype>SHARED_LOCK ){
    reserved = 1;
  }

  /* Otherwise see if some other process holds it.
  */
#ifndef __DJGPP__
  if( !reserved ){
    struct flock lock;
    lock.l_whence = SEEK_SET;
    lock.l_start = RESERVED_BYTE;
    lock.l_len = 1;
    lock.l_type = F_WRLCK;
    if (-1 == fcntl(pFile->h, F_GETLK, &lock)) {
      int tErrno = errno;
      rc = sqliteErrorFromPosixError(tErrno, SQLITE_IOERR_CHECKRESERVEDLOCK);
      pFile->lastErrno = tErrno;
    } else if( lock.l_type!=F_UNLCK ){
      reserved = 1;
    }
  }
#endif
  
  unixLeaveMutex();
  OSTRACE4("TEST WR-LOCK %d %d %d\n", pFile->h, rc, reserved);

  *pResOut = reserved;
  return rc;
}

/*
** Lock the file with the lock specified by parameter locktype - one
** of the following:
**
**     (1) SHARED_LOCK
**     (2) RESERVED_LOCK
**     (3) PENDING_LOCK
**     (4) EXCLUSIVE_LOCK
**
** Sometimes when requesting one lock state, additional lock states
** are inserted in between.  The locking might fail on one of the later
** transitions leaving the lock state different from what it started but
** still short of its goal.  The following chart shows the allowed
** transitions and the inserted intermediate states:
**
**    UNLOCKED -> SHARED
**    SHARED -> RESERVED
**    SHARED -> (PENDING) -> EXCLUSIVE
**    RESERVED -> (PENDING) -> EXCLUSIVE
**    PENDING -> EXCLUSIVE
**
** This routine will only increase a lock.  Use the sqlite3OsUnlock()
** routine to lower a locking level.
*/
static int unixLock(sqlite3_file *id, int locktype){
  /* The following describes the implementation of the various locks and
  ** lock transitions in terms of the POSIX advisory shared and exclusive
  ** lock primitives (called read-locks and write-locks below, to avoid
  ** confusion with SQLite lock names). The algorithms are complicated
  ** slightly in order to be compatible with windows systems simultaneously
  ** accessing the same database file, in case that is ever required.
  **
  ** Symbols defined in os.h indentify the 'pending byte' and the 'reserved
  ** byte', each single bytes at well known offsets, and the 'shared byte
  ** range', a range of 510 bytes at a well known offset.
  **
  ** To obtain a SHARED lock, a read-lock is obtained on the 'pending
  ** byte'.  If this is successful, a random byte from the 'shared byte
  ** range' is read-locked and the lock on the 'pending byte' released.
  **
  ** A process may only obtain a RESERVED lock after it has a SHARED lock.
  ** A RESERVED lock is implemented by grabbing a write-lock on the
  ** 'reserved byte'. 
  **
  ** A process may only obtain a PENDING lock after it has obtained a
  ** SHARED lock. A PENDING lock is implemented by obtaining a write-lock
  ** on the 'pending byte'. This ensures that no new SHARED locks can be
  ** obtained, but existing SHARED locks are allowed to persist. A process
  ** does not have to obtain a RESERVED lock on the way to a PENDING lock.
  ** This property is used by the algorithm for rolling back a journal file
  ** after a crash.
  **
  ** An EXCLUSIVE lock, obtained after a PENDING lock is held, is
  ** implemented by obtaining a write-lock on the entire 'shared byte
  ** range'. Since all other locks require a read-lock on one of the bytes
  ** within this range, this ensures that no other locks are held on the
  ** database. 
  **
  ** The reason a single byte cannot be used instead of the 'shared byte
  ** range' is that some versions of windows do not support read-locks. By
  ** locking a random byte from a range, concurrent SHARED locks may exist
  ** even if the locking primitive used is always a write-lock.
  */
  int rc = SQLITE_OK;
  unixFile *pFile = (unixFile*)id;
  struct unixLockInfo *pLock = pFile->pLock;