os_unix.c

这是一个开源的数据库系统,值得学习啊, 里面用了SQL语句,与微软的SQL SERVIER,差不了多少

    s = fcntl(pFile->h, F_SETLK, &lock);
    if( s==(-1) ){
      rc = (errno==EINVAL) ? SQLITE_NOLFS : SQLITE_BUSY;
    }
  }
  
  if( rc==SQLITE_OK ){
    pFile->locktype = locktype;
    pLock->locktype = locktype;
  }else if( locktype==EXCLUSIVE_LOCK ){
    pFile->locktype = PENDING_LOCK;
    pLock->locktype = PENDING_LOCK;
  }

end_lock:
  leaveMutex();
  OSTRACE4("LOCK    %d %s %s\n", pFile->h, locktypeName(locktype), 
      rc==SQLITE_OK ? "ok" : "failed");
  return rc;
}

/*
** 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(sqlite3_file *id, int locktype){
  struct lockInfo *pLock;
  struct flock lock;
  int rc = SQLITE_OK;
  unixFile *pFile = (unixFile*)id;

  assert( pFile );
  OSTRACE7("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;
  }
  enterMutex();
  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;
      lock.l_whence = SEEK_SET;
      lock.l_start = SHARED_FIRST;
      lock.l_len = SHARED_SIZE;
      if( fcntl(pFile->h, F_SETLK, &lock)==(-1) ){
        /* This should never happen */
        rc = SQLITE_IOERR_RDLOCK;
      }
    }
    lock.l_type = F_UNLCK;
    lock.l_whence = SEEK_SET;
    lock.l_start = PENDING_BYTE;
    lock.l_len = 2L;  assert( PENDING_BYTE+1==RESERVED_BYTE );
    if( fcntl(pFile->h, F_SETLK, &lock)!=(-1) ){
      pLock->locktype = SHARED_LOCK;
    }else{
      rc = SQLITE_IOERR_UNLOCK;  /* This should never happen */
    }
  }
  if( locktype==NO_LOCK ){
    struct openCnt *pOpen;

    /* Decrement the shared lock counter.  Release the lock using an
    ** OS call only when all threads in this same process have released
    ** the lock.
    */
    pLock->cnt--;
    if( pLock->cnt==0 ){
      lock.l_type = F_UNLCK;
      lock.l_whence = SEEK_SET;
      lock.l_start = lock.l_len = 0L;
      if( fcntl(pFile->h, F_SETLK, &lock)!=(-1) ){
        pLock->locktype = NO_LOCK;
      }else{
        rc = SQLITE_IOERR_UNLOCK;  /* This should never happen */
      }
    }

    /* Decrement the count of locks against this same file.  When the
    ** count reaches zero, close any other file descriptors whose close
    ** was deferred because of outstanding locks.
    */
    pOpen = pFile->pOpen;
    pOpen->nLock--;
    assert( pOpen->nLock>=0 );
    if( pOpen->nLock==0 && pOpen->nPending>0 ){
      int i;
      for(i=0; i<pOpen->nPending; i++){
        close(pOpen->aPending[i]);
      }
      free(pOpen->aPending);
      pOpen->nPending = 0;
      pOpen->aPending = 0;
    }
  }
  leaveMutex();
  pFile->locktype = locktype;
  return rc;
}

/*
** Close a file.
*/
static int unixClose(sqlite3_file *id){
  unixFile *pFile = (unixFile *)id;
  if( !pFile ) return SQLITE_OK;
  unixUnlock(id, NO_LOCK);
  if( pFile->dirfd>=0 ) close(pFile->dirfd);
  pFile->dirfd = -1;
  enterMutex();

  if( pFile->pOpen->nLock ){
    /* If there are outstanding locks, do not actually close the file just
    ** yet because that would clear those locks.  Instead, add the file
    ** descriptor to pOpen->aPending.  It will be automatically closed when
    ** the last lock is cleared.
    */
    int *aNew;
    struct openCnt *pOpen = pFile->pOpen;
    aNew = realloc( pOpen->aPending, (pOpen->nPending+1)*sizeof(int) );
    if( aNew==0 ){
      /* If a malloc fails, just leak the file descriptor */
    }else{
      pOpen->aPending = aNew;
      pOpen->aPending[pOpen->nPending] = pFile->h;
      pOpen->nPending++;
    }
  }else{
    /* There are no outstanding locks so we can close the file immediately */
    close(pFile->h);
  }
  releaseLockInfo(pFile->pLock);
  releaseOpenCnt(pFile->pOpen);

  leaveMutex();
  pFile->isOpen = 0;
  OSTRACE2("CLOSE   %-3d\n", pFile->h);
  OpenCounter(-1);
  memset(pFile, 0, sizeof(unixFile));
  return SQLITE_OK;
}


#ifdef SQLITE_ENABLE_LOCKING_STYLE
#pragma mark AFP Support

/*
 ** The afpLockingContext structure contains all afp lock specific state
 */
typedef struct afpLockingContext afpLockingContext;
struct afpLockingContext {
  unsigned long long sharedLockByte;
  char *filePath;
};

struct ByteRangeLockPB2
{
  unsigned long long offset;        /* offset to first byte to lock */
  unsigned long long length;        /* nbr of bytes to lock */
  unsigned long long retRangeStart; /* nbr of 1st byte locked if successful */
  unsigned char unLockFlag;         /* 1 = unlock, 0 = lock */
  unsigned char startEndFlag;       /* 1=rel to end of fork, 0=rel to start */
  int fd;                           /* file desc to assoc this lock with */
};

#define afpfsByteRangeLock2FSCTL        _IOWR('z', 23, struct ByteRangeLockPB2)

/* 
** Return 0 on success, 1 on failure.  To match the behavior of the 
** normal posix file locking (used in unixLock for example), we should 
** provide 'richer' return codes - specifically to differentiate between
** 'file busy' and 'file system error' results.
*/
static int _AFPFSSetLock(
  const char *path, 
  int fd, 
  unsigned long long offset, 
  unsigned long long length, 
  int setLockFlag
){
  struct ByteRangeLockPB2       pb;
  int                     err;
  
  pb.unLockFlag = setLockFlag ? 0 : 1;
  pb.startEndFlag = 0;
  pb.offset = offset;
  pb.length = length; 
  pb.fd = fd;
  OSTRACE5("AFPLOCK setting lock %s for %d in range %llx:%llx\n", 
    (setLockFlag?"ON":"OFF"), fd, offset, length);
  err = fsctl(path, afpfsByteRangeLock2FSCTL, &pb, 0);
  if ( err==-1 ) {
    OSTRACE4("AFPLOCK failed to fsctl() '%s' %d %s\n", path, errno, 
      strerror(errno));
    return 1; /* error */
  } else {
    return 0;
  }
}

/*
 ** 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, return
 ** non-zero.  If the file is unlocked or holds only SHARED locks, then
 ** return zero.
 */
static int afpUnixCheckReservedLock(sqlite3_file *id){
  int r = 0;
  unixFile *pFile = (unixFile*)id;
  
  assert( pFile ); 
  afpLockingContext *context = (afpLockingContext *) pFile->lockingContext;
  
  /* Check if a thread in this process holds such a lock */
  if( pFile->locktype>SHARED_LOCK ){
    r = 1;
  }
  
  /* Otherwise see if some other process holds it.
   */
  if ( !r ) {
    /* lock the byte */
    int failed = _AFPFSSetLock(context->filePath, pFile->h, RESERVED_BYTE, 1,1);  
    if (failed) {
      /* if we failed to get the lock then someone else must have it */
      r = 1;
    } else {
      /* if we succeeded in taking the reserved lock, unlock it to restore
      ** the original state */
      _AFPFSSetLock(context->filePath, pFile->h, RESERVED_BYTE, 1, 0);
    }
  }
  OSTRACE3("TEST WR-LOCK %d %d\n", pFile->h, r);
  
  return r;
}

/* AFP-style locking following the behavior of unixLock, see the unixLock 
** function comments for details of lock management. */
static int afpUnixLock(sqlite3_file *id, int locktype)
{
  int rc = SQLITE_OK;
  unixFile *pFile = (unixFile*)id;
  afpLockingContext *context = (afpLockingContext *) pFile->lockingContext;
  int gotPendingLock = 0;
  
  assert( pFile );
  OSTRACE5("LOCK    %d %s was %s pid=%d\n", pFile->h,
         locktypeName(locktype), locktypeName(pFile->locktype), getpid());  
  /* If there is already a lock of this type or more restrictive on the
    ** unixFile, do nothing. Don't use the afp_end_lock: exit path, as
    ** enterMutex() hasn't been called yet.
    */
  if( pFile->locktype>=locktype ){
    OSTRACE3("LOCK    %d %s ok (already held)\n", pFile->h,
           locktypeName(locktype));
    return SQLITE_OK;
  }

  /* Make sure the locking sequence is correct
    */
  assert( pFile->locktype!=NO_LOCK || locktype==SHARED_LOCK );
  assert( locktype!=PENDING_LOCK );
  assert( locktype!=RESERVED_LOCK || pFile->locktype==SHARED_LOCK );
  
  /* This mutex is needed because pFile->pLock is shared across threads
    */
  enterMutex();

  /* Make sure the current thread owns the pFile.
    */
  rc = transferOwnership(pFile);
  if( rc!=SQLITE_OK ){
    leaveMutex();
    return rc;
  }
    
  /* A PENDING lock is needed before acquiring a SHARED lock and before
    ** acquiring an EXCLUSIVE lock.  For the SHARED lock, the PENDING will
    ** be released.
    */
  if( locktype==SHARED_LOCK 
      || (locktype==EXCLUSIVE_LOCK && pFile->locktype<PENDING_LOCK)
      ){
    int failed = _AFPFSSetLock(context->filePath, pFile->h, 
      PENDING_BYTE, 1, 1);
    if (failed) {
      rc = SQLITE_BUSY;
      goto afp_end_lock;
    }
  }
  
  /* If control gets to this point, then actually go ahead and make
    ** operating system calls for the specified lock.
    */
  if( locktype==SHARED_LOCK ){
    int lk, failed;
    int tries = 0;
    
    /* Now get the read-lock */
    /* note that the quality of the randomness doesn't matter that much */
    lk = random(); 
    context->sharedLockByte = (lk & 0x7fffffff)%(SHARED_SIZE - 1);
    failed = _AFPFSSetLock(context->filePath, pFile->h, 
      SHARED_FIRST+context->sharedLockByte, 1, 1);
    
    /* Drop the temporary PENDING lock */
    if (_AFPFSSetLock(context->filePath, pFile->h, PENDING_BYTE, 1, 0)) {
      rc = SQLITE_IOERR_UNLOCK;  /* This should never happen */
      goto afp_end_lock;
    }
    
    if( failed ){
      rc = SQLITE_BUSY;
    } else {
      pFile->locktype = SHARED_LOCK;
    }
  }else{
    /* The request was for a RESERVED or EXCLUSIVE lock.  It is
    ** assumed that there is a SHARED or greater lock on the file
    ** already.
    */
    int failed = 0;
    assert( 0!=pFile->locktype );
    if (locktype >= RESERVED_LOCK && pFile->locktype < RESERVED_LOCK) {
        /* Acquire a RESERVED lock */
        failed = _AFPFSSetLock(context->filePath, pFile->h, RESERVED_BYTE, 1,1);
    }
    if (!failed && locktype == EXCLUSIVE_LOCK) {
      /* Acquire an EXCLUSIVE lock */
        
      /* Remove the shared lock before trying the range.  we'll need to 
      ** reestablish the shared lock if we can't get the  afpUnixUnlock
      */
      if (!_AFPFSSetLock(context->filePath, pFile->h, SHARED_FIRST +
                         context->sharedLockByte, 1, 0)) {
        /* now attemmpt to get the exclusive lock range */
        failed = _AFPFSSetLock(context->filePath, pFile->h, SHARED_FIRST, 
                               SHARED_SIZE, 1);
        if (failed && _AFPFSSetLock(context->filePath, pFile->h, SHARED_FIRST +
                                    context->sharedLockByte, 1, 1)) {
          rc = SQLITE_IOERR_RDLOCK; /* this should never happen */
        }
      } else {
        /* */
        rc = SQLITE_IOERR_UNLOCK; /* this should never happen */
      }
    }
    if( failed && rc == SQLITE_OK){
      rc = SQLITE_BUSY;
    }
  }
  
  if( rc==SQLITE_OK ){
    pFile->locktype = locktype;
  }else if( locktype==EXCLUSIVE_LOCK ){
    pFile->locktype = PENDING_LOCK;
  }
  
afp_end_lock:
    leaveMutex();
  OSTRACE4("LOCK    %d %s %s\n", pFile->h, locktypeName(locktype), 
         rc==SQLITE_OK ? "ok" : "failed");
  return rc;
}

/*
 ** 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 afpUnixUnlock(sqlite3_file *id, int locktype) {
  struct flock lock;
  int rc = SQLITE_OK;
  unixFile *pFile = (unixFile*)id;
  afpLockingContext *context = (afpLockingContext *) pFile->lockingContext;

  assert( pFile );
  OSTRACE5("UNLOCK  %d %d was %d pid=%d\n", pFile->h, locktype,
         pFile->locktype, getpid());
  
  assert( locktype<=SHARED_LOCK );
  if( pFile->locktype<=locktype ){
    return SQLITE_OK;
  }
  if( CHECK_THREADID(pFile) ){
    return SQLITE_MISUSE;
  }
  enterMutex();
  if( pFile->locktype>SHARED_LOCK ){
    if( locktype==SHARED_LOCK ){
      int failed = 0;

      /* unlock the exclusive range - then re-establish the shared lock */
      if (pFile->locktype==EXCLUSIVE_LOCK) {
        failed = _AFPFSSetLock(context->filePath, pFile->h, SHARED_FIRST, 
                                 SHARED_SIZE, 0);
        if (!failed) {
          /* successfully removed the exclusive lock */
          if (_AFPFSSetLock(context->filePath, pFile->h, SHARED_FIRST+
                            context->sharedLockByte, 1, 1)) {
            /* failed to re-establish our shared lock */
            rc = SQLITE_IOERR_RDLOCK; /* This should never happen */
          }
        } else {
          /* This should never happen - failed to unlock the exclusive range */
          rc = SQLITE_IOERR_UNLOCK;
        } 
      }
    }
    if (rc == SQLITE_OK && pFile->locktype>=PENDING_LOCK) {
      if (_AFPFSSetLock(context->filePath, pFile->h, PENDING_BYTE, 1, 0)){
        /* failed to release the pending lock */
        rc = SQLITE_IOERR_UNLOCK; /* This should never happen */
      }
    } 
    if (rc == SQLITE_OK && pFile->locktype>=RESERVED_LOCK) {