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(fd, op, p);
    sqlite3DebugPrintf("fcntl unknown %d %d %d\n", fd, op, s);
    return s;
  }
  if( p->l_type==F_RDLCK ){
    zType = "RDLCK";
  }else if( p->l_type==F_WRLCK ){
    zType = "WRLCK";
  }else if( p->l_type==F_UNLCK ){
    zType = "UNLCK";
  }else{
    assert( 0 );
  }
  assert( p->l_whence==SEEK_SET );
  s = fcntl(fd, op, p);
  savedErrno = errno;
  sqlite3DebugPrintf("fcntl %d %d %s %s %d %d %d %d\n",
     threadid, fd, zOpName, zType, (int)p->l_start, (int)p->l_len,
     (int)p->l_pid, s);
  if( s==(-1) && op==F_SETLK && (p->l_type==F_RDLCK || p->l_type==F_WRLCK) ){
    struct flock l2;
    l2 = *p;
    fcntl(fd, F_GETLK, &l2);
    if( l2.l_type==F_RDLCK ){
      zType = "RDLCK";
    }else if( l2.l_type==F_WRLCK ){
      zType = "WRLCK";
    }else if( l2.l_type==F_UNLCK ){
      zType = "UNLCK";
    }else{
      assert( 0 );
    }
    sqlite3DebugPrintf("fcntl-failure-reason: %s %d %d %d\n",
       zType, (int)l2.l_start, (int)l2.l_len, (int)l2.l_pid);
  }
  errno = savedErrno;
  return s;
}
#define fcntl lockTrace
#endif /* SQLITE_LOCK_TRACE */

/*
** The testThreadLockingBehavior() routine launches two separate
** threads on this routine.  This routine attempts to lock a file
** descriptor then returns.  The success or failure of that attempt
** allows the testThreadLockingBehavior() procedure to determine
** whether or not threads can override each others locks.
*/
static void *threadLockingTest(void *pArg){
  struct threadTestData *pData = (struct threadTestData*)pArg;
  pData->result = fcntl(pData->fd, F_SETLK, &pData->lock);
  return pArg;
}

/*
** 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;
  struct threadTestData d[2];
  pthread_t t[2];

  fd = dup(fd_orig);
  if( fd<0 ) return;
  memset(d, 0, sizeof(d));
  d[0].fd = fd;
  d[0].lock.l_type = F_RDLCK;
  d[0].lock.l_len = 1;
  d[0].lock.l_start = 0;
  d[0].lock.l_whence = SEEK_SET;
  d[1] = d[0];
  d[1].lock.l_type = F_WRLCK;
  pthread_create(&t[0], 0, threadLockingTest, &d[0]);
  pthread_create(&t[1], 0, threadLockingTest, &d[1]);
  pthread_join(t[0], 0);
  pthread_join(t[1], 0);
  close(fd);
  threadsOverrideEachOthersLocks =  d[0].result==0 && d[1].result==0;
}
#endif /* SQLITE_THREADSAFE */

/*
** Release a lockInfo structure previously allocated by findLockInfo().
*/
static void releaseLockInfo(struct lockInfo *pLock){
  if (pLock == NULL)
    return;
  pLock->nRef--;
  if( pLock->nRef==0 ){
    sqlite3HashInsert(&lockHash, &pLock->key, sizeof(pLock->key), 0);
    sqlite3_free(pLock);
  }
}

/*
** Release a openCnt structure previously allocated by findLockInfo().
*/
static void releaseOpenCnt(struct openCnt *pOpen){
  if (pOpen == NULL)
    return;
  pOpen->nRef--;
  if( pOpen->nRef==0 ){
    sqlite3HashInsert(&openHash, &pOpen->key, sizeof(pOpen->key), 0);
    free(pOpen->aPending);
    sqlite3_free(pOpen);
  }
}

#ifdef SQLITE_ENABLE_LOCKING_STYLE
/*
** Tests a byte-range locking query to see if byte range locks are 
** supported, if not we fall back to dotlockLockingStyle.
*/
static sqlite3LockingStyle sqlite3TestLockingStyle(
  const char *filePath, 
  int fd
){
  /* test byte-range lock using fcntl */
  struct flock lockInfo;
  
  lockInfo.l_len = 1;
  lockInfo.l_start = 0;
  lockInfo.l_whence = SEEK_SET;
  lockInfo.l_type = F_RDLCK;
  
  if( fcntl(fd, F_GETLK, &lockInfo)!=-1 ) {
    return posixLockingStyle;
  } 
  
  /* testing for flock can give false positives.  So if if the above test
  ** fails, then we fall back to using dot-lock style locking.
  */  
  return dotlockLockingStyle;
}

/* 
** Examines the f_fstypename entry in the statfs structure as returned by 
** stat() for the file system hosting the database file, assigns the 
** appropriate locking style based on it's value.  These values and 
** assignments are based on Darwin/OSX behavior and have not been tested on 
** other systems.
*/
static sqlite3LockingStyle sqlite3DetectLockingStyle(
  const char *filePath, 
  int fd
){

#ifdef SQLITE_FIXED_LOCKING_STYLE
  return (sqlite3LockingStyle)SQLITE_FIXED_LOCKING_STYLE;
#else
  struct statfs fsInfo;

  if (statfs(filePath, &fsInfo) == -1)
    return sqlite3TestLockingStyle(filePath, fd);
  
  if (fsInfo.f_flags & MNT_RDONLY)
    return noLockingStyle;
  
  if( (!strcmp(fsInfo.f_fstypename, "hfs")) ||
    (!strcmp(fsInfo.f_fstypename, "ufs")) )
                return posixLockingStyle;
  
  if(!strcmp(fsInfo.f_fstypename, "afpfs"))
    return afpLockingStyle;
  
  if(!strcmp(fsInfo.f_fstypename, "nfs")) 
    return sqlite3TestLockingStyle(filePath, fd);
  
  if(!strcmp(fsInfo.f_fstypename, "smbfs"))
    return flockLockingStyle;
  
  if(!strcmp(fsInfo.f_fstypename, "msdos"))
    return dotlockLockingStyle;
  
  if(!strcmp(fsInfo.f_fstypename, "webdav"))
    return unsupportedLockingStyle;
  
  return sqlite3TestLockingStyle(filePath, fd);  
#endif /* SQLITE_FIXED_LOCKING_STYLE */
}

#endif /* SQLITE_ENABLE_LOCKING_STYLE */

/*
** Given a file descriptor, locate lockInfo and openCnt structures that
** describes that file descriptor.  Create new ones if necessary.  The
** return values might be uninitialized if an error occurs.
**
** Return the number of errors.
*/
static int findLockInfo(
  int fd,                      /* The file descriptor used in the key */
  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 ) return 1;

  memset(&key1, 0, sizeof(key1));
  key1.dev = statbuf.st_dev;
  key1.ino = statbuf.st_ino;
#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;
  key2.ino = statbuf.st_ino;
  pLock = (struct lockInfo*)sqlite3HashFind(&lockHash, &key1, sizeof(key1));
  if( pLock==0 ){
    struct lockInfo *pOld;
    pLock = sqlite3_malloc( sizeof(*pLock) );
    if( pLock==0 ){
      rc = 1;
      goto exit_findlockinfo;
    }
    pLock->key = key1;
    pLock->nRef = 1;
    pLock->cnt = 0;
    pLock->locktype = 0;
    pOld = sqlite3HashInsert(&lockHash, &pLock->key, sizeof(key1), pLock);
    if( pOld!=0 ){
      assert( pOld==pLock );
      sqlite3_free(pLock);
      rc = 1;
      goto exit_findlockinfo;
    }
  }else{
    pLock->nRef++;
  }
  *ppLock = pLock;
  if( ppOpen!=0 ){
    pOpen = (struct openCnt*)sqlite3HashFind(&openHash, &key2, sizeof(key2));
    if( pOpen==0 ){
      struct openCnt *pOld;
      pOpen = sqlite3_malloc( sizeof(*pOpen) );
      if( pOpen==0 ){
        releaseLockInfo(pLock);
        rc = 1;
        goto exit_findlockinfo;
      }
      pOpen->key = key2;
      pOpen->nRef = 1;
      pOpen->nLock = 0;
      pOpen->nPending = 0;
      pOpen->aPending = 0;
      pOld = sqlite3HashInsert(&openHash, &pOpen->key, sizeof(key2), pOpen);
      if( pOld!=0 ){
        assert( pOld==pOpen );
        sqlite3_free(pOpen);
        releaseLockInfo(pLock);
        rc = 1;
        goto exit_findlockinfo;
      }
    }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);
    rc = findLockInfo(pFile->h, &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
  /* 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.
*/
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 %d\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{
    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 %d\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 );