SQLite

  struct unixLockInfo *pLock;      /* Info about locks on this inode */
  int h;                           /* The file descriptor */
  int dirfd;                       /* File descriptor for the directory */
  unsigned char locktype;          /* The type of lock held on this fd */
  int lastErrno;                   /* The unix errno from the last I/O error */
  void *lockingContext;            /* Locking style specific state */
  UnixUnusedFd *pUnused;           /* Pre-allocated UnixUnusedFd */
  int fileFlags;                   /* Miscellanous flags */
#if SQLITE_ENABLE_LOCKING_STYLE
  int openFlags;                   /* The flags specified at open() */
#endif
#if SQLITE_THREADSAFE && defined(__linux__)
  pthread_t tid;                   /* The thread that "owns" this unixFile */
#endif
#if OS_VXWORKS

  /* In test mode, increase the size of this structure a bit so that 
  ** it is larger than the struct CrashFile defined in test6.c.
  */
  char aPadding[32];
#endif
};

/*
** The following macros define bits in unixFile.fileFlags
*/
#define SQLITE_WHOLE_FILE_LOCKING  0x0001   /* Use whole-file locking */

/*
** Include code that is common to all os_*.c files
*/
#include "os_common.h"

/*
** Define various macros that are missing from some systems.

  
  unixLeaveMutex();
  OSTRACE4("TEST WR-LOCK %d %d %d\n", pFile->h, rc, reserved);

  *pResOut = reserved;
  return rc;
}

/*
** Perform a file locking operation on a range of bytes in a file.
** The "op" parameter should be one of F_RDLCK, F_WRLCK, or F_UNLCK.
** Return 0 on success or -1 for failure.  On failure, write the error
** code into *pErrcode.
**
** If the SQLITE_WHOLE_FILE_LOCKING bit is clear, then only lock
** the range of bytes on the locking page between SHARED_FIRST and
** SHARED_SIZE.  If SQLITE_WHOLE_FILE_LOCKING is set, then lock all
** bytes from 0 up to but not including PENDING_BYTE, and all bytes
** that follow SHARED_FIRST.
**
** In other words, of SQLITE_WHOLE_FILE_LOCKING if false (the historical
** default case) then only lock a small range of bytes from SHARED_FIRST
** through SHARED_FIRST+SHARED_SIZE-1.  But if SQLITE_WHOLE_FILE_LOCKING is
** true then lock every byte in the file except for PENDING_BYTE and
** RESERVED_BYTE.
**
** SQLITE_WHOLE_FILE_LOCKING=true overlaps SQLITE_WHOLE_FILE_LOCKING=false
** and so the locking schemes are compatible.  One type of lock will
** effectively exclude the other type.  The reason for using the
** SQLITE_WHOLE_FILE_LOCKING=true is that by indicating the full range
** of bytes to be read or written, we give hints to NFS to help it
** maintain cache coherency.  On the other hand, whole file locking
** is slower, so we don't want to use it except for NFS.
*/
static int rangeLock(unixFile *pFile, int op, int *pErrcode){
  struct flock lock;
  int rc;
  lock.l_type = op;
  lock.l_start = SHARED_FIRST;
  lock.l_whence = SEEK_SET;
  if( (pFile->fileFlags & SQLITE_WHOLE_FILE_LOCKING)==0 ){
    lock.l_len = SHARED_SIZE;
    rc = fcntl(pFile->h, F_SETLK, &lock);
    *pErrcode = errno;
  }else{
    lock.l_len = 0;
    rc = fcntl(pFile->h, F_SETLK, &lock);
    *pErrcode = errno;
    if( NEVER(op==F_UNLCK) || rc!=(-1) ){
      lock.l_start = 0;
      lock.l_len = PENDING_BYTE;
      rc = fcntl(pFile->h, F_SETLK, &lock);
      if( ALWAYS(op!=F_UNLCK) && rc==(-1) ){
        *pErrcode = errno;
        lock.l_type = F_UNLCK;
        lock.l_start = SHARED_FIRST;
        lock.l_len = 0;
        fcntl(pFile->h, F_SETLK, &lock);
      }
    }
  }
  return rc;
}

/*
** Lock the file with the lock specified by parameter locktype - one
** of the following:
**
**     (1) SHARED_LOCK
**     (2) RESERVED_LOCK

  ** even if the locking primitive used is always a write-lock.
  */
  int rc = SQLITE_OK;
  unixFile *pFile = (unixFile*)id;
  struct unixLockInfo *pLock = pFile->pLock;
  struct flock lock;
  int s;
  int tErrno;

  assert( pFile );
  OSTRACE7("LOCK    %d %s was %s(%s,%d) pid=%d\n", pFile->h,
      locktypeName(locktype), locktypeName(pFile->locktype),
      locktypeName(pLock->locktype), pLock->cnt , getpid());

  /* If there is already a lock of this type or more restrictive on the
  ** unixFile, do nothing. Don't use the end_lock: exit path, as
  ** unixEnterMutex() 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.
  **  (1) We never move from unlocked to anything higher than shared lock.
  **  (2) SQLite never explicitly requests a pendig lock.
  **  (3) A shared lock is always held when a reserve lock is requested.
  */
  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
  */

    assert( pLock->cnt>0 );
    pFile->locktype = SHARED_LOCK;
    pLock->cnt++;
    pFile->pOpen->nLock++;
    goto end_lock;
  }





  /* 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.
  */
  lock.l_len = 1L;
  lock.l_whence = SEEK_SET;
  if( locktype==SHARED_LOCK 
      || (locktype==EXCLUSIVE_LOCK && pFile->locktype<PENDING_LOCK)
  ){
    lock.l_type = (locktype==SHARED_LOCK?F_RDLCK:F_WRLCK);
    lock.l_start = PENDING_BYTE;
    s = fcntl(pFile->h, F_SETLK, &lock);
    if( s==(-1) ){
      tErrno = errno;
      rc = sqliteErrorFromPosixError(tErrno, SQLITE_IOERR_LOCK);
      if( IS_LOCK_ERROR(rc) ){
        pFile->lastErrno = tErrno;
      }
      goto 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 ){

    assert( pLock->cnt==0 );
    assert( pLock->locktype==0 );

    /* Now get the read-lock */



    s = rangeLock(pFile, F_RDLCK, &tErrno);

    /* Drop the temporary PENDING lock */
    lock.l_start = PENDING_BYTE;
    lock.l_len = 1L;
    lock.l_type = F_UNLCK;
    if( fcntl(pFile->h, F_SETLK, &lock)!=0 ){
      if( s != -1 ){
        /* This could happen with a network mount */

    ** already.
    */
    assert( 0!=pFile->locktype );
    lock.l_type = F_WRLCK;
    switch( locktype ){
      case RESERVED_LOCK:
        lock.l_start = RESERVED_BYTE;
        s = fcntl(pFile->h, F_SETLK, &lock);
        tErrno = errno;
        break;
      case EXCLUSIVE_LOCK:


        s = rangeLock(pFile, F_WRLCK, &tErrno);
        break;
      default:
        assert(0);
    }

    if( s==(-1) ){

      rc = sqliteErrorFromPosixError(tErrno, SQLITE_IOERR_LOCK);
      if( IS_LOCK_ERROR(rc) ){
        pFile->lastErrno = tErrno;
      }
    }
  }
  

** 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){
  unixFile *pFile = (unixFile*)id; /* The open file */
  struct unixLockInfo *pLock;      /* Structure describing current lock state */
  struct flock lock;               /* Information passed into fcntl() */
  int rc = SQLITE_OK;              /* Return code from this interface */

  int h;                           /* The underlying file descriptor */
  int tErrno;                      /* Error code from system call errors */

  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 ){

         || pFile->dbUpdate==0
         || pFile->transCntrChng==1 );
    pFile->inNormalWrite = 0;
#endif


    if( locktype==SHARED_LOCK ){




      if( rangeLock(pFile, F_RDLCK, &tErrno)==(-1) ){

        rc = sqliteErrorFromPosixError(tErrno, SQLITE_IOERR_RDLOCK);
        if( IS_LOCK_ERROR(rc) ){
          pFile->lastErrno = tErrno;
        }
        goto end_unlock;
      }
    }
    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(h, F_SETLK, &lock)!=(-1) ){
      pLock->locktype = SHARED_LOCK;
    }else{
      tErrno = errno;
      rc = sqliteErrorFromPosixError(tErrno, SQLITE_IOERR_UNLOCK);
      if( IS_LOCK_ERROR(rc) ){
        pFile->lastErrno = tErrno;
      }
      goto end_unlock;
    }
  }

      lock.l_start = lock.l_len = 0L;
      SimulateIOErrorBenign(1);
      SimulateIOError( h=(-1) )
      SimulateIOErrorBenign(0);
      if( fcntl(h, F_SETLK, &lock)!=(-1) ){
        pLock->locktype = NO_LOCK;
      }else{
        tErrno = errno;
        rc = sqliteErrorFromPosixError(tErrno, SQLITE_IOERR_UNLOCK);
        if( IS_LOCK_ERROR(rc) ){
          pFile->lastErrno = tErrno;
        }
        pLock->locktype = NO_LOCK;
        pFile->locktype = NO_LOCK;
      }

/******************* End of the no-op lock implementation *********************
******************************************************************************/

/******************************************************************************
************************* Begin dot-file Locking ******************************
**
** The dotfile locking implementation uses the existance of separate lock
** files in order to control access to the database.  This works on just
** about every filesystem imaginable.  But there are serious downsides:
**
**    (1)  There is zero concurrency.  A single reader blocks all other
**         connections from reading or writing the database.
**
**    (2)  An application crash or power loss can leave stale lock files

   LOCK,                       /* xLock */                                   \
   UNLOCK,                     /* xUnlock */                                 \
   CKLOCK,                     /* xCheckReservedLock */                      \
   unixFileControl,            /* xFileControl */                            \
   unixSectorSize,             /* xSectorSize */                             \
   unixDeviceCharacteristics   /* xDeviceCapabilities */                     \
};                                                                           \
static const sqlite3_io_methods *FINDER##Impl(const char *z, unixFile *p){   \
  UNUSED_PARAMETER(z); UNUSED_PARAMETER(p);                                  \
  return &METHOD;                                                            \
}                                                                            \
static const sqlite3_io_methods *(*const FINDER)(const char*,unixFile *p)    \
    = FINDER##Impl;

/*
** Here are all of the sqlite3_io_methods objects for each of the
** locking strategies.  Functions that return pointers to these methods
** are also created.
*/

  afpClose,                 /* xClose method */
  afpLock,                  /* xLock method */
  afpUnlock,                /* xUnlock method */
  afpCheckReservedLock      /* xCheckReservedLock method */
)
#endif

/*
** The "Whole File Locking" finder returns the same set of methods as
** the posix locking finder.  But it also sets the SQLITE_WHOLE_FILE_LOCKING
** flag to force the posix advisory locks to cover the whole file instead
** of just a small span of bytes near the 1GiB boundary.  Whole File Locking
** is useful on NFS-mounted files since it helps NFS to maintain cache
** coherency.  But it is a detriment to other filesystems since it runs
** slower.
*/
static const sqlite3_io_methods *posixWflIoFinderImpl(const char*z, unixFile*p){
  UNUSED_PARAMETER(z);
  p->fileFlags = SQLITE_WHOLE_FILE_LOCKING;
  return &posixIoMethods;
}
static const sqlite3_io_methods 
  *(*const posixWflIoFinder)(const char*,unixFile *p) = posixWflIoFinderImpl;

/*
** The proxy locking method is a "super-method" in the sense that it
** opens secondary file descriptors for the conch and lock files and
** it uses proxy, dot-file, AFP, and flock() locking methods on those
** secondary files.  For this reason, the division that implements
** proxy locking is located much further down in the file.  But we need
** to go ahead and define the sqlite3_io_methods and finder function

** for the database file "filePath".  It then returns the sqlite3_io_methods
** object that implements that strategy.
**
** This is for MacOSX only.
*/
static const sqlite3_io_methods *autolockIoFinderImpl(
  const char *filePath,    /* name of the database file */
  unixFile *pNew           /* open file object for the database file */
){
  static const struct Mapping {
    const char *zFilesystem;              /* Filesystem type name */
    const sqlite3_io_methods *pMethods;   /* Appropriate locking method */
  } aMap[] = {
    { "hfs",    &posixIoMethods },
    { "ufs",    &posixIoMethods },

  ** Test byte-range lock using fcntl(). If the call succeeds, 
  ** assume that the file-system supports POSIX style locks. 
  */
  lockInfo.l_len = 1;
  lockInfo.l_start = 0;
  lockInfo.l_whence = SEEK_SET;
  lockInfo.l_type = F_RDLCK;
  if( fcntl(pNew->h, F_GETLK, &lockInfo)!=-1 ) {
    pNew->fileFlags = SQLITE_WHOLE_FILE_LOCKING;
    return &posixIoMethods;
  }else{
    return &dotlockIoMethods;
  }
}
static const sqlite3_io_methods 
  *(*const autolockIoFinder)(const char*,unixFile*) = autolockIoFinderImpl;

#endif /* defined(__APPLE__) && SQLITE_ENABLE_LOCKING_STYLE */

#if OS_VXWORKS && SQLITE_ENABLE_LOCKING_STYLE
/* 
** This "finder" function attempts to determine the best locking strategy 
** for the database file "filePath".  It then returns the sqlite3_io_methods
** object that implements that strategy.
**
** This is for VXWorks only.
*/
static const sqlite3_io_methods *autolockIoFinderImpl(
  const char *filePath,    /* name of the database file */
  unixFile *pNew           /* the open file object */
){
  struct flock lockInfo;

  if( !filePath ){
    /* If filePath==NULL that means we are dealing with a transient file
    ** that does not need to be locked. */
    return &nolockIoMethods;
  }

  /* Test if fcntl() is supported and use POSIX style locks.
  ** Otherwise fall back to the named semaphore method.
  */
  lockInfo.l_len = 1;
  lockInfo.l_start = 0;
  lockInfo.l_whence = SEEK_SET;
  lockInfo.l_type = F_RDLCK;
  if( fcntl(pNew->h, F_GETLK, &lockInfo)!=-1 ) {
    return &posixIoMethods;
  }else{
    return &semIoMethods;
  }
}
static const sqlite3_io_methods 
  *(*const autolockIoFinder)(const char*,unixFile*) = autolockIoFinderImpl;

#endif /* OS_VXWORKS && SQLITE_ENABLE_LOCKING_STYLE */

/*
** An abstract type for a pointer to a IO method finder function:
*/
typedef const sqlite3_io_methods *(*finder_type)(const char*,unixFile*);


/****************************************************************************
**************************** sqlite3_vfs methods ****************************
**
** This division contains the implementation of methods on the
** sqlite3_vfs object.

  */
  UNUSED_PARAMETER(isDelete);

  OSTRACE3("OPEN    %-3d %s\n", h, zFilename);    
  pNew->h = h;
  pNew->dirfd = dirfd;
  SET_THREADID(pNew);
  pNew->fileFlags = 0;

#if OS_VXWORKS
  pNew->pId = vxworksFindFileId(zFilename);
  if( pNew->pId==0 ){
    noLock = 1;
    rc = SQLITE_NOMEM;
  }
#endif

  if( noLock ){
    pLockingStyle = &nolockIoMethods;
  }else{
    pLockingStyle = (**(finder_type*)pVfs->pAppData)(zFilename, pNew);
#if SQLITE_ENABLE_LOCKING_STYLE
    /* Cache zFilename in the locking context (AFP and dotlock override) for
    ** proxyLock activation is possible (remote proxy is based on db name)
    ** zFilename remains valid until file is closed, to support */
    pNew->lockingContext = (void*)zFilename;
#endif
  }

#if SQLITE_ENABLE_LOCKING_STYLE && (OS_VXWORKS || defined(__APPLE__))
    UNIXVFS("unix",          autolockIoFinder ),
#else
    UNIXVFS("unix",          posixIoFinder ),
#endif
    UNIXVFS("unix-none",     nolockIoFinder ),
    UNIXVFS("unix-dotfile",  dotlockIoFinder ),
    UNIXVFS("unix-wfl",      posixWflIoFinder ),
#if OS_VXWORKS
    UNIXVFS("unix-namedsem", semIoFinder ),
#endif
#if SQLITE_ENABLE_LOCKING_STYLE
    UNIXVFS("unix-posix",    posixIoFinder ),
#if !OS_VXWORKS
    UNIXVFS("unix-flock",    flockIoFinder ),