#ifndef O_NOFOLLOW
# define O_NOFOLLOW 0
#endif
#ifndef O_BINARY
# define O_BINARY 0
#endif











/*
** The threadid macro resolves to the thread-id or to 0.  Used for
** testing and debugging only.
*/
#if SQLITE_THREADSAFE
#define threadid pthread_self()
#else
#define threadid 0
#endif

/*
** Many system calls are accessed through pointer-to-functions so that
** they may be overridden at runtime to facilitate fault injection during
** testing and sandboxing.  The following array holds the names and pointers
** to all overrideable system calls.
*/
static struct unix_syscall {
  const char *zName;      /* Name of the sytem call */
  void *pCurrent;         /* Current value of the system call */
  void *pDefault;         /* Default value */
} aSyscall[] = {
  { "open",         (void*)open,       0  },
#define osOpen      ((int(*)(const char*,int,int))aSyscall[0].pCurrent)

  { "close",        (void*)close,      0  },
#define osClose     ((int(*)(int))aSyscall[1].pCurrent)

  { "access",       (void*)access,     0  },
#define osAccess    ((int(*)(const char*,int))aSyscall[2].pCurrent)

  { "getcwd",       (void*)getcwd,     0  },
#define osGetcwd    ((char*(*)(char*,size_t))aSyscall[3].pCurrent)

  { "stat",         (void*)stat,       0  },
#define osStat      ((int(*)(const char*,struct stat*))aSyscall[4].pCurrent)

/*
** The DJGPP compiler environment looks mostly like Unix, but it
** lacks the fcntl() system call.  So redefine fcntl() to be something
** that always succeeds.  This means that locking does not occur under
** DJGPP.  But it is DOS - what did you expect?
*/
#ifdef __DJGPP__
  { "fstat",        0,                 0  },
#define osFstat(a,b,c)    0
#else     
  { "fstat",        (void*)fstat,      0  },
#define osFstat     ((int(*)(int,struct stat*))aSyscall[5].pCurrent)
#endif

  { "ftruncate",    (void*)ftruncate,  0  },
#define osFtruncate ((int(*)(int,off_t))aSyscall[6].pCurrent)

  { "fcntl",        (void*)fcntl,      0  },
#define osFcntl     ((int(*)(int,int,...))aSyscall[7].pCurrent)

  { "read",         (void*)read,       0  },
#define osRead      ((ssize_t(*)(int,void*,size_t))aSyscall[8].pCurrent)

#if defined(USE_PREAD) || defined(SQLITE_ENABLE_LOCKING_STYLE)
  { "pread",        (void*)pread,      0  },
#else
  { "pread",        (void*)0,          0  },
#endif
#define osPread     ((ssize_t(*)(int,void*,size_t,off_t))aSyscall[9].pCurrent)

#if defined(USE_PREAD64)
  { "pread64",      (void*)pread64,    0  },
#else
  { "pread64",      (void*)0,          0  },
#endif
#define osPread64   ((ssize_t(*)(int,void*,size_t,off_t))aSyscall[10].pCurrent)

  { "write",        (void*)write,      0  },
#define osWrite     ((ssize_t(*)(int,const void*,size_t))aSyscall[11].pCurrent)

#if defined(USE_PREAD) || defined(SQLITE_ENABLE_LOCKING_STYLE)
  { "pwrite",       (void*)pwrite,     0  },
#else
  { "pwrite",       (void*)0,          0  },
#endif
#define osPwrite    ((ssize_t(*)(int,const void*,size_t,off_t))\
                    aSyscall[12].pCurrent)

#if defined(USE_PREAD64)
  { "pwrite64",     (void*)pwrite64,   0  },
#else
  { "pwrite64",     (void*)0,          0  },
#endif
#define osPwrite64  ((ssize_t(*)(int,const void*,size_t,off_t))\
                    aSyscall[13].pCurrent)

  { "fchmod",       (void*)fchmod,     0  },
#define osFchmod    ((int(*)(int,mode_t))aSyscall[14].pCurrent)

#if defined(HAVE_POSIX_FALLOCATE) && HAVE_POSIX_FALLOCATE
  { "fallocate",    (void*)posix_fallocate,  0 },
#else
  { "fallocate",    (void*)0,                0 },
#endif
#define osFallocate ((int(*)(int,off_t,off_t)aSyscall[15].pCurrent)

}; /* End of the overrideable system calls */

/*
** This is the xSetSystemCall() method of sqlite3_vfs for all of the
** "unix" VFSes.  Return SQLITE_OK opon successfully updating the
** system call pointer, or SQLITE_NOTFOUND if there is no configurable
** system call named zName.
*/
static int unixSetSystemCall(
  sqlite3_vfs *pNotUsed,     /* The VFS pointer.  Not used */
  const char *zName,         /* Name of system call to override */
  void *pNewFunc             /* Pointer to new system call value */
){
  int i;
  int rc = SQLITE_NOTFOUND;
  if( zName==0 ){
    /* If no zName is given, restore all system calls to their default
    ** settings and return NULL
    */
    for(i=0; i<sizeof(aSyscall)/sizeof(aSyscall[0]); i++){
      if( aSyscall[i].pDefault ){
        aSyscall[i].pCurrent = aSyscall[i].pDefault;
        rc = SQLITE_OK;
      }
    }
  }else{
    /* If zName is specified, operate on only the one system call
    ** specified.
    */
    for(i=0; i<sizeof(aSyscall)/sizeof(aSyscall[0]); i++){
      if( strcmp(zName, aSyscall[i].zName)==0 ){
        if( aSyscall[i].pDefault==0 ){
          aSyscall[i].pDefault = aSyscall[i].pCurrent;
        }
        rc = SQLITE_OK;
        if( pNewFunc==0 ) pNewFunc = aSyscall[i].pDefault;
        aSyscall[i].pCurrent = pNewFunc;
        break;
      }
    }
  }
  return rc;
}

/*
** Return the value of a system call.  Return NULL if zName is not a
** recognized system call name.  NULL is also returned if the system call
** is currently undefined.
*/
static void *unixGetSystemCall(sqlite3_vfs *pNotUsed, const char *zName){
  int i;
  for(i=0; i<sizeof(aSyscall)/sizeof(aSyscall[0]); i++){
    if( strcmp(zName, aSyscall[i].zName)==0 ) return aSyscall[i].pCurrent;
  }
  return 0;
}

/*
** Return the name of the first system call after zName.  If zName==NULL
** then return the name of the first system call.  Return NULL if zName
** is the last system call or if zName is not the name of a valid
** system call.
*/
static const char *unixNextSystemCall(sqlite3_vfs *p, const char *zName){
  int i;
  if( zName==0 ){
    i = -1;
  }else{
    for(i=0; i<sizeof(aSyscall)/sizeof(aSyscall[0])-1; i++){
      if( strcmp(zName, aSyscall[0].zName)==0 ) break;
    }
  }
  for(i++; i<sizeof(aSyscall)/sizeof(aSyscall[0]); i++){
    if( aSyscall[0].pCurrent!=0 ) return aSyscall[0].zName;
  }
  return 0;
}

/*
** Retry open() calls that fail due to EINTR
*/
static int robust_open(const char *z, int f, int m){
  int rc;
  do{ rc = osOpen(z,f,m); }while( rc<0 && errno==EINTR );
  return rc;
}

/*
** Helper functions to obtain and relinquish the global mutex. The
** global mutex is used to protect the unixInodeInfo and
** vxworksFileId objects used by this file, all of which may be 
** shared by multiple threads.
**

  int s;
  int savedErrno;
  if( op==F_GETLK ){
    zOpName = "GETLK";
  }else if( op==F_SETLK ){
    zOpName = "SETLK";
  }else{
    s = osFcntl(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 = osFcntl(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;
    osFcntl(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;
}
#undef osFcntl
#define osFcntl lockTrace
#endif /* SQLITE_LOCK_TRACE */


/*
** Retry ftruncate() calls that fail due to EINTR
*/

static int robust_ftruncate(int h, sqlite3_int64 sz){
  int rc;
  do{ rc = osFtruncate(h,sz); }while( rc<0 && errno==EINTR );
  return rc;
}





/*
** This routine translates a standard POSIX errno code into something
** useful to the clients of the sqlite3 functions.  Specifically, it is
** intended to translate a variety of "try again" errors into SQLITE_BUSY
** and a variety of "please close the file descriptor NOW" errors into 
** SQLITE_IOERR

**
** Note that it is not safe to retry close() after EINTR since the
** file descriptor might have already been reused by another thread.
** So we don't even try to recover from an EINTR.  Just log the error
** and move on.
*/
static void robust_close(unixFile *pFile, int h, int lineno){
  if( osClose(h) ){
    unixLogErrorAtLine(SQLITE_IOERR_CLOSE, "close",
                       pFile ? pFile->zPath : 0, lineno);
  }
}

/*
** Close all file descriptors accumuated in the unixInodeInfo->pUnused list.

  assert( unixMutexHeld() );

  /* Get low-level information about the file that we can used to
  ** create a unique name for the file.
  */
  fd = pFile->h;
  rc = osFstat(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 && (pFile->fsFlags & SQLITE_FSFLAGS_IS_MSDOS)!=0 ){
    do{ rc = osWrite(fd, "S", 1); }while( rc<0 && errno==EINTR );
    if( rc!=1 ){
      pFile->lastErrno = errno;
      return SQLITE_IOERR;
    }
    rc = osFstat(fd, &statbuf);
    if( rc!=0 ){
      pFile->lastErrno = errno;
      return SQLITE_IOERR;
    }
  }
#endif

#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 == osFcntl(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;
    }
  }

  lock.l_len = 1L;
  lock.l_whence = SEEK_SET;
  if( eFileLock==SHARED_LOCK 
      || (eFileLock==EXCLUSIVE_LOCK && pFile->eFileLock<PENDING_LOCK)
  ){
    lock.l_type = (eFileLock==SHARED_LOCK?F_RDLCK:F_WRLCK);
    lock.l_start = PENDING_BYTE;
    s = osFcntl(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( eFileLock==SHARED_LOCK ){
    assert( pInode->nShared==0 );
    assert( pInode->eFileLock==0 );

    /* Now get the read-lock */
    lock.l_start = SHARED_FIRST;
    lock.l_len = SHARED_SIZE;
    if( (s = osFcntl(pFile->h, F_SETLK, &lock))==(-1) ){
      tErrno = errno;
    }
    /* Drop the temporary PENDING lock */
    lock.l_start = PENDING_BYTE;
    lock.l_len = 1L;
    lock.l_type = F_UNLCK;
    if( osFcntl(pFile->h, F_SETLK, &lock)!=0 ){
      if( s != -1 ){
        /* This could happen with a network mount */
        tErrno = errno; 
        rc = sqliteErrorFromPosixError(tErrno, SQLITE_IOERR_UNLOCK); 
        if( IS_LOCK_ERROR(rc) ){
          pFile->lastErrno = tErrno;
        }

      case EXCLUSIVE_LOCK:
        lock.l_start = SHARED_FIRST;
        lock.l_len = SHARED_SIZE;
        break;
      default:
        assert(0);
    }
    s = osFcntl(pFile->h, F_SETLK, &lock);
    if( s==(-1) ){
      tErrno = errno;
      rc = sqliteErrorFromPosixError(tErrno, SQLITE_IOERR_LOCK);
      if( IS_LOCK_ERROR(rc) ){
        pFile->lastErrno = tErrno;
      }
    }

      if( handleNFSUnlock ){
        off_t divSize = SHARED_SIZE - 1;
        
        lock.l_type = F_UNLCK;
        lock.l_whence = SEEK_SET;
        lock.l_start = SHARED_FIRST;
        lock.l_len = divSize;
        if( osFcntl(h, F_SETLK, &lock)==(-1) ){
          tErrno = errno;
          rc = sqliteErrorFromPosixError(tErrno, SQLITE_IOERR_UNLOCK);
          if( IS_LOCK_ERROR(rc) ){
            pFile->lastErrno = tErrno;
          }
          goto end_unlock;
        }
        lock.l_type = F_RDLCK;
        lock.l_whence = SEEK_SET;
        lock.l_start = SHARED_FIRST;
        lock.l_len = divSize;
        if( osFcntl(h, F_SETLK, &lock)==(-1) ){
          tErrno = errno;
          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 = SHARED_FIRST+divSize;
        lock.l_len = SHARED_SIZE-divSize;
        if( osFcntl(h, F_SETLK, &lock)==(-1) ){
          tErrno = errno;
          rc = sqliteErrorFromPosixError(tErrno, SQLITE_IOERR_UNLOCK);
          if( IS_LOCK_ERROR(rc) ){
            pFile->lastErrno = tErrno;
          }
          goto end_unlock;
        }
      }else
#endif /* defined(__APPLE__) && SQLITE_ENABLE_LOCKING_STYLE */
      {
        lock.l_type = F_RDLCK;
        lock.l_whence = SEEK_SET;
        lock.l_start = SHARED_FIRST;
        lock.l_len = SHARED_SIZE;
        if( osFcntl(h, F_SETLK, &lock)==(-1) ){
          tErrno = errno;
          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( osFcntl(h, F_SETLK, &lock)!=(-1) ){
      pInode->eFileLock = SHARED_LOCK;
    }else{
      tErrno = errno;
      rc = sqliteErrorFromPosixError(tErrno, SQLITE_IOERR_UNLOCK);
      if( IS_LOCK_ERROR(rc) ){
        pFile->lastErrno = tErrno;
      }

    if( pInode->nShared==0 ){
      lock.l_type = F_UNLCK;
      lock.l_whence = SEEK_SET;
      lock.l_start = lock.l_len = 0L;
      SimulateIOErrorBenign(1);
      SimulateIOError( h=(-1) )
      SimulateIOErrorBenign(0);
      if( osFcntl(h, F_SETLK, &lock)!=(-1) ){
        pInode->eFileLock = NO_LOCK;
      }else{
        tErrno = errno;
        rc = sqliteErrorFromPosixError(tErrno, SQLITE_IOERR_UNLOCK);
        if( IS_LOCK_ERROR(rc) ){
          pFile->lastErrno = tErrno;
        }

  if( pFile->eFileLock>SHARED_LOCK ){
    /* Either this connection or some other connection in the same process
    ** holds a lock on the file.  No need to check further. */
    reserved = 1;
  }else{
    /* The lock is held if and only if the lockfile exists */
    const char *zLockFile = (const char*)pFile->lockingContext;
    reserved = osAccess(zLockFile, 0)==0;
  }
  OSTRACE(("TEST WR-LOCK %d %d %d (dotlock)\n", pFile->h, rc, reserved));
  *pResOut = reserved;
  return rc;
}

/*

    /* Always update the timestamp on the old file */
    utimes(zLockFile, NULL);
#endif
    return SQLITE_OK;
  }
  
  /* grab an exclusive lock */
  fd = robust_open(zLockFile,O_RDONLY|O_CREAT|O_EXCL,0600);
  if( fd<0 ){
    /* failed to open/create the file, someone else may have stolen the lock */
    int tErrno = errno;
    if( EEXIST == tErrno ){
      rc = SQLITE_BUSY;
    } else {
      rc = sqliteErrorFromPosixError(tErrno, SQLITE_IOERR_LOCK);

static int seekAndRead(unixFile *id, sqlite3_int64 offset, void *pBuf, int cnt){
  int got;
#if (!defined(USE_PREAD) && !defined(USE_PREAD64))
  i64 newOffset;
#endif
  TIMER_START;
#if defined(USE_PREAD)
  do{ got = osPread(id->h, pBuf, cnt, offset); }while( got<0 && errno==EINTR );
  SimulateIOError( got = -1 );
#elif defined(USE_PREAD64)
  do{ got = osPread64(id->h, pBuf, cnt, offset); }while( got<0 && errno==EINTR);
  SimulateIOError( got = -1 );
#else
  newOffset = lseek(id->h, offset, SEEK_SET);
  SimulateIOError( newOffset-- );
  if( newOffset!=offset ){
    if( newOffset == -1 ){
      ((unixFile*)id)->lastErrno = errno;
    }else{
      ((unixFile*)id)->lastErrno = 0;			
    }
    return -1;
  }
  do{ got = osRead(id->h, pBuf, cnt); }while( got<0 && errno==EINTR );
#endif
  TIMER_END;
  if( got<0 ){
    ((unixFile*)id)->lastErrno = errno;
  }
  OSTRACE(("READ    %-3d %5d %7lld %llu\n", id->h, got, offset, TIMER_ELAPSED));
  return got;

static int seekAndWrite(unixFile *id, i64 offset, const void *pBuf, int cnt){
  int got;
#if (!defined(USE_PREAD) && !defined(USE_PREAD64))
  i64 newOffset;
#endif
  TIMER_START;
#if defined(USE_PREAD)
  do{ got = osPwrite(id->h, pBuf, cnt, offset); }while( got<0 && errno==EINTR );
#elif defined(USE_PREAD64)
  do{ got = osPwrite64(id->h, pBuf, cnt, offset);}while( got<0 && errno==EINTR);
#else
  newOffset = lseek(id->h, offset, SEEK_SET);
  if( newOffset!=offset ){
    if( newOffset == -1 ){
      ((unixFile*)id)->lastErrno = errno;
    }else{
      ((unixFile*)id)->lastErrno = 0;			
    }
    return -1;
  }
  do{ got = osWrite(id->h, pBuf, cnt); }while( got<0 && errno==EINTR );
#endif
  TIMER_END;
  if( got<0 ){
    ((unixFile*)id)->lastErrno = errno;
  }

  OSTRACE(("WRITE   %-3d %5d %7lld %llu\n", id->h, got, offset, TIMER_ELAPSED));

  /* If we compiled with the SQLITE_NO_SYNC flag, then syncing is a
  ** no-op
  */
#ifdef SQLITE_NO_SYNC
  rc = SQLITE_OK;
#elif HAVE_FULLFSYNC
  if( fullSync ){
    rc = osFcntl(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 

/*
** Determine the current size of a file in bytes
*/
static int unixFileSize(sqlite3_file *id, i64 *pSize){
  int rc;
  struct stat buf;
  assert( id );
  rc = osFstat(((unixFile*)id)->h, &buf);
  SimulateIOError( rc=1 );
  if( rc!=0 ){
    ((unixFile*)id)->lastErrno = errno;
    return SQLITE_IOERR_FSTAT;
  }
  *pSize = buf.st_size;

** SQLITE_FCNTL_SIZE_HINT operation is a no-op for Unix.
*/
static int fcntlSizeHint(unixFile *pFile, i64 nByte){
  if( pFile->szChunk ){
    i64 nSize;                    /* Required file size */
    struct stat buf;              /* Used to hold return values of fstat() */
   
    if( osFstat(pFile->h, &buf) ) return SQLITE_IOERR_FSTAT;

    nSize = ((nByte+pFile->szChunk-1) / pFile->szChunk) * pFile->szChunk;
    if( nSize>(i64)buf.st_size ){
#if defined(HAVE_POSIX_FALLOCATE) && HAVE_POSIX_FALLOCATE
      int rc;
      do{
        rc = osFallocate(pFile->.h, buf.st_size, nSize-buf.st_size;
      }while( rc<0 && errno=EINTR );
      if( rc ) return SQLITE_IOERR_WRITE;
#else
      /* If the OS does not have posix_fallocate(), fake it. First use
      ** ftruncate() to set the file size, then write a single byte to
      ** the last byte in each block within the extended region. This
      ** is the same technique used by glibc to implement posix_fallocate()

  /* Initialize the locking parameters */
  memset(&f, 0, sizeof(f));
  f.l_type = lockType;
  f.l_whence = SEEK_SET;
  f.l_start = ofst;
  f.l_len = n;

  rc = osFcntl(pShmNode->h, F_SETLK, &f);
  rc = (rc!=(-1)) ? SQLITE_OK : SQLITE_BUSY;

  /* Update the global lock state and do debug tracing */
#ifdef SQLITE_DEBUG
  { u16 mask;
  OSTRACE(("SHM-LOCK "));
  mask = (1<<(ofst+n)) - (1<<ofst);

    struct stat sStat;                 /* fstat() info for database file */

    /* Call fstat() to figure out the permissions on the database file. If
    ** a new *-shm file is created, an attempt will be made to create it
    ** with the same permissions. The actual permissions the file is created
    ** with are subject to the current umask setting.
    */
    if( osFstat(pDbFd->h, &sStat) ){
      rc = SQLITE_IOERR_FSTAT;
      goto shm_open_err;
    }

#ifdef SQLITE_SHM_DIRECTORY
    nShmFilename = sizeof(SQLITE_SHM_DIRECTORY) + 30;
#else

    pShmNode->pInode = pDbFd->pInode;
    pShmNode->mutex = sqlite3_mutex_alloc(SQLITE_MUTEX_FAST);
    if( pShmNode->mutex==0 ){
      rc = SQLITE_NOMEM;
      goto shm_open_err;
    }

    pShmNode->h = robust_open(zShmFilename, O_RDWR|O_CREAT,
                             (sStat.st_mode & 0777));
    if( pShmNode->h<0 ){
      rc = unixLogError(SQLITE_CANTOPEN_BKPT, "open", zShmFilename);
      goto shm_open_err;
    }

    /* Check to see if another process is holding the dead-man switch.
    ** If not, truncate the file to zero length. 

    pShmNode->szRegion = szRegion;

    /* The requested region is not mapped into this processes address space.
    ** Check to see if it has been allocated (i.e. if the wal-index file is
    ** large enough to contain the requested region).
    */
    if( osFstat(pShmNode->h, &sStat) ){
      rc = SQLITE_IOERR_SHMSIZE;
      goto shmpage_out;
    }

    if( sStat.st_size<nByte ){
      /* The requested memory region does not exist. If bExtend is set to
      ** false, exit early. *pp will be set to NULL and SQLITE_OK returned.

  ** 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( osFcntl(pNew->h, F_GETLK, &lockInfo)!=-1 ) {
    if( strcmp(fsInfo.f_fstypename, "nfs")==0 ){
      return &nfsIoMethods;
    } else {
      return &posixIoMethods;
    }
  }else{
    return &dotlockIoMethods;

  /* 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( osFcntl(pNew->h, F_GETLK, &lockInfo)!=-1 ) {
    return &posixIoMethods;
  }else{
    return &semIoMethods;
  }
}
static const sqlite3_io_methods 
  *(*const autolockIoFinder)(const char*,unixFile*) = autolockIoFinderImpl;

  int fd = -1;
  char zDirname[MAX_PATHNAME+1];

  sqlite3_snprintf(MAX_PATHNAME, zDirname, "%s", zFilename);
  for(ii=(int)strlen(zDirname); ii>1 && zDirname[ii]!='/'; ii--);
  if( ii>0 ){
    zDirname[ii] = '\0';
    fd = robust_open(zDirname, O_RDONLY|O_BINARY, 0);
    if( fd>=0 ){
#ifdef FD_CLOEXEC
      osFcntl(fd, F_SETFD, osFcntl(fd, F_GETFD, 0) | FD_CLOEXEC);
#endif
      OSTRACE(("OPENDIR %-3d %s\n", fd, zDirname));
    }
  }
  *pFd = fd;
  return (fd>=0?SQLITE_OK:unixLogError(SQLITE_CANTOPEN_BKPT, "open", zDirname));
}

  struct stat buf;
  const char *zDir = 0;

  azDirs[0] = sqlite3_temp_directory;
  if( !azDirs[1] ) azDirs[1] = getenv("TMPDIR");
  for(i=0; i<sizeof(azDirs)/sizeof(azDirs[0]); zDir=azDirs[i++]){
    if( zDir==0 ) continue;
    if( osStat(zDir, &buf) ) continue;
    if( !S_ISDIR(buf.st_mode) ) continue;
    if( osAccess(zDir, 07) ) continue;
    break;
  }
  return zDir;
}

/*
** Create a temporary file name in zBuf.  zBuf must be allocated

    sqlite3_snprintf(nBuf-17, zBuf, "%s/"SQLITE_TEMP_FILE_PREFIX, zDir);
    j = (int)strlen(zBuf);
    sqlite3_randomness(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( osAccess(zBuf,0)==0 );
  return SQLITE_OK;
}

#if SQLITE_ENABLE_LOCKING_STYLE && defined(__APPLE__)
/*
** Routine to transform a unixFile into a proxy-locking unixFile.
** Implementation in the proxy-lock division, but used by unixOpen()

    mode_t openMode;              /* Permissions to create file with */
    rc = findCreateFileMode(zName, flags, &openMode);
    if( rc!=SQLITE_OK ){
      assert( !p->pUnused );
      assert( eType==SQLITE_OPEN_WAL || eType==SQLITE_OPEN_MAIN_JOURNAL );
      return rc;
    }
    fd = robust_open(zName, openFlags, openMode);
    OSTRACE(("OPENX   %-3d %s 0%o\n", fd, zName, openFlags));
    if( fd<0 && errno!=EISDIR && isReadWrite && !isExclusive ){
      /* Failed to open the file for read/write access. Try read-only. */
      flags &= ~(SQLITE_OPEN_READWRITE|SQLITE_OPEN_CREATE);
      openFlags &= ~(O_RDWR|O_CREAT);
      flags |= SQLITE_OPEN_READONLY;
      openFlags |= O_RDONLY;
      fd = robust_open(zName, openFlags, openMode);
    }
    if( fd<0 ){
      rc = unixLogError(SQLITE_CANTOPEN_BKPT, "open", zName);
      goto open_finished;
    }
  }
  assert( fd>=0 );

      assert( eType!=SQLITE_OPEN_MAIN_DB );
      robust_close(p, fd, __LINE__);
      goto open_finished;
    }
  }

#ifdef FD_CLOEXEC
  osFcntl(fd, F_SETFD, osFcntl(fd, F_GETFD, 0) | FD_CLOEXEC);
#endif

  noLock = eType!=SQLITE_OPEN_MAIN_DB;

  
#if defined(__APPLE__) || SQLITE_ENABLE_LOCKING_STYLE
  struct statfs fsInfo;

    case SQLITE_ACCESS_READ:
      amode = R_OK;
      break;

    default:
      assert(!"Invalid flags argument");
  }
  *pResOut = (osAccess(zPath, amode)==0);
  if( flags==SQLITE_ACCESS_EXISTS && *pResOut ){
    struct stat buf;
    if( 0==stat(zPath, &buf) && buf.st_size==0 ){
      *pResOut = 0;
    }
  }
  return SQLITE_OK;

  UNUSED_PARAMETER(pVfs);

  zOut[nOut-1] = '\0';
  if( zPath[0]=='/' ){
    sqlite3_snprintf(nOut, zOut, "%s", zPath);
  }else{
    int nCwd;
    if( osGetcwd(zOut, nOut-1)==0 ){
      return unixLogError(SQLITE_CANTOPEN_BKPT, "getcwd", zPath);
    }
    nCwd = (int)strlen(zOut);
    sqlite3_snprintf(nOut-nCwd, &zOut[nCwd], "/%s", zPath);
  }
  return SQLITE_OK;
}

  ** that we always use the same random number sequence.  This makes the
  ** tests repeatable.
  */
  memset(zBuf, 0, nBuf);
#if !defined(SQLITE_TEST)
  {
    int pid, fd;
    fd = robust_open("/dev/urandom", O_RDONLY, 0);
    if( fd<0 ){
      time_t t;
      time(&t);
      memcpy(zBuf, &t, sizeof(t));
      pid = getpid();
      memcpy(&zBuf[sizeof(t)], &pid, sizeof(pid));
      assert( sizeof(t)+sizeof(pid)<=(size_t)nBuf );
      nBuf = sizeof(t) + sizeof(pid);
    }else{
      do{ nBuf = osRead(fd, zBuf, nBuf); }while( nBuf<0 && errno==EINTR );
      robust_close(0, fd, __LINE__);
    }
  }
#endif
  return nBuf;
}

  }else{
    pUnused = sqlite3_malloc(sizeof(*pUnused));
    if( !pUnused ){
      return SQLITE_NOMEM;
    }
  }
  if( fd<0 ){
    fd = robust_open(path, openFlags, SQLITE_DEFAULT_FILE_PERMISSIONS);
    terrno = errno;
    if( fd<0 && errno==ENOENT && islockfile ){
      if( proxyCreateLockPath(path) == SQLITE_OK ){
        fd = robust_open(path, openFlags, SQLITE_DEFAULT_FILE_PERMISSIONS);
      }
    }
  }
  if( fd<0 ){
    openFlags = O_RDONLY;
    fd = robust_open(path, openFlags, SQLITE_DEFAULT_FILE_PERMISSIONS);
    terrno = errno;
  }
  if( fd<0 ){
    if( islockfile ){
      return SQLITE_BUSY;
    }
    switch (terrno) {

  pathLen = strlcpy(tPath, cPath, MAXPATHLEN);
  if( pathLen>MAXPATHLEN || pathLen<6 || 
     (strlcpy(&tPath[pathLen-5], "break", 6) != 5) ){
    sqlite3_snprintf(sizeof(errmsg),errmsg,"path error (len %d)",(int)pathLen);
    goto end_breaklock;
  }
  /* read the conch content */
  readLen = osPread(conchFile->h, buf, PROXY_MAXCONCHLEN, 0);
  if( readLen<PROXY_PATHINDEX ){
    sqlite3_snprintf(sizeof(errmsg),errmsg,"read error (len %d)",(int)readLen);
    goto end_breaklock;
  }
  /* write it out to the temporary break file */
  fd = robust_open(tPath, (O_RDWR|O_CREAT|O_EXCL),
                   SQLITE_DEFAULT_FILE_PERMISSIONS);
  if( fd<0 ){
    sqlite3_snprintf(sizeof(errmsg), errmsg, "create failed (%d)", errno);
    goto end_breaklock;
  }
  if( osPwrite(fd, buf, readLen, 0) != (ssize_t)readLen ){
    sqlite3_snprintf(sizeof(errmsg), errmsg, "write failed (%d)", errno);
    goto end_breaklock;
  }
  if( rename(tPath, cPath) ){
    sqlite3_snprintf(sizeof(errmsg), errmsg, "rename failed (%d)", errno);
    goto end_breaklock;
  }

      /* If the lock failed (busy):
       * 1st try: get the mod time of the conch, wait 0.5s and try again. 
       * 2nd try: fail if the mod time changed or host id is different, wait 
       *           10 sec and try again
       * 3rd try: break the lock unless the mod time has changed.
       */
      struct stat buf;
      if( osFstat(conchFile->h, &buf) ){
        pFile->lastErrno = errno;
        return SQLITE_IOERR_LOCK;
      }
      
      if( nTries==1 ){
        conchModTime = buf.st_mtimespec;
        usleep(500000); /* wait 0.5 sec and try the lock again*/
        continue;  
      }

      assert( nTries>1 );
      if( conchModTime.tv_sec != buf.st_mtimespec.tv_sec || 
         conchModTime.tv_nsec != buf.st_mtimespec.tv_nsec ){
        return SQLITE_BUSY;
      }
      
      if( nTries==2 ){  
        char tBuf[PROXY_MAXCONCHLEN];
        int len = osPread(conchFile->h, tBuf, PROXY_MAXCONCHLEN, 0);
        if( len<0 ){
          pFile->lastErrno = errno;
          return SQLITE_IOERR_LOCK;
        }
        if( len>PROXY_PATHINDEX && tBuf[0]==(char)PROXY_CONCHVERSION){
          /* don't break the lock if the host id doesn't match */
          if( 0!=memcmp(&tBuf[PROXY_HEADERLEN], myHostID, PROXY_HOSTIDLEN) ){

        rc = unixWrite((sqlite3_file *)conchFile, writeBuffer, writeSize, 0);
        fsync(conchFile->h);
        /* If we created a new conch file (not just updated the contents of a 
         ** valid conch file), try to match the permissions of the database 
         */
        if( rc==SQLITE_OK && createConch ){
          struct stat buf;
          int err = osFstat(pFile->h, &buf);
          if( err==0 ){
            mode_t cmode = buf.st_mode&(S_IRUSR|S_IWUSR | S_IRGRP|S_IWGRP |
                                        S_IROTH|S_IWOTH);
            /* try to match the database file R/W permissions, ignore failure */
#ifndef SQLITE_PROXY_DEBUG
            osFchmod(conchFile->h, cmode);
#else
            do{
              rc = osFchmod(conchFile->h, cmode);
            }while( rc==(-1) && errno==EINTR );
            if( rc!=0 ){
              int code = errno;
              fprintf(stderr, "fchmod %o FAILED with %d %s\n",
                      cmode, code, strerror(code));
            } else {
              fprintf(stderr, "fchmod %o SUCCEDED\n",cmode);

    end_takeconch:
      OSTRACE(("TRANSPROXY: CLOSE  %d\n", pFile->h));
      if( rc==SQLITE_OK && pFile->openFlags ){
        if( pFile->h>=0 ){
          robust_close(pFile, pFile->h, __LINE__);
        }
        pFile->h = -1;
        int fd = robust_open(pCtx->dbPath, pFile->openFlags,
                      SQLITE_DEFAULT_FILE_PERMISSIONS);
        OSTRACE(("TRANSPROXY: OPEN  %d\n", fd));
        if( fd>=0 ){
          pFile->h = fd;
        }else{
          rc=SQLITE_CANTOPEN_BKPT; /* SQLITE_BUSY? proxyTakeConch called
           during locking */

      ** Ugh, since O_RDONLY==0x0000 we test for !O_RDWR since unixOpen asserts
      ** that openFlags will have only one of O_RDONLY or O_RDWR.
      */
      struct statfs fsInfo;
      struct stat conchInfo;
      int goLockless = 0;

      if( osStat(pCtx->conchFilePath, &conchInfo) == -1 ) {
        int err = errno;
        if( (err==ENOENT) && (statfs(dbPath, &fsInfo) != -1) ){
          goLockless = (fsInfo.f_flags&MNT_RDONLY) == MNT_RDONLY;
        }
      }
      if( goLockless ){
        pCtx->conchHeld = -1; /* read only FS/ lockless */

  ** Most finders simply return a pointer to a fixed sqlite3_io_methods
  ** object.  But the "autolockIoFinder" available on MacOSX does a little
  ** more than that; it looks at the filesystem type that hosts the 
  ** database file and tries to choose an locking method appropriate for
  ** that filesystem time.
  */
  #define UNIXVFS(VFSNAME, FINDER) {                        \
    3,                    /* iVersion */                    \
    sizeof(unixFile),     /* szOsFile */                    \
    MAX_PATHNAME,         /* mxPathname */                  \
    0,                    /* pNext */                       \
    VFSNAME,              /* zName */                       \
    (void*)&FINDER,       /* pAppData */                    \
    unixOpen,             /* xOpen */                       \
    unixDelete,           /* xDelete */                     \
    unixAccess,           /* xAccess */                     \
    unixFullPathname,     /* xFullPathname */               \
    unixDlOpen,           /* xDlOpen */                     \
    unixDlError,          /* xDlError */                    \
    unixDlSym,            /* xDlSym */                      \
    unixDlClose,          /* xDlClose */                    \
    unixRandomness,       /* xRandomness */                 \
    unixSleep,            /* xSleep */                      \
    unixCurrentTime,      /* xCurrentTime */                \
    unixGetLastError,     /* xGetLastError */               \
    unixCurrentTimeInt64, /* xCurrentTimeInt64 */           \
    unixSetSystemCall,    /* xSetSystemCall */              \
    unixGetSystemCall,    /* xGetSystemCall */              \
    unixNextSystemCall,   /* xNextSystemCall */             \
  }

  /*
  ** All default VFSes for unix are contained in the following array.
  **
  ** Note that the sqlite3_vfs.pNext field of the VFS object is modified
  ** by the SQLite core when the VFS is registered.  So the following

/*
** Initialize and deinitialize the operating system interface.
*/
int sqlite3_os_init(void){
  static sqlite3_vfs winVfs = {
    3,                   /* iVersion */
    sizeof(winFile),     /* szOsFile */
    MAX_PATH,            /* mxPathname */
    0,                   /* pNext */
    "win32",             /* zName */
    0,                   /* pAppData */
    winOpen,             /* xOpen */
    winDelete,           /* xDelete */
    winAccess,           /* xAccess */
    winFullPathname,     /* xFullPathname */
    winDlOpen,           /* xDlOpen */
    winDlError,          /* xDlError */
    winDlSym,            /* xDlSym */
    winDlClose,          /* xDlClose */
    winRandomness,       /* xRandomness */
    winSleep,            /* xSleep */
    winCurrentTime,      /* xCurrentTime */
    winGetLastError,     /* xGetLastError */
    winCurrentTimeInt64, /* xCurrentTimeInt64 */
    0,                   /* xSetSystemCall */
    0,                   /* xGetSystemCall */
    0,                   /* xNextSystemCall */
  };

#ifndef SQLITE_OMIT_WAL
  /* get memory map allocation granularity */
  memset(&winSysInfo, 0, sizeof(SYSTEM_INFO));
  GetSystemInfo(&winSysInfo);
  assert(winSysInfo.dwAllocationGranularity > 0);

** ^SQLite will use the xCurrentTimeInt64() method to get the current
** date and time if that method is available (if iVersion is 2 or 
** greater and the function pointer is not NULL) and will fall back
** to xCurrentTime() if xCurrentTimeInt64() is unavailable.
*/
typedef struct sqlite3_vfs sqlite3_vfs;
struct sqlite3_vfs {
  int iVersion;            /* Structure version number (currently 3) */
  int szOsFile;            /* Size of subclassed sqlite3_file */
  int mxPathname;          /* Maximum file pathname length */
  sqlite3_vfs *pNext;      /* Next registered VFS */
  const char *zName;       /* Name of this virtual file system */
  void *pAppData;          /* Pointer to application-specific data */
  int (*xOpen)(sqlite3_vfs*, const char *zName, sqlite3_file*,
               int flags, int *pOutFlags);

  /*
  ** The methods above are in version 1 of the sqlite_vfs object
  ** definition.  Those that follow are added in version 2 or later
  */
  int (*xCurrentTimeInt64)(sqlite3_vfs*, sqlite3_int64*);
  /*
  ** The methods above are in versions 1 and 2 of the sqlite_vfs object.
  ** Those below are for version 3 and greater.
  */
  int (*xSetSystemCall)(sqlite3_vfs*, const char *zName, void *pFunc);
  void *(*xGetSystemCall)(sqlite3_vfs*, const char *zName);
  const char *(*xNextSystemCall)(sqlite3_vfs*, const char *zName);
  /*
  ** The methods above are in versions 1 through 3 of the sqlite_vfs object.
  ** New fields may be appended in figure versions.  The iVersion
  ** value will increment whenever this happens. 
  */
};

/*
** CAPI3REF: Flags for the xAccess VFS method