SQLite

/*
** Only set the lastErrno if the error code is a real error and not 
** a normal expected return code of SQLITE_BUSY or SQLITE_OK
*/
#define IS_LOCK_ERROR(x)  ((x != SQLITE_OK) && (x != SQLITE_BUSY))

/* Forward references */
typedef struct unixShm unixShm;               /* Connection shared memory */
typedef struct unixShmNode unixShmNode;       /* Shared memory instance */
typedef struct unixInodeInfo unixInodeInfo;   /* An i-node */
typedef struct UnixUnusedFd UnixUnusedFd;     /* An unused file descriptor */

/*
** Sometimes, after a file handle is closed by SQLite, the file descriptor
** cannot be closed immediately. In these cases, instances of the following
** structure are used to store the file descriptor while waiting for an
** opportunity to either close or reuse it.
*/

struct UnixUnusedFd {
  int fd;                   /* File descriptor to close */
  int flags;                /* Flags this file descriptor was opened with */
  UnixUnusedFd *pNext;      /* Next unused file descriptor on same file */
};

/*
** The unixFile structure is subclass of sqlite3_file specific to the unix
** VFS implementations.
*/
typedef struct unixFile unixFile;
struct unixFile {
  sqlite3_io_methods const *pMethod;  /* Always the first entry */
  unixInodeInfo *pInode;              /* Info about locks on this inode */
  int h;                              /* The file descriptor */
  int dirfd;                          /* File descriptor for the directory */
  unsigned char eFileLock;            /* The type of lock held on this fd */
  int lastErrno;                      /* The unix errno from last I/O error */
  void *lockingContext;               /* Locking style specific state */
  UnixUnusedFd *pUnused;              /* Pre-allocated UnixUnusedFd */
  int fileFlags;                      /* Miscellanous flags */

** object keeps a count of the number of unixFile pointing to it.
*/
struct unixInodeInfo {
  struct unixFileId fileId;       /* The lookup key */
  int nShared;                    /* Number of SHARED locks held */
  int eFileLock;                  /* One of SHARED_LOCK, RESERVED_LOCK etc. */
  int nRef;                       /* Number of pointers to this structure */
  unixShmNode *pShmNode;          /* Shared memory associated with this inode */
  int nLock;                      /* Number of outstanding file locks */
  UnixUnusedFd *pUnused;          /* Unused file descriptors to close */
  unixInodeInfo *pNext;           /* List of all unixInodeInfo objects */
  unixInodeInfo *pPrev;           /*    .... doubly linked */
#if defined(SQLITE_ENABLE_LOCKING_STYLE)
  unsigned long long sharedByte;  /* for AFP simulated shared lock */
#endif


#if OS_VXWORKS
  sem_t *pSem;                    /* Named POSIX semaphore */
  char aSemName[MAX_PATHNAME+2];  /* Name of that semaphore */
#endif


};

/*
** A lists of all unixInodeInfo objects.
*/
static unixInodeInfo *inodeList = 0;

/*
** Release a unixInodeInfo structure previously allocated by findInodeInfo().
**
** The mutex entered using the unixEnterMutex() function must be held
** when this function is called.
*/
static void releaseInodeInfo(unixInodeInfo *pInode){
  assert( unixMutexHeld() );
  if( pInode ){
    pInode->nRef--;
    if( pInode->nRef==0 ){
      assert( pInode->pShmNode==0 );
      if( pInode->pPrev ){
        assert( pInode->pPrev->pNext==pInode );
        pInode->pPrev->pNext = pInode->pNext;
      }else{
        assert( inodeList==pInode );
        inodeList = pInode->pNext;
      }

** The mutex entered using the unixEnterMutex() function must be held
** when this function is called.
**
** Return an appropriate error code.
*/
static int findInodeInfo(
  unixFile *pFile,               /* Unix file with file desc used in the key */
  unixInodeInfo **ppInode        /* Return the unixInodeInfo object here */
){
  int rc;                        /* System call return code */
  int fd;                        /* The file descriptor for pFile */
  struct unixFileId fileId;      /* Lookup key for the unixInodeInfo */
  struct stat statbuf;           /* Low-level file information */
  unixInodeInfo *pInode = 0;     /* Candidate unixInodeInfo object */

  assert( unixMutexHeld() );

  /* Get low-level information about the file that we can used to
  ** create a unique name for the file.
  */
  fd = pFile->h;

  ** 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;
  unixInodeInfo *pInode = pFile->pInode;
  struct flock lock;
  int s = 0;
  int tErrno = 0;

  assert( pFile );
  OSTRACE(("LOCK    %d %s was %s(%s,%d) pid=%d (unix)\n", pFile->h,
      azFileLock(eFileLock), azFileLock(pFile->eFileLock),

**
** Otherwise, if an error occurs, then successfully closed file descriptor
** entries are removed from the list, and SQLITE_IOERR_CLOSE returned. 
** not deleted and SQLITE_IOERR_CLOSE returned.
*/ 
static int closePendingFds(unixFile *pFile){
  int rc = SQLITE_OK;
  unixInodeInfo *pInode = pFile->pInode;
  UnixUnusedFd *pError = 0;
  UnixUnusedFd *p;
  UnixUnusedFd *pNext;
  for(p=pInode->pUnused; p; p=pNext){
    pNext = p->pNext;
    if( close(p->fd) ){
      pFile->lastErrno = errno;

}

/*
** Add the file descriptor used by file handle pFile to the corresponding
** pUnused list.
*/
static void setPendingFd(unixFile *pFile){
  unixInodeInfo *pInode = pFile->pInode;
  UnixUnusedFd *p = pFile->pUnused;
  p->pNext = pInode->pUnused;
  pInode->pUnused = p;
  pFile->h = -1;
  pFile->pUnused = 0;
}

** the byte range is divided into 2 parts and the first part is unlocked then
** set to a read lock, then the other part is simply unlocked.  This works 
** around a bug in BSD NFS lockd (also seen on MacOSX 10.3+) that fails to 
** remove the write lock on a region when a read lock is set.
*/
static int _posixUnlock(sqlite3_file *id, int eFileLock, int handleNFSUnlock){
  unixFile *pFile = (unixFile*)id;
  unixInodeInfo *pInode;
  struct flock lock;
  int rc = SQLITE_OK;
  int h;
  int tErrno;                      /* Error code from system call errors */

  assert( pFile );
  OSTRACE(("UNLOCK  %d %d was %d(%d,%d) pid=%d (unix)\n", pFile->h, eFileLock,

**
** This routine will only increase a lock.  Use the sqlite3OsUnlock()
** routine to lower a locking level.
*/
static int afpLock(sqlite3_file *id, int eFileLock){
  int rc = SQLITE_OK;
  unixFile *pFile = (unixFile*)id;
  unixInodeInfo *pInode = pFile->pInode;
  afpLockingContext *context = (afpLockingContext *) pFile->lockingContext;
  
  assert( pFile );
  OSTRACE(("LOCK    %d %s was %s(%s,%d) pid=%d (afp)\n", pFile->h,
           azFileLock(eFileLock), azFileLock(pFile->eFileLock),
           azFileLock(pInode->eFileLock), pInode->nShared , getpid()));

**
** 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 afpUnlock(sqlite3_file *id, int eFileLock) {
  int rc = SQLITE_OK;
  unixFile *pFile = (unixFile*)id;
  unixInodeInfo *pInode;
  afpLockingContext *context = (afpLockingContext *) pFile->lockingContext;
  int skipShared = 0;
#ifdef SQLITE_TEST
  int h = pFile->h;
#endif

  assert( pFile );

  return 0;
}

#ifndef SQLITE_OMIT_WAL


/*
** Object used to represent an shared memory buffer.  
**
** When multiple threads all reference the same wal-index, each thread
** has its own unixShm object, but they all point to a single instance
** of this unixShmNode object.  In other words, each wal-index is opened
** only once per process.
**
** Each unixShmNode object is connected to a single unixInodeInfo object.
** We could coalesce this object into unixInodeInfo, but that would mean
** every open file that does not use shared memory (in other words, most
** open files) would have to carry around this extra information.  So
** the unixInodeInfo object contains a pointer to this unixShmNode object
** and the unixShmNode object is created only when needed.
**
** unixMutexHeld() must be true when creating or destroying
** this object or while reading or writing the following fields:
**
**      nRef

**
** The following fields are read-only after the object is created:
** 
**      fid
**      zFilename
**
** Either unixShmNode.mutex must be held or unixShmNode.nRef==0 and
** unixMutexHeld() is true when reading or writing any other field
** in this structure.
**
** To avoid deadlocks, mutex and mutexBuf are always released in the
** reverse order that they are acquired.  mutexBuf is always acquired
** first and released last.  This invariant is check by asserting
** sqlite3_mutex_notheld() on mutex whenever mutexBuf is acquired or
** released.
*/
struct unixShmNode {
  unixInodeInfo *pInode;     /* unixInodeInfo that owns this SHM node */
  sqlite3_mutex *mutex;      /* Mutex to access this object */
  sqlite3_mutex *mutexBuf;   /* Mutex to access zBuf[] */
  char *zFilename;           /* Name of the mmapped file */
  int h;                     /* Open file descriptor */
  int szMap;                 /* Size of the mapping into memory */
  char *pMMapBuf;            /* Where currently mmapped().  NULL if unmapped */
  int nRef;                  /* Number of unixShm objects pointing to this */
  unixShm *pFirst;           /* All unixShm objects pointing to this */

#ifdef SQLITE_DEBUG
  u8 exclMask;               /* Mask of exclusive locks held */
  u8 sharedMask;             /* Mask of shared locks held */
  u8 nextShmId;              /* Next available unixShm.id value */
#endif
};








/*
** Structure used internally by this VFS to record the state of an
** open shared memory connection.
**
** The following fields are initialized when this object is created and
** are read-only thereafter:
**
**    unixShm.pFile
**    unixShm.id
**
** All other fields are read/write.  The unixShm.pFile->mutex must be held
** while accessing any read/write fields.
*/
struct unixShm {
  unixShmNode *pShmNode;     /* The underlying unixShmNode object */
  unixShm *pNext;            /* Next unixShm with the same unixShmNode */
  u8 lockState;              /* Current lock state */
  u8 hasMutex;               /* True if holding the unixShmNode mutex */
  u8 hasMutexBuf;            /* True if holding pFile->mutexBuf */
  u8 sharedMask;             /* Mask of shared locks held */
  u8 exclMask;               /* Mask of exclusive locks held */
#ifdef SQLITE_DEBUG
  u8 id;                     /* Id of this connection within its unixShmNode */
#endif
};

/*
** Size increment by which shared memory grows
*/
#define SQLITE_UNIX_SHM_INCR  4096

** lockMask might contain multiple bits but all bits are guaranteed
** to be contiguous.
**
** Locks block if the mask is exactly UNIX_SHM_C and are non-blocking
** otherwise.
*/
static int unixShmSystemLock(
  unixShmNode *pShmNode, /* Apply locks to this open shared-memory segment */
  int lockType,          /* F_UNLCK, F_RDLCK, or F_WRLCK */
  u8 lockMask            /* Which bytes to lock or unlock */
){
  struct flock f;       /* The posix advisory locking structure */
  int lockOp;           /* The opcode for fcntl() */
  int i;                /* Offset into the locking byte range */
  int rc;               /* Result code form fcntl() */
  u8 mask;              /* Mask of bits in lockMask */

  /* Access to the unixShmNode object is serialized by the caller */
  assert( sqlite3_mutex_held(pShmNode->mutex) || pShmNode->nRef==0 );

  /* Initialize the locking parameters */
  memset(&f, 0, sizeof(f));
  f.l_type = lockType;
  f.l_whence = SEEK_SET;
  if( lockMask==UNIX_SHM_C && lockType!=F_UNLCK ){
    lockOp = F_SETLKW;

    mask <<= 1;
  }

  /* Verify that all bits set in lockMask are contiguous */
  assert( mask==0 || (lockMask & ~(mask | (mask-1)))==0 );

  /* Acquire the system-level lock */
  rc = fcntl(pShmNode->h, lockOp, &f);
  rc = (rc!=(-1)) ? SQLITE_OK : SQLITE_BUSY;

  /* Update the global lock state and do debug tracing */
#ifdef SQLITE_DEBUG
  OSTRACE(("SHM-LOCK "));
  if( rc==SQLITE_OK ){
    if( lockType==F_UNLCK ){
      OSTRACE(("unlock ok"));
      pShmNode->exclMask &= ~lockMask;
      pShmNode->sharedMask &= ~lockMask;
    }else if( lockType==F_RDLCK ){
      OSTRACE(("read-lock ok"));
      pShmNode->exclMask &= ~lockMask;
      pShmNode->sharedMask |= lockMask;
    }else{
      assert( lockType==F_WRLCK );
      OSTRACE(("write-lock ok"));
      pShmNode->exclMask |= lockMask;
      pShmNode->sharedMask &= ~lockMask;
    }
  }else{
    if( lockType==F_UNLCK ){
      OSTRACE(("unlock failed"));
    }else if( lockType==F_RDLCK ){
      OSTRACE(("read-lock failed"));
    }else{
      assert( lockType==F_WRLCK );
      OSTRACE(("write-lock failed"));
    }
  }
  OSTRACE((" - change requested %s - afterwards %s:%s\n",
           unixShmLockString(lockMask),
           unixShmLockString(pShmNode->sharedMask),
           unixShmLockString(pShmNode->exclMask)));
#endif

  return rc;        
}

/*
** For connection p, unlock all of the locks identified by the unlockMask
** parameter.
*/
static int unixShmUnlock(
  unixShmNode *pShmNode,   /* The underlying shared-memory file */
  unixShm *p,              /* The connection to be unlocked */
  u8 unlockMask            /* Mask of locks to be unlocked */
){
  int rc;      /* Result code */
  unixShm *pX; /* For looping over all sibling connections */
  u8 allMask;  /* Union of locks held by connections other than "p" */

  /* Access to the unixShmNode object is serialized by the caller */
  assert( sqlite3_mutex_held(pShmNode->mutex) );

  /* Compute locks held by sibling connections */
  allMask = 0;
  for(pX=pShmNode->pFirst; pX; pX=pX->pNext){
    if( pX==p ) continue;
    assert( (pX->exclMask & (p->exclMask|p->sharedMask))==0 );
    allMask |= pX->sharedMask;
  }

  /* Unlock the system-level locks */
  if( (unlockMask & allMask)!=unlockMask ){
    rc = unixShmSystemLock(pShmNode, F_UNLCK, unlockMask & ~allMask);
  }else{
    rc = SQLITE_OK;
  }

  /* Undo the local locks */
  if( rc==SQLITE_OK ){
    p->exclMask &= ~unlockMask;
    p->sharedMask &= ~unlockMask;
  } 
  return rc;
}

/*
** Get reader locks for connection p on all locks in the readMask parameter.
*/
static int unixShmSharedLock(
  unixShmNode *pShmNode,   /* The underlying shared-memory file */
  unixShm *p,              /* The connection to get the shared locks */
  u8 readMask              /* Mask of shared locks to be acquired */
){
  int rc;        /* Result code */
  unixShm *pX;   /* For looping over all sibling connections */
  u8 allShared;  /* Union of locks held by connections other than "p" */

  /* Access to the unixShmNode object is serialized by the caller */
  assert( sqlite3_mutex_held(pShmNode->mutex) );

  /* Find out which shared locks are already held by sibling connections.
  ** If any sibling already holds an exclusive lock, go ahead and return
  ** SQLITE_BUSY.
  */
  allShared = 0;
  for(pX=pShmNode->pFirst; pX; pX=pX->pNext){
    if( pX==p ) continue;
    if( (pX->exclMask & readMask)!=0 ) return SQLITE_BUSY;
    allShared |= pX->sharedMask;
  }

  /* Get shared locks at the system level, if necessary */
  if( (~allShared) & readMask ){
    rc = unixShmSystemLock(pShmNode, F_RDLCK, readMask);
  }else{
    rc = SQLITE_OK;
  }

  /* Get the local shared locks */
  if( rc==SQLITE_OK ){
    p->sharedMask |= readMask;
  }
  return rc;
}

/*
** For connection p, get an exclusive lock on all locks identified in
** the writeMask parameter.
*/
static int unixShmExclusiveLock(
  unixShmNode *pShmNode,    /* The underlying shared-memory file */
  unixShm *p,               /* The connection to get the exclusive locks */
  u8 writeMask              /* Mask of exclusive locks to be acquired */
){
  int rc;        /* Result code */
  unixShm *pX;   /* For looping over all sibling connections */

  /* Access to the unixShmNode object is serialized by the caller */
  assert( sqlite3_mutex_held(pShmNode->mutex) );

  /* Make sure no sibling connections hold locks that will block this
  ** lock.  If any do, return SQLITE_BUSY right away.
  */
  for(pX=pShmNode->pFirst; pX; pX=pX->pNext){
    if( pX==p ) continue;
    if( (pX->exclMask & writeMask)!=0 ) return SQLITE_BUSY;
    if( (pX->sharedMask & writeMask)!=0 ) return SQLITE_BUSY;
  }

  /* Get the exclusive locks at the system level.  Then if successful
  ** also mark the local connection as being locked.
  */
  rc = unixShmSystemLock(pShmNode, F_WRLCK, writeMask);
  if( rc==SQLITE_OK ){
    p->sharedMask &= ~writeMask;
    p->exclMask |= writeMask;
  }
  return rc;
}

/*
** Purge the unixShmNodeList list of all entries with unixShmNode.nRef==0.
**
** This is not a VFS shared-memory method; it is a utility function called
** by VFS shared-memory methods.
*/
static void unixShmPurge(unixFile *pFd){

  unixShmNode *p = pFd->pInode->pShmNode;
  assert( unixMutexHeld() );


  if( p && p->nRef==0 ){
    assert( p->pInode==pFd->pInode );
    if( p->mutex ) sqlite3_mutex_free(p->mutex);
    if( p->mutexBuf ) sqlite3_mutex_free(p->mutexBuf);
    if( p->h>=0 ) close(p->h);
    p->pInode->pShmNode = 0;
    sqlite3_free(p);



  }
}

/*
** Open a shared-memory area.  This particular implementation uses
** mmapped files.
**

** file are currently open, in this process or in other processes, then
** the file must be truncated to zero length or have its header cleared.
*/
static int unixShmOpen(
  sqlite3_file *fd      /* The file descriptor of the associated database */
){
  struct unixShm *p = 0;             /* The connection to be opened */
  struct unixShmNode *pShmNode = 0;  /* The underlying mmapped file */
  int rc;                            /* Result code */



  struct unixFile *pDbFd;            /* Underlying database file */
  int nPath;                         /* Size of pDbFd->zPath in bytes */

  /* Allocate space for the new sqlite3_shm object.

  */
  p = sqlite3_malloc( sizeof(*p) );
  if( p==0 ) return SQLITE_NOMEM;
  memset(p, 0, sizeof(*p));
  pDbFd = (struct unixFile*)fd;
  assert( pDbFd->pShm==0 );









  /* Check to see if a unixShmNode object already exists.  Reuse an existing

  ** one if present.  Create a new one if necessary.
  */
  unixEnterMutex();

  pShmNode = pDbFd->pInode->pShmNode;
  if( pShmNode==0 ){
    nPath = strlen(pDbFd->zPath);
    pShmNode = sqlite3_malloc( sizeof(*pShmNode) + nPath + 15 );
    if( pShmNode==0 ){
      rc = SQLITE_NOMEM;


      goto shm_open_err;
    }

    memset(pShmNode, 0, sizeof(*pShmNode));
    pShmNode->zFilename = (char*)&pShmNode[1];
    sqlite3_snprintf(nPath+15, pShmNode->zFilename,

                     "%s-wal-index", pDbFd->zPath);

    pShmNode->h = -1;

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

    pShmNode->h = open(pShmNode->zFilename, O_RDWR|O_CREAT, 0664);
    if( pShmNode->h<0 ){
      rc = SQLITE_CANTOPEN_BKPT;
      goto shm_open_err;
    }









    /* Check to see if another process is holding the dead-man switch.
    ** If not, truncate the file to zero length. 
    */
    rc = SQLITE_OK;
    if( unixShmSystemLock(pShmNode, F_WRLCK, UNIX_SHM_DMS)==SQLITE_OK ){
      if( ftruncate(pShmNode->h, 0) ){
        rc = SQLITE_IOERR;
      }
    }
    if( rc==SQLITE_OK ){
      rc = unixShmSystemLock(pShmNode, F_RDLCK, UNIX_SHM_DMS);
    }
    if( rc ) goto shm_open_err;
  }

  /* Make the new connection a child of the unixShmNode */
  p->pShmNode = pShmNode;
  p->pNext = pShmNode->pFirst;
#ifdef SQLITE_DEBUG
  p->id = pShmNode->nextShmId++;
#endif
  pShmNode->pFirst = p;
  pShmNode->nRef++;
  pDbFd->pShm = p;
  unixLeaveMutex();
  return SQLITE_OK;

  /* Jump here on any error */
shm_open_err:
  unixShmPurge(pDbFd);       /* This call frees pShmNode if required */
  sqlite3_free(p);

  unixLeaveMutex();
  return rc;
}

/*
** Close a connection to shared-memory.  Delete the underlying 
** storage if deleteFlag is true.
*/
static int unixShmClose(
  sqlite3_file *fd,          /* The underlying database file */
  int deleteFlag             /* Delete shared-memory if true */
){
  unixShm *p;            /* The connection to be closed */
  unixShmNode *pShmNode; /* The underlying shared-memory file */
  unixShm **pp;          /* For looping over sibling connections */
  unixFile *pDbFd;       /* The underlying database file */

  pDbFd = (unixFile*)fd;
  p = pDbFd->pShm;
  if( p==0 ) return SQLITE_OK;
  pShmNode = p->pShmNode;

  assert( pShmNode==pDbFd->pInode->pShmNode );
  assert( pShmNode->pInode==pDbFd->pInode );

  /* Verify that the connection being closed holds no locks */
  assert( p->exclMask==0 );
  assert( p->sharedMask==0 );

  /* Remove connection p from the set of connections associated
  ** with pShmNode */
  sqlite3_mutex_enter(pShmNode->mutex);
  for(pp=&pShmNode->pFirst; (*pp)!=p; pp = &(*pp)->pNext){}
  *pp = p->pNext;

  /* Free the connection p */
  sqlite3_free(p);
  pDbFd->pShm = 0;
  sqlite3_mutex_leave(pShmNode->mutex);

  /* If pShmNode->nRef has reached 0, then close the underlying
  ** shared-memory file, too */
  unixEnterMutex();
  assert( pShmNode->nRef>0 );
  pShmNode->nRef--;
  if( pShmNode->nRef==0 ){
    if( deleteFlag ) unlink(pShmNode->zFilename);
    unixShmPurge(pDbFd);
  }
  unixLeaveMutex();

  return SQLITE_OK;
}

/*

static int unixShmSize(
  sqlite3_file *fd,         /* The open database file holding SHM */
  int reqSize,              /* Requested size.  -1 for query only */
  int *pNewSize             /* Write new size here */
){
  unixFile *pDbFd = (unixFile*)fd;
  unixShm *p = pDbFd->pShm;
  unixShmNode *pShmNode = p->pShmNode;
  int rc = SQLITE_OK;
  struct stat sStat;

  assert( pShmNode==pDbFd->pInode->pShmNode );
  assert( pShmNode->pInode==pDbFd->pInode );

  /* On a query, this loop runs once.  When reqSize>=0, the loop potentially
  ** runs twice, except if the actual size is already greater than or equal
  ** to the requested size, reqSize is set to -1 on the first iteration and
  ** the loop only runs once.
  */
  while( 1 ){
    if( fstat(pShmNode->h, &sStat)==0 ){
      *pNewSize = (int)sStat.st_size;
      if( reqSize>=0 && reqSize<=(int)sStat.st_size ) break;
    }else{
      *pNewSize = 0;
      rc = SQLITE_IOERR;
      break;
    }
    if( reqSize<0 ) break;
    reqSize = (reqSize + SQLITE_UNIX_SHM_INCR - 1)/SQLITE_UNIX_SHM_INCR;
    reqSize *= SQLITE_UNIX_SHM_INCR;
    rc = ftruncate(pShmNode->h, reqSize);
    reqSize = -1;
  }
  return rc;
}


/*

  sqlite3_file *fd,        /* Database file holding shared memory */
  int reqMapSize,          /* Requested size of mapping. -1 means don't care */
  int *pNewMapSize,        /* Write new size of mapping here */
  void **ppBuf             /* Write mapping buffer origin here */
){
  unixFile *pDbFd = (unixFile*)fd;
  unixShm *p = pDbFd->pShm;
  unixShmNode *pShmNode = p->pShmNode;
  int rc = SQLITE_OK;

  assert( pShmNode==pDbFd->pInode->pShmNode );
  assert( pShmNode->pInode==pDbFd->pInode );

  if( p->lockState!=SQLITE_SHM_CHECKPOINT && p->hasMutexBuf==0 ){
    assert( sqlite3_mutex_notheld(pShmNode->mutex) );
    sqlite3_mutex_enter(pShmNode->mutexBuf);
    p->hasMutexBuf = 1;
  }
  sqlite3_mutex_enter(pShmNode->mutex);
  if( pShmNode->szMap==0 || reqMapSize>pShmNode->szMap ){
    int actualSize;
    if( unixShmSize(fd, -1, &actualSize)==SQLITE_OK
     && reqMapSize<actualSize
    ){
      reqMapSize = actualSize;
    }
    if( pShmNode->pMMapBuf ){
      munmap(pShmNode->pMMapBuf, pShmNode->szMap);
    }
    pShmNode->pMMapBuf = mmap(0, reqMapSize, PROT_READ|PROT_WRITE, MAP_SHARED,
                           pShmNode->h, 0);
    pShmNode->szMap = pShmNode->pMMapBuf ? reqMapSize : 0;
  }
  *pNewMapSize = pShmNode->szMap;
  *ppBuf = pShmNode->pMMapBuf;
  sqlite3_mutex_leave(pShmNode->mutex);
  return rc;
}

/*
** Release the lock held on the shared memory segment to that other
** threads are free to resize it if necessary.
**
** If the lock is not currently held, this routine is a harmless no-op.
**
** If the shared-memory object is in lock state RECOVER, then we do not
** really want to release the lock, so in that case too, this routine
** is a no-op.
*/
static int unixShmRelease(sqlite3_file *fd){
  unixFile *pDbFd = (unixFile*)fd;
  unixShm *p = pDbFd->pShm;

  if( p->hasMutexBuf && p->lockState!=SQLITE_SHM_RECOVER ){
    assert( sqlite3_mutex_notheld(p->pShmNode->mutex) );
    sqlite3_mutex_leave(p->pShmNode->mutexBuf);
    p->hasMutexBuf = 0;
  }
  return SQLITE_OK;
}

/*
** Symbolic names for LOCK states used for debugging.

static int unixShmLock(
  sqlite3_file *fd,          /* Database file holding the shared memory */
  int desiredLock,           /* One of SQLITE_SHM_xxxxx locking states */
  int *pGotLock              /* The lock you actually got */
){
  unixFile *pDbFd = (unixFile*)fd;
  unixShm *p = pDbFd->pShm;
  unixShmNode *pShmNode = p->pShmNode;
  int rc = SQLITE_PROTOCOL;

  assert( pShmNode==pDbFd->pInode->pShmNode );
  assert( pShmNode->pInode==pDbFd->pInode );

  /* Note that SQLITE_SHM_READ_FULL and SQLITE_SHM_PENDING are never
  ** directly requested; they are side effects from requesting
  ** SQLITE_SHM_READ and SQLITE_SHM_CHECKPOINT, respectively.
  */
  assert( desiredLock==SQLITE_SHM_UNLOCK
       || desiredLock==SQLITE_SHM_READ

    return SQLITE_OK;
  }

  OSTRACE(("SHM-LOCK shmid-%d, pid-%d request %s->%s\n",
            p->id, getpid(), azLkName[p->lockState], azLkName[desiredLock]));
  
  if( desiredLock==SQLITE_SHM_RECOVER && !p->hasMutexBuf ){
    assert( sqlite3_mutex_notheld(pShmNode->mutex) );
    sqlite3_mutex_enter(pShmNode->mutexBuf);
    p->hasMutexBuf = 1;
  }
  sqlite3_mutex_enter(pShmNode->mutex);
  switch( desiredLock ){
    case SQLITE_SHM_UNLOCK: {
      assert( p->lockState!=SQLITE_SHM_RECOVER );
      unixShmUnlock(pShmNode, p, UNIX_SHM_A|UNIX_SHM_B|UNIX_SHM_C|UNIX_SHM_D);
      rc = SQLITE_OK;
      p->lockState = SQLITE_SHM_UNLOCK;
      break;
    }
    case SQLITE_SHM_READ: {
      if( p->lockState==SQLITE_SHM_UNLOCK ){
        int nAttempt;
        rc = SQLITE_BUSY;
        assert( p->lockState==SQLITE_SHM_UNLOCK );
        for(nAttempt=0; nAttempt<5 && rc==SQLITE_BUSY; nAttempt++){
          rc = unixShmSharedLock(pShmNode, p, UNIX_SHM_A|UNIX_SHM_B);
          if( rc==SQLITE_BUSY ){
            rc = unixShmSharedLock(pShmNode, p, UNIX_SHM_D);
            if( rc==SQLITE_OK ){
              p->lockState = SQLITE_SHM_READ_FULL;
            }
          }else{
            unixShmUnlock(pShmNode, p, UNIX_SHM_B);
            p->lockState = SQLITE_SHM_READ;
          }
        }
      }else{
       assert( p->lockState==SQLITE_SHM_WRITE
               || p->lockState==SQLITE_SHM_RECOVER );
        rc = unixShmSharedLock(pShmNode, p, UNIX_SHM_A);
        unixShmUnlock(pShmNode, p, UNIX_SHM_C|UNIX_SHM_D);
        p->lockState = SQLITE_SHM_READ;
      }
      break;
    }
    case SQLITE_SHM_WRITE: {
      assert( p->lockState==SQLITE_SHM_READ 
              || p->lockState==SQLITE_SHM_READ_FULL );
      rc = unixShmExclusiveLock(pShmNode, p, UNIX_SHM_C|UNIX_SHM_D);
      if( rc==SQLITE_OK ){
        p->lockState = SQLITE_SHM_WRITE;
      }
      break;
    }
    case SQLITE_SHM_CHECKPOINT: {
      assert( p->lockState==SQLITE_SHM_UNLOCK
           || p->lockState==SQLITE_SHM_PENDING
      );
      if( p->lockState==SQLITE_SHM_UNLOCK ){
        rc = unixShmExclusiveLock(pShmNode, p, UNIX_SHM_B|UNIX_SHM_C);
        if( rc==SQLITE_OK ){
          p->lockState = SQLITE_SHM_PENDING;
        }
      }
      if( p->lockState==SQLITE_SHM_PENDING ){
        rc = unixShmExclusiveLock(pShmNode, p, UNIX_SHM_A);
        if( rc==SQLITE_OK ){
          p->lockState = SQLITE_SHM_CHECKPOINT;
        }
      }
      break;
    }
    default: {
      assert( desiredLock==SQLITE_SHM_RECOVER );
      assert( p->lockState==SQLITE_SHM_READ
           || p->lockState==SQLITE_SHM_READ_FULL
      );
      assert( sqlite3_mutex_held(pShmNode->mutexBuf) );
      rc = unixShmExclusiveLock(pShmNode, p, UNIX_SHM_C);
      if( rc==SQLITE_OK ){
        p->lockState = SQLITE_SHM_RECOVER;
      }
      break;
    }
  }
  sqlite3_mutex_leave(pShmNode->mutex);
  OSTRACE(("SHM-LOCK shmid-%d, pid-%d got %s\n",
           p->id, getpid(), azLkName[p->lockState]));
  if( pGotLock ) *pGotLock = p->lockState;
  return rc;
}

#else

  ** For this reason, if an error occurs in the stat() call here, it is
  ** ignored and -1 is returned. The caller will try to open a new file
  ** descriptor on the same path, fail, and return an error to SQLite.
  **
  ** Even if a subsequent open() call does succeed, the consequences of
  ** not searching for a resusable file descriptor are not dire.  */
  if( 0==stat(zPath, &sStat) ){
    unixInodeInfo *pInode;

    unixEnterMutex();
    pInode = inodeList;
    while( pInode && (pInode->fileId.dev!=sStat.st_dev
                     || pInode->fileId.ino!=sStat.st_ino) ){
       pInode = pInode->pNext;
    }