SQLite

** 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:
** 
**      hShm
**      zFilename
**
** Either unixShmNode.pShmMutex must be held or unixShmNode.nRef==0 and
** unixMutexHeld() is true when reading or writing any other field
** in this structure.
*/
struct unixShmNode {
  unixInodeInfo *pInode;     /* unixInodeInfo that owns this SHM node */
  sqlite3_mutex *pShmMutex;  /* Mutex to access this object */
  char *zFilename;           /* Name of the mmapped file */
  int hShm;                  /* Open file descriptor */
  int szRegion;              /* Size of shared-memory regions */
  u16 nRegion;               /* Size of array apRegion */
  u8 isReadonly;             /* True if read-only */
  u8 isUnlocked;             /* True if no DMS lock held */
  char **apRegion;           /* Array of mapped shared-memory regions */
  int nRef;                  /* Number of unixShm objects pointing to this */
  unixShm *pFirst;           /* All unixShm objects pointing to this */

  /* Shared locks never span more than one byte */
  assert( n==1 || lockType!=F_RDLCK );

  /* Locks are within range */
  assert( n>=1 && n<=SQLITE_SHM_NLOCK );

  if( pShmNode->hShm>=0 ){
    /* Initialize the locking parameters */
    f.l_type = lockType;
    f.l_whence = SEEK_SET;
    f.l_start = ofst;
    f.l_len = n;
    rc = osSetPosixAdvisoryLock(pShmNode->hShm, &f, pFile);
    rc = (rc!=(-1)) ? SQLITE_OK : SQLITE_BUSY;
  }

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

  assert( unixMutexHeld() );
  if( p && ALWAYS(p->nRef==0) ){
    int nShmPerMap = unixShmRegionPerMap();
    int i;
    assert( p->pInode==pFd->pInode );
    sqlite3_mutex_free(p->pShmMutex);
    for(i=0; i<p->nRegion; i+=nShmPerMap){
      if( p->hShm>=0 ){
        osMunmap(p->apRegion[i], p->szRegion);
      }else{
        sqlite3_free(p->apRegion[i]);
      }
    }
    sqlite3_free(p->apRegion);
    if( p->hShm>=0 ){
      robust_close(pFd, p->hShm, __LINE__);
      p->hShm = -1;
    }
    p->pInode->pShmNode = 0;
    sqlite3_free(p);
  }
}

/*

  ** process might open and use the *-shm file without truncating it.
  ** And if the *-shm file has been corrupted by a power failure or
  ** system crash, the database itself may also become corrupt.  */
  lock.l_whence = SEEK_SET;
  lock.l_start = UNIX_SHM_DMS;
  lock.l_len = 1;
  lock.l_type = F_WRLCK;
  if( osFcntl(pShmNode->hShm, F_GETLK, &lock)!=0 ) {
    rc = SQLITE_IOERR_LOCK;
  }else if( lock.l_type==F_UNLCK ){
    if( pShmNode->isReadonly ){
      pShmNode->isUnlocked = 1;
      rc = SQLITE_READONLY_CANTINIT;
    }else{
      rc = unixShmSystemLock(pDbFd, F_WRLCK, UNIX_SHM_DMS, 1);
      if( rc==SQLITE_OK && robust_ftruncate(pShmNode->hShm, 0) ){
        rc = unixLogError(SQLITE_IOERR_SHMOPEN,"ftruncate",pShmNode->zFilename);
      }
    }
  }else if( lock.l_type==F_WRLCK ){
    rc = SQLITE_BUSY;
  }

    sqlite3_snprintf(nShmFilename, zShm, 
                     SQLITE_SHM_DIRECTORY "/sqlite-shm-%x-%x",
                     (u32)sStat.st_ino, (u32)sStat.st_dev);
#else
    sqlite3_snprintf(nShmFilename, zShm, "%s-shm", zBasePath);
    sqlite3FileSuffix3(pDbFd->zPath, zShm);
#endif
    pShmNode->hShm = -1;
    pDbFd->pInode->pShmNode = pShmNode;
    pShmNode->pInode = pDbFd->pInode;
    if( sqlite3GlobalConfig.bCoreMutex ){
      pShmNode->pShmMutex = sqlite3_mutex_alloc(SQLITE_MUTEX_FAST);
      if( pShmNode->pShmMutex==0 ){
        rc = SQLITE_NOMEM_BKPT;
        goto shm_open_err;
      }
    }

    if( pInode->bProcessLock==0 ){
      if( 0==sqlite3_uri_boolean(pDbFd->zPath, "readonly_shm", 0) ){
        pShmNode->hShm = robust_open(zShm, O_RDWR|O_CREAT,(sStat.st_mode&0777));
      }
      if( pShmNode->hShm<0 ){
        pShmNode->hShm = robust_open(zShm, O_RDONLY, (sStat.st_mode&0777));
        if( pShmNode->hShm<0 ){
          rc = unixLogError(SQLITE_CANTOPEN_BKPT, "open", zShm);
          goto shm_open_err;
        }
        pShmNode->isReadonly = 1;
      }

      /* If this process is running as root, make sure that the SHM file
      ** is owned by the same user that owns the original database.  Otherwise,
      ** the original owner will not be able to connect.
      */
      robustFchown(pShmNode->hShm, sStat.st_uid, sStat.st_gid);

      rc = unixLockSharedMemory(pDbFd, pShmNode);
      if( rc!=SQLITE_OK && rc!=SQLITE_READONLY_CANTINIT ) goto shm_open_err;
    }
  }

  /* Make the new connection a child of the unixShmNode */

  if( pShmNode->isUnlocked ){
    rc = unixLockSharedMemory(pDbFd, pShmNode);
    if( rc!=SQLITE_OK ) goto shmpage_out;
    pShmNode->isUnlocked = 0;
  }
  assert( szRegion==pShmNode->szRegion || pShmNode->nRegion==0 );
  assert( pShmNode->pInode==pDbFd->pInode );
  assert( pShmNode->hShm>=0 || pDbFd->pInode->bProcessLock==1 );
  assert( pShmNode->hShm<0 || pDbFd->pInode->bProcessLock==0 );

  /* Minimum number of regions required to be mapped. */
  nReqRegion = ((iRegion+nShmPerMap) / nShmPerMap) * nShmPerMap;

  if( pShmNode->nRegion<nReqRegion ){
    char **apNew;                      /* New apRegion[] array */
    int nByte = nReqRegion*szRegion;   /* Minimum required file size */
    struct stat sStat;                 /* Used by fstat() */

    pShmNode->szRegion = szRegion;

    if( pShmNode->hShm>=0 ){
      /* 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->hShm, &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.

          static const int pgsz = 4096;
          int iPg;

          /* Write to the last byte of each newly allocated or extended page */
          assert( (nByte % pgsz)==0 );
          for(iPg=(sStat.st_size/pgsz); iPg<(nByte/pgsz); iPg++){
            int x = 0;
            if( seekAndWriteFd(pShmNode->hShm, iPg*pgsz + pgsz-1,"",1,&x)!=1 ){
              const char *zFile = pShmNode->zFilename;
              rc = unixLogError(SQLITE_IOERR_SHMSIZE, "write", zFile);
              goto shmpage_out;
            }
          }
        }
      }

      goto shmpage_out;
    }
    pShmNode->apRegion = apNew;
    while( pShmNode->nRegion<nReqRegion ){
      int nMap = szRegion*nShmPerMap;
      int i;
      void *pMem;
      if( pShmNode->hShm>=0 ){
        pMem = osMmap(0, nMap,
            pShmNode->isReadonly ? PROT_READ : PROT_READ|PROT_WRITE, 
            MAP_SHARED, pShmNode->hShm, szRegion*(i64)pShmNode->nRegion
        );
        if( pMem==MAP_FAILED ){
          rc = unixLogError(SQLITE_IOERR_SHMMAP, "mmap", pShmNode->zFilename);
          goto shmpage_out;
        }
      }else{
        pMem = sqlite3_malloc64(szRegion);

  assert( ofst>=0 && ofst+n<=SQLITE_SHM_NLOCK );
  assert( n>=1 );
  assert( flags==(SQLITE_SHM_LOCK | SQLITE_SHM_SHARED)
       || flags==(SQLITE_SHM_LOCK | SQLITE_SHM_EXCLUSIVE)
       || flags==(SQLITE_SHM_UNLOCK | SQLITE_SHM_SHARED)
       || flags==(SQLITE_SHM_UNLOCK | SQLITE_SHM_EXCLUSIVE) );
  assert( n==1 || (flags & SQLITE_SHM_EXCLUSIVE)!=0 );
  assert( pShmNode->hShm>=0 || pDbFd->pInode->bProcessLock==1 );
  assert( pShmNode->hShm<0 || pDbFd->pInode->bProcessLock==0 );

  mask = (1<<(ofst+n)) - (1<<ofst);
  assert( n>1 || mask==(1<<ofst) );
  sqlite3_mutex_enter(pShmNode->pShmMutex);
  if( flags & SQLITE_SHM_UNLOCK ){
    u16 allMask = 0; /* Mask of locks held by siblings */

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

  return SQLITE_OK;