SQLite

Check-in [534aab837e]
Login

Many hyperlinks are disabled.
Use anonymous login to enable hyperlinks.

Overview
Comment:Optionally support shared-memory in /dev/shm or other tmpfs directory using an unsupported compile-time option.
Downloads: Tarball | ZIP archive
Timelines: family | ancestors | descendants | both | trunk
Files: files | file ages | folders
SHA1: 534aab837e5a34fd58963ab2b505bfbb9da2e8ee
User & Date: drh 2010-07-15 18:32:41.000
References
2010-07-15
18:38
Previous check-in [534aab837e] accidently reverted some changes from [a121cd80c5]. This check-in restores those changes. (check-in: abff795f38 user: drh tags: trunk)
Context
2010-07-15
18:38
Previous check-in [534aab837e] accidently reverted some changes from [a121cd80c5]. This check-in restores those changes. (check-in: abff795f38 user: drh tags: trunk)
18:32
Optionally support shared-memory in /dev/shm or other tmpfs directory using an unsupported compile-time option. (check-in: 534aab837e user: drh tags: trunk)
18:20
Changes to wal.c so that SQLite can read even if the WAL file is opened read-only, provided the wal-index (shm file) is opened read/write. (check-in: 932d19da73 user: dan tags: trunk)
Changes
Unified Diff Ignore Whitespace Patch
Changes to src/os_unix.c.
3278
3279
3280
3281
3282
3283
3284
3285


3286
3287
3288
3289
3290
3291







3292
3293
3294
3295
3296
3297
3298
** Open a shared-memory area associated with open database file pDbFd.  
** This particular implementation uses mmapped files.
**
** The file used to implement shared-memory is in the same directory
** as the open database file and has the same name as the open database
** file with the "-shm" suffix added.  For example, if the database file
** is "/home/user1/config.db" then the file that is created and mmapped
** for shared memory will be called "/home/user1/config.db-shm".  We


** experimented with using files in /dev/tmp or an some other tmpfs mount.
** But if a file in a different directory from the database file is used,
** then differing access permissions or a chroot() might cause two different 
** processes on the same database to end up using different files for 
** shared memory - meaning that their memory would not really be shared - 
** resulting in database corruption.







**
** When opening a new shared-memory file, if no other instances of that
** 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 unixOpenSharedMemory(unixFile *pDbFd){
  struct unixShm *p = 0;          /* The connection to be opened */







|
>
>
|
|
|
|
|
|
>
>
>
>
>
>
>







3278
3279
3280
3281
3282
3283
3284
3285
3286
3287
3288
3289
3290
3291
3292
3293
3294
3295
3296
3297
3298
3299
3300
3301
3302
3303
3304
3305
3306
3307
** Open a shared-memory area associated with open database file pDbFd.  
** This particular implementation uses mmapped files.
**
** The file used to implement shared-memory is in the same directory
** as the open database file and has the same name as the open database
** file with the "-shm" suffix added.  For example, if the database file
** is "/home/user1/config.db" then the file that is created and mmapped
** for shared memory will be called "/home/user1/config.db-shm".  
**
** Another approach to is to use files in /dev/shm or /dev/tmp or an
** some other tmpfs mount. But if a file in a different directory
** from the database file is used, then differing access permissions
** or a chroot() might cause two different processes on the same
** database to end up using different files for shared memory - 
** meaning that their memory would not really be shared - resulting
** in database corruption.  Nevertheless, this tmpfs file usage
** can be enabled at compile-time using -DSQLITE_SHM_DIRECTORY="/dev/shm"
** or the equivalent.  The use of the SQLITE_SHM_DIRECTORY compile-time
** option results in an incompatible build of SQLite;  builds of SQLite
** that with differing SQLITE_SHM_DIRECTORY settings attempt to use the
** same database file at the same time, database corruption will likely
** result. The SQLITE_SHM_DIRECTORY compile-time option is considered
** "unsupported" and may go away in a future SQLite release.
**
** When opening a new shared-memory file, if no other instances of that
** 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 unixOpenSharedMemory(unixFile *pDbFd){
  struct unixShm *p = 0;          /* The connection to be opened */
3323
3324
3325
3326
3327
3328
3329



3330

3331
3332
3333
3334
3335
3336
3337





3338

3339
3340
3341
3342
3343
3344
3345
    ** with are subject to the current umask setting.
    */
    if( fstat(pDbFd->h, &sStat) ){
      rc = SQLITE_IOERR_FSTAT;
      goto shm_open_err;
    }




    nShmFilename = 5 + (int)strlen(pDbFd->zPath);

    pShmNode = sqlite3_malloc( sizeof(*pShmNode) + nShmFilename );
    if( pShmNode==0 ){
      rc = SQLITE_NOMEM;
      goto shm_open_err;
    }
    memset(pShmNode, 0, sizeof(*pShmNode));
    zShmFilename = pShmNode->zFilename = (char*)&pShmNode[1];





    sqlite3_snprintf(nShmFilename, zShmFilename, "%s-shm", 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;







>
>
>

>







>
>
>
>
>

>







3332
3333
3334
3335
3336
3337
3338
3339
3340
3341
3342
3343
3344
3345
3346
3347
3348
3349
3350
3351
3352
3353
3354
3355
3356
3357
3358
3359
3360
3361
3362
3363
3364
    ** with are subject to the current umask setting.
    */
    if( fstat(pDbFd->h, &sStat) ){
      rc = SQLITE_IOERR_FSTAT;
      goto shm_open_err;
    }

#ifdef SQLITE_SHM_DIRECTORY
    nShmFilename = sizeof(SQLITE_SHM_DIRECTORY) + 30;
#else
    nShmFilename = 5 + (int)strlen(pDbFd->zPath);
#endif
    pShmNode = sqlite3_malloc( sizeof(*pShmNode) + nShmFilename );
    if( pShmNode==0 ){
      rc = SQLITE_NOMEM;
      goto shm_open_err;
    }
    memset(pShmNode, 0, sizeof(*pShmNode));
    zShmFilename = pShmNode->zFilename = (char*)&pShmNode[1];
#ifdef SQLITE_SHM_DIRECTORY
    sqlite3_snprintf(nShmFilename, zShmFilename, 
                     SQLITE_SHM_DIRECTORY "/sqlite-shm-%x-%x",
                     (u32)sStat.st_ino, (u32)sStat.st_dev);
#else
    sqlite3_snprintf(nShmFilename, zShmFilename, "%s-shm", pDbFd->zPath);
#endif
    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;
4307
4308
4309
4310
4311
4312
4313
4314
4315
4316
4317
4318
4319
4320
4321
4322
4323
4324
4325
4326
4327
4328
4329
4330
4331
4332
4333
4334
4335
4336
4337
4338
4339
4340
4341
** to create new files with. If no error occurs, then SQLITE_OK is returned
** and a value suitable for passing as the third argument to open(2) is
** written to *pMode. If an IO error occurs, an SQLite error code is 
** returned and the value of *pMode is not modified.
**
** If the file being opened is a temporary file, it is always created with
** the octal permissions 0600 (read/writable by owner only). If the file
** is a database or master journal file, it is created with the permissions 
** mask SQLITE_DEFAULT_FILE_PERMISSIONS.
**
** Finally, if the file being opened is a WAL or regular journal file, then 
** this function queries the file-system for the permissions on the 
** corresponding database file and sets *pMode to this value. Whenever 
** possible, WAL and journal files are created using the same permissions 
** as the associated database file.
*/
static int findCreateFileMode(
  const char *zPath,              /* Path of file (possibly) being created */
  int flags,                      /* Flags passed as 4th argument to xOpen() */
  mode_t *pMode                   /* OUT: Permissions to open file with */
){
  int rc = SQLITE_OK;             /* Return Code */
  if( flags & (SQLITE_OPEN_WAL|SQLITE_OPEN_MAIN_JOURNAL) ){
    char zDb[MAX_PATHNAME+1];     /* Database file path */
    int nDb;                      /* Number of valid bytes in zDb */
    struct stat sStat;            /* Output of stat() on database file */

    nDb = sqlite3Strlen30(zPath) - ((flags & SQLITE_OPEN_WAL) ? 4 : 8);
    memcpy(zDb, zPath, nDb);
    zDb[nDb] = '\0';
    if( 0==stat(zDb, &sStat) ){
      *pMode = sStat.st_mode & 0777;
    }else{
      rc = SQLITE_IOERR_FSTAT;
    }







|
|

|
|
|
<
|







|




|







4326
4327
4328
4329
4330
4331
4332
4333
4334
4335
4336
4337
4338

4339
4340
4341
4342
4343
4344
4345
4346
4347
4348
4349
4350
4351
4352
4353
4354
4355
4356
4357
4358
4359
** to create new files with. If no error occurs, then SQLITE_OK is returned
** and a value suitable for passing as the third argument to open(2) is
** written to *pMode. If an IO error occurs, an SQLite error code is 
** returned and the value of *pMode is not modified.
**
** If the file being opened is a temporary file, it is always created with
** the octal permissions 0600 (read/writable by owner only). If the file
** is a database, journal or master journal file, it is created with the
** permissions mask SQLITE_DEFAULT_FILE_PERMISSIONS.
**
** Finally, if the file being opened is a WAL file, then this function
** queries the file-system for the permissions on the corresponding database
** file and sets *pMode to this value. Whenever possible, WAL files are 

** created using the same permissions as the associated database file.
*/
static int findCreateFileMode(
  const char *zPath,              /* Path of file (possibly) being created */
  int flags,                      /* Flags passed as 4th argument to xOpen() */
  mode_t *pMode                   /* OUT: Permissions to open file with */
){
  int rc = SQLITE_OK;             /* Return Code */
  if( flags & SQLITE_OPEN_WAL ){
    char zDb[MAX_PATHNAME+1];     /* Database file path */
    int nDb;                      /* Number of valid bytes in zDb */
    struct stat sStat;            /* Output of stat() on database file */

    nDb = sqlite3Strlen30(zPath) - 4;
    memcpy(zDb, zPath, nDb);
    zDb[nDb] = '\0';
    if( 0==stat(zDb, &sStat) ){
      *pMode = sStat.st_mode & 0777;
    }else{
      rc = SQLITE_IOERR_FSTAT;
    }
4470
4471
4472
4473
4474
4475
4476
4477
4478
4479
4480
4481
4482
4483
4484
  openFlags |= (O_LARGEFILE|O_BINARY);

  if( fd<0 ){
    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 = 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);







<







4488
4489
4490
4491
4492
4493
4494

4495
4496
4497
4498
4499
4500
4501
  openFlags |= (O_LARGEFILE|O_BINARY);

  if( fd<0 ){
    mode_t openMode;              /* Permissions to create file with */
    rc = findCreateFileMode(zName, flags, &openMode);
    if( rc!=SQLITE_OK ){
      assert( !p->pUnused );

      return rc;
    }
    fd = 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);