2 Unix SMB/CIFS implementation.
4 trivial database library
6 Copyright (C) Andrew Tridgell 1999-2005
7 Copyright (C) Paul `Rusty' Russell 2000
8 Copyright (C) Jeremy Allison 2000-2003
10 ** NOTE! The following LGPL license applies to the tdb
11 ** library. This does NOT imply that all of Samba is released
14 This library is free software; you can redistribute it and/or
15 modify it under the terms of the GNU Lesser General Public
16 License as published by the Free Software Foundation; either
17 version 3 of the License, or (at your option) any later version.
19 This library is distributed in the hope that it will be useful,
20 but WITHOUT ANY WARRANTY; without even the implied warranty of
21 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
22 Lesser General Public License for more details.
24 You should have received a copy of the GNU Lesser General Public
25 License along with this library; if not, see <http://www.gnu.org/licenses/>.
28 #include "tdb_private.h"
30 /* all contexts, to ensure no double-opens (fcntl locks don't nest!) */
31 static struct tdb_context
*tdbs
= NULL
;
33 /* We use two hashes to double-check they're using the right hash function. */
34 void tdb_header_hash(struct tdb_context
*tdb
,
35 uint32_t *magic1_hash
, uint32_t *magic2_hash
)
38 uint32_t tdb_magic
= TDB_MAGIC
;
40 hash_key
.dptr
= discard_const_p(unsigned char, TDB_MAGIC_FOOD
);
41 hash_key
.dsize
= sizeof(TDB_MAGIC_FOOD
);
42 *magic1_hash
= tdb
->hash_fn(&hash_key
);
44 hash_key
.dptr
= (unsigned char *)CONVERT(tdb_magic
);
45 hash_key
.dsize
= sizeof(tdb_magic
);
46 *magic2_hash
= tdb
->hash_fn(&hash_key
);
48 /* Make sure at least one hash is non-zero! */
49 if (*magic1_hash
== 0 && *magic2_hash
== 0)
53 /* initialise a new database with a specified hash size */
54 static int tdb_new_database(struct tdb_context
*tdb
, int hash_size
)
56 struct tdb_header
*newdb
;
60 /* We make it up in memory, then write it out if not internal */
61 size
= sizeof(struct tdb_header
) + (hash_size
+1)*sizeof(tdb_off_t
);
62 if (!(newdb
= (struct tdb_header
*)calloc(size
, 1))) {
63 tdb
->ecode
= TDB_ERR_OOM
;
67 /* Fill in the header */
68 newdb
->version
= TDB_VERSION
;
69 newdb
->hash_size
= hash_size
;
71 tdb_header_hash(tdb
, &newdb
->magic1_hash
, &newdb
->magic2_hash
);
73 /* Make sure older tdbs (which don't check the magic hash fields)
74 * will refuse to open this TDB. */
75 if (tdb
->flags
& TDB_INCOMPATIBLE_HASH
)
76 newdb
->rwlocks
= TDB_HASH_RWLOCK_MAGIC
;
78 if (tdb
->flags
& TDB_INTERNAL
) {
80 tdb
->map_ptr
= (char *)newdb
;
81 memcpy(&tdb
->header
, newdb
, sizeof(tdb
->header
));
82 /* Convert the `ondisk' version if asked. */
86 if (lseek(tdb
->fd
, 0, SEEK_SET
) == -1)
89 if (ftruncate(tdb
->fd
, 0) == -1)
92 /* This creates an endian-converted header, as if read from disk */
94 memcpy(&tdb
->header
, newdb
, sizeof(tdb
->header
));
95 /* Don't endian-convert the magic food! */
96 memcpy(newdb
->magic_food
, TDB_MAGIC_FOOD
, strlen(TDB_MAGIC_FOOD
)+1);
97 /* we still have "ret == -1" here */
98 if (tdb_write_all(tdb
->fd
, newdb
, size
))
108 static int tdb_already_open(dev_t device
,
111 struct tdb_context
*i
;
113 for (i
= tdbs
; i
; i
= i
->next
) {
114 if (i
->device
== device
&& i
->inode
== ino
) {
122 /* open the database, creating it if necessary
124 The open_flags and mode are passed straight to the open call on the
125 database file. A flags value of O_WRONLY is invalid. The hash size
126 is advisory, use zero for a default value.
128 Return is NULL on error, in which case errno is also set. Don't
129 try to call tdb_error or tdb_errname, just do strerror(errno).
131 @param name may be NULL for internal databases. */
132 _PUBLIC_
struct tdb_context
*tdb_open(const char *name
, int hash_size
, int tdb_flags
,
133 int open_flags
, mode_t mode
)
135 return tdb_open_ex(name
, hash_size
, tdb_flags
, open_flags
, mode
, NULL
, NULL
);
138 /* a default logging function */
139 static void null_log_fn(struct tdb_context
*tdb
, enum tdb_debug_level level
, const char *fmt
, ...) PRINTF_ATTRIBUTE(3, 4);
140 static void null_log_fn(struct tdb_context
*tdb
, enum tdb_debug_level level
, const char *fmt
, ...)
144 static bool check_header_hash(struct tdb_context
*tdb
,
145 bool default_hash
, uint32_t *m1
, uint32_t *m2
)
147 tdb_header_hash(tdb
, m1
, m2
);
148 if (tdb
->header
.magic1_hash
== *m1
&&
149 tdb
->header
.magic2_hash
== *m2
) {
153 /* If they explicitly set a hash, always respect it. */
157 /* Otherwise, try the other inbuilt hash. */
158 if (tdb
->hash_fn
== tdb_old_hash
)
159 tdb
->hash_fn
= tdb_jenkins_hash
;
161 tdb
->hash_fn
= tdb_old_hash
;
162 return check_header_hash(tdb
, false, m1
, m2
);
165 _PUBLIC_
struct tdb_context
*tdb_open_ex(const char *name
, int hash_size
, int tdb_flags
,
166 int open_flags
, mode_t mode
,
167 const struct tdb_logging_context
*log_ctx
,
168 tdb_hash_func hash_fn
)
170 struct tdb_context
*tdb
;
172 int rev
= 0, locked
= 0;
176 const char *hash_alg
;
177 uint32_t magic1
, magic2
;
179 if (!(tdb
= (struct tdb_context
*)calloc(1, sizeof *tdb
))) {
191 tdb
->flags
= tdb_flags
;
192 tdb
->open_flags
= open_flags
;
196 tdb
->log
.log_fn
= null_log_fn
;
197 tdb
->log
.log_private
= NULL
;
200 if (name
== NULL
&& (tdb_flags
& TDB_INTERNAL
)) {
201 name
= "__TDB_INTERNAL__";
205 tdb
->name
= discard_const_p(char, "__NULL__");
206 TDB_LOG((tdb
, TDB_DEBUG_FATAL
, "tdb_open_ex: called with name == NULL\n"));
212 /* now make a copy of the name, as the caller memory might went away */
213 if (!(tdb
->name
= (char *)strdup(name
))) {
215 * set the name as the given string, so that tdb_name() will
216 * work in case of an error.
218 tdb
->name
= discard_const_p(char, name
);
219 TDB_LOG((tdb
, TDB_DEBUG_ERROR
, "tdb_open_ex: can't strdup(%s)\n",
227 tdb
->hash_fn
= hash_fn
;
228 hash_alg
= "the user defined";
230 /* This controls what we use when creating a tdb. */
231 if (tdb
->flags
& TDB_INCOMPATIBLE_HASH
) {
232 tdb
->hash_fn
= tdb_jenkins_hash
;
234 tdb
->hash_fn
= tdb_old_hash
;
236 hash_alg
= "either default";
239 /* cache the page size */
240 tdb
->page_size
= getpagesize();
241 if (tdb
->page_size
<= 0) {
242 tdb
->page_size
= 0x2000;
245 tdb
->max_dead_records
= (tdb_flags
& TDB_VOLATILE
) ? 5 : 0;
247 if ((open_flags
& O_ACCMODE
) == O_WRONLY
) {
248 TDB_LOG((tdb
, TDB_DEBUG_ERROR
, "tdb_open_ex: can't open tdb %s write-only\n",
255 hash_size
= DEFAULT_HASH_SIZE
;
256 if ((open_flags
& O_ACCMODE
) == O_RDONLY
) {
258 /* read only databases don't do locking or clear if first */
259 tdb
->flags
|= TDB_NOLOCK
;
260 tdb
->flags
&= ~TDB_CLEAR_IF_FIRST
;
263 if ((tdb
->flags
& TDB_ALLOW_NESTING
) &&
264 (tdb
->flags
& TDB_DISALLOW_NESTING
)) {
265 tdb
->ecode
= TDB_ERR_NESTING
;
266 TDB_LOG((tdb
, TDB_DEBUG_FATAL
, "tdb_open_ex: "
267 "allow_nesting and disallow_nesting are not allowed together!"));
272 if (getenv("TDB_NO_FSYNC")) {
273 tdb
->flags
|= TDB_NOSYNC
;
277 * TDB_ALLOW_NESTING is the default behavior.
278 * Note: this may change in future versions!
280 if (!(tdb
->flags
& TDB_DISALLOW_NESTING
)) {
281 tdb
->flags
|= TDB_ALLOW_NESTING
;
284 /* internal databases don't mmap or lock, and start off cleared */
285 if (tdb
->flags
& TDB_INTERNAL
) {
286 tdb
->flags
|= (TDB_NOLOCK
| TDB_NOMMAP
);
287 tdb
->flags
&= ~TDB_CLEAR_IF_FIRST
;
288 if (tdb_new_database(tdb
, hash_size
) != 0) {
289 TDB_LOG((tdb
, TDB_DEBUG_ERROR
, "tdb_open_ex: tdb_new_database failed!"));
295 if ((tdb
->fd
= open(name
, open_flags
, mode
)) == -1) {
296 TDB_LOG((tdb
, TDB_DEBUG_WARNING
, "tdb_open_ex: could not open file %s: %s\n",
297 name
, strerror(errno
)));
298 goto fail
; /* errno set by open(2) */
301 /* on exec, don't inherit the fd */
302 v
= fcntl(tdb
->fd
, F_GETFD
, 0);
303 fcntl(tdb
->fd
, F_SETFD
, v
| FD_CLOEXEC
);
305 /* ensure there is only one process initialising at once */
306 if (tdb_nest_lock(tdb
, OPEN_LOCK
, F_WRLCK
, TDB_LOCK_WAIT
) == -1) {
307 TDB_LOG((tdb
, TDB_DEBUG_ERROR
, "tdb_open_ex: failed to get open lock on %s: %s\n",
308 name
, strerror(errno
)));
309 goto fail
; /* errno set by tdb_brlock */
312 /* we need to zero database if we are the only one with it open */
313 if ((tdb_flags
& TDB_CLEAR_IF_FIRST
) &&
315 (locked
= (tdb_nest_lock(tdb
, ACTIVE_LOCK
, F_WRLCK
, TDB_LOCK_NOWAIT
|TDB_LOCK_PROBE
) == 0))) {
317 ret
= tdb_brlock(tdb
, F_WRLCK
, FREELIST_TOP
, 0,
320 TDB_LOG((tdb
, TDB_DEBUG_FATAL
, "tdb_open_ex: "
321 "tdb_brlock failed for %s: %s\n",
322 name
, strerror(errno
)));
325 ret
= tdb_new_database(tdb
, hash_size
);
327 TDB_LOG((tdb
, TDB_DEBUG_FATAL
, "tdb_open_ex: "
328 "tdb_new_database failed for %s: %s\n",
329 name
, strerror(errno
)));
333 ret
= tdb_brunlock(tdb
, F_WRLCK
, FREELIST_TOP
, 0);
335 TDB_LOG((tdb
, TDB_DEBUG_FATAL
, "tdb_open_ex: "
336 "tdb_unlockall failed for %s: %s\n",
337 name
, strerror(errno
)));
340 ret
= lseek(tdb
->fd
, 0, SEEK_SET
);
342 TDB_LOG((tdb
, TDB_DEBUG_FATAL
, "tdb_open_ex: "
343 "lseek failed for %s: %s\n",
344 name
, strerror(errno
)));
350 if (read(tdb
->fd
, &tdb
->header
, sizeof(tdb
->header
)) != sizeof(tdb
->header
)
351 || strcmp(tdb
->header
.magic_food
, TDB_MAGIC_FOOD
) != 0) {
352 if (!(open_flags
& O_CREAT
) || tdb_new_database(tdb
, hash_size
) == -1) {
354 errno
= EIO
; /* ie bad format or something */
358 rev
= (tdb
->flags
& TDB_CONVERT
);
359 } else if (tdb
->header
.version
!= TDB_VERSION
360 && !(rev
= (tdb
->header
.version
==TDB_BYTEREV(TDB_VERSION
)))) {
365 vp
= (unsigned char *)&tdb
->header
.version
;
366 vertest
= (((uint32_t)vp
[0]) << 24) | (((uint32_t)vp
[1]) << 16) |
367 (((uint32_t)vp
[2]) << 8) | (uint32_t)vp
[3];
368 tdb
->flags
|= (vertest
==TDB_VERSION
) ? TDB_BIGENDIAN
: 0;
370 tdb
->flags
&= ~TDB_CONVERT
;
372 tdb
->flags
|= TDB_CONVERT
;
373 tdb_convert(&tdb
->header
, sizeof(tdb
->header
));
375 if (fstat(tdb
->fd
, &st
) == -1)
378 if (tdb
->header
.rwlocks
!= 0 &&
379 tdb
->header
.rwlocks
!= TDB_HASH_RWLOCK_MAGIC
) {
380 TDB_LOG((tdb
, TDB_DEBUG_ERROR
, "tdb_open_ex: spinlocks no longer supported\n"));
384 if ((tdb
->header
.magic1_hash
== 0) && (tdb
->header
.magic2_hash
== 0)) {
385 /* older TDB without magic hash references */
386 tdb
->hash_fn
= tdb_old_hash
;
387 } else if (!check_header_hash(tdb
, !hash_fn
, &magic1
, &magic2
)) {
388 TDB_LOG((tdb
, TDB_DEBUG_FATAL
, "tdb_open_ex: "
389 "%s was not created with %s hash function we are using\n"
390 "magic1_hash[0x%08X %s 0x%08X] "
391 "magic2_hash[0x%08X %s 0x%08X]\n",
393 tdb
->header
.magic1_hash
,
394 (tdb
->header
.magic1_hash
== magic1
) ? "==" : "!=",
396 tdb
->header
.magic2_hash
,
397 (tdb
->header
.magic2_hash
== magic2
) ? "==" : "!=",
403 /* Is it already in the open list? If so, fail. */
404 if (tdb_already_open(st
.st_dev
, st
.st_ino
)) {
405 TDB_LOG((tdb
, TDB_DEBUG_ERROR
, "tdb_open_ex: "
406 "%s (%d,%d) is already open in this process\n",
407 name
, (int)st
.st_dev
, (int)st
.st_ino
));
412 /* Beware truncation! */
413 tdb
->map_size
= st
.st_size
;
414 if (tdb
->map_size
!= st
.st_size
) {
415 /* Ensure ecode is set for log fn. */
416 tdb
->ecode
= TDB_ERR_IO
;
417 TDB_LOG((tdb
, TDB_DEBUG_FATAL
, "tdb_open_ex: "
418 "len %llu too large!\n", (long long)st
.st_size
));
423 tdb
->device
= st
.st_dev
;
424 tdb
->inode
= st
.st_ino
;
427 if (tdb_nest_unlock(tdb
, ACTIVE_LOCK
, F_WRLCK
, false) == -1) {
428 TDB_LOG((tdb
, TDB_DEBUG_ERROR
, "tdb_open_ex: "
429 "failed to release ACTIVE_LOCK on %s: %s\n",
430 name
, strerror(errno
)));
436 /* We always need to do this if the CLEAR_IF_FIRST flag is set, even if
437 we didn't get the initial exclusive lock as we need to let all other
438 users know we're using it. */
440 if (tdb_flags
& TDB_CLEAR_IF_FIRST
) {
441 /* leave this lock in place to indicate it's in use */
442 if (tdb_nest_lock(tdb
, ACTIVE_LOCK
, F_RDLCK
, TDB_LOCK_WAIT
) == -1) {
447 /* if needed, run recovery */
448 if (tdb_transaction_recover(tdb
) == -1) {
454 char tracefile
[strlen(name
) + 32];
456 snprintf(tracefile
, sizeof(tracefile
),
457 "%s.trace.%li", name
, (long)getpid());
458 tdb
->tracefd
= open(tracefile
, O_WRONLY
|O_CREAT
|O_EXCL
, 0600);
459 if (tdb
->tracefd
>= 0) {
460 tdb_enable_seqnum(tdb
);
461 tdb_trace_open(tdb
, "tdb_open", hash_size
, tdb_flags
,
464 TDB_LOG((tdb
, TDB_DEBUG_ERROR
, "tdb_open_ex: failed to open trace file %s!\n", tracefile
));
469 /* Internal (memory-only) databases skip all the code above to
470 * do with disk files, and resume here by releasing their
471 * open lock and hooking into the active list. */
472 if (tdb_nest_unlock(tdb
, OPEN_LOCK
, F_WRLCK
, false) == -1) {
480 { int save_errno
= errno
;
489 if (tdb
->flags
& TDB_INTERNAL
)
490 SAFE_FREE(tdb
->map_ptr
);
495 if (close(tdb
->fd
) != 0)
496 TDB_LOG((tdb
, TDB_DEBUG_ERROR
, "tdb_open_ex: failed to close tdb->fd on error!\n"));
497 SAFE_FREE(tdb
->lockrecs
);
498 SAFE_FREE(tdb
->name
);
506 * Set the maximum number of dead records per hash chain
509 _PUBLIC_
void tdb_set_max_dead(struct tdb_context
*tdb
, int max_dead
)
511 tdb
->max_dead_records
= max_dead
;
517 * @returns -1 for error; 0 for success.
519 _PUBLIC_
int tdb_close(struct tdb_context
*tdb
)
521 struct tdb_context
**i
;
524 if (tdb
->transaction
) {
525 tdb_transaction_cancel(tdb
);
527 tdb_trace(tdb
, "tdb_close");
530 if (tdb
->flags
& TDB_INTERNAL
)
531 SAFE_FREE(tdb
->map_ptr
);
535 SAFE_FREE(tdb
->name
);
537 ret
= close(tdb
->fd
);
540 SAFE_FREE(tdb
->lockrecs
);
542 /* Remove from contexts list */
543 for (i
= &tdbs
; *i
; i
= &(*i
)->next
) {
553 memset(tdb
, 0, sizeof(*tdb
));
559 /* register a loging function */
560 _PUBLIC_
void tdb_set_logging_function(struct tdb_context
*tdb
,
561 const struct tdb_logging_context
*log_ctx
)
566 _PUBLIC_
void *tdb_get_logging_private(struct tdb_context
*tdb
)
568 return tdb
->log
.log_private
;
571 static int tdb_reopen_internal(struct tdb_context
*tdb
, bool active_lock
)
573 #if !defined(LIBREPLACE_PREAD_NOT_REPLACED) || \
574 !defined(LIBREPLACE_PWRITE_NOT_REPLACED)
578 if (tdb
->flags
& TDB_INTERNAL
) {
579 return 0; /* Nothing to do. */
582 if (tdb_have_extra_locks(tdb
)) {
583 TDB_LOG((tdb
, TDB_DEBUG_ERROR
, "tdb_reopen: reopen not allowed with locks held\n"));
587 if (tdb
->transaction
!= 0) {
588 TDB_LOG((tdb
, TDB_DEBUG_ERROR
, "tdb_reopen: reopen not allowed inside a transaction\n"));
592 /* If we have real pread & pwrite, we can skip reopen. */
593 #if !defined(LIBREPLACE_PREAD_NOT_REPLACED) || \
594 !defined(LIBREPLACE_PWRITE_NOT_REPLACED)
595 if (tdb_munmap(tdb
) != 0) {
596 TDB_LOG((tdb
, TDB_DEBUG_FATAL
, "tdb_reopen: munmap failed (%s)\n", strerror(errno
)));
599 if (close(tdb
->fd
) != 0)
600 TDB_LOG((tdb
, TDB_DEBUG_FATAL
, "tdb_reopen: WARNING closing tdb->fd failed!\n"));
601 tdb
->fd
= open(tdb
->name
, tdb
->open_flags
& ~(O_CREAT
|O_TRUNC
), 0);
603 TDB_LOG((tdb
, TDB_DEBUG_FATAL
, "tdb_reopen: open failed (%s)\n", strerror(errno
)));
606 if (fstat(tdb
->fd
, &st
) != 0) {
607 TDB_LOG((tdb
, TDB_DEBUG_FATAL
, "tdb_reopen: fstat failed (%s)\n", strerror(errno
)));
610 if (st
.st_ino
!= tdb
->inode
|| st
.st_dev
!= tdb
->device
) {
611 TDB_LOG((tdb
, TDB_DEBUG_FATAL
, "tdb_reopen: file dev/inode has changed!\n"));
614 if (tdb_mmap(tdb
) != 0) {
617 #endif /* fake pread or pwrite */
619 /* We may still think we hold the active lock. */
620 tdb
->num_lockrecs
= 0;
621 SAFE_FREE(tdb
->lockrecs
);
623 if (active_lock
&& tdb_nest_lock(tdb
, ACTIVE_LOCK
, F_RDLCK
, TDB_LOCK_WAIT
) == -1) {
624 TDB_LOG((tdb
, TDB_DEBUG_FATAL
, "tdb_reopen: failed to obtain active lock\n"));
635 /* reopen a tdb - this can be used after a fork to ensure that we have an independent
636 seek pointer from our parent and to re-establish locks */
637 _PUBLIC_
int tdb_reopen(struct tdb_context
*tdb
)
639 return tdb_reopen_internal(tdb
, tdb
->flags
& TDB_CLEAR_IF_FIRST
);
642 /* reopen all tdb's */
643 _PUBLIC_
int tdb_reopen_all(int parent_longlived
)
645 struct tdb_context
*tdb
;
647 for (tdb
=tdbs
; tdb
; tdb
= tdb
->next
) {
648 bool active_lock
= (tdb
->flags
& TDB_CLEAR_IF_FIRST
);
651 * If the parent is longlived (ie. a
652 * parent daemon architecture), we know
653 * it will keep it's active lock on a
654 * tdb opened with CLEAR_IF_FIRST. Thus
655 * for child processes we don't have to
656 * add an active lock. This is essential
657 * to improve performance on systems that
658 * keep POSIX locks as a non-scalable data
659 * structure in the kernel.
661 if (parent_longlived
) {
662 /* Ensure no clear-if-first. */
666 if (tdb_reopen_internal(tdb
, active_lock
) != 0)