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 #define TDB_MARK_LOCK 0x80000000
32 void tdb_setalarm_sigptr(struct tdb_context *tdb, volatile sig_atomic_t *ptr)
34 tdb->interrupt_sig_ptr = ptr;
37 /* a byte range locking function - return 0 on success
38 this functions locks/unlocks 1 byte at the specified offset.
40 On error, errno is also set so that errors are passed back properly
43 note that a len of zero means lock to end of file
45 int tdb_brlock(struct tdb_context *tdb, tdb_off_t offset,
46 int rw_type, int lck_type, int probe, size_t len)
51 if (tdb->flags & TDB_NOLOCK) {
55 if ((rw_type == F_WRLCK) && (tdb->read_only || tdb->traverse_read)) {
56 tdb->ecode = TDB_ERR_RDONLY;
61 fl.l_whence = SEEK_SET;
67 ret = fcntl(tdb->fd,lck_type,&fl);
69 /* Check for a sigalarm break. */
70 if (ret == -1 && errno == EINTR &&
71 tdb->interrupt_sig_ptr &&
72 *tdb->interrupt_sig_ptr) {
75 } while (ret == -1 && errno == EINTR);
78 /* Generic lock error. errno set by fcntl.
79 * EAGAIN is an expected return from non-blocking
81 if (!probe && lck_type != F_SETLK) {
82 /* Ensure error code is set for log fun to examine. */
83 tdb->ecode = TDB_ERR_LOCK;
84 TDB_LOG((tdb, TDB_DEBUG_TRACE,"tdb_brlock failed (fd=%d) at offset %d rw_type=%d lck_type=%d len=%d\n",
85 tdb->fd, offset, rw_type, lck_type, (int)len));
94 upgrade a read lock to a write lock. This needs to be handled in a
95 special way as some OSes (such as solaris) have too conservative
96 deadlock detection and claim a deadlock when progress can be
97 made. For those OSes we may loop for a while.
99 int tdb_brlock_upgrade(struct tdb_context *tdb, tdb_off_t offset, size_t len)
104 if (tdb_brlock(tdb, offset, F_WRLCK, F_SETLKW, 1, len) == 0) {
107 if (errno != EDEADLK) {
110 /* sleep for as short a time as we can - more portable than usleep() */
113 select(0, NULL, NULL, NULL, &tv);
115 TDB_LOG((tdb, TDB_DEBUG_TRACE,"tdb_brlock_upgrade failed at offset %d\n", offset));
120 /* lock a list in the database. list -1 is the alloc list */
121 static int _tdb_lock(struct tdb_context *tdb, int list, int ltype, int op)
123 struct tdb_lock_type *new_lck;
125 bool mark_lock = ((ltype & TDB_MARK_LOCK) == TDB_MARK_LOCK);
127 ltype &= ~TDB_MARK_LOCK;
129 /* a global lock allows us to avoid per chain locks */
130 if (tdb->global_lock.count &&
131 (ltype == tdb->global_lock.ltype || ltype == F_RDLCK)) {
135 if (tdb->global_lock.count) {
136 tdb->ecode = TDB_ERR_LOCK;
140 if (list < -1 || list >= (int)tdb->header.hash_size) {
141 tdb->ecode = TDB_ERR_LOCK;
142 TDB_LOG((tdb, TDB_DEBUG_ERROR,"tdb_lock: invalid list %d for ltype=%d\n",
146 if (tdb->flags & TDB_NOLOCK)
149 for (i=0; i<tdb->num_lockrecs; i++) {
150 if (tdb->lockrecs[i].list == list) {
151 if (tdb->lockrecs[i].count == 0) {
153 * Can't happen, see tdb_unlock(). It should
156 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_lock: "
157 "lck->count == 0 for list %d", list));
160 * Just increment the in-memory struct, posix locks
163 tdb->lockrecs[i].count++;
168 new_lck = (struct tdb_lock_type *)realloc(
170 sizeof(*tdb->lockrecs) * (tdb->num_lockrecs+1));
171 if (new_lck == NULL) {
175 tdb->lockrecs = new_lck;
177 /* Since fcntl locks don't nest, we do a lock for the first one,
178 and simply bump the count for future ones */
180 tdb->methods->tdb_brlock(tdb,FREELIST_TOP+4*list, ltype, op,
187 tdb->lockrecs[tdb->num_lockrecs].list = list;
188 tdb->lockrecs[tdb->num_lockrecs].count = 1;
189 tdb->lockrecs[tdb->num_lockrecs].ltype = ltype;
190 tdb->num_lockrecs += 1;
195 /* lock a list in the database. list -1 is the alloc list */
196 int tdb_lock(struct tdb_context *tdb, int list, int ltype)
199 ret = _tdb_lock(tdb, list, ltype, F_SETLKW);
201 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_lock failed on list %d "
202 "ltype=%d (%s)\n", list, ltype, strerror(errno)));
207 /* lock a list in the database. list -1 is the alloc list. non-blocking lock */
208 int tdb_lock_nonblock(struct tdb_context *tdb, int list, int ltype)
210 return _tdb_lock(tdb, list, ltype, F_SETLK);
214 /* unlock the database: returns void because it's too late for errors. */
215 /* changed to return int it may be interesting to know there
216 has been an error --simo */
217 int tdb_unlock(struct tdb_context *tdb, int list, int ltype)
221 struct tdb_lock_type *lck = NULL;
222 bool mark_lock = ((ltype & TDB_MARK_LOCK) == TDB_MARK_LOCK);
224 ltype &= ~TDB_MARK_LOCK;
226 /* a global lock allows us to avoid per chain locks */
227 if (tdb->global_lock.count &&
228 (ltype == tdb->global_lock.ltype || ltype == F_RDLCK)) {
232 if (tdb->global_lock.count) {
233 tdb->ecode = TDB_ERR_LOCK;
237 if (tdb->flags & TDB_NOLOCK)
241 if (list < -1 || list >= (int)tdb->header.hash_size) {
242 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_unlock: list %d invalid (%d)\n", list, tdb->header.hash_size));
246 for (i=0; i<tdb->num_lockrecs; i++) {
247 if (tdb->lockrecs[i].list == list) {
248 lck = &tdb->lockrecs[i];
253 if ((lck == NULL) || (lck->count == 0)) {
254 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_unlock: count is 0\n"));
258 if (lck->count > 1) {
264 * This lock has count==1 left, so we need to unlock it in the
265 * kernel. We don't bother with decrementing the in-memory array
266 * element, we're about to overwrite it with the last array element
273 ret = tdb->methods->tdb_brlock(tdb, FREELIST_TOP+4*list, F_UNLCK,
279 * Shrink the array by overwriting the element just unlocked with the
280 * last array element.
283 if (tdb->num_lockrecs > 1) {
284 *lck = tdb->lockrecs[tdb->num_lockrecs-1];
286 tdb->num_lockrecs -= 1;
289 * We don't bother with realloc when the array shrinks, but if we have
290 * a completely idle tdb we should get rid of the locked array.
293 if (tdb->num_lockrecs == 0) {
294 SAFE_FREE(tdb->lockrecs);
298 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_unlock: An error occurred unlocking!\n"));
303 get the transaction lock
305 int tdb_transaction_lock(struct tdb_context *tdb, int ltype)
307 if (tdb->global_lock.count) {
310 if (tdb->transaction_lock_count > 0) {
311 tdb->transaction_lock_count++;
315 if (tdb->methods->tdb_brlock(tdb, TRANSACTION_LOCK, ltype,
316 F_SETLKW, 0, 1) == -1) {
317 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_transaction_lock: failed to get transaction lock\n"));
318 tdb->ecode = TDB_ERR_LOCK;
321 tdb->transaction_lock_count++;
326 release the transaction lock
328 int tdb_transaction_unlock(struct tdb_context *tdb)
331 if (tdb->global_lock.count) {
334 if (tdb->transaction_lock_count > 1) {
335 tdb->transaction_lock_count--;
338 ret = tdb->methods->tdb_brlock(tdb, TRANSACTION_LOCK, F_UNLCK, F_SETLKW, 0, 1);
340 tdb->transaction_lock_count = 0;
348 /* lock/unlock entire database */
349 static int _tdb_lockall(struct tdb_context *tdb, int ltype, int op)
351 bool mark_lock = ((ltype & TDB_MARK_LOCK) == TDB_MARK_LOCK);
353 ltype &= ~TDB_MARK_LOCK;
355 /* There are no locks on read-only dbs */
356 if (tdb->read_only || tdb->traverse_read) {
357 tdb->ecode = TDB_ERR_LOCK;
361 if (tdb->global_lock.count && tdb->global_lock.ltype == ltype) {
362 tdb->global_lock.count++;
366 if (tdb->global_lock.count) {
367 /* a global lock of a different type exists */
368 tdb->ecode = TDB_ERR_LOCK;
372 if (tdb->num_locks != 0) {
373 /* can't combine global and chain locks */
374 tdb->ecode = TDB_ERR_LOCK;
379 tdb->methods->tdb_brlock(tdb, FREELIST_TOP, ltype, op,
380 0, 4*tdb->header.hash_size)) {
381 if (op == F_SETLKW) {
382 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_lockall failed (%s)\n", strerror(errno)));
387 tdb->global_lock.count = 1;
388 tdb->global_lock.ltype = ltype;
395 /* unlock entire db */
396 static int _tdb_unlockall(struct tdb_context *tdb, int ltype)
398 bool mark_lock = ((ltype & TDB_MARK_LOCK) == TDB_MARK_LOCK);
400 ltype &= ~TDB_MARK_LOCK;
402 /* There are no locks on read-only dbs */
403 if (tdb->read_only || tdb->traverse_read) {
404 tdb->ecode = TDB_ERR_LOCK;
408 if (tdb->global_lock.ltype != ltype || tdb->global_lock.count == 0) {
409 tdb->ecode = TDB_ERR_LOCK;
413 if (tdb->global_lock.count > 1) {
414 tdb->global_lock.count--;
419 tdb->methods->tdb_brlock(tdb, FREELIST_TOP, F_UNLCK, F_SETLKW,
420 0, 4*tdb->header.hash_size)) {
421 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_unlockall failed (%s)\n", strerror(errno)));
425 tdb->global_lock.count = 0;
426 tdb->global_lock.ltype = 0;
431 /* lock entire database with write lock */
432 int tdb_lockall(struct tdb_context *tdb)
434 tdb_trace(tdb, "tdb_lockall");
435 return _tdb_lockall(tdb, F_WRLCK, F_SETLKW);
438 /* lock entire database with write lock - mark only */
439 int tdb_lockall_mark(struct tdb_context *tdb)
441 tdb_trace(tdb, "tdb_lockall_mark");
442 return _tdb_lockall(tdb, F_WRLCK | TDB_MARK_LOCK, F_SETLKW);
445 /* unlock entire database with write lock - unmark only */
446 int tdb_lockall_unmark(struct tdb_context *tdb)
448 tdb_trace(tdb, "tdb_lockall_unmark");
449 return _tdb_unlockall(tdb, F_WRLCK | TDB_MARK_LOCK);
452 /* lock entire database with write lock - nonblocking varient */
453 int tdb_lockall_nonblock(struct tdb_context *tdb)
455 int ret = _tdb_lockall(tdb, F_WRLCK, F_SETLK);
456 tdb_trace_ret(tdb, "tdb_lockall_nonblock", ret);
460 /* unlock entire database with write lock */
461 int tdb_unlockall(struct tdb_context *tdb)
463 tdb_trace(tdb, "tdb_unlockall");
464 return _tdb_unlockall(tdb, F_WRLCK);
467 /* lock entire database with read lock */
468 int tdb_lockall_read(struct tdb_context *tdb)
470 tdb_trace(tdb, "tdb_lockall_read");
471 return _tdb_lockall(tdb, F_RDLCK, F_SETLKW);
474 /* lock entire database with read lock - nonblock varient */
475 int tdb_lockall_read_nonblock(struct tdb_context *tdb)
477 int ret = _tdb_lockall(tdb, F_RDLCK, F_SETLK);
478 tdb_trace_ret(tdb, "tdb_lockall_read_nonblock", ret);
482 /* unlock entire database with read lock */
483 int tdb_unlockall_read(struct tdb_context *tdb)
485 tdb_trace(tdb, "tdb_unlockall_read");
486 return _tdb_unlockall(tdb, F_RDLCK);
489 /* lock/unlock one hash chain. This is meant to be used to reduce
490 contention - it cannot guarantee how many records will be locked */
491 int tdb_chainlock(struct tdb_context *tdb, TDB_DATA key)
493 int ret = tdb_lock(tdb, BUCKET(tdb->hash_fn(&key)), F_WRLCK);
494 tdb_trace_1rec(tdb, "tdb_chainlock", key);
498 /* lock/unlock one hash chain, non-blocking. This is meant to be used
499 to reduce contention - it cannot guarantee how many records will be
501 int tdb_chainlock_nonblock(struct tdb_context *tdb, TDB_DATA key)
503 int ret = tdb_lock_nonblock(tdb, BUCKET(tdb->hash_fn(&key)), F_WRLCK);
504 tdb_trace_1rec_ret(tdb, "tdb_chainlock_nonblock", key, ret);
508 /* mark a chain as locked without actually locking it. Warning! use with great caution! */
509 int tdb_chainlock_mark(struct tdb_context *tdb, TDB_DATA key)
511 int ret = tdb_lock(tdb, BUCKET(tdb->hash_fn(&key)), F_WRLCK | TDB_MARK_LOCK);
512 tdb_trace_1rec(tdb, "tdb_chainlock_mark", key);
516 /* unmark a chain as locked without actually locking it. Warning! use with great caution! */
517 int tdb_chainlock_unmark(struct tdb_context *tdb, TDB_DATA key)
519 tdb_trace_1rec(tdb, "tdb_chainlock_unmark", key);
520 return tdb_unlock(tdb, BUCKET(tdb->hash_fn(&key)), F_WRLCK | TDB_MARK_LOCK);
523 int tdb_chainunlock(struct tdb_context *tdb, TDB_DATA key)
525 tdb_trace_1rec(tdb, "tdb_chainunlock", key);
526 return tdb_unlock(tdb, BUCKET(tdb->hash_fn(&key)), F_WRLCK);
529 int tdb_chainlock_read(struct tdb_context *tdb, TDB_DATA key)
532 ret = tdb_lock(tdb, BUCKET(tdb->hash_fn(&key)), F_RDLCK);
533 tdb_trace_1rec(tdb, "tdb_chainlock_read", key);
537 int tdb_chainunlock_read(struct tdb_context *tdb, TDB_DATA key)
539 tdb_trace_1rec(tdb, "tdb_chainunlock_read", key);
540 return tdb_unlock(tdb, BUCKET(tdb->hash_fn(&key)), F_RDLCK);
543 /* record lock stops delete underneath */
544 int tdb_lock_record(struct tdb_context *tdb, tdb_off_t off)
546 if (tdb->global_lock.count) {
549 return off ? tdb->methods->tdb_brlock(tdb, off, F_RDLCK, F_SETLKW, 0, 1) : 0;
553 Write locks override our own fcntl readlocks, so check it here.
554 Note this is meant to be F_SETLK, *not* F_SETLKW, as it's not
555 an error to fail to get the lock here.
557 int tdb_write_lock_record(struct tdb_context *tdb, tdb_off_t off)
559 struct tdb_traverse_lock *i;
560 for (i = &tdb->travlocks; i; i = i->next)
563 return tdb->methods->tdb_brlock(tdb, off, F_WRLCK, F_SETLK, 1, 1);
567 Note this is meant to be F_SETLK, *not* F_SETLKW, as it's not
568 an error to fail to get the lock here.
570 int tdb_write_unlock_record(struct tdb_context *tdb, tdb_off_t off)
572 return tdb->methods->tdb_brlock(tdb, off, F_UNLCK, F_SETLK, 0, 1);
575 /* fcntl locks don't stack: avoid unlocking someone else's */
576 int tdb_unlock_record(struct tdb_context *tdb, tdb_off_t off)
578 struct tdb_traverse_lock *i;
581 if (tdb->global_lock.count) {
587 for (i = &tdb->travlocks; i; i = i->next)
590 return (count == 1 ? tdb->methods->tdb_brlock(tdb, off, F_UNLCK, F_SETLKW, 0, 1) : 0);