source: vendor/current/lib/tdb/common/lock.c

Last change on this file was 988, checked in by Silvan Scherrer, 9 years ago

Samba Server: update vendor to version 4.4.3
File size: 23.8 KB
Line 
1 /*
2 Unix SMB/CIFS implementation.
3
4 trivial database library
5
6 Copyright (C) Andrew Tridgell 1999-2005
7 Copyright (C) Paul `Rusty' Russell 2000
8 Copyright (C) Jeremy Allison 2000-2003
9
10 ** NOTE! The following LGPL license applies to the tdb
11 ** library. This does NOT imply that all of Samba is released
12 ** under the LGPL
13
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.
18
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.
23
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/>.
26*/
27
28#include "tdb_private.h"
29
30_PUBLIC_ void tdb_setalarm_sigptr(struct tdb_context *tdb, volatile sig_atomic_t *ptr)
31{
32 tdb->interrupt_sig_ptr = ptr;
33}
34
35static int fcntl_lock(struct tdb_context *tdb,
36 int rw, off_t off, off_t len, bool waitflag)
37{
38 struct flock fl;
39 int cmd;
40
41#ifdef USE_TDB_MUTEX_LOCKING
42 {
43 int ret;
44 if (tdb_mutex_lock(tdb, rw, off, len, waitflag, &ret)) {
45 return ret;
46 }
47 }
48#endif
49
50 fl.l_type = rw;
51 fl.l_whence = SEEK_SET;
52 fl.l_start = off;
53 fl.l_len = len;
54 fl.l_pid = 0;
55
56 cmd = waitflag ? F_SETLKW : F_SETLK;
57
58 return fcntl(tdb->fd, cmd, &fl);
59}
60
61static int fcntl_unlock(struct tdb_context *tdb, int rw, off_t off, off_t len)
62{
63 struct flock fl;
64#if 0 /* Check they matched up locks and unlocks correctly. */
65 char line[80];
66 FILE *locks;
67 bool found = false;
68
69 locks = fopen("/proc/locks", "r");
70
71 while (fgets(line, 80, locks)) {
72 char *p;
73 int type, start, l;
74
75 /* eg. 1: FLOCK ADVISORY WRITE 2440 08:01:2180826 0 EOF */
76 p = strchr(line, ':') + 1;
77 if (strncmp(p, " POSIX ADVISORY ", strlen(" POSIX ADVISORY ")))
78 continue;
79 p += strlen(" FLOCK ADVISORY ");
80 if (strncmp(p, "READ ", strlen("READ ")) == 0)
81 type = F_RDLCK;
82 else if (strncmp(p, "WRITE ", strlen("WRITE ")) == 0)
83 type = F_WRLCK;
84 else
85 abort();
86 p += 6;
87 if (atoi(p) != getpid())
88 continue;
89 p = strchr(strchr(p, ' ') + 1, ' ') + 1;
90 start = atoi(p);
91 p = strchr(p, ' ') + 1;
92 if (strncmp(p, "EOF", 3) == 0)
93 l = 0;
94 else
95 l = atoi(p) - start + 1;
96
97 if (off == start) {
98 if (len != l) {
99 fprintf(stderr, "Len %u should be %u: %s",
100 (int)len, l, line);
101 abort();
102 }
103 if (type != rw) {
104 fprintf(stderr, "Type %s wrong: %s",
105 rw == F_RDLCK ? "READ" : "WRITE", line);
106 abort();
107 }
108 found = true;
109 break;
110 }
111 }
112
113 if (!found) {
114 fprintf(stderr, "Unlock on %u@%u not found!\n",
115 (int)off, (int)len);
116 abort();
117 }
118
119 fclose(locks);
120#endif
121
122#ifdef USE_TDB_MUTEX_LOCKING
123 {
124 int ret;
125 if (tdb_mutex_unlock(tdb, rw, off, len, &ret)) {
126 return ret;
127 }
128 }
129#endif
130
131 fl.l_type = F_UNLCK;
132 fl.l_whence = SEEK_SET;
133 fl.l_start = off;
134 fl.l_len = len;
135 fl.l_pid = 0;
136
137 return fcntl(tdb->fd, F_SETLKW, &fl);
138}
139
140/* list -1 is the alloc list, otherwise a hash chain. */
141static tdb_off_t lock_offset(int list)
142{
143 return FREELIST_TOP + 4*list;
144}
145
146/* a byte range locking function - return 0 on success
147 this functions locks/unlocks "len" byte at the specified offset.
148
149 On error, errno is also set so that errors are passed back properly
150 through tdb_open().
151
152 note that a len of zero means lock to end of file
153*/
154int tdb_brlock(struct tdb_context *tdb,
155 int rw_type, tdb_off_t offset, size_t len,
156 enum tdb_lock_flags flags)
157{
158 int ret;
159
160 if (tdb->flags & TDB_NOLOCK) {
161 return 0;
162 }
163
164 if (flags & TDB_LOCK_MARK_ONLY) {
165 return 0;
166 }
167
168 if ((rw_type == F_WRLCK) && (tdb->read_only || tdb->traverse_read)) {
169 tdb->ecode = TDB_ERR_RDONLY;
170 return -1;
171 }
172
173 do {
174 ret = fcntl_lock(tdb, rw_type, offset, len,
175 flags & TDB_LOCK_WAIT);
176 /* Check for a sigalarm break. */
177 if (ret == -1 && errno == EINTR &&
178 tdb->interrupt_sig_ptr &&
179 *tdb->interrupt_sig_ptr) {
180 break;
181 }
182 } while (ret == -1 && errno == EINTR);
183
184 if (ret == -1) {
185 tdb->ecode = TDB_ERR_LOCK;
186 /* Generic lock error. errno set by fcntl.
187 * EAGAIN is an expected return from non-blocking
188 * locks. */
189 if (!(flags & TDB_LOCK_PROBE) && errno != EAGAIN) {
190 TDB_LOG((tdb, TDB_DEBUG_TRACE,"tdb_brlock failed (fd=%d) at offset %u rw_type=%d flags=%d len=%zu\n",
191 tdb->fd, offset, rw_type, flags, len));
192 }
193 return -1;
194 }
195 return 0;
196}
197
198int tdb_brunlock(struct tdb_context *tdb,
199 int rw_type, tdb_off_t offset, size_t len)
200{
201 int ret;
202
203 if (tdb->flags & TDB_NOLOCK) {
204 return 0;
205 }
206
207 do {
208 ret = fcntl_unlock(tdb, rw_type, offset, len);
209 } while (ret == -1 && errno == EINTR);
210
211 if (ret == -1) {
212 TDB_LOG((tdb, TDB_DEBUG_TRACE,"tdb_brunlock failed (fd=%d) at offset %u rw_type=%u len=%zu\n",
213 tdb->fd, offset, rw_type, len));
214 }
215 return ret;
216}
217
218/*
219 * Do a tdb_brlock in a loop. Some OSes (such as solaris) have too
220 * conservative deadlock detection and claim a deadlock when progress can be
221 * made. For those OSes we may loop for a while.
222 */
223
224static int tdb_brlock_retry(struct tdb_context *tdb,
225 int rw_type, tdb_off_t offset, size_t len,
226 enum tdb_lock_flags flags)
227{
228 int count = 1000;
229
230 while (count--) {
231 struct timeval tv;
232 int ret;
233
234 ret = tdb_brlock(tdb, rw_type, offset, len, flags);
235 if (ret == 0) {
236 return 0;
237 }
238 if (errno != EDEADLK) {
239 break;
240 }
241 /* sleep for as short a time as we can - more portable than usleep() */
242 tv.tv_sec = 0;
243 tv.tv_usec = 1;
244 select(0, NULL, NULL, NULL, &tv);
245 }
246 return -1;
247}
248
249/*
250 upgrade a read lock to a write lock.
251*/
252int tdb_allrecord_upgrade(struct tdb_context *tdb)
253{
254 int ret;
255
256 if (tdb->allrecord_lock.count != 1) {
257 TDB_LOG((tdb, TDB_DEBUG_ERROR,
258 "tdb_allrecord_upgrade failed: count %u too high\n",
259 tdb->allrecord_lock.count));
260 return -1;
261 }
262
263 if (tdb->allrecord_lock.off != 1) {
264 TDB_LOG((tdb, TDB_DEBUG_ERROR,
265 "tdb_allrecord_upgrade failed: already upgraded?\n"));
266 return -1;
267 }
268
269 if (tdb_have_mutexes(tdb)) {
270 ret = tdb_mutex_allrecord_upgrade(tdb);
271 if (ret == -1) {
272 goto fail;
273 }
274 ret = tdb_brlock_retry(tdb, F_WRLCK, lock_offset(tdb->hash_size),
275 0, TDB_LOCK_WAIT|TDB_LOCK_PROBE);
276 if (ret == -1) {
277 tdb_mutex_allrecord_downgrade(tdb);
278 }
279 } else {
280 ret = tdb_brlock_retry(tdb, F_WRLCK, FREELIST_TOP, 0,
281 TDB_LOCK_WAIT|TDB_LOCK_PROBE);
282 }
283
284 if (ret == 0) {
285 tdb->allrecord_lock.ltype = F_WRLCK;
286 tdb->allrecord_lock.off = 0;
287 return 0;
288 }
289fail:
290 TDB_LOG((tdb, TDB_DEBUG_TRACE,"tdb_allrecord_upgrade failed\n"));
291 return -1;
292}
293
294static struct tdb_lock_type *find_nestlock(struct tdb_context *tdb,
295 tdb_off_t offset)
296{
297 unsigned int i;
298
299 for (i=0; i<tdb->num_lockrecs; i++) {
300 if (tdb->lockrecs[i].off == offset) {
301 return &tdb->lockrecs[i];
302 }
303 }
304 return NULL;
305}
306
307/* lock an offset in the database. */
308int tdb_nest_lock(struct tdb_context *tdb, uint32_t offset, int ltype,
309 enum tdb_lock_flags flags)
310{
311 struct tdb_lock_type *new_lck;
312
313 if (offset >= lock_offset(tdb->hash_size)) {
314 tdb->ecode = TDB_ERR_LOCK;
315 TDB_LOG((tdb, TDB_DEBUG_ERROR,"tdb_lock: invalid offset %u for ltype=%d\n",
316 offset, ltype));
317 return -1;
318 }
319 if (tdb->flags & TDB_NOLOCK)
320 return 0;
321
322 new_lck = find_nestlock(tdb, offset);
323 if (new_lck) {
324 /*
325 * Just increment the in-memory struct, posix locks
326 * don't stack.
327 */
328 new_lck->count++;
329 return 0;
330 }
331
332 if (tdb->num_lockrecs == tdb->lockrecs_array_length) {
333 new_lck = (struct tdb_lock_type *)realloc(
334 tdb->lockrecs,
335 sizeof(*tdb->lockrecs) * (tdb->num_lockrecs+1));
336 if (new_lck == NULL) {
337 errno = ENOMEM;
338 return -1;
339 }
340 tdb->lockrecs_array_length = tdb->num_lockrecs+1;
341 tdb->lockrecs = new_lck;
342 }
343
344 /* Since fcntl locks don't nest, we do a lock for the first one,
345 and simply bump the count for future ones */
346 if (tdb_brlock(tdb, ltype, offset, 1, flags)) {
347 return -1;
348 }
349
350 new_lck = &tdb->lockrecs[tdb->num_lockrecs];
351
352 new_lck->off = offset;
353 new_lck->count = 1;
354 new_lck->ltype = ltype;
355 tdb->num_lockrecs++;
356
357 return 0;
358}
359
360static int tdb_lock_and_recover(struct tdb_context *tdb)
361{
362 int ret;
363
364 /* We need to match locking order in transaction commit. */
365 if (tdb_brlock(tdb, F_WRLCK, FREELIST_TOP, 0, TDB_LOCK_WAIT)) {
366 return -1;
367 }
368
369 if (tdb_brlock(tdb, F_WRLCK, OPEN_LOCK, 1, TDB_LOCK_WAIT)) {
370 tdb_brunlock(tdb, F_WRLCK, FREELIST_TOP, 0);
371 return -1;
372 }
373
374 ret = tdb_transaction_recover(tdb);
375
376 tdb_brunlock(tdb, F_WRLCK, OPEN_LOCK, 1);
377 tdb_brunlock(tdb, F_WRLCK, FREELIST_TOP, 0);
378
379 return ret;
380}
381
382static bool have_data_locks(const struct tdb_context *tdb)
383{
384 unsigned int i;
385
386 for (i = 0; i < tdb->num_lockrecs; i++) {
387 if (tdb->lockrecs[i].off >= lock_offset(-1))
388 return true;
389 }
390 return false;
391}
392
393/*
394 * A allrecord lock allows us to avoid per chain locks. Check if the allrecord
395 * lock is strong enough.
396 */
397static int tdb_lock_covered_by_allrecord_lock(struct tdb_context *tdb,
398 int ltype)
399{
400 if (ltype == F_RDLCK) {
401 /*
402 * The allrecord_lock is equal (F_RDLCK) or stronger
403 * (F_WRLCK). Pass.
404 */
405 return 0;
406 }
407
408 if (tdb->allrecord_lock.ltype == F_RDLCK) {
409 /*
410 * We ask for ltype==F_WRLCK, but the allrecord_lock
411 * is too weak. We can't upgrade here, so fail.
412 */
413 tdb->ecode = TDB_ERR_LOCK;
414 return -1;
415 }
416
417 /*
418 * Asking for F_WRLCK, allrecord is F_WRLCK as well. Pass.
419 */
420 return 0;
421}
422
423static int tdb_lock_list(struct tdb_context *tdb, int list, int ltype,
424 enum tdb_lock_flags waitflag)
425{
426 int ret;
427 bool check = false;
428
429 if (tdb->allrecord_lock.count) {
430 return tdb_lock_covered_by_allrecord_lock(tdb, ltype);
431 }
432
433 /*
434 * Check for recoveries: Someone might have kill -9'ed a process
435 * during a commit.
436 */
437 check = !have_data_locks(tdb);
438 ret = tdb_nest_lock(tdb, lock_offset(list), ltype, waitflag);
439
440 if (ret == 0 && check && tdb_needs_recovery(tdb)) {
441 tdb_nest_unlock(tdb, lock_offset(list), ltype, false);
442
443 if (tdb_lock_and_recover(tdb) == -1) {
444 return -1;
445 }
446 return tdb_lock_list(tdb, list, ltype, waitflag);
447 }
448 return ret;
449}
450
451/* lock a list in the database. list -1 is the alloc list */
452int tdb_lock(struct tdb_context *tdb, int list, int ltype)
453{
454 int ret;
455
456 ret = tdb_lock_list(tdb, list, ltype, TDB_LOCK_WAIT);
457 if (ret) {
458 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_lock failed on list %d "
459 "ltype=%d (%s)\n", list, ltype, strerror(errno)));
460 }
461 return ret;
462}
463
464/* lock a list in the database. list -1 is the alloc list. non-blocking lock */
465_PUBLIC_ int tdb_lock_nonblock(struct tdb_context *tdb, int list, int ltype)
466{
467 return tdb_lock_list(tdb, list, ltype, TDB_LOCK_NOWAIT);
468}
469
470
471int tdb_nest_unlock(struct tdb_context *tdb, uint32_t offset, int ltype,
472 bool mark_lock)
473{
474 int ret = -1;
475 struct tdb_lock_type *lck;
476
477 if (tdb->flags & TDB_NOLOCK)
478 return 0;
479
480 /* Sanity checks */
481 if (offset >= lock_offset(tdb->hash_size)) {
482 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_unlock: offset %u invalid (%d)\n", offset, tdb->hash_size));
483 return ret;
484 }
485
486 lck = find_nestlock(tdb, offset);
487 if ((lck == NULL) || (lck->count == 0)) {
488 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_unlock: count is 0\n"));
489 return -1;
490 }
491
492 if (lck->count > 1) {
493 lck->count--;
494 return 0;
495 }
496
497 /*
498 * This lock has count==1 left, so we need to unlock it in the
499 * kernel. We don't bother with decrementing the in-memory array
500 * element, we're about to overwrite it with the last array element
501 * anyway.
502 */
503
504 if (mark_lock) {
505 ret = 0;
506 } else {
507 ret = tdb_brunlock(tdb, ltype, offset, 1);
508 }
509
510 /*
511 * Shrink the array by overwriting the element just unlocked with the
512 * last array element.
513 */
514 *lck = tdb->lockrecs[--tdb->num_lockrecs];
515
516 /*
517 * We don't bother with realloc when the array shrinks, but if we have
518 * a completely idle tdb we should get rid of the locked array.
519 */
520
521 if (ret)
522 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_unlock: An error occurred unlocking!\n"));
523 return ret;
524}
525
526_PUBLIC_ int tdb_unlock(struct tdb_context *tdb, int list, int ltype)
527{
528 /* a global lock allows us to avoid per chain locks */
529 if (tdb->allrecord_lock.count) {
530 return tdb_lock_covered_by_allrecord_lock(tdb, ltype);
531 }
532
533 return tdb_nest_unlock(tdb, lock_offset(list), ltype, false);
534}
535
536/*
537 get the transaction lock
538 */
539int tdb_transaction_lock(struct tdb_context *tdb, int ltype,
540 enum tdb_lock_flags lockflags)
541{
542 return tdb_nest_lock(tdb, TRANSACTION_LOCK, ltype, lockflags);
543}
544
545/*
546 release the transaction lock
547 */
548int tdb_transaction_unlock(struct tdb_context *tdb, int ltype)
549{
550 return tdb_nest_unlock(tdb, TRANSACTION_LOCK, ltype, false);
551}
552
553/* Returns 0 if all done, -1 if error, 1 if ok. */
554static int tdb_allrecord_check(struct tdb_context *tdb, int ltype,
555 enum tdb_lock_flags flags, bool upgradable)
556{
557 /* There are no locks on read-only dbs */
558 if (tdb->read_only || tdb->traverse_read) {
559 tdb->ecode = TDB_ERR_LOCK;
560 return -1;
561 }
562
563 if (tdb->allrecord_lock.count && tdb->allrecord_lock.ltype == ltype) {
564 tdb->allrecord_lock.count++;
565 return 0;
566 }
567
568 if (tdb->allrecord_lock.count) {
569 /* a global lock of a different type exists */
570 tdb->ecode = TDB_ERR_LOCK;
571 return -1;
572 }
573
574 if (tdb_have_extra_locks(tdb)) {
575 /* can't combine global and chain locks */
576 tdb->ecode = TDB_ERR_LOCK;
577 return -1;
578 }
579
580 if (upgradable && ltype != F_RDLCK) {
581 /* tdb error: you can't upgrade a write lock! */
582 tdb->ecode = TDB_ERR_LOCK;
583 return -1;
584 }
585 return 1;
586}
587
588/* We only need to lock individual bytes, but Linux merges consecutive locks
589 * so we lock in contiguous ranges. */
590static int tdb_chainlock_gradual(struct tdb_context *tdb,
591 int ltype, enum tdb_lock_flags flags,
592 size_t off, size_t len)
593{
594 int ret;
595 enum tdb_lock_flags nb_flags = (flags & ~TDB_LOCK_WAIT);
596
597 if (len <= 4) {
598 /* Single record. Just do blocking lock. */
599 return tdb_brlock(tdb, ltype, off, len, flags);
600 }
601
602 /* First we try non-blocking. */
603 ret = tdb_brlock(tdb, ltype, off, len, nb_flags);
604 if (ret == 0) {
605 return 0;
606 }
607
608 /* Try locking first half, then second. */
609 ret = tdb_chainlock_gradual(tdb, ltype, flags, off, len / 2);
610 if (ret == -1)
611 return -1;
612
613 ret = tdb_chainlock_gradual(tdb, ltype, flags,
614 off + len / 2, len - len / 2);
615 if (ret == -1) {
616 tdb_brunlock(tdb, ltype, off, len / 2);
617 return -1;
618 }
619 return 0;
620}
621
622/* lock/unlock entire database. It can only be upgradable if you have some
623 * other way of guaranteeing exclusivity (ie. transaction write lock).
624 * We do the locking gradually to avoid being starved by smaller locks. */
625int tdb_allrecord_lock(struct tdb_context *tdb, int ltype,
626 enum tdb_lock_flags flags, bool upgradable)
627{
628 int ret;
629
630 switch (tdb_allrecord_check(tdb, ltype, flags, upgradable)) {
631 case -1:
632 return -1;
633 case 0:
634 return 0;
635 }
636
637 /* We cover two kinds of locks:
638 * 1) Normal chain locks. Taken for almost all operations.
639 * 2) Individual records locks. Taken after normal or free
640 * chain locks.
641 *
642 * It is (1) which cause the starvation problem, so we're only
643 * gradual for that. */
644
645 if (tdb_have_mutexes(tdb)) {
646 ret = tdb_mutex_allrecord_lock(tdb, ltype, flags);
647 } else {
648 ret = tdb_chainlock_gradual(tdb, ltype, flags, FREELIST_TOP,
649 tdb->hash_size * 4);
650 }
651
652 if (ret == -1) {
653 return -1;
654 }
655
656 /* Grab individual record locks. */
657 if (tdb_brlock(tdb, ltype, lock_offset(tdb->hash_size), 0,
658 flags) == -1) {
659 if (tdb_have_mutexes(tdb)) {
660 tdb_mutex_allrecord_unlock(tdb);
661 } else {
662 tdb_brunlock(tdb, ltype, FREELIST_TOP,
663 tdb->hash_size * 4);
664 }
665 return -1;
666 }
667
668 tdb->allrecord_lock.count = 1;
669 /* If it's upgradable, it's actually exclusive so we can treat
670 * it as a write lock. */
671 tdb->allrecord_lock.ltype = upgradable ? F_WRLCK : ltype;
672 tdb->allrecord_lock.off = upgradable;
673
674 if (tdb_needs_recovery(tdb)) {
675 bool mark = flags & TDB_LOCK_MARK_ONLY;
676 tdb_allrecord_unlock(tdb, ltype, mark);
677 if (mark) {
678 tdb->ecode = TDB_ERR_LOCK;
679 TDB_LOG((tdb, TDB_DEBUG_ERROR,
680 "tdb_lockall_mark cannot do recovery\n"));
681 return -1;
682 }
683 if (tdb_lock_and_recover(tdb) == -1) {
684 return -1;
685 }
686 return tdb_allrecord_lock(tdb, ltype, flags, upgradable);
687 }
688
689 return 0;
690}
691
692
693
694/* unlock entire db */
695int tdb_allrecord_unlock(struct tdb_context *tdb, int ltype, bool mark_lock)
696{
697 /* There are no locks on read-only dbs */
698 if (tdb->read_only || tdb->traverse_read) {
699 tdb->ecode = TDB_ERR_LOCK;
700 return -1;
701 }
702
703 if (tdb->allrecord_lock.count == 0) {
704 tdb->ecode = TDB_ERR_LOCK;
705 return -1;
706 }
707
708 /* Upgradable locks are marked as write locks. */
709 if (tdb->allrecord_lock.ltype != ltype
710 && (!tdb->allrecord_lock.off || ltype != F_RDLCK)) {
711 tdb->ecode = TDB_ERR_LOCK;
712 return -1;
713 }
714
715 if (tdb->allrecord_lock.count > 1) {
716 tdb->allrecord_lock.count--;
717 return 0;
718 }
719
720 if (!mark_lock) {
721 int ret;
722
723 if (tdb_have_mutexes(tdb)) {
724 ret = tdb_mutex_allrecord_unlock(tdb);
725 if (ret == 0) {
726 ret = tdb_brunlock(tdb, ltype,
727 lock_offset(tdb->hash_size),
728 0);
729 }
730 } else {
731 ret = tdb_brunlock(tdb, ltype, FREELIST_TOP, 0);
732 }
733
734 if (ret != 0) {
735 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_unlockall failed "
736 "(%s)\n", strerror(errno)));
737 return -1;
738 }
739 }
740
741 tdb->allrecord_lock.count = 0;
742 tdb->allrecord_lock.ltype = 0;
743
744 return 0;
745}
746
747/* lock entire database with write lock */
748_PUBLIC_ int tdb_lockall(struct tdb_context *tdb)
749{
750 tdb_trace(tdb, "tdb_lockall");
751 return tdb_allrecord_lock(tdb, F_WRLCK, TDB_LOCK_WAIT, false);
752}
753
754/* lock entire database with write lock - mark only */
755_PUBLIC_ int tdb_lockall_mark(struct tdb_context *tdb)
756{
757 tdb_trace(tdb, "tdb_lockall_mark");
758 return tdb_allrecord_lock(tdb, F_WRLCK, TDB_LOCK_MARK_ONLY, false);
759}
760
761/* unlock entire database with write lock - unmark only */
762_PUBLIC_ int tdb_lockall_unmark(struct tdb_context *tdb)
763{
764 tdb_trace(tdb, "tdb_lockall_unmark");
765 return tdb_allrecord_unlock(tdb, F_WRLCK, true);
766}
767
768/* lock entire database with write lock - nonblocking varient */
769_PUBLIC_ int tdb_lockall_nonblock(struct tdb_context *tdb)
770{
771 int ret = tdb_allrecord_lock(tdb, F_WRLCK, TDB_LOCK_NOWAIT, false);
772 tdb_trace_ret(tdb, "tdb_lockall_nonblock", ret);
773 return ret;
774}
775
776/* unlock entire database with write lock */
777_PUBLIC_ int tdb_unlockall(struct tdb_context *tdb)
778{
779 tdb_trace(tdb, "tdb_unlockall");
780 return tdb_allrecord_unlock(tdb, F_WRLCK, false);
781}
782
783/* lock entire database with read lock */
784_PUBLIC_ int tdb_lockall_read(struct tdb_context *tdb)
785{
786 tdb_trace(tdb, "tdb_lockall_read");
787 return tdb_allrecord_lock(tdb, F_RDLCK, TDB_LOCK_WAIT, false);
788}
789
790/* lock entire database with read lock - nonblock varient */
791_PUBLIC_ int tdb_lockall_read_nonblock(struct tdb_context *tdb)
792{
793 int ret = tdb_allrecord_lock(tdb, F_RDLCK, TDB_LOCK_NOWAIT, false);
794 tdb_trace_ret(tdb, "tdb_lockall_read_nonblock", ret);
795 return ret;
796}
797
798/* unlock entire database with read lock */
799_PUBLIC_ int tdb_unlockall_read(struct tdb_context *tdb)
800{
801 tdb_trace(tdb, "tdb_unlockall_read");
802 return tdb_allrecord_unlock(tdb, F_RDLCK, false);
803}
804
805/* lock/unlock one hash chain. This is meant to be used to reduce
806 contention - it cannot guarantee how many records will be locked */
807_PUBLIC_ int tdb_chainlock(struct tdb_context *tdb, TDB_DATA key)
808{
809 int ret = tdb_lock(tdb, BUCKET(tdb->hash_fn(&key)), F_WRLCK);
810 tdb_trace_1rec(tdb, "tdb_chainlock", key);
811 return ret;
812}
813
814/* lock/unlock one hash chain, non-blocking. This is meant to be used
815 to reduce contention - it cannot guarantee how many records will be
816 locked */
817_PUBLIC_ int tdb_chainlock_nonblock(struct tdb_context *tdb, TDB_DATA key)
818{
819 int ret = tdb_lock_nonblock(tdb, BUCKET(tdb->hash_fn(&key)), F_WRLCK);
820 tdb_trace_1rec_ret(tdb, "tdb_chainlock_nonblock", key, ret);
821 return ret;
822}
823
824/* mark a chain as locked without actually locking it. Warning! use with great caution! */
825_PUBLIC_ int tdb_chainlock_mark(struct tdb_context *tdb, TDB_DATA key)
826{
827 int ret = tdb_nest_lock(tdb, lock_offset(BUCKET(tdb->hash_fn(&key))),
828 F_WRLCK, TDB_LOCK_MARK_ONLY);
829 tdb_trace_1rec(tdb, "tdb_chainlock_mark", key);
830 return ret;
831}
832
833/* unmark a chain as locked without actually locking it. Warning! use with great caution! */
834_PUBLIC_ int tdb_chainlock_unmark(struct tdb_context *tdb, TDB_DATA key)
835{
836 tdb_trace_1rec(tdb, "tdb_chainlock_unmark", key);
837 return tdb_nest_unlock(tdb, lock_offset(BUCKET(tdb->hash_fn(&key))),
838 F_WRLCK, true);
839}
840
841_PUBLIC_ int tdb_chainunlock(struct tdb_context *tdb, TDB_DATA key)
842{
843 tdb_trace_1rec(tdb, "tdb_chainunlock", key);
844 return tdb_unlock(tdb, BUCKET(tdb->hash_fn(&key)), F_WRLCK);
845}
846
847_PUBLIC_ int tdb_chainlock_read(struct tdb_context *tdb, TDB_DATA key)
848{
849 int ret;
850 ret = tdb_lock(tdb, BUCKET(tdb->hash_fn(&key)), F_RDLCK);
851 tdb_trace_1rec(tdb, "tdb_chainlock_read", key);
852 return ret;
853}
854
855_PUBLIC_ int tdb_chainunlock_read(struct tdb_context *tdb, TDB_DATA key)
856{
857 tdb_trace_1rec(tdb, "tdb_chainunlock_read", key);
858 return tdb_unlock(tdb, BUCKET(tdb->hash_fn(&key)), F_RDLCK);
859}
860
861_PUBLIC_ int tdb_chainlock_read_nonblock(struct tdb_context *tdb, TDB_DATA key)
862{
863 int ret = tdb_lock_nonblock(tdb, BUCKET(tdb->hash_fn(&key)), F_RDLCK);
864 tdb_trace_1rec_ret(tdb, "tdb_chainlock_read_nonblock", key, ret);
865 return ret;
866}
867
868/* record lock stops delete underneath */
869int tdb_lock_record(struct tdb_context *tdb, tdb_off_t off)
870{
871 if (tdb->allrecord_lock.count) {
872 return 0;
873 }
874 return off ? tdb_brlock(tdb, F_RDLCK, off, 1, TDB_LOCK_WAIT) : 0;
875}
876
877/*
878 Write locks override our own fcntl readlocks, so check it here.
879 Note this is meant to be F_SETLK, *not* F_SETLKW, as it's not
880 an error to fail to get the lock here.
881*/
882int tdb_write_lock_record(struct tdb_context *tdb, tdb_off_t off)
883{
884 struct tdb_traverse_lock *i;
885 for (i = &tdb->travlocks; i; i = i->next)
886 if (i->off == off)
887 return -1;
888 if (tdb->allrecord_lock.count) {
889 if (tdb->allrecord_lock.ltype == F_WRLCK) {
890 return 0;
891 }
892 return -1;
893 }
894 return tdb_brlock(tdb, F_WRLCK, off, 1, TDB_LOCK_NOWAIT|TDB_LOCK_PROBE);
895}
896
897int tdb_write_unlock_record(struct tdb_context *tdb, tdb_off_t off)
898{
899 if (tdb->allrecord_lock.count) {
900 return 0;
901 }
902 return tdb_brunlock(tdb, F_WRLCK, off, 1);
903}
904
905/* fcntl locks don't stack: avoid unlocking someone else's */
906int tdb_unlock_record(struct tdb_context *tdb, tdb_off_t off)
907{
908 struct tdb_traverse_lock *i;
909 uint32_t count = 0;
910
911 if (tdb->allrecord_lock.count) {
912 return 0;
913 }
914
915 if (off == 0)
916 return 0;
917 for (i = &tdb->travlocks; i; i = i->next)
918 if (i->off == off)
919 count++;
920 return (count == 1 ? tdb_brunlock(tdb, F_RDLCK, off, 1) : 0);
921}
922
923bool tdb_have_extra_locks(struct tdb_context *tdb)
924{
925 unsigned int extra = tdb->num_lockrecs;
926
927 /* A transaction holds the lock for all records. */
928 if (!tdb->transaction && tdb->allrecord_lock.count) {
929 return true;
930 }
931
932 /* We always hold the active lock if CLEAR_IF_FIRST. */
933 if (find_nestlock(tdb, ACTIVE_LOCK)) {
934 extra--;
935 }
936
937 /* In a transaction, we expect to hold the transaction lock */
938 if (tdb->transaction && find_nestlock(tdb, TRANSACTION_LOCK)) {
939 extra--;
940 }
941
942 return extra;
943}
944
945/* The transaction code uses this to remove all locks. */
946void tdb_release_transaction_locks(struct tdb_context *tdb)
947{
948 unsigned int i, active = 0;
949
950 if (tdb->allrecord_lock.count != 0) {
951 tdb_allrecord_unlock(tdb, tdb->allrecord_lock.ltype, false);
952 tdb->allrecord_lock.count = 0;
953 }
954
955 for (i=0;i<tdb->num_lockrecs;i++) {
956 struct tdb_lock_type *lck = &tdb->lockrecs[i];
957
958 /* Don't release the active lock! Copy it to first entry. */
959 if (lck->off == ACTIVE_LOCK) {
960 tdb->lockrecs[active++] = *lck;
961 } else {
962 tdb_brunlock(tdb, lck->ltype, lck->off, 1);
963 }
964 }
965 tdb->num_lockrecs = active;
966}
967
968/* Following functions are added specifically to support CTDB. */
969
970/* Don't do actual fcntl locking, just mark tdb locked */
971int tdb_transaction_write_lock_mark(struct tdb_context *tdb);
972_PUBLIC_ int tdb_transaction_write_lock_mark(struct tdb_context *tdb)
973{
974 return tdb_transaction_lock(tdb, F_WRLCK, TDB_LOCK_MARK_ONLY);
975}
976
977/* Don't do actual fcntl unlocking, just mark tdb unlocked */
978int tdb_transaction_write_lock_unmark(struct tdb_context *tdb);
979_PUBLIC_ int tdb_transaction_write_lock_unmark(struct tdb_context *tdb)
980{
981 return tdb_nest_unlock(tdb, TRANSACTION_LOCK, F_WRLCK, true);
982}
Note: See TracBrowser for help on using the repository browser.