ctdb-locking: make process_callbacks() more robust
[Samba.git] / ctdb / server / ctdb_lock.c
blob6a79e209232431f2068404598d12e007b323e8ab
1 /*
2 ctdb lock handling
3 provide API to do non-blocking locks for single or all databases
5 Copyright (C) Amitay Isaacs 2012
7 This program is free software; you can redistribute it and/or modify
8 it under the terms of the GNU General Public License as published by
9 the Free Software Foundation; either version 3 of the License, or
10 (at your option) any later version.
12 This program is distributed in the hope that it will be useful,
13 but WITHOUT ANY WARRANTY; without even the implied warranty of
14 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 GNU General Public License for more details.
17 You should have received a copy of the GNU General Public License
18 along with this program; if not, see <http://www.gnu.org/licenses/>.
20 #include "includes.h"
21 #include "include/ctdb_private.h"
22 #include "include/ctdb_protocol.h"
23 #include "tevent.h"
24 #include "tdb.h"
25 #include "lib/tdb_wrap/tdb_wrap.h"
26 #include "system/filesys.h"
27 #include "lib/util/dlinklist.h"
30 * Non-blocking Locking API
32 * 1. Create a child process to do blocking locks.
33 * 2. Once the locks are obtained, signal parent process via fd.
34 * 3. Invoke registered callback routine with locking status.
35 * 4. If the child process cannot get locks within certain time,
36 * execute an external script to debug.
38 * ctdb_lock_record() - get a lock on a record
39 * ctdb_lock_db() - get a lock on a DB
40 * ctdb_lock_alldb_prio() - get a lock on all DBs with given priority
41 * ctdb_lock_alldb() - get a lock on all DBs
43 * auto_mark - whether to mark/unmark DBs in before/after callback
44 * = false is used for freezing databases for
45 * recovery since the recovery cannot start till
46 * databases are locked on all the nodes.
47 * = true is used for record locks.
50 enum lock_type {
51 LOCK_RECORD,
52 LOCK_DB,
53 LOCK_ALLDB_PRIO,
54 LOCK_ALLDB,
57 static const char * const lock_type_str[] = {
58 "lock_record",
59 "lock_db",
60 "lock_alldb_prio",
61 "lock_alldb",
64 struct lock_request;
66 /* lock_context is the common part for a lock request */
67 struct lock_context {
68 struct lock_context *next, *prev;
69 enum lock_type type;
70 struct ctdb_context *ctdb;
71 struct ctdb_db_context *ctdb_db;
72 TDB_DATA key;
73 uint32_t priority;
74 bool auto_mark;
75 struct lock_request *request;
76 pid_t child;
77 int fd[2];
78 struct tevent_fd *tfd;
79 struct tevent_timer *ttimer;
80 struct timeval start_time;
81 uint32_t key_hash;
82 bool can_schedule;
85 /* lock_request is the client specific part for a lock request */
86 struct lock_request {
87 struct lock_context *lctx;
88 void (*callback)(void *, bool);
89 void *private_data;
94 * Support samba 3.6.x (and older) versions which do not set db priority.
96 * By default, all databases are set to priority 1. So only when priority
97 * is set to 1, check for databases that need higher priority.
99 static bool later_db(struct ctdb_context *ctdb, const char *name)
101 if (ctdb->tunable.samba3_hack == 0) {
102 return false;
105 if (strstr(name, "brlock") ||
106 strstr(name, "g_lock") ||
107 strstr(name, "notify_onelevel") ||
108 strstr(name, "serverid") ||
109 strstr(name, "xattr_tdb")) {
110 return true;
113 return false;
116 typedef int (*db_handler_t)(struct ctdb_db_context *ctdb_db,
117 uint32_t priority,
118 void *private_data);
120 static int ctdb_db_iterator(struct ctdb_context *ctdb, uint32_t priority,
121 db_handler_t handler, void *private_data)
123 struct ctdb_db_context *ctdb_db;
124 int ret;
126 for (ctdb_db = ctdb->db_list; ctdb_db; ctdb_db = ctdb_db->next) {
127 if (ctdb_db->priority != priority) {
128 continue;
130 if (later_db(ctdb, ctdb_db->db_name)) {
131 continue;
133 ret = handler(ctdb_db, priority, private_data);
134 if (ret != 0) {
135 return -1;
139 /* If priority != 1, later_db check is not required and can return */
140 if (priority != 1) {
141 return 0;
144 for (ctdb_db = ctdb->db_list; ctdb_db; ctdb_db = ctdb_db->next) {
145 if (!later_db(ctdb, ctdb_db->db_name)) {
146 continue;
148 ret = handler(ctdb_db, priority, private_data);
149 if (ret != 0) {
150 return -1;
154 return 0;
159 * lock all databases - mark only
161 static int db_lock_mark_handler(struct ctdb_db_context *ctdb_db, uint32_t priority,
162 void *private_data)
164 int tdb_transaction_write_lock_mark(struct tdb_context *);
166 DEBUG(DEBUG_INFO, ("marking locked database %s, priority:%u\n",
167 ctdb_db->db_name, priority));
169 if (tdb_transaction_write_lock_mark(ctdb_db->ltdb->tdb) != 0) {
170 DEBUG(DEBUG_ERR, ("Failed to mark (transaction lock) database %s\n",
171 ctdb_db->db_name));
172 return -1;
175 if (tdb_lockall_mark(ctdb_db->ltdb->tdb) != 0) {
176 DEBUG(DEBUG_ERR, ("Failed to mark (all lock) database %s\n",
177 ctdb_db->db_name));
178 return -1;
181 return 0;
184 int ctdb_lockall_mark_prio(struct ctdb_context *ctdb, uint32_t priority)
187 * This function is only used by the main dameon during recovery.
188 * At this stage, the databases have already been locked, by a
189 * dedicated child process. The freeze_mode variable is used to track
190 * whether the actual locks are held by the child process or not.
193 if (ctdb->freeze_mode[priority] != CTDB_FREEZE_FROZEN) {
194 DEBUG(DEBUG_ERR, ("Attempt to mark all databases locked when not frozen\n"));
195 return -1;
198 return ctdb_db_iterator(ctdb, priority, db_lock_mark_handler, NULL);
201 static int ctdb_lockall_mark(struct ctdb_context *ctdb)
203 uint32_t priority;
205 for (priority=1; priority<=NUM_DB_PRIORITIES; priority++) {
206 if (ctdb_db_iterator(ctdb, priority, db_lock_mark_handler, NULL) != 0) {
207 return -1;
211 return 0;
216 * lock all databases - unmark only
218 static int db_lock_unmark_handler(struct ctdb_db_context *ctdb_db, uint32_t priority,
219 void *private_data)
221 int tdb_transaction_write_lock_unmark(struct tdb_context *);
223 DEBUG(DEBUG_INFO, ("unmarking locked database %s, priority:%u\n",
224 ctdb_db->db_name, priority));
226 if (tdb_transaction_write_lock_unmark(ctdb_db->ltdb->tdb) != 0) {
227 DEBUG(DEBUG_ERR, ("Failed to unmark (transaction lock) database %s\n",
228 ctdb_db->db_name));
229 return -1;
232 if (tdb_lockall_unmark(ctdb_db->ltdb->tdb) != 0) {
233 DEBUG(DEBUG_ERR, ("Failed to unmark (all lock) database %s\n",
234 ctdb_db->db_name));
235 return -1;
238 return 0;
241 int ctdb_lockall_unmark_prio(struct ctdb_context *ctdb, uint32_t priority)
244 * This function is only used by the main daemon during recovery.
245 * At this stage, the databases have already been locked, by a
246 * dedicated child process. The freeze_mode variable is used to track
247 * whether the actual locks are held by the child process or not.
250 if (ctdb->freeze_mode[priority] != CTDB_FREEZE_FROZEN) {
251 DEBUG(DEBUG_ERR, ("Attempt to unmark all databases locked when not frozen\n"));
252 return -1;
255 return ctdb_db_iterator(ctdb, priority, db_lock_unmark_handler, NULL);
258 static int ctdb_lockall_unmark(struct ctdb_context *ctdb)
260 uint32_t priority;
262 for (priority=NUM_DB_PRIORITIES; priority>0; priority--) {
263 if (ctdb_db_iterator(ctdb, priority, db_lock_unmark_handler, NULL) != 0) {
264 return -1;
268 return 0;
272 static void ctdb_lock_schedule(struct ctdb_context *ctdb);
275 * Destructor to kill the child locking process
277 static int ctdb_lock_context_destructor(struct lock_context *lock_ctx)
279 if (lock_ctx->request) {
280 lock_ctx->request->lctx = NULL;
282 if (lock_ctx->child > 0) {
283 ctdb_kill(lock_ctx->ctdb, lock_ctx->child, SIGKILL);
284 if (lock_ctx->type == LOCK_RECORD) {
285 DLIST_REMOVE(lock_ctx->ctdb_db->lock_current, lock_ctx);
286 } else {
287 DLIST_REMOVE(lock_ctx->ctdb->lock_current, lock_ctx);
289 if (lock_ctx->ctdb_db) {
290 lock_ctx->ctdb_db->lock_num_current--;
292 CTDB_DECREMENT_STAT(lock_ctx->ctdb, locks.num_current);
293 if (lock_ctx->ctdb_db) {
294 CTDB_DECREMENT_DB_STAT(lock_ctx->ctdb_db, locks.num_current);
296 } else {
297 if (lock_ctx->type == LOCK_RECORD) {
298 DLIST_REMOVE(lock_ctx->ctdb_db->lock_pending, lock_ctx);
299 } else {
300 DLIST_REMOVE(lock_ctx->ctdb->lock_pending, lock_ctx);
302 CTDB_DECREMENT_STAT(lock_ctx->ctdb, locks.num_pending);
303 if (lock_ctx->ctdb_db) {
304 CTDB_DECREMENT_DB_STAT(lock_ctx->ctdb_db, locks.num_pending);
308 ctdb_lock_schedule(lock_ctx->ctdb);
310 return 0;
315 * Destructor to remove lock request
317 static int ctdb_lock_request_destructor(struct lock_request *lock_request)
319 if (lock_request->lctx == NULL) {
320 return 0;
323 lock_request->lctx->request = NULL;
324 TALLOC_FREE(lock_request->lctx);
326 return 0;
330 * Process all the callbacks waiting for lock
332 * If lock has failed, callback is executed with locked=false
334 static void process_callbacks(struct lock_context *lock_ctx, bool locked)
336 struct lock_request *request;
337 bool auto_mark = lock_ctx->auto_mark;
339 if (auto_mark && locked) {
340 switch (lock_ctx->type) {
341 case LOCK_RECORD:
342 tdb_chainlock_mark(lock_ctx->ctdb_db->ltdb->tdb, lock_ctx->key);
343 break;
345 case LOCK_DB:
346 tdb_lockall_mark(lock_ctx->ctdb_db->ltdb->tdb);
347 break;
349 case LOCK_ALLDB_PRIO:
350 ctdb_lockall_mark_prio(lock_ctx->ctdb, lock_ctx->priority);
351 break;
353 case LOCK_ALLDB:
354 ctdb_lockall_mark(lock_ctx->ctdb);
355 break;
359 request = lock_ctx->request;
360 if (auto_mark) {
361 /* Since request may be freed in the callback, unset the lock
362 * context, so request destructor will not free lock context.
364 request->lctx = NULL;
367 /* Since request may be freed in the callback, unset the request */
368 lock_ctx->request = NULL;
370 request->callback(request->private_data, locked);
372 if (!auto_mark) {
373 return;
376 if (locked) {
377 switch (lock_ctx->type) {
378 case LOCK_RECORD:
379 tdb_chainlock_unmark(lock_ctx->ctdb_db->ltdb->tdb, lock_ctx->key);
380 break;
382 case LOCK_DB:
383 tdb_lockall_unmark(lock_ctx->ctdb_db->ltdb->tdb);
384 break;
386 case LOCK_ALLDB_PRIO:
387 ctdb_lockall_unmark_prio(lock_ctx->ctdb, lock_ctx->priority);
388 break;
390 case LOCK_ALLDB:
391 ctdb_lockall_unmark(lock_ctx->ctdb);
392 break;
398 static int lock_bucket_id(double t)
400 double ms = 1.e-3, s = 1;
401 int id;
403 if (t < 1*ms) {
404 id = 0;
405 } else if (t < 10*ms) {
406 id = 1;
407 } else if (t < 100*ms) {
408 id = 2;
409 } else if (t < 1*s) {
410 id = 3;
411 } else if (t < 2*s) {
412 id = 4;
413 } else if (t < 4*s) {
414 id = 5;
415 } else if (t < 8*s) {
416 id = 6;
417 } else if (t < 16*s) {
418 id = 7;
419 } else if (t < 32*s) {
420 id = 8;
421 } else if (t < 64*s) {
422 id = 9;
423 } else {
424 id = 10;
427 return id;
431 * Callback routine when the required locks are obtained.
432 * Called from parent context
434 static void ctdb_lock_handler(struct tevent_context *ev,
435 struct tevent_fd *tfd,
436 uint16_t flags,
437 void *private_data)
439 struct lock_context *lock_ctx;
440 TALLOC_CTX *tmp_ctx = NULL;
441 char c;
442 bool locked;
443 double t;
444 int id;
446 lock_ctx = talloc_get_type_abort(private_data, struct lock_context);
448 /* cancel the timeout event */
449 TALLOC_FREE(lock_ctx->ttimer);
451 t = timeval_elapsed(&lock_ctx->start_time);
452 id = lock_bucket_id(t);
454 if (lock_ctx->auto_mark) {
455 tmp_ctx = talloc_new(ev);
456 talloc_steal(tmp_ctx, lock_ctx);
459 /* Read the status from the child process */
460 if (sys_read(lock_ctx->fd[0], &c, 1) != 1) {
461 locked = false;
462 } else {
463 locked = (c == 0 ? true : false);
466 /* Update statistics */
467 CTDB_INCREMENT_STAT(lock_ctx->ctdb, locks.num_calls);
468 if (lock_ctx->ctdb_db) {
469 CTDB_INCREMENT_DB_STAT(lock_ctx->ctdb_db, locks.num_calls);
472 if (locked) {
473 if (lock_ctx->ctdb_db) {
474 CTDB_INCREMENT_STAT(lock_ctx->ctdb, locks.buckets[id]);
475 CTDB_UPDATE_LATENCY(lock_ctx->ctdb, lock_ctx->ctdb_db,
476 lock_type_str[lock_ctx->type], locks.latency,
477 lock_ctx->start_time);
479 CTDB_UPDATE_DB_LATENCY(lock_ctx->ctdb_db, lock_type_str[lock_ctx->type], locks.latency, t);
480 CTDB_INCREMENT_DB_STAT(lock_ctx->ctdb_db, locks.buckets[id]);
482 } else {
483 CTDB_INCREMENT_STAT(lock_ctx->ctdb, locks.num_failed);
484 if (lock_ctx->ctdb_db) {
485 CTDB_INCREMENT_DB_STAT(lock_ctx->ctdb_db, locks.num_failed);
489 process_callbacks(lock_ctx, locked);
491 if (lock_ctx->auto_mark) {
492 talloc_free(tmp_ctx);
498 * Callback routine when required locks are not obtained within timeout
499 * Called from parent context
501 static void ctdb_lock_timeout_handler(struct tevent_context *ev,
502 struct tevent_timer *ttimer,
503 struct timeval current_time,
504 void *private_data)
506 static char debug_locks[PATH_MAX+1] = "";
507 struct lock_context *lock_ctx;
508 struct ctdb_context *ctdb;
509 pid_t pid;
510 double elapsed_time;
511 int new_timer;
513 lock_ctx = talloc_get_type_abort(private_data, struct lock_context);
514 ctdb = lock_ctx->ctdb;
516 /* If a node stopped/banned, don't spam the logs */
517 if (ctdb->nodes[ctdb->pnn]->flags & NODE_FLAGS_INACTIVE) {
518 lock_ctx->ttimer = NULL;
519 return;
522 elapsed_time = timeval_elapsed(&lock_ctx->start_time);
523 if (lock_ctx->ctdb_db) {
524 DEBUG(DEBUG_WARNING,
525 ("Unable to get %s lock on database %s for %.0lf seconds\n",
526 (lock_ctx->type == LOCK_RECORD ? "RECORD" : "DB"),
527 lock_ctx->ctdb_db->db_name, elapsed_time));
528 } else {
529 DEBUG(DEBUG_WARNING,
530 ("Unable to get ALLDB locks for %.0lf seconds\n",
531 elapsed_time));
534 if (ctdb_set_helper("lock debugging helper",
535 debug_locks, sizeof(debug_locks),
536 "CTDB_DEBUG_LOCKS",
537 getenv("CTDB_BASE"), "debug_locks.sh")) {
538 pid = vfork();
539 if (pid == 0) {
540 execl(debug_locks, debug_locks, NULL);
541 _exit(0);
543 ctdb_track_child(ctdb, pid);
544 } else {
545 DEBUG(DEBUG_WARNING,
546 (__location__
547 " Unable to setup lock debugging\n"));
550 /* Back-off logging if lock is not obtained for a long time */
551 if (elapsed_time < 100.0) {
552 new_timer = 10;
553 } else if (elapsed_time < 1000.0) {
554 new_timer = 100;
555 } else {
556 new_timer = 1000;
559 /* reset the timeout timer */
560 // talloc_free(lock_ctx->ttimer);
561 lock_ctx->ttimer = tevent_add_timer(ctdb->ev,
562 lock_ctx,
563 timeval_current_ofs(new_timer, 0),
564 ctdb_lock_timeout_handler,
565 (void *)lock_ctx);
569 static int db_count_handler(struct ctdb_db_context *ctdb_db, uint32_t priority,
570 void *private_data)
572 int *count = (int *)private_data;
574 (*count) += 2;
576 return 0;
579 static int db_flags(struct ctdb_db_context *ctdb_db)
581 int tdb_flags = TDB_DEFAULT;
583 #ifdef TDB_MUTEX_LOCKING
584 if (!ctdb_db->persistent && ctdb_db->ctdb->tunable.mutex_enabled) {
585 tdb_flags = (TDB_MUTEX_LOCKING | TDB_CLEAR_IF_FIRST);
587 #endif
588 return tdb_flags;
591 struct db_namelist {
592 const char **names;
593 int n;
596 static int db_name_handler(struct ctdb_db_context *ctdb_db, uint32_t priority,
597 void *private_data)
599 struct db_namelist *list = (struct db_namelist *)private_data;
601 list->names[list->n] = talloc_strdup(list->names, ctdb_db->db_path);
602 list->names[list->n+1] = talloc_asprintf(list->names, "0x%x",
603 db_flags(ctdb_db));
604 list->n += 2;
606 return 0;
609 static bool lock_helper_args(TALLOC_CTX *mem_ctx,
610 struct lock_context *lock_ctx, int fd,
611 int *argc, const char ***argv)
613 struct ctdb_context *ctdb = lock_ctx->ctdb;
614 const char **args = NULL;
615 int nargs, i;
616 int priority;
617 struct db_namelist list;
619 switch (lock_ctx->type) {
620 case LOCK_RECORD:
621 nargs = 6;
622 break;
624 case LOCK_DB:
625 nargs = 5;
626 break;
628 case LOCK_ALLDB_PRIO:
629 nargs = 3;
630 ctdb_db_iterator(ctdb, lock_ctx->priority, db_count_handler, &nargs);
631 break;
633 case LOCK_ALLDB:
634 nargs = 3;
635 for (priority=1; priority<NUM_DB_PRIORITIES; priority++) {
636 ctdb_db_iterator(ctdb, priority, db_count_handler, &nargs);
638 break;
641 /* Add extra argument for null termination */
642 nargs++;
644 args = talloc_array(mem_ctx, const char *, nargs);
645 if (args == NULL) {
646 return false;
649 args[0] = talloc_asprintf(args, "%d", getpid());
650 args[1] = talloc_asprintf(args, "%d", fd);
652 switch (lock_ctx->type) {
653 case LOCK_RECORD:
654 args[2] = talloc_strdup(args, "RECORD");
655 args[3] = talloc_strdup(args, lock_ctx->ctdb_db->db_path);
656 args[4] = talloc_asprintf(args, "0x%x",
657 db_flags(lock_ctx->ctdb_db));
658 if (lock_ctx->key.dsize == 0) {
659 args[5] = talloc_strdup(args, "NULL");
660 } else {
661 args[5] = hex_encode_talloc(args, lock_ctx->key.dptr, lock_ctx->key.dsize);
663 break;
665 case LOCK_DB:
666 args[2] = talloc_strdup(args, "DB");
667 args[3] = talloc_strdup(args, lock_ctx->ctdb_db->db_path);
668 args[4] = talloc_asprintf(args, "0x%x",
669 db_flags(lock_ctx->ctdb_db));
670 break;
672 case LOCK_ALLDB_PRIO:
673 args[2] = talloc_strdup(args, "DB");
674 list.names = args;
675 list.n = 3;
676 ctdb_db_iterator(ctdb, lock_ctx->priority, db_name_handler, &list);
677 break;
679 case LOCK_ALLDB:
680 args[2] = talloc_strdup(args, "DB");
681 list.names = args;
682 list.n = 3;
683 for (priority=1; priority<NUM_DB_PRIORITIES; priority++) {
684 ctdb_db_iterator(ctdb, priority, db_name_handler, &list);
686 break;
689 /* Make sure last argument is NULL */
690 args[nargs-1] = NULL;
692 for (i=0; i<nargs-1; i++) {
693 if (args[i] == NULL) {
694 talloc_free(args);
695 return false;
699 *argc = nargs;
700 *argv = args;
701 return true;
705 * Find a lock request that can be scheduled
707 static struct lock_context *ctdb_find_lock_context(struct ctdb_context *ctdb)
709 struct lock_context *lock_ctx, *next_ctx;
710 struct ctdb_db_context *ctdb_db;
712 /* First check if there are database lock requests */
714 for (lock_ctx = ctdb->lock_pending; lock_ctx != NULL;
715 lock_ctx = next_ctx) {
717 if (lock_ctx->request != NULL) {
718 /* Found a lock context with a request */
719 return lock_ctx;
722 next_ctx = lock_ctx->next;
724 DEBUG(DEBUG_INFO, ("Removing lock context without lock "
725 "request\n"));
726 DLIST_REMOVE(ctdb->lock_pending, lock_ctx);
727 CTDB_DECREMENT_STAT(ctdb, locks.num_pending);
728 if (lock_ctx->ctdb_db) {
729 CTDB_DECREMENT_DB_STAT(lock_ctx->ctdb_db,
730 locks.num_pending);
732 talloc_free(lock_ctx);
735 /* Next check database queues */
736 for (ctdb_db = ctdb->db_list; ctdb_db; ctdb_db = ctdb_db->next) {
737 if (ctdb_db->lock_num_current ==
738 ctdb->tunable.lock_processes_per_db) {
739 continue;
742 for (lock_ctx = ctdb_db->lock_pending; lock_ctx != NULL;
743 lock_ctx = next_ctx) {
745 next_ctx = lock_ctx->next;
747 if (lock_ctx->request != NULL) {
748 return lock_ctx;
751 DEBUG(DEBUG_INFO, ("Removing lock context without "
752 "lock request\n"));
753 DLIST_REMOVE(ctdb_db->lock_pending, lock_ctx);
754 CTDB_DECREMENT_STAT(ctdb, locks.num_pending);
755 CTDB_DECREMENT_DB_STAT(ctdb_db, locks.num_pending);
756 talloc_free(lock_ctx);
760 return NULL;
764 * Schedule a new lock child process
765 * Set up callback handler and timeout handler
767 static void ctdb_lock_schedule(struct ctdb_context *ctdb)
769 struct lock_context *lock_ctx;
770 int ret, argc;
771 TALLOC_CTX *tmp_ctx;
772 static char prog[PATH_MAX+1] = "";
773 const char **args;
775 if (!ctdb_set_helper("lock helper",
776 prog, sizeof(prog),
777 "CTDB_LOCK_HELPER",
778 CTDB_HELPER_BINDIR, "ctdb_lock_helper")) {
779 ctdb_die(ctdb, __location__
780 " Unable to set lock helper\n");
783 /* Find a lock context with requests */
784 lock_ctx = ctdb_find_lock_context(ctdb);
785 if (lock_ctx == NULL) {
786 return;
789 lock_ctx->child = -1;
790 ret = pipe(lock_ctx->fd);
791 if (ret != 0) {
792 DEBUG(DEBUG_ERR, ("Failed to create pipe in ctdb_lock_schedule\n"));
793 return;
796 set_close_on_exec(lock_ctx->fd[0]);
798 /* Create data for child process */
799 tmp_ctx = talloc_new(lock_ctx);
800 if (tmp_ctx == NULL) {
801 DEBUG(DEBUG_ERR, ("Failed to allocate memory for helper args\n"));
802 close(lock_ctx->fd[0]);
803 close(lock_ctx->fd[1]);
804 return;
807 /* Create arguments for lock helper */
808 if (!lock_helper_args(tmp_ctx, lock_ctx, lock_ctx->fd[1],
809 &argc, &args)) {
810 DEBUG(DEBUG_ERR, ("Failed to create lock helper args\n"));
811 close(lock_ctx->fd[0]);
812 close(lock_ctx->fd[1]);
813 talloc_free(tmp_ctx);
814 return;
817 if (!ctdb_vfork_with_logging(lock_ctx, ctdb, "lock_helper",
818 prog, argc, (const char **)args,
819 NULL, NULL, &lock_ctx->child)) {
820 DEBUG(DEBUG_ERR, ("Failed to create a child in ctdb_lock_schedule\n"));
821 close(lock_ctx->fd[0]);
822 close(lock_ctx->fd[1]);
823 talloc_free(tmp_ctx);
824 return;
827 /* Parent process */
828 close(lock_ctx->fd[1]);
830 talloc_free(tmp_ctx);
832 /* Set up timeout handler */
833 lock_ctx->ttimer = tevent_add_timer(ctdb->ev,
834 lock_ctx,
835 timeval_current_ofs(10, 0),
836 ctdb_lock_timeout_handler,
837 (void *)lock_ctx);
838 if (lock_ctx->ttimer == NULL) {
839 ctdb_kill(ctdb, lock_ctx->child, SIGKILL);
840 lock_ctx->child = -1;
841 close(lock_ctx->fd[0]);
842 return;
845 /* Set up callback */
846 lock_ctx->tfd = tevent_add_fd(ctdb->ev,
847 lock_ctx,
848 lock_ctx->fd[0],
849 EVENT_FD_READ,
850 ctdb_lock_handler,
851 (void *)lock_ctx);
852 if (lock_ctx->tfd == NULL) {
853 TALLOC_FREE(lock_ctx->ttimer);
854 ctdb_kill(ctdb, lock_ctx->child, SIGKILL);
855 lock_ctx->child = -1;
856 close(lock_ctx->fd[0]);
857 return;
859 tevent_fd_set_auto_close(lock_ctx->tfd);
861 /* Move the context from pending to current */
862 if (lock_ctx->type == LOCK_RECORD) {
863 DLIST_REMOVE(lock_ctx->ctdb_db->lock_pending, lock_ctx);
864 DLIST_ADD_END(lock_ctx->ctdb_db->lock_current, lock_ctx, NULL);
865 } else {
866 DLIST_REMOVE(ctdb->lock_pending, lock_ctx);
867 DLIST_ADD_END(ctdb->lock_current, lock_ctx, NULL);
869 CTDB_DECREMENT_STAT(lock_ctx->ctdb, locks.num_pending);
870 CTDB_INCREMENT_STAT(lock_ctx->ctdb, locks.num_current);
871 if (lock_ctx->ctdb_db) {
872 lock_ctx->ctdb_db->lock_num_current++;
873 CTDB_DECREMENT_DB_STAT(lock_ctx->ctdb_db, locks.num_pending);
874 CTDB_INCREMENT_DB_STAT(lock_ctx->ctdb_db, locks.num_current);
880 * Lock record / db depending on type
882 static struct lock_request *ctdb_lock_internal(TALLOC_CTX *mem_ctx,
883 struct ctdb_context *ctdb,
884 struct ctdb_db_context *ctdb_db,
885 TDB_DATA key,
886 uint32_t priority,
887 void (*callback)(void *, bool),
888 void *private_data,
889 enum lock_type type,
890 bool auto_mark)
892 struct lock_context *lock_ctx = NULL;
893 struct lock_request *request;
895 if (callback == NULL) {
896 DEBUG(DEBUG_WARNING, ("No callback function specified, not locking\n"));
897 return NULL;
900 lock_ctx = talloc_zero(ctdb, struct lock_context);
901 if (lock_ctx == NULL) {
902 DEBUG(DEBUG_ERR, ("Failed to create a new lock context\n"));
903 return NULL;
906 if ((request = talloc_zero(mem_ctx, struct lock_request)) == NULL) {
907 talloc_free(lock_ctx);
908 return NULL;
911 lock_ctx->type = type;
912 lock_ctx->ctdb = ctdb;
913 lock_ctx->ctdb_db = ctdb_db;
914 lock_ctx->key.dsize = key.dsize;
915 if (key.dsize > 0) {
916 lock_ctx->key.dptr = talloc_memdup(lock_ctx, key.dptr, key.dsize);
917 if (lock_ctx->key.dptr == NULL) {
918 DEBUG(DEBUG_ERR, (__location__ "Memory allocation error\n"));
919 talloc_free(lock_ctx);
920 talloc_free(request);
921 return NULL;
923 lock_ctx->key_hash = ctdb_hash(&key);
924 } else {
925 lock_ctx->key.dptr = NULL;
927 lock_ctx->priority = priority;
928 lock_ctx->auto_mark = auto_mark;
930 lock_ctx->request = request;
931 lock_ctx->child = -1;
933 /* Non-record locks are required by recovery and should be scheduled
934 * immediately, so keep them at the head of the pending queue.
936 if (lock_ctx->type == LOCK_RECORD) {
937 DLIST_ADD_END(ctdb_db->lock_pending, lock_ctx, NULL);
938 } else {
939 DLIST_ADD_END(ctdb->lock_pending, lock_ctx, NULL);
941 CTDB_INCREMENT_STAT(ctdb, locks.num_pending);
942 if (ctdb_db) {
943 CTDB_INCREMENT_DB_STAT(ctdb_db, locks.num_pending);
946 /* Start the timer when we activate the context */
947 lock_ctx->start_time = timeval_current();
949 request->lctx = lock_ctx;
950 request->callback = callback;
951 request->private_data = private_data;
953 talloc_set_destructor(request, ctdb_lock_request_destructor);
954 talloc_set_destructor(lock_ctx, ctdb_lock_context_destructor);
956 ctdb_lock_schedule(ctdb);
958 return request;
963 * obtain a lock on a record in a database
965 struct lock_request *ctdb_lock_record(TALLOC_CTX *mem_ctx,
966 struct ctdb_db_context *ctdb_db,
967 TDB_DATA key,
968 bool auto_mark,
969 void (*callback)(void *, bool),
970 void *private_data)
972 return ctdb_lock_internal(mem_ctx,
973 ctdb_db->ctdb,
974 ctdb_db,
975 key,
977 callback,
978 private_data,
979 LOCK_RECORD,
980 auto_mark);
985 * obtain a lock on a database
987 struct lock_request *ctdb_lock_db(TALLOC_CTX *mem_ctx,
988 struct ctdb_db_context *ctdb_db,
989 bool auto_mark,
990 void (*callback)(void *, bool),
991 void *private_data)
993 return ctdb_lock_internal(mem_ctx,
994 ctdb_db->ctdb,
995 ctdb_db,
996 tdb_null,
998 callback,
999 private_data,
1000 LOCK_DB,
1001 auto_mark);
1006 * obtain locks on all databases of specified priority
1008 struct lock_request *ctdb_lock_alldb_prio(TALLOC_CTX *mem_ctx,
1009 struct ctdb_context *ctdb,
1010 uint32_t priority,
1011 bool auto_mark,
1012 void (*callback)(void *, bool),
1013 void *private_data)
1015 if (priority < 1 || priority > NUM_DB_PRIORITIES) {
1016 DEBUG(DEBUG_ERR, ("Invalid db priority: %u\n", priority));
1017 return NULL;
1020 return ctdb_lock_internal(mem_ctx,
1021 ctdb,
1022 NULL,
1023 tdb_null,
1024 priority,
1025 callback,
1026 private_data,
1027 LOCK_ALLDB_PRIO,
1028 auto_mark);
1033 * obtain locks on all databases
1035 struct lock_request *ctdb_lock_alldb(TALLOC_CTX *mem_ctx,
1036 struct ctdb_context *ctdb,
1037 bool auto_mark,
1038 void (*callback)(void *, bool),
1039 void *private_data)
1041 return ctdb_lock_internal(mem_ctx,
1042 ctdb,
1043 NULL,
1044 tdb_null,
1046 callback,
1047 private_data,
1048 LOCK_ALLDB,
1049 auto_mark);