selftest: Add setup_fileserver()
[Samba.git] / ctdb / server / ctdb_lock.c
blob82295ea5dcb8003baf9cacbfb34309fba970f2ec
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;
396 talloc_free(lock_ctx);
400 static int lock_bucket_id(double t)
402 double ms = 1.e-3, s = 1;
403 int id;
405 if (t < 1*ms) {
406 id = 0;
407 } else if (t < 10*ms) {
408 id = 1;
409 } else if (t < 100*ms) {
410 id = 2;
411 } else if (t < 1*s) {
412 id = 3;
413 } else if (t < 2*s) {
414 id = 4;
415 } else if (t < 4*s) {
416 id = 5;
417 } else if (t < 8*s) {
418 id = 6;
419 } else if (t < 16*s) {
420 id = 7;
421 } else if (t < 32*s) {
422 id = 8;
423 } else if (t < 64*s) {
424 id = 9;
425 } else {
426 id = 10;
429 return id;
433 * Callback routine when the required locks are obtained.
434 * Called from parent context
436 static void ctdb_lock_handler(struct tevent_context *ev,
437 struct tevent_fd *tfd,
438 uint16_t flags,
439 void *private_data)
441 struct lock_context *lock_ctx;
442 char c;
443 bool locked;
444 double t;
445 int id;
447 lock_ctx = talloc_get_type_abort(private_data, struct lock_context);
449 /* cancel the timeout event */
450 TALLOC_FREE(lock_ctx->ttimer);
452 t = timeval_elapsed(&lock_ctx->start_time);
453 id = lock_bucket_id(t);
455 /* Read the status from the child process */
456 if (sys_read(lock_ctx->fd[0], &c, 1) != 1) {
457 locked = false;
458 } else {
459 locked = (c == 0 ? true : false);
462 /* Update statistics */
463 CTDB_INCREMENT_STAT(lock_ctx->ctdb, locks.num_calls);
464 if (lock_ctx->ctdb_db) {
465 CTDB_INCREMENT_DB_STAT(lock_ctx->ctdb_db, locks.num_calls);
468 if (locked) {
469 if (lock_ctx->ctdb_db) {
470 CTDB_INCREMENT_STAT(lock_ctx->ctdb, locks.buckets[id]);
471 CTDB_UPDATE_LATENCY(lock_ctx->ctdb, lock_ctx->ctdb_db,
472 lock_type_str[lock_ctx->type], locks.latency,
473 lock_ctx->start_time);
475 CTDB_UPDATE_DB_LATENCY(lock_ctx->ctdb_db, lock_type_str[lock_ctx->type], locks.latency, t);
476 CTDB_INCREMENT_DB_STAT(lock_ctx->ctdb_db, locks.buckets[id]);
478 } else {
479 CTDB_INCREMENT_STAT(lock_ctx->ctdb, locks.num_failed);
480 if (lock_ctx->ctdb_db) {
481 CTDB_INCREMENT_DB_STAT(lock_ctx->ctdb_db, locks.num_failed);
485 process_callbacks(lock_ctx, locked);
490 * Callback routine when required locks are not obtained within timeout
491 * Called from parent context
493 static void ctdb_lock_timeout_handler(struct tevent_context *ev,
494 struct tevent_timer *ttimer,
495 struct timeval current_time,
496 void *private_data)
498 static char debug_locks[PATH_MAX+1] = "";
499 struct lock_context *lock_ctx;
500 struct ctdb_context *ctdb;
501 pid_t pid;
502 double elapsed_time;
503 int new_timer;
505 lock_ctx = talloc_get_type_abort(private_data, struct lock_context);
506 ctdb = lock_ctx->ctdb;
508 /* If a node stopped/banned, don't spam the logs */
509 if (ctdb->nodes[ctdb->pnn]->flags & NODE_FLAGS_INACTIVE) {
510 lock_ctx->ttimer = NULL;
511 return;
514 elapsed_time = timeval_elapsed(&lock_ctx->start_time);
515 if (lock_ctx->ctdb_db) {
516 DEBUG(DEBUG_WARNING,
517 ("Unable to get %s lock on database %s for %.0lf seconds\n",
518 (lock_ctx->type == LOCK_RECORD ? "RECORD" : "DB"),
519 lock_ctx->ctdb_db->db_name, elapsed_time));
520 } else {
521 DEBUG(DEBUG_WARNING,
522 ("Unable to get ALLDB locks for %.0lf seconds\n",
523 elapsed_time));
526 if (ctdb_set_helper("lock debugging helper",
527 debug_locks, sizeof(debug_locks),
528 "CTDB_DEBUG_LOCKS",
529 getenv("CTDB_BASE"), "debug_locks.sh")) {
530 pid = vfork();
531 if (pid == 0) {
532 execl(debug_locks, debug_locks, NULL);
533 _exit(0);
535 ctdb_track_child(ctdb, pid);
536 } else {
537 DEBUG(DEBUG_WARNING,
538 (__location__
539 " Unable to setup lock debugging\n"));
542 /* Back-off logging if lock is not obtained for a long time */
543 if (elapsed_time < 100.0) {
544 new_timer = 10;
545 } else if (elapsed_time < 1000.0) {
546 new_timer = 100;
547 } else {
548 new_timer = 1000;
551 /* reset the timeout timer */
552 // talloc_free(lock_ctx->ttimer);
553 lock_ctx->ttimer = tevent_add_timer(ctdb->ev,
554 lock_ctx,
555 timeval_current_ofs(new_timer, 0),
556 ctdb_lock_timeout_handler,
557 (void *)lock_ctx);
561 static int db_count_handler(struct ctdb_db_context *ctdb_db, uint32_t priority,
562 void *private_data)
564 int *count = (int *)private_data;
566 (*count) += 2;
568 return 0;
571 static int db_flags(struct ctdb_db_context *ctdb_db)
573 int tdb_flags = TDB_DEFAULT;
575 #ifdef TDB_MUTEX_LOCKING
576 if (!ctdb_db->persistent && ctdb_db->ctdb->tunable.mutex_enabled) {
577 tdb_flags = (TDB_MUTEX_LOCKING | TDB_CLEAR_IF_FIRST);
579 #endif
580 return tdb_flags;
583 struct db_namelist {
584 const char **names;
585 int n;
588 static int db_name_handler(struct ctdb_db_context *ctdb_db, uint32_t priority,
589 void *private_data)
591 struct db_namelist *list = (struct db_namelist *)private_data;
593 list->names[list->n] = talloc_strdup(list->names, ctdb_db->db_path);
594 list->names[list->n+1] = talloc_asprintf(list->names, "0x%x",
595 db_flags(ctdb_db));
596 list->n += 2;
598 return 0;
601 static bool lock_helper_args(TALLOC_CTX *mem_ctx,
602 struct lock_context *lock_ctx, int fd,
603 int *argc, const char ***argv)
605 struct ctdb_context *ctdb = lock_ctx->ctdb;
606 const char **args = NULL;
607 int nargs, i;
608 int priority;
609 struct db_namelist list;
611 switch (lock_ctx->type) {
612 case LOCK_RECORD:
613 nargs = 6;
614 break;
616 case LOCK_DB:
617 nargs = 5;
618 break;
620 case LOCK_ALLDB_PRIO:
621 nargs = 3;
622 ctdb_db_iterator(ctdb, lock_ctx->priority, db_count_handler, &nargs);
623 break;
625 case LOCK_ALLDB:
626 nargs = 3;
627 for (priority=1; priority<NUM_DB_PRIORITIES; priority++) {
628 ctdb_db_iterator(ctdb, priority, db_count_handler, &nargs);
630 break;
633 /* Add extra argument for null termination */
634 nargs++;
636 args = talloc_array(mem_ctx, const char *, nargs);
637 if (args == NULL) {
638 return false;
641 args[0] = talloc_asprintf(args, "%d", getpid());
642 args[1] = talloc_asprintf(args, "%d", fd);
644 switch (lock_ctx->type) {
645 case LOCK_RECORD:
646 args[2] = talloc_strdup(args, "RECORD");
647 args[3] = talloc_strdup(args, lock_ctx->ctdb_db->db_path);
648 args[4] = talloc_asprintf(args, "0x%x",
649 db_flags(lock_ctx->ctdb_db));
650 if (lock_ctx->key.dsize == 0) {
651 args[5] = talloc_strdup(args, "NULL");
652 } else {
653 args[5] = hex_encode_talloc(args, lock_ctx->key.dptr, lock_ctx->key.dsize);
655 break;
657 case LOCK_DB:
658 args[2] = talloc_strdup(args, "DB");
659 args[3] = talloc_strdup(args, lock_ctx->ctdb_db->db_path);
660 args[4] = talloc_asprintf(args, "0x%x",
661 db_flags(lock_ctx->ctdb_db));
662 break;
664 case LOCK_ALLDB_PRIO:
665 args[2] = talloc_strdup(args, "DB");
666 list.names = args;
667 list.n = 3;
668 ctdb_db_iterator(ctdb, lock_ctx->priority, db_name_handler, &list);
669 break;
671 case LOCK_ALLDB:
672 args[2] = talloc_strdup(args, "DB");
673 list.names = args;
674 list.n = 3;
675 for (priority=1; priority<NUM_DB_PRIORITIES; priority++) {
676 ctdb_db_iterator(ctdb, priority, db_name_handler, &list);
678 break;
681 /* Make sure last argument is NULL */
682 args[nargs-1] = NULL;
684 for (i=0; i<nargs-1; i++) {
685 if (args[i] == NULL) {
686 talloc_free(args);
687 return false;
691 *argc = nargs;
692 *argv = args;
693 return true;
697 * Find a lock request that can be scheduled
699 static struct lock_context *ctdb_find_lock_context(struct ctdb_context *ctdb)
701 struct lock_context *lock_ctx, *next_ctx;
702 struct ctdb_db_context *ctdb_db;
704 /* First check if there are database lock requests */
706 for (lock_ctx = ctdb->lock_pending; lock_ctx != NULL;
707 lock_ctx = next_ctx) {
709 if (lock_ctx->request != NULL) {
710 /* Found a lock context with a request */
711 return lock_ctx;
714 next_ctx = lock_ctx->next;
716 DEBUG(DEBUG_INFO, ("Removing lock context without lock "
717 "request\n"));
718 DLIST_REMOVE(ctdb->lock_pending, lock_ctx);
719 CTDB_DECREMENT_STAT(ctdb, locks.num_pending);
720 if (lock_ctx->ctdb_db) {
721 CTDB_DECREMENT_DB_STAT(lock_ctx->ctdb_db,
722 locks.num_pending);
724 talloc_free(lock_ctx);
727 /* Next check database queues */
728 for (ctdb_db = ctdb->db_list; ctdb_db; ctdb_db = ctdb_db->next) {
729 if (ctdb_db->lock_num_current ==
730 ctdb->tunable.lock_processes_per_db) {
731 continue;
734 for (lock_ctx = ctdb_db->lock_pending; lock_ctx != NULL;
735 lock_ctx = next_ctx) {
737 next_ctx = lock_ctx->next;
739 if (lock_ctx->request != NULL) {
740 return lock_ctx;
743 DEBUG(DEBUG_INFO, ("Removing lock context without "
744 "lock request\n"));
745 DLIST_REMOVE(ctdb_db->lock_pending, lock_ctx);
746 CTDB_DECREMENT_STAT(ctdb, locks.num_pending);
747 CTDB_DECREMENT_DB_STAT(ctdb_db, locks.num_pending);
748 talloc_free(lock_ctx);
752 return NULL;
756 * Schedule a new lock child process
757 * Set up callback handler and timeout handler
759 static void ctdb_lock_schedule(struct ctdb_context *ctdb)
761 struct lock_context *lock_ctx;
762 int ret, argc;
763 TALLOC_CTX *tmp_ctx;
764 static char prog[PATH_MAX+1] = "";
765 const char **args;
767 if (!ctdb_set_helper("lock helper",
768 prog, sizeof(prog),
769 "CTDB_LOCK_HELPER",
770 CTDB_HELPER_BINDIR, "ctdb_lock_helper")) {
771 ctdb_die(ctdb, __location__
772 " Unable to set lock helper\n");
775 /* Find a lock context with requests */
776 lock_ctx = ctdb_find_lock_context(ctdb);
777 if (lock_ctx == NULL) {
778 return;
781 lock_ctx->child = -1;
782 ret = pipe(lock_ctx->fd);
783 if (ret != 0) {
784 DEBUG(DEBUG_ERR, ("Failed to create pipe in ctdb_lock_schedule\n"));
785 return;
788 set_close_on_exec(lock_ctx->fd[0]);
790 /* Create data for child process */
791 tmp_ctx = talloc_new(lock_ctx);
792 if (tmp_ctx == NULL) {
793 DEBUG(DEBUG_ERR, ("Failed to allocate memory for helper args\n"));
794 close(lock_ctx->fd[0]);
795 close(lock_ctx->fd[1]);
796 return;
799 /* Create arguments for lock helper */
800 if (!lock_helper_args(tmp_ctx, lock_ctx, lock_ctx->fd[1],
801 &argc, &args)) {
802 DEBUG(DEBUG_ERR, ("Failed to create lock helper args\n"));
803 close(lock_ctx->fd[0]);
804 close(lock_ctx->fd[1]);
805 talloc_free(tmp_ctx);
806 return;
809 if (!ctdb_vfork_with_logging(lock_ctx, ctdb, "lock_helper",
810 prog, argc, (const char **)args,
811 NULL, NULL, &lock_ctx->child)) {
812 DEBUG(DEBUG_ERR, ("Failed to create a child in ctdb_lock_schedule\n"));
813 close(lock_ctx->fd[0]);
814 close(lock_ctx->fd[1]);
815 talloc_free(tmp_ctx);
816 return;
819 /* Parent process */
820 close(lock_ctx->fd[1]);
822 talloc_free(tmp_ctx);
824 /* Set up timeout handler */
825 lock_ctx->ttimer = tevent_add_timer(ctdb->ev,
826 lock_ctx,
827 timeval_current_ofs(10, 0),
828 ctdb_lock_timeout_handler,
829 (void *)lock_ctx);
830 if (lock_ctx->ttimer == NULL) {
831 ctdb_kill(ctdb, lock_ctx->child, SIGKILL);
832 lock_ctx->child = -1;
833 close(lock_ctx->fd[0]);
834 return;
837 /* Set up callback */
838 lock_ctx->tfd = tevent_add_fd(ctdb->ev,
839 lock_ctx,
840 lock_ctx->fd[0],
841 EVENT_FD_READ,
842 ctdb_lock_handler,
843 (void *)lock_ctx);
844 if (lock_ctx->tfd == NULL) {
845 TALLOC_FREE(lock_ctx->ttimer);
846 ctdb_kill(ctdb, lock_ctx->child, SIGKILL);
847 lock_ctx->child = -1;
848 close(lock_ctx->fd[0]);
849 return;
851 tevent_fd_set_auto_close(lock_ctx->tfd);
853 /* Move the context from pending to current */
854 if (lock_ctx->type == LOCK_RECORD) {
855 DLIST_REMOVE(lock_ctx->ctdb_db->lock_pending, lock_ctx);
856 DLIST_ADD_END(lock_ctx->ctdb_db->lock_current, lock_ctx, NULL);
857 } else {
858 DLIST_REMOVE(ctdb->lock_pending, lock_ctx);
859 DLIST_ADD_END(ctdb->lock_current, lock_ctx, NULL);
861 CTDB_DECREMENT_STAT(lock_ctx->ctdb, locks.num_pending);
862 CTDB_INCREMENT_STAT(lock_ctx->ctdb, locks.num_current);
863 if (lock_ctx->ctdb_db) {
864 lock_ctx->ctdb_db->lock_num_current++;
865 CTDB_DECREMENT_DB_STAT(lock_ctx->ctdb_db, locks.num_pending);
866 CTDB_INCREMENT_DB_STAT(lock_ctx->ctdb_db, locks.num_current);
872 * Lock record / db depending on type
874 static struct lock_request *ctdb_lock_internal(TALLOC_CTX *mem_ctx,
875 struct ctdb_context *ctdb,
876 struct ctdb_db_context *ctdb_db,
877 TDB_DATA key,
878 uint32_t priority,
879 void (*callback)(void *, bool),
880 void *private_data,
881 enum lock_type type,
882 bool auto_mark)
884 struct lock_context *lock_ctx = NULL;
885 struct lock_request *request;
887 if (callback == NULL) {
888 DEBUG(DEBUG_WARNING, ("No callback function specified, not locking\n"));
889 return NULL;
892 lock_ctx = talloc_zero(ctdb, struct lock_context);
893 if (lock_ctx == NULL) {
894 DEBUG(DEBUG_ERR, ("Failed to create a new lock context\n"));
895 return NULL;
898 if ((request = talloc_zero(mem_ctx, struct lock_request)) == NULL) {
899 talloc_free(lock_ctx);
900 return NULL;
903 lock_ctx->type = type;
904 lock_ctx->ctdb = ctdb;
905 lock_ctx->ctdb_db = ctdb_db;
906 lock_ctx->key.dsize = key.dsize;
907 if (key.dsize > 0) {
908 lock_ctx->key.dptr = talloc_memdup(lock_ctx, key.dptr, key.dsize);
909 if (lock_ctx->key.dptr == NULL) {
910 DEBUG(DEBUG_ERR, (__location__ "Memory allocation error\n"));
911 talloc_free(lock_ctx);
912 talloc_free(request);
913 return NULL;
915 lock_ctx->key_hash = ctdb_hash(&key);
916 } else {
917 lock_ctx->key.dptr = NULL;
919 lock_ctx->priority = priority;
920 lock_ctx->auto_mark = auto_mark;
922 lock_ctx->request = request;
923 lock_ctx->child = -1;
925 /* Non-record locks are required by recovery and should be scheduled
926 * immediately, so keep them at the head of the pending queue.
928 if (lock_ctx->type == LOCK_RECORD) {
929 DLIST_ADD_END(ctdb_db->lock_pending, lock_ctx, NULL);
930 } else {
931 DLIST_ADD_END(ctdb->lock_pending, lock_ctx, NULL);
933 CTDB_INCREMENT_STAT(ctdb, locks.num_pending);
934 if (ctdb_db) {
935 CTDB_INCREMENT_DB_STAT(ctdb_db, locks.num_pending);
938 /* Start the timer when we activate the context */
939 lock_ctx->start_time = timeval_current();
941 request->lctx = lock_ctx;
942 request->callback = callback;
943 request->private_data = private_data;
945 talloc_set_destructor(request, ctdb_lock_request_destructor);
946 talloc_set_destructor(lock_ctx, ctdb_lock_context_destructor);
948 ctdb_lock_schedule(ctdb);
950 return request;
955 * obtain a lock on a record in a database
957 struct lock_request *ctdb_lock_record(TALLOC_CTX *mem_ctx,
958 struct ctdb_db_context *ctdb_db,
959 TDB_DATA key,
960 bool auto_mark,
961 void (*callback)(void *, bool),
962 void *private_data)
964 return ctdb_lock_internal(mem_ctx,
965 ctdb_db->ctdb,
966 ctdb_db,
967 key,
969 callback,
970 private_data,
971 LOCK_RECORD,
972 auto_mark);
977 * obtain a lock on a database
979 struct lock_request *ctdb_lock_db(TALLOC_CTX *mem_ctx,
980 struct ctdb_db_context *ctdb_db,
981 bool auto_mark,
982 void (*callback)(void *, bool),
983 void *private_data)
985 return ctdb_lock_internal(mem_ctx,
986 ctdb_db->ctdb,
987 ctdb_db,
988 tdb_null,
990 callback,
991 private_data,
992 LOCK_DB,
993 auto_mark);
998 * obtain locks on all databases of specified priority
1000 struct lock_request *ctdb_lock_alldb_prio(TALLOC_CTX *mem_ctx,
1001 struct ctdb_context *ctdb,
1002 uint32_t priority,
1003 bool auto_mark,
1004 void (*callback)(void *, bool),
1005 void *private_data)
1007 if (priority < 1 || priority > NUM_DB_PRIORITIES) {
1008 DEBUG(DEBUG_ERR, ("Invalid db priority: %u\n", priority));
1009 return NULL;
1012 return ctdb_lock_internal(mem_ctx,
1013 ctdb,
1014 NULL,
1015 tdb_null,
1016 priority,
1017 callback,
1018 private_data,
1019 LOCK_ALLDB_PRIO,
1020 auto_mark);
1025 * obtain locks on all databases
1027 struct lock_request *ctdb_lock_alldb(TALLOC_CTX *mem_ctx,
1028 struct ctdb_context *ctdb,
1029 bool auto_mark,
1030 void (*callback)(void *, bool),
1031 void *private_data)
1033 return ctdb_lock_internal(mem_ctx,
1034 ctdb,
1035 NULL,
1036 tdb_null,
1038 callback,
1039 private_data,
1040 LOCK_ALLDB,
1041 auto_mark);