ctdb-locking: Implement active lock requests limit per database
[Samba/wip.git] / ctdb / server / ctdb_lock.c
blob23be5f52e4cbb93b0197058f4f446b6210bda689
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 "db_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 * diagnose using /proc/locks and log warning message
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
46 /* FIXME: Add a tunable max_lock_processes_per_db */
47 #define MAX_LOCK_PROCESSES_PER_DB (100)
49 enum lock_type {
50 LOCK_RECORD,
51 LOCK_DB,
52 LOCK_ALLDB_PRIO,
53 LOCK_ALLDB,
56 static const char * const lock_type_str[] = {
57 "lock_record",
58 "lock_db",
59 "lock_alldb_prio",
60 "lock_db",
63 struct lock_request;
65 /* lock_context is the common part for a lock request */
66 struct lock_context {
67 struct lock_context *next, *prev;
68 enum lock_type type;
69 struct ctdb_context *ctdb;
70 struct ctdb_db_context *ctdb_db;
71 TDB_DATA key;
72 uint32_t priority;
73 bool auto_mark;
74 struct lock_request *req_queue;
75 pid_t child;
76 int fd[2];
77 struct tevent_fd *tfd;
78 struct tevent_timer *ttimer;
79 pid_t block_child;
80 int block_fd[2];
81 struct timeval start_time;
84 /* lock_request is the client specific part for a lock request */
85 struct lock_request {
86 struct lock_request *next, *prev;
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 dameon 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->child > 0) {
280 ctdb_kill(lock_ctx->ctdb, lock_ctx->child, SIGKILL);
281 DLIST_REMOVE(lock_ctx->ctdb->lock_current, lock_ctx);
282 if (lock_ctx->ctdb_db) {
283 lock_ctx->ctdb_db->lock_num_current--;
285 CTDB_DECREMENT_STAT(lock_ctx->ctdb, locks.num_current);
286 if (lock_ctx->type == LOCK_RECORD || lock_ctx->type == LOCK_DB) {
287 CTDB_DECREMENT_DB_STAT(lock_ctx->ctdb_db, locks.num_current);
289 } else {
290 DLIST_REMOVE(lock_ctx->ctdb->lock_pending, lock_ctx);
291 lock_ctx->ctdb->lock_num_pending--;
292 CTDB_DECREMENT_STAT(lock_ctx->ctdb, locks.num_pending);
293 if (lock_ctx->type == LOCK_RECORD || lock_ctx->type == LOCK_DB) {
294 CTDB_DECREMENT_DB_STAT(lock_ctx->ctdb_db, locks.num_pending);
298 ctdb_lock_schedule(lock_ctx->ctdb);
300 return 0;
305 * Destructor to remove lock request
307 static int ctdb_lock_request_destructor(struct lock_request *lock_request)
309 DLIST_REMOVE(lock_request->lctx->req_queue, lock_request);
310 return 0;
314 void ctdb_lock_free_request_context(struct lock_request *lock_req)
316 struct lock_context *lock_ctx;
318 lock_ctx = lock_req->lctx;
319 talloc_free(lock_req);
320 talloc_free(lock_ctx);
325 * Process all the callbacks waiting for lock
327 * If lock has failed, callback is executed with locked=false
329 static void process_callbacks(struct lock_context *lock_ctx, bool locked)
331 struct lock_request *request, *next;
333 if (lock_ctx->auto_mark && locked) {
334 switch (lock_ctx->type) {
335 case LOCK_RECORD:
336 tdb_chainlock_mark(lock_ctx->ctdb_db->ltdb->tdb, lock_ctx->key);
337 break;
339 case LOCK_DB:
340 tdb_lockall_mark(lock_ctx->ctdb_db->ltdb->tdb);
341 break;
343 case LOCK_ALLDB_PRIO:
344 ctdb_lockall_mark_prio(lock_ctx->ctdb, lock_ctx->priority);
345 break;
347 case LOCK_ALLDB:
348 ctdb_lockall_mark(lock_ctx->ctdb);
349 break;
353 /* Iterate through all callbacks */
354 request = lock_ctx->req_queue;
355 while (request) {
356 if (lock_ctx->auto_mark) {
357 /* Reset the destructor, so request is not removed from the list */
358 talloc_set_destructor(request, NULL);
361 /* In case, callback frees the request, store next */
362 next = request->next;
363 request->callback(request->private_data, locked);
364 request = next;
367 if (lock_ctx->auto_mark && locked) {
368 switch (lock_ctx->type) {
369 case LOCK_RECORD:
370 tdb_chainlock_unmark(lock_ctx->ctdb_db->ltdb->tdb, lock_ctx->key);
371 break;
373 case LOCK_DB:
374 tdb_lockall_unmark(lock_ctx->ctdb_db->ltdb->tdb);
375 break;
377 case LOCK_ALLDB_PRIO:
378 ctdb_lockall_unmark_prio(lock_ctx->ctdb, lock_ctx->priority);
379 break;
381 case LOCK_ALLDB:
382 ctdb_lockall_unmark(lock_ctx->ctdb);
383 break;
389 static int lock_bucket_id(double t)
391 double ms = 1.e-3, s = 1;
392 int id;
394 if (t < 1*ms) {
395 id = 0;
396 } else if (t < 10*ms) {
397 id = 1;
398 } else if (t < 100*ms) {
399 id = 2;
400 } else if (t < 1*s) {
401 id = 3;
402 } else if (t < 2*s) {
403 id = 4;
404 } else if (t < 4*s) {
405 id = 5;
406 } else if (t < 8*s) {
407 id = 6;
408 } else if (t < 16*s) {
409 id = 7;
410 } else if (t < 32*s) {
411 id = 8;
412 } else if (t < 64*s) {
413 id = 9;
414 } else {
415 id = 10;
418 return id;
422 * Callback routine when the required locks are obtained.
423 * Called from parent context
425 static void ctdb_lock_handler(struct tevent_context *ev,
426 struct tevent_fd *tfd,
427 uint16_t flags,
428 void *private_data)
430 struct lock_context *lock_ctx;
431 TALLOC_CTX *tmp_ctx = NULL;
432 char c;
433 bool locked;
434 double t;
435 int id;
437 lock_ctx = talloc_get_type_abort(private_data, struct lock_context);
439 /* cancel the timeout event */
440 if (lock_ctx->ttimer) {
441 TALLOC_FREE(lock_ctx->ttimer);
444 t = timeval_elapsed(&lock_ctx->start_time);
445 id = lock_bucket_id(t);
447 if (lock_ctx->auto_mark) {
448 tmp_ctx = talloc_new(ev);
449 talloc_steal(tmp_ctx, lock_ctx);
452 /* Read the status from the child process */
453 if (read(lock_ctx->fd[0], &c, 1) != 1) {
454 locked = false;
455 } else {
456 locked = (c == 0 ? true : false);
459 /* Update statistics */
460 CTDB_DECREMENT_STAT(lock_ctx->ctdb, locks.num_pending);
461 CTDB_INCREMENT_STAT(lock_ctx->ctdb, locks.num_calls);
462 if (lock_ctx->ctdb_db) {
463 CTDB_DECREMENT_DB_STAT(lock_ctx->ctdb_db, locks.num_pending);
464 CTDB_INCREMENT_DB_STAT(lock_ctx->ctdb_db, locks.num_calls);
467 if (locked) {
468 if (lock_ctx->ctdb_db) {
469 CTDB_INCREMENT_STAT(lock_ctx->ctdb, locks.num_current);
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_INCREMENT_DB_STAT(lock_ctx->ctdb_db, locks.num_current);
476 CTDB_UPDATE_DB_LATENCY(lock_ctx->ctdb_db, lock_type_str[lock_ctx->type], locks.latency, t);
477 CTDB_INCREMENT_DB_STAT(lock_ctx->ctdb_db, locks.buckets[id]);
479 } else {
480 CTDB_INCREMENT_STAT(lock_ctx->ctdb, locks.num_failed);
481 if (lock_ctx->ctdb_db) {
482 CTDB_INCREMENT_DB_STAT(lock_ctx->ctdb_db, locks.num_failed);
486 process_callbacks(lock_ctx, locked);
488 if (lock_ctx->auto_mark) {
489 talloc_free(tmp_ctx);
495 * Callback routine when required locks are not obtained within timeout
496 * Called from parent context
498 static void ctdb_lock_timeout_handler(struct tevent_context *ev,
499 struct tevent_timer *ttimer,
500 struct timeval current_time,
501 void *private_data)
503 static const char * debug_locks = NULL;
504 struct lock_context *lock_ctx;
505 struct ctdb_context *ctdb;
506 pid_t pid;
508 lock_ctx = talloc_get_type_abort(private_data, struct lock_context);
509 ctdb = lock_ctx->ctdb;
511 if (lock_ctx->type == LOCK_RECORD || lock_ctx->type == LOCK_DB) {
512 DEBUG(DEBUG_WARNING,
513 ("Unable to get %s lock on database %s for %.0lf seconds\n",
514 (lock_ctx->type == LOCK_RECORD ? "RECORD" : "DB"),
515 lock_ctx->ctdb_db->db_name,
516 timeval_elapsed(&lock_ctx->start_time)));
517 } else {
518 DEBUG(DEBUG_WARNING,
519 ("Unable to get ALLDB locks for %.0lf seconds\n",
520 timeval_elapsed(&lock_ctx->start_time)));
523 /* Fire a child process to find the blocking process. */
524 if (debug_locks == NULL) {
525 debug_locks = getenv("CTDB_DEBUG_LOCKS");
526 if (debug_locks == NULL) {
527 debug_locks = talloc_asprintf(ctdb,
528 "%s/debug_locks.sh",
529 getenv("CTDB_BASE"));
532 if (debug_locks != NULL) {
533 pid = fork();
534 if (pid == 0) {
535 execl(debug_locks, debug_locks, NULL);
537 } else {
538 DEBUG(DEBUG_WARNING,
539 (__location__
540 " Unable to setup lock debugging - no memory?\n"));
543 /* reset the timeout timer */
544 // talloc_free(lock_ctx->ttimer);
545 lock_ctx->ttimer = tevent_add_timer(ctdb->ev,
546 lock_ctx,
547 timeval_current_ofs(10, 0),
548 ctdb_lock_timeout_handler,
549 (void *)lock_ctx);
553 static int db_count_handler(struct ctdb_db_context *ctdb_db, uint32_t priority,
554 void *private_data)
556 int *count = (int *)private_data;
558 (*count)++;
560 return 0;
563 struct db_namelist {
564 char **names;
565 int n;
568 static int db_name_handler(struct ctdb_db_context *ctdb_db, uint32_t priority,
569 void *private_data)
571 struct db_namelist *list = (struct db_namelist *)private_data;
573 list->names[list->n] = talloc_strdup(list->names, ctdb_db->db_path);
574 list->n++;
576 return 0;
579 static char **lock_helper_args(TALLOC_CTX *mem_ctx, struct lock_context *lock_ctx, int fd)
581 struct ctdb_context *ctdb = lock_ctx->ctdb;
582 char **args = NULL;
583 int nargs, i;
584 int priority;
585 struct db_namelist list;
587 switch (lock_ctx->type) {
588 case LOCK_RECORD:
589 nargs = 6;
590 break;
592 case LOCK_DB:
593 nargs = 5;
594 break;
596 case LOCK_ALLDB_PRIO:
597 nargs = 4;
598 ctdb_db_iterator(ctdb, lock_ctx->priority, db_count_handler, &nargs);
599 break;
601 case LOCK_ALLDB:
602 nargs = 4;
603 for (priority=1; priority<NUM_DB_PRIORITIES; priority++) {
604 ctdb_db_iterator(ctdb, priority, db_count_handler, &nargs);
606 break;
609 /* Add extra argument for null termination */
610 nargs++;
612 args = talloc_array(mem_ctx, char *, nargs);
613 if (args == NULL) {
614 return NULL;
617 args[0] = talloc_strdup(args, "ctdb_lock_helper");
618 args[1] = talloc_asprintf(args, "%d", getpid());
619 args[2] = talloc_asprintf(args, "%d", fd);
621 switch (lock_ctx->type) {
622 case LOCK_RECORD:
623 args[3] = talloc_strdup(args, "RECORD");
624 args[4] = talloc_strdup(args, lock_ctx->ctdb_db->db_path);
625 if (lock_ctx->key.dsize == 0) {
626 args[5] = talloc_strdup(args, "NULL");
627 } else {
628 args[5] = hex_encode_talloc(args, lock_ctx->key.dptr, lock_ctx->key.dsize);
630 break;
632 case LOCK_DB:
633 args[3] = talloc_strdup(args, "DB");
634 args[4] = talloc_strdup(args, lock_ctx->ctdb_db->db_path);
635 break;
637 case LOCK_ALLDB_PRIO:
638 args[3] = talloc_strdup(args, "DB");
639 list.names = args;
640 list.n = 4;
641 ctdb_db_iterator(ctdb, lock_ctx->priority, db_name_handler, &list);
642 break;
644 case LOCK_ALLDB:
645 args[3] = talloc_strdup(args, "DB");
646 list.names = args;
647 list.n = 4;
648 for (priority=1; priority<NUM_DB_PRIORITIES; priority++) {
649 ctdb_db_iterator(ctdb, priority, db_name_handler, &list);
651 break;
654 /* Make sure last argument is NULL */
655 args[nargs-1] = NULL;
657 for (i=0; i<nargs-1; i++) {
658 if (args[i] == NULL) {
659 talloc_free(args);
660 return NULL;
664 return args;
669 * Find the lock context of a given type
671 static struct lock_context *find_lock_context(struct lock_context *lock_list,
672 struct ctdb_db_context *ctdb_db,
673 TDB_DATA key,
674 uint32_t priority,
675 enum lock_type type)
677 struct lock_context *lock_ctx;
679 /* Search active locks */
680 for (lock_ctx=lock_list; lock_ctx; lock_ctx=lock_ctx->next) {
681 if (lock_ctx->type != type) {
682 continue;
685 switch (lock_ctx->type) {
686 case LOCK_RECORD:
687 if (ctdb_db == lock_ctx->ctdb_db &&
688 key.dsize == lock_ctx->key.dsize &&
689 memcmp(key.dptr, lock_ctx->key.dptr, key.dsize) == 0) {
690 goto done;
692 break;
694 case LOCK_DB:
695 if (ctdb_db == lock_ctx->ctdb_db) {
696 goto done;
698 break;
700 case LOCK_ALLDB_PRIO:
701 if (priority == lock_ctx->priority) {
702 goto done;
704 break;
706 case LOCK_ALLDB:
707 goto done;
708 break;
712 /* Did not find the lock context we are searching for */
713 lock_ctx = NULL;
715 done:
716 return lock_ctx;
722 * Schedule a new lock child process
723 * Set up callback handler and timeout handler
725 static void ctdb_lock_schedule(struct ctdb_context *ctdb)
727 struct lock_context *lock_ctx, *next_ctx, *active_ctx;
728 int ret;
729 TALLOC_CTX *tmp_ctx;
730 const char *helper = BINDIR "/ctdb_lock_helper";
731 static const char *prog = NULL;
732 char **args;
734 if (prog == NULL) {
735 const char *t;
737 t = getenv("CTDB_LOCK_HELPER");
738 if (t != NULL) {
739 prog = talloc_strdup(ctdb, t);
740 } else {
741 prog = talloc_strdup(ctdb, helper);
743 CTDB_NO_MEMORY_VOID(ctdb, prog);
746 if (ctdb->lock_pending == NULL) {
747 return;
750 /* Find a lock context with requests */
751 lock_ctx = ctdb->lock_pending;
752 while (lock_ctx != NULL) {
753 next_ctx = lock_ctx->next;
754 if (! lock_ctx->req_queue) {
755 DEBUG(DEBUG_INFO, ("Removing lock context without lock requests\n"));
756 DLIST_REMOVE(ctdb->lock_pending, lock_ctx);
757 ctdb->lock_num_pending--;
758 CTDB_DECREMENT_STAT(ctdb, locks.num_pending);
759 if (lock_ctx->ctdb_db) {
760 CTDB_DECREMENT_DB_STAT(lock_ctx->ctdb_db, locks.num_pending);
762 talloc_free(lock_ctx);
763 } else {
764 active_ctx = find_lock_context(ctdb->lock_current, lock_ctx->ctdb_db,
765 lock_ctx->key, lock_ctx->priority,
766 lock_ctx->type);
767 if (active_ctx == NULL) {
768 if (lock_ctx->ctdb_db == NULL ||
769 lock_ctx->ctdb_db->lock_num_current < MAX_LOCK_PROCESSES_PER_DB) {
770 /* Found a lock context with lock requests */
771 break;
775 /* There is already a child waiting for the
776 * same key. So don't schedule another child
777 * just yet.
780 lock_ctx = next_ctx;
783 if (lock_ctx == NULL) {
784 return;
787 lock_ctx->child = -1;
788 ret = pipe(lock_ctx->fd);
789 if (ret != 0) {
790 DEBUG(DEBUG_ERR, ("Failed to create pipe in ctdb_lock_schedule\n"));
791 return;
794 set_close_on_exec(lock_ctx->fd[0]);
796 /* Create data for child process */
797 tmp_ctx = talloc_new(lock_ctx);
798 if (tmp_ctx == NULL) {
799 DEBUG(DEBUG_ERR, ("Failed to allocate memory for helper args\n"));
800 close(lock_ctx->fd[0]);
801 close(lock_ctx->fd[1]);
802 return;
805 /* Create arguments for lock helper */
806 args = lock_helper_args(tmp_ctx, lock_ctx, lock_ctx->fd[1]);
807 if (args == NULL) {
808 DEBUG(DEBUG_ERR, ("Failed to create lock helper args\n"));
809 close(lock_ctx->fd[0]);
810 close(lock_ctx->fd[1]);
811 talloc_free(tmp_ctx);
812 return;
815 lock_ctx->child = ctdb_fork(ctdb);
817 if (lock_ctx->child == (pid_t)-1) {
818 DEBUG(DEBUG_ERR, ("Failed to create a child in ctdb_lock_schedule\n"));
819 close(lock_ctx->fd[0]);
820 close(lock_ctx->fd[1]);
821 talloc_free(tmp_ctx);
822 return;
826 /* Child process */
827 if (lock_ctx->child == 0) {
828 ret = execv(prog, args);
829 if (ret < 0) {
830 DEBUG(DEBUG_ERR, ("Failed to execute helper %s (%d, %s)\n",
831 prog, errno, strerror(errno)));
833 _exit(1);
836 /* Parent process */
837 close(lock_ctx->fd[1]);
839 talloc_set_destructor(lock_ctx, ctdb_lock_context_destructor);
841 talloc_free(tmp_ctx);
843 /* Set up timeout handler */
844 lock_ctx->ttimer = tevent_add_timer(ctdb->ev,
845 lock_ctx,
846 timeval_current_ofs(10, 0),
847 ctdb_lock_timeout_handler,
848 (void *)lock_ctx);
849 if (lock_ctx->ttimer == NULL) {
850 ctdb_kill(ctdb, lock_ctx->child, SIGKILL);
851 lock_ctx->child = -1;
852 talloc_set_destructor(lock_ctx, NULL);
853 close(lock_ctx->fd[0]);
854 return;
857 /* Set up callback */
858 lock_ctx->tfd = tevent_add_fd(ctdb->ev,
859 lock_ctx,
860 lock_ctx->fd[0],
861 EVENT_FD_READ,
862 ctdb_lock_handler,
863 (void *)lock_ctx);
864 if (lock_ctx->tfd == NULL) {
865 TALLOC_FREE(lock_ctx->ttimer);
866 ctdb_kill(ctdb, lock_ctx->child, SIGKILL);
867 lock_ctx->child = -1;
868 talloc_set_destructor(lock_ctx, NULL);
869 close(lock_ctx->fd[0]);
870 return;
872 tevent_fd_set_auto_close(lock_ctx->tfd);
874 /* Move the context from pending to current */
875 DLIST_REMOVE(ctdb->lock_pending, lock_ctx);
876 ctdb->lock_num_pending--;
877 DLIST_ADD_END(ctdb->lock_current, lock_ctx, NULL);
878 if (lock_ctx->ctdb_db) {
879 lock_ctx->ctdb_db->lock_num_current++;
885 * Lock record / db depending on type
887 static struct lock_request *ctdb_lock_internal(struct ctdb_context *ctdb,
888 struct ctdb_db_context *ctdb_db,
889 TDB_DATA key,
890 uint32_t priority,
891 void (*callback)(void *, bool),
892 void *private_data,
893 enum lock_type type,
894 bool auto_mark)
896 struct lock_context *lock_ctx;
897 struct lock_request *request;
899 if (callback == NULL) {
900 DEBUG(DEBUG_WARNING, ("No callback function specified, not locking\n"));
901 return NULL;
904 /* get a context for this key - search only the pending contexts,
905 * current contexts might in the middle of processing callbacks */
906 lock_ctx = find_lock_context(ctdb->lock_pending, ctdb_db, key, priority, type);
908 /* No existing context, create one */
909 if (lock_ctx == NULL) {
910 lock_ctx = talloc_zero(ctdb, struct lock_context);
911 if (lock_ctx == NULL) {
912 DEBUG(DEBUG_ERR, ("Failed to create a new lock context\n"));
913 return NULL;
916 lock_ctx->type = type;
917 lock_ctx->ctdb = ctdb;
918 lock_ctx->ctdb_db = ctdb_db;
919 lock_ctx->key.dsize = key.dsize;
920 if (key.dsize > 0) {
921 lock_ctx->key.dptr = talloc_memdup(lock_ctx, key.dptr, key.dsize);
922 } else {
923 lock_ctx->key.dptr = NULL;
925 lock_ctx->priority = priority;
926 lock_ctx->auto_mark = auto_mark;
928 lock_ctx->child = -1;
929 lock_ctx->block_child = -1;
931 DLIST_ADD_END(ctdb->lock_pending, lock_ctx, NULL);
932 ctdb->lock_num_pending++;
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();
942 if ((request = talloc_zero(lock_ctx, struct lock_request)) == NULL) {
943 return NULL;
946 request->lctx = lock_ctx;
947 request->callback = callback;
948 request->private_data = private_data;
950 talloc_set_destructor(request, ctdb_lock_request_destructor);
951 DLIST_ADD_END(lock_ctx->req_queue, request, NULL);
953 ctdb_lock_schedule(ctdb);
955 return request;
960 * obtain a lock on a record in a database
962 struct lock_request *ctdb_lock_record(struct ctdb_db_context *ctdb_db,
963 TDB_DATA key,
964 bool auto_mark,
965 void (*callback)(void *, bool),
966 void *private_data)
968 return ctdb_lock_internal(ctdb_db->ctdb,
969 ctdb_db,
970 key,
972 callback,
973 private_data,
974 LOCK_RECORD,
975 auto_mark);
980 * obtain a lock on a database
982 struct lock_request *ctdb_lock_db(struct ctdb_db_context *ctdb_db,
983 bool auto_mark,
984 void (*callback)(void *, bool),
985 void *private_data)
987 return ctdb_lock_internal(ctdb_db->ctdb,
988 ctdb_db,
989 tdb_null,
991 callback,
992 private_data,
993 LOCK_DB,
994 auto_mark);
999 * obtain locks on all databases of specified priority
1001 struct lock_request *ctdb_lock_alldb_prio(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(ctdb,
1013 NULL,
1014 tdb_null,
1015 priority,
1016 callback,
1017 private_data,
1018 LOCK_ALLDB_PRIO,
1019 auto_mark);
1024 * obtain locks on all databases
1026 struct lock_request *ctdb_lock_alldb(struct ctdb_context *ctdb,
1027 bool auto_mark,
1028 void (*callback)(void *, bool),
1029 void *private_data)
1031 return ctdb_lock_internal(ctdb,
1032 NULL,
1033 tdb_null,
1035 callback,
1036 private_data,
1037 LOCK_ALLDB,
1038 auto_mark);