2 ctdb ltdb code - server side
4 Copyright (C) Andrew Tridgell 2007
6 This program is free software; you can redistribute it and/or modify
7 it under the terms of the GNU General Public License as published by
8 the Free Software Foundation; either version 3 of the License, or
9 (at your option) any later version.
11 This program is distributed in the hope that it will be useful,
12 but WITHOUT ANY WARRANTY; without even the implied warranty of
13 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 GNU General Public License for more details.
16 You should have received a copy of the GNU General Public License
17 along with this program; if not, see <http://www.gnu.org/licenses/>.
22 #include "system/network.h"
23 #include "system/filesys.h"
24 #include "system/dir.h"
25 #include "system/time.h"
26 #include "../include/ctdb_private.h"
27 #include "../common/rb_tree.h"
29 #include "lib/util/dlinklist.h"
32 #define PERSISTENT_HEALTH_TDB "persistent_health.tdb"
35 * write a record to a normal database
37 * This is the server-variant of the ctdb_ltdb_store function.
38 * It contains logic to determine whether a record should be
39 * stored or deleted. It also sends SCHEDULE_FOR_DELETION
40 * controls to the local ctdb daemon if apporpriate.
42 static int ctdb_ltdb_store_server(struct ctdb_db_context *ctdb_db,
44 struct ctdb_ltdb_header *header,
47 struct ctdb_context *ctdb = ctdb_db->ctdb;
50 bool seqnum_suppressed = false;
52 bool schedule_for_deletion = false;
53 bool remove_from_delete_queue = false;
56 if (ctdb->flags & CTDB_FLAG_TORTURE) {
57 struct ctdb_ltdb_header *h2;
58 rec = tdb_fetch(ctdb_db->ltdb->tdb, key);
59 h2 = (struct ctdb_ltdb_header *)rec.dptr;
60 if (rec.dptr && rec.dsize >= sizeof(h2) && h2->rsn > header->rsn) {
61 DEBUG(DEBUG_CRIT,("RSN regression! %llu %llu\n",
62 (unsigned long long)h2->rsn, (unsigned long long)header->rsn));
64 if (rec.dptr) free(rec.dptr);
67 if (ctdb->vnn_map == NULL) {
69 * Called from a client: always store the record
70 * Also don't call ctdb_lmaster since it uses the vnn_map!
76 lmaster = ctdb_lmaster(ctdb_db->ctdb, &key);
79 * If we migrate an empty record off to another node
80 * and the record has not been migrated with data,
81 * delete the record instead of storing the empty record.
83 if (data.dsize != 0) {
85 } else if (header->flags & CTDB_REC_RO_FLAGS) {
87 } else if (ctdb_db->persistent) {
89 } else if (header->flags & CTDB_REC_FLAG_AUTOMATIC) {
91 * The record is not created by the client but
92 * automatically by the ctdb_ltdb_fetch logic that
93 * creates a record with an initial header in the
94 * ltdb before trying to migrate the record from
95 * the current lmaster. Keep it instead of trying
96 * to delete the non-existing record...
99 schedule_for_deletion = true;
100 } else if (header->flags & CTDB_REC_FLAG_MIGRATED_WITH_DATA) {
102 } else if (ctdb_db->ctdb->pnn == lmaster) {
104 * If we are lmaster, then we usually keep the record.
105 * But if we retrieve the dmaster role by a VACUUM_MIGRATE
106 * and the record is empty and has never been migrated
107 * with data, then we should delete it instead of storing it.
108 * This is part of the vacuuming process.
110 * The reason that we usually need to store even empty records
111 * on the lmaster is that a client operating directly on the
112 * lmaster (== dmaster) expects the local copy of the record to
113 * exist after successful ctdb migrate call. If the record does
114 * not exist, the client goes into a migrate loop and eventually
115 * fails. So storing the empty record makes sure that we do not
116 * need to change the client code.
118 if (!(header->flags & CTDB_REC_FLAG_VACUUM_MIGRATED)) {
120 } else if (ctdb_db->ctdb->pnn != header->dmaster) {
123 } else if (ctdb_db->ctdb->pnn == header->dmaster) {
128 if (!ctdb_db->persistent &&
129 (ctdb_db->ctdb->pnn == header->dmaster) &&
130 !(header->flags & CTDB_REC_RO_FLAGS))
134 if (data.dsize == 0) {
135 schedule_for_deletion = true;
138 remove_from_delete_queue = !schedule_for_deletion;
143 * The VACUUM_MIGRATED flag is only set temporarily for
144 * the above logic when the record was retrieved by a
145 * VACUUM_MIGRATE call and should not be stored in the
148 * The VACUUM_MIGRATE call is triggered by a vacuum fetch,
149 * and there are two cases in which the corresponding record
150 * is stored in the local database:
151 * 1. The record has been migrated with data in the past
152 * (the MIGRATED_WITH_DATA record flag is set).
153 * 2. The record has been filled with data again since it
154 * had been submitted in the VACUUM_FETCH message to the
156 * For such records it is important to not store the
157 * VACUUM_MIGRATED flag in the database.
159 header->flags &= ~CTDB_REC_FLAG_VACUUM_MIGRATED;
162 * Similarly, clear the AUTOMATIC flag which should not enter
163 * the local database copy since this would require client
164 * modifications to clear the flag when the client stores
167 header->flags &= ~CTDB_REC_FLAG_AUTOMATIC;
169 rec.dsize = sizeof(*header) + data.dsize;
170 rec.dptr = talloc_size(ctdb, rec.dsize);
171 CTDB_NO_MEMORY(ctdb, rec.dptr);
173 memcpy(rec.dptr, header, sizeof(*header));
174 memcpy(rec.dptr + sizeof(*header), data.dptr, data.dsize);
176 /* Databases with seqnum updates enabled only get their seqnum
177 changes when/if we modify the data */
178 if (ctdb_db->seqnum_update != NULL) {
180 old = tdb_fetch(ctdb_db->ltdb->tdb, key);
182 if ( (old.dsize == rec.dsize)
183 && !memcmp(old.dptr+sizeof(struct ctdb_ltdb_header),
184 rec.dptr+sizeof(struct ctdb_ltdb_header),
185 rec.dsize-sizeof(struct ctdb_ltdb_header)) ) {
186 tdb_remove_flags(ctdb_db->ltdb->tdb, TDB_SEQNUM);
187 seqnum_suppressed = true;
189 if (old.dptr) free(old.dptr);
192 DEBUG(DEBUG_DEBUG, (__location__ " db[%s]: %s record: hash[0x%08x]\n",
194 keep?"storing":"deleting",
198 ret = tdb_store(ctdb_db->ltdb->tdb, key, rec, TDB_REPLACE);
200 ret = tdb_delete(ctdb_db->ltdb->tdb, key);
207 tdb_error(ctdb_db->ltdb->tdb) == TDB_ERR_NOEXIST)
212 DEBUG(lvl, (__location__ " db[%s]: Failed to %s record: "
215 keep?"store":"delete", ret,
216 tdb_errorstr(ctdb_db->ltdb->tdb)));
218 schedule_for_deletion = false;
219 remove_from_delete_queue = false;
221 if (seqnum_suppressed) {
222 tdb_add_flags(ctdb_db->ltdb->tdb, TDB_SEQNUM);
225 talloc_free(rec.dptr);
227 if (schedule_for_deletion) {
229 ret2 = ctdb_local_schedule_for_deletion(ctdb_db, header, key);
231 DEBUG(DEBUG_ERR, (__location__ " ctdb_local_schedule_for_deletion failed.\n"));
235 if (remove_from_delete_queue) {
236 ctdb_local_remove_from_delete_queue(ctdb_db, header, key);
242 struct lock_fetch_state {
243 struct ctdb_context *ctdb;
244 void (*recv_pkt)(void *, struct ctdb_req_header *);
246 struct ctdb_req_header *hdr;
248 bool ignore_generation;
252 called when we should retry the operation
254 static void lock_fetch_callback(void *p, bool locked)
256 struct lock_fetch_state *state = talloc_get_type(p, struct lock_fetch_state);
257 if (!state->ignore_generation &&
258 state->generation != state->ctdb->vnn_map->generation) {
259 DEBUG(DEBUG_NOTICE,("Discarding previous generation lockwait packet\n"));
260 talloc_free(state->hdr);
263 state->recv_pkt(state->recv_context, state->hdr);
264 DEBUG(DEBUG_INFO,(__location__ " PACKET REQUEUED\n"));
269 do a non-blocking ltdb_lock, deferring this ctdb request until we
272 It does the following:
274 1) tries to get the chainlock. If it succeeds, then it returns 0
276 2) if it fails to get a chainlock immediately then it sets up a
277 non-blocking chainlock via ctdb_lock_record, and when it gets the
278 chainlock it re-submits this ctdb request to the main packet
281 This effectively queues all ctdb requests that cannot be
282 immediately satisfied until it can get the lock. This means that
283 the main ctdb daemon will not block waiting for a chainlock held by
286 There are 3 possible return values:
288 0: means that it got the lock immediately.
289 -1: means that it failed to get the lock, and won't retry
290 -2: means that it failed to get the lock immediately, but will retry
292 int ctdb_ltdb_lock_requeue(struct ctdb_db_context *ctdb_db,
293 TDB_DATA key, struct ctdb_req_header *hdr,
294 void (*recv_pkt)(void *, struct ctdb_req_header *),
295 void *recv_context, bool ignore_generation)
298 struct tdb_context *tdb = ctdb_db->ltdb->tdb;
299 struct lock_request *lreq;
300 struct lock_fetch_state *state;
302 ret = tdb_chainlock_nonblock(tdb, key);
305 !(errno == EACCES || errno == EAGAIN || errno == EDEADLK)) {
306 /* a hard failure - don't try again */
310 /* when torturing, ensure we test the contended path */
311 if ((ctdb_db->ctdb->flags & CTDB_FLAG_TORTURE) &&
314 tdb_chainunlock(tdb, key);
317 /* first the non-contended path */
322 state = talloc(hdr, struct lock_fetch_state);
323 state->ctdb = ctdb_db->ctdb;
325 state->recv_pkt = recv_pkt;
326 state->recv_context = recv_context;
327 state->generation = ctdb_db->ctdb->vnn_map->generation;
328 state->ignore_generation = ignore_generation;
330 /* now the contended path */
331 lreq = ctdb_lock_record(ctdb_db, key, true, lock_fetch_callback, state);
336 /* we need to move the packet off the temporary context in ctdb_input_pkt(),
337 so it won't be freed yet */
338 talloc_steal(state, hdr);
340 /* now tell the caller than we will retry asynchronously */
345 a varient of ctdb_ltdb_lock_requeue that also fetches the record
347 int ctdb_ltdb_lock_fetch_requeue(struct ctdb_db_context *ctdb_db,
348 TDB_DATA key, struct ctdb_ltdb_header *header,
349 struct ctdb_req_header *hdr, TDB_DATA *data,
350 void (*recv_pkt)(void *, struct ctdb_req_header *),
351 void *recv_context, bool ignore_generation)
355 ret = ctdb_ltdb_lock_requeue(ctdb_db, key, hdr, recv_pkt,
356 recv_context, ignore_generation);
358 ret = ctdb_ltdb_fetch(ctdb_db, key, header, hdr, data);
361 uret = ctdb_ltdb_unlock(ctdb_db, key);
363 DEBUG(DEBUG_ERR,(__location__ " ctdb_ltdb_unlock() failed with error %d\n", uret));
372 paraoid check to see if the db is empty
374 static void ctdb_check_db_empty(struct ctdb_db_context *ctdb_db)
376 struct tdb_context *tdb = ctdb_db->ltdb->tdb;
377 int count = tdb_traverse_read(tdb, NULL, NULL);
379 DEBUG(DEBUG_ALERT,(__location__ " tdb '%s' not empty on attach! aborting\n",
381 ctdb_fatal(ctdb_db->ctdb, "database not empty on attach");
385 int ctdb_load_persistent_health(struct ctdb_context *ctdb,
386 struct ctdb_db_context *ctdb_db)
388 struct tdb_context *tdb = ctdb->db_persistent_health->tdb;
394 key.dptr = discard_const_p(uint8_t, ctdb_db->db_name);
395 key.dsize = strlen(ctdb_db->db_name);
397 old = ctdb_db->unhealthy_reason;
398 ctdb_db->unhealthy_reason = NULL;
400 val = tdb_fetch(tdb, key);
402 reason = talloc_strndup(ctdb_db,
403 (const char *)val.dptr,
405 if (reason == NULL) {
406 DEBUG(DEBUG_ALERT,(__location__ " talloc_strndup(%d) failed\n",
408 ctdb_db->unhealthy_reason = old;
419 ctdb_db->unhealthy_reason = reason;
423 int ctdb_update_persistent_health(struct ctdb_context *ctdb,
424 struct ctdb_db_context *ctdb_db,
425 const char *given_reason,/* NULL means healthy */
426 int num_healthy_nodes)
428 struct tdb_context *tdb = ctdb->db_persistent_health->tdb;
432 char *new_reason = NULL;
433 char *old_reason = NULL;
435 ret = tdb_transaction_start(tdb);
437 DEBUG(DEBUG_ALERT,(__location__ " tdb_transaction_start('%s') failed: %d - %s\n",
438 tdb_name(tdb), ret, tdb_errorstr(tdb)));
442 ret = ctdb_load_persistent_health(ctdb, ctdb_db);
444 DEBUG(DEBUG_ALERT,(__location__ " ctdb_load_persistent_health('%s') failed: %d\n",
445 ctdb_db->db_name, ret));
448 old_reason = ctdb_db->unhealthy_reason;
450 key.dptr = discard_const_p(uint8_t, ctdb_db->db_name);
451 key.dsize = strlen(ctdb_db->db_name);
454 new_reason = talloc_strdup(ctdb_db, given_reason);
455 if (new_reason == NULL) {
456 DEBUG(DEBUG_ALERT,(__location__ " talloc_strdup(%s) failed\n",
460 } else if (old_reason && num_healthy_nodes == 0) {
462 * If the reason indicates ok, but there where no healthy nodes
463 * available, that it means, we have not recovered valid content
464 * of the db. So if there's an old reason, prefix it with
465 * "NO-HEALTHY-NODES - "
469 #define _TMP_PREFIX "NO-HEALTHY-NODES - "
470 ret = strncmp(_TMP_PREFIX, old_reason, strlen(_TMP_PREFIX));
472 prefix = _TMP_PREFIX;
476 new_reason = talloc_asprintf(ctdb_db, "%s%s",
478 if (new_reason == NULL) {
479 DEBUG(DEBUG_ALERT,(__location__ " talloc_asprintf(%s%s) failed\n",
480 prefix, old_reason));
487 val.dptr = discard_const_p(uint8_t, new_reason);
488 val.dsize = strlen(new_reason);
490 ret = tdb_store(tdb, key, val, TDB_REPLACE);
492 tdb_transaction_cancel(tdb);
493 DEBUG(DEBUG_ALERT,(__location__ " tdb_store('%s', %s, %s) failed: %d - %s\n",
494 tdb_name(tdb), ctdb_db->db_name, new_reason,
495 ret, tdb_errorstr(tdb)));
496 talloc_free(new_reason);
499 DEBUG(DEBUG_ALERT,("Updated db health for db(%s) to: %s\n",
500 ctdb_db->db_name, new_reason));
501 } else if (old_reason) {
502 ret = tdb_delete(tdb, key);
504 tdb_transaction_cancel(tdb);
505 DEBUG(DEBUG_ALERT,(__location__ " tdb_delete('%s', %s) failed: %d - %s\n",
506 tdb_name(tdb), ctdb_db->db_name,
507 ret, tdb_errorstr(tdb)));
508 talloc_free(new_reason);
511 DEBUG(DEBUG_NOTICE,("Updated db health for db(%s): OK\n",
515 ret = tdb_transaction_commit(tdb);
516 if (ret != TDB_SUCCESS) {
517 DEBUG(DEBUG_ALERT,(__location__ " tdb_transaction_commit('%s') failed: %d - %s\n",
518 tdb_name(tdb), ret, tdb_errorstr(tdb)));
519 talloc_free(new_reason);
523 talloc_free(old_reason);
524 ctdb_db->unhealthy_reason = new_reason;
529 static int ctdb_backup_corrupted_tdb(struct ctdb_context *ctdb,
530 struct ctdb_db_context *ctdb_db)
532 time_t now = time(NULL);
540 /* formatted like: foo.tdb.0.corrupted.20091204160825.0Z */
541 new_path = talloc_asprintf(ctdb_db, "%s.corrupted."
542 "%04u%02u%02u%02u%02u%02u.0Z",
544 tm->tm_year+1900, tm->tm_mon+1,
545 tm->tm_mday, tm->tm_hour, tm->tm_min,
547 if (new_path == NULL) {
548 DEBUG(DEBUG_CRIT,(__location__ " talloc_asprintf() failed\n"));
552 new_reason = talloc_asprintf(ctdb_db,
553 "ERROR - Backup of corrupted TDB in '%s'",
555 if (new_reason == NULL) {
556 DEBUG(DEBUG_CRIT,(__location__ " talloc_asprintf() failed\n"));
559 ret = ctdb_update_persistent_health(ctdb, ctdb_db, new_reason, 0);
560 talloc_free(new_reason);
562 DEBUG(DEBUG_CRIT,(__location__
563 ": ctdb_backup_corrupted_tdb(%s) not implemented yet\n",
568 ret = rename(ctdb_db->db_path, new_path);
570 DEBUG(DEBUG_CRIT,(__location__
571 ": ctdb_backup_corrupted_tdb(%s) rename to %s failed: %d - %s\n",
572 ctdb_db->db_path, new_path,
573 errno, strerror(errno)));
574 talloc_free(new_path);
578 DEBUG(DEBUG_CRIT,(__location__
579 ": ctdb_backup_corrupted_tdb(%s) renamed to %s\n",
580 ctdb_db->db_path, new_path));
581 talloc_free(new_path);
585 int ctdb_recheck_persistent_health(struct ctdb_context *ctdb)
587 struct ctdb_db_context *ctdb_db;
592 for (ctdb_db = ctdb->db_list; ctdb_db; ctdb_db = ctdb_db->next) {
593 if (!ctdb_db->persistent) {
597 ret = ctdb_load_persistent_health(ctdb, ctdb_db);
599 DEBUG(DEBUG_ALERT,(__location__
600 " load persistent health for '%s' failed\n",
605 if (ctdb_db->unhealthy_reason == NULL) {
607 DEBUG(DEBUG_INFO,(__location__
608 " persistent db '%s' healthy\n",
614 DEBUG(DEBUG_ALERT,(__location__
615 " persistent db '%s' unhealthy: %s\n",
617 ctdb_db->unhealthy_reason));
619 DEBUG((fail!=0)?DEBUG_ALERT:DEBUG_NOTICE,
620 ("ctdb_recheck_presistent_health: OK[%d] FAIL[%d]\n",
632 mark a database - as healthy
634 int32_t ctdb_control_db_set_healthy(struct ctdb_context *ctdb, TDB_DATA indata)
636 uint32_t db_id = *(uint32_t *)indata.dptr;
637 struct ctdb_db_context *ctdb_db;
639 bool may_recover = false;
641 ctdb_db = find_ctdb_db(ctdb, db_id);
643 DEBUG(DEBUG_ERR,(__location__ " Unknown db 0x%x\n", db_id));
647 if (ctdb_db->unhealthy_reason) {
651 ret = ctdb_update_persistent_health(ctdb, ctdb_db, NULL, 1);
653 DEBUG(DEBUG_ERR,(__location__
654 " ctdb_update_persistent_health(%s) failed\n",
659 if (may_recover && ctdb->runstate == CTDB_RUNSTATE_STARTUP) {
660 DEBUG(DEBUG_ERR, (__location__ " db %s become healthy - force recovery for startup\n",
662 ctdb->recovery_mode = CTDB_RECOVERY_ACTIVE;
668 int32_t ctdb_control_db_get_health(struct ctdb_context *ctdb,
672 uint32_t db_id = *(uint32_t *)indata.dptr;
673 struct ctdb_db_context *ctdb_db;
676 ctdb_db = find_ctdb_db(ctdb, db_id);
678 DEBUG(DEBUG_ERR,(__location__ " Unknown db 0x%x\n", db_id));
682 ret = ctdb_load_persistent_health(ctdb, ctdb_db);
684 DEBUG(DEBUG_ERR,(__location__
685 " ctdb_load_persistent_health(%s) failed\n",
691 if (ctdb_db->unhealthy_reason) {
692 outdata->dptr = (uint8_t *)ctdb_db->unhealthy_reason;
693 outdata->dsize = strlen(ctdb_db->unhealthy_reason)+1;
700 int ctdb_set_db_readonly(struct ctdb_context *ctdb, struct ctdb_db_context *ctdb_db)
704 if (ctdb_db->readonly) {
708 if (ctdb_db->persistent) {
709 DEBUG(DEBUG_ERR,("Persistent databases do not support readonly property\n"));
713 ropath = talloc_asprintf(ctdb_db, "%s.RO", ctdb_db->db_path);
714 if (ropath == NULL) {
715 DEBUG(DEBUG_CRIT,("Failed to asprintf the tracking database\n"));
718 ctdb_db->rottdb = tdb_open(ropath,
719 ctdb->tunable.database_hash_size,
720 TDB_NOLOCK|TDB_CLEAR_IF_FIRST|TDB_NOSYNC,
722 if (ctdb_db->rottdb == NULL) {
723 DEBUG(DEBUG_CRIT,("Failed to open/create the tracking database '%s'\n", ropath));
728 DEBUG(DEBUG_NOTICE,("OPENED tracking database : '%s'\n", ropath));
730 ctdb_db->readonly = true;
732 DEBUG(DEBUG_NOTICE, ("Readonly property set on DB %s\n", ctdb_db->db_name));
739 attach to a database, handling both persistent and non-persistent databases
740 return 0 on success, -1 on failure
742 static int ctdb_local_attach(struct ctdb_context *ctdb, const char *db_name,
743 bool persistent, const char *unhealthy_reason,
746 struct ctdb_db_context *ctdb_db, *tmp_db;
751 int remaining_tries = 0;
753 ctdb_db = talloc_zero(ctdb, struct ctdb_db_context);
754 CTDB_NO_MEMORY(ctdb, ctdb_db);
756 ctdb_db->priority = 1;
757 ctdb_db->ctdb = ctdb;
758 ctdb_db->db_name = talloc_strdup(ctdb_db, db_name);
759 CTDB_NO_MEMORY(ctdb, ctdb_db->db_name);
761 key.dsize = strlen(db_name)+1;
762 key.dptr = discard_const(db_name);
763 ctdb_db->db_id = ctdb_hash(&key);
764 ctdb_db->persistent = persistent;
766 if (!ctdb_db->persistent) {
767 ctdb_db->delete_queue = trbt_create(ctdb_db, 0);
768 if (ctdb_db->delete_queue == NULL) {
769 CTDB_NO_MEMORY(ctdb, ctdb_db->delete_queue);
772 ctdb_db->ctdb_ltdb_store_fn = ctdb_ltdb_store_server;
775 /* check for hash collisions */
776 for (tmp_db=ctdb->db_list;tmp_db;tmp_db=tmp_db->next) {
777 if (tmp_db->db_id == ctdb_db->db_id) {
778 DEBUG(DEBUG_CRIT,("db_id 0x%x hash collision. name1='%s' name2='%s'\n",
779 tmp_db->db_id, db_name, tmp_db->db_name));
780 talloc_free(ctdb_db);
786 if (unhealthy_reason) {
787 ret = ctdb_update_persistent_health(ctdb, ctdb_db,
788 unhealthy_reason, 0);
790 DEBUG(DEBUG_ALERT,(__location__ " ctdb_update_persistent_health('%s','%s') failed: %d\n",
791 ctdb_db->db_name, unhealthy_reason, ret));
792 talloc_free(ctdb_db);
797 if (ctdb->max_persistent_check_errors > 0) {
800 if (ctdb->runstate == CTDB_RUNSTATE_RUNNING) {
804 ret = ctdb_load_persistent_health(ctdb, ctdb_db);
806 DEBUG(DEBUG_ALERT,(__location__ " ctdb_load_persistent_health('%s') failed: %d\n",
807 ctdb_db->db_name, ret));
808 talloc_free(ctdb_db);
813 if (ctdb_db->unhealthy_reason && remaining_tries == 0) {
814 DEBUG(DEBUG_ALERT,(__location__ "ERROR: tdb %s is marked as unhealthy: %s\n",
815 ctdb_db->db_name, ctdb_db->unhealthy_reason));
816 talloc_free(ctdb_db);
820 if (ctdb_db->unhealthy_reason) {
821 /* this is just a warning, but we want that in the log file! */
822 DEBUG(DEBUG_ALERT,(__location__ "Warning: tdb %s is marked as unhealthy: %s\n",
823 ctdb_db->db_name, ctdb_db->unhealthy_reason));
826 /* open the database */
827 ctdb_db->db_path = talloc_asprintf(ctdb_db, "%s/%s.%u",
828 persistent?ctdb->db_directory_persistent:ctdb->db_directory,
831 tdb_flags = persistent? TDB_DEFAULT : TDB_CLEAR_IF_FIRST | TDB_NOSYNC;
832 if (ctdb->valgrinding) {
833 tdb_flags |= TDB_NOMMAP;
835 tdb_flags |= TDB_DISALLOW_NESTING;
837 tdb_flags |= TDB_INCOMPATIBLE_HASH;
841 ctdb_db->ltdb = tdb_wrap_open(ctdb, ctdb_db->db_path,
842 ctdb->tunable.database_hash_size,
844 O_CREAT|O_RDWR, mode);
845 if (ctdb_db->ltdb == NULL) {
847 int saved_errno = errno;
850 DEBUG(DEBUG_CRIT,("Failed to open tdb '%s': %d - %s\n",
853 strerror(saved_errno)));
854 talloc_free(ctdb_db);
858 if (remaining_tries == 0) {
859 DEBUG(DEBUG_CRIT,(__location__
860 "Failed to open persistent tdb '%s': %d - %s\n",
863 strerror(saved_errno)));
864 talloc_free(ctdb_db);
868 ret = stat(ctdb_db->db_path, &st);
870 DEBUG(DEBUG_CRIT,(__location__
871 "Failed to open persistent tdb '%s': %d - %s\n",
874 strerror(saved_errno)));
875 talloc_free(ctdb_db);
879 ret = ctdb_backup_corrupted_tdb(ctdb, ctdb_db);
881 DEBUG(DEBUG_CRIT,(__location__
882 "Failed to open persistent tdb '%s': %d - %s\n",
885 strerror(saved_errno)));
886 talloc_free(ctdb_db);
896 ctdb_check_db_empty(ctdb_db);
898 ret = tdb_check(ctdb_db->ltdb->tdb, NULL, NULL);
903 DEBUG(DEBUG_CRIT,("tdb_check(%s) failed: %d - %s\n",
904 ctdb_db->db_path, ret,
905 tdb_errorstr(ctdb_db->ltdb->tdb)));
906 if (remaining_tries == 0) {
907 talloc_free(ctdb_db);
911 fd = tdb_fd(ctdb_db->ltdb->tdb);
912 ret = fstat(fd, &st);
914 DEBUG(DEBUG_CRIT,(__location__
915 "Failed to fstat() persistent tdb '%s': %d - %s\n",
919 talloc_free(ctdb_db);
924 talloc_free(ctdb_db->ltdb);
925 ctdb_db->ltdb = NULL;
927 ret = ctdb_backup_corrupted_tdb(ctdb, ctdb_db);
929 DEBUG(DEBUG_CRIT,("Failed to backup corrupted tdb '%s'\n",
931 talloc_free(ctdb_db);
941 /* set up a rb tree we can use to track which records we have a
942 fetch-lock in-flight for so we can defer any additional calls
945 ctdb_db->deferred_fetch = trbt_create(ctdb_db, 0);
946 if (ctdb_db->deferred_fetch == NULL) {
947 DEBUG(DEBUG_ERR,("Failed to create deferred fetch rb tree for ctdb database\n"));
948 talloc_free(ctdb_db);
952 DLIST_ADD(ctdb->db_list, ctdb_db);
954 /* setting this can help some high churn databases */
955 tdb_set_max_dead(ctdb_db->ltdb->tdb, ctdb->tunable.database_max_dead);
958 all databases support the "null" function. we need this in
959 order to do forced migration of records
961 ret = ctdb_daemon_set_call(ctdb, ctdb_db->db_id, ctdb_null_func, CTDB_NULL_FUNC);
963 DEBUG(DEBUG_CRIT,("Failed to setup null function for '%s'\n", ctdb_db->db_name));
964 talloc_free(ctdb_db);
969 all databases support the "fetch" function. we need this
970 for efficient Samba3 ctdb fetch
972 ret = ctdb_daemon_set_call(ctdb, ctdb_db->db_id, ctdb_fetch_func, CTDB_FETCH_FUNC);
974 DEBUG(DEBUG_CRIT,("Failed to setup fetch function for '%s'\n", ctdb_db->db_name));
975 talloc_free(ctdb_db);
980 all databases support the "fetch_with_header" function. we need this
981 for efficient readonly record fetches
983 ret = ctdb_daemon_set_call(ctdb, ctdb_db->db_id, ctdb_fetch_with_header_func, CTDB_FETCH_WITH_HEADER_FUNC);
985 DEBUG(DEBUG_CRIT,("Failed to setup fetch function for '%s'\n", ctdb_db->db_name));
986 talloc_free(ctdb_db);
990 ret = ctdb_vacuum_init(ctdb_db);
992 DEBUG(DEBUG_CRIT,("Failed to setup vacuuming for "
993 "database '%s'\n", ctdb_db->db_name));
994 talloc_free(ctdb_db);
999 DEBUG(DEBUG_INFO,("Attached to database '%s'\n", ctdb_db->db_path));
1006 struct ctdb_deferred_attach_context {
1007 struct ctdb_deferred_attach_context *next, *prev;
1008 struct ctdb_context *ctdb;
1009 struct ctdb_req_control *c;
1013 static int ctdb_deferred_attach_destructor(struct ctdb_deferred_attach_context *da_ctx)
1015 DLIST_REMOVE(da_ctx->ctdb->deferred_attach, da_ctx);
1020 static void ctdb_deferred_attach_timeout(struct event_context *ev, struct timed_event *te, struct timeval t, void *private_data)
1022 struct ctdb_deferred_attach_context *da_ctx = talloc_get_type(private_data, struct ctdb_deferred_attach_context);
1023 struct ctdb_context *ctdb = da_ctx->ctdb;
1025 ctdb_request_control_reply(ctdb, da_ctx->c, NULL, -1, NULL);
1026 talloc_free(da_ctx);
1029 static void ctdb_deferred_attach_callback(struct event_context *ev, struct timed_event *te, struct timeval t, void *private_data)
1031 struct ctdb_deferred_attach_context *da_ctx = talloc_get_type(private_data, struct ctdb_deferred_attach_context);
1032 struct ctdb_context *ctdb = da_ctx->ctdb;
1034 /* This talloc-steals the packet ->c */
1035 ctdb_input_pkt(ctdb, (struct ctdb_req_header *)da_ctx->c);
1036 talloc_free(da_ctx);
1039 int ctdb_process_deferred_attach(struct ctdb_context *ctdb)
1041 struct ctdb_deferred_attach_context *da_ctx;
1043 /* call it from the main event loop as soon as the current event
1046 while ((da_ctx = ctdb->deferred_attach) != NULL) {
1047 DLIST_REMOVE(ctdb->deferred_attach, da_ctx);
1048 event_add_timed(ctdb->ev, da_ctx, timeval_current_ofs(1,0), ctdb_deferred_attach_callback, da_ctx);
1055 a client has asked to attach a new database
1057 int32_t ctdb_control_db_attach(struct ctdb_context *ctdb, TDB_DATA indata,
1058 TDB_DATA *outdata, uint64_t tdb_flags,
1059 bool persistent, uint32_t client_id,
1060 struct ctdb_req_control *c,
1063 const char *db_name = (const char *)indata.dptr;
1064 struct ctdb_db_context *db;
1065 struct ctdb_node *node = ctdb->nodes[ctdb->pnn];
1066 struct ctdb_client *client = NULL;
1068 if (ctdb->tunable.allow_client_db_attach == 0) {
1069 DEBUG(DEBUG_ERR, ("DB Attach to database %s denied by tunable "
1070 "AllowClientDBAccess == 0\n", db_name));
1074 /* dont allow any local clients to attach while we are in recovery mode
1075 * except for the recovery daemon.
1076 * allow all attach from the network since these are always from remote
1079 if (client_id != 0) {
1080 client = ctdb_reqid_find(ctdb, client_id, struct ctdb_client);
1082 if (client != NULL) {
1083 /* If the node is inactive it is not part of the cluster
1084 and we should not allow clients to attach to any
1087 if (node->flags & NODE_FLAGS_INACTIVE) {
1088 DEBUG(DEBUG_ERR,("DB Attach to database %s refused since node is inactive (flags=0x%x)\n", db_name, node->flags));
1092 if (ctdb->recovery_mode == CTDB_RECOVERY_ACTIVE &&
1093 client->pid != ctdb->recoverd_pid &&
1094 ctdb->runstate < CTDB_RUNSTATE_RUNNING) {
1095 struct ctdb_deferred_attach_context *da_ctx = talloc(client, struct ctdb_deferred_attach_context);
1097 if (da_ctx == NULL) {
1098 DEBUG(DEBUG_ERR,("DB Attach to database %s deferral for client with pid:%d failed due to OOM.\n", db_name, client->pid));
1102 da_ctx->ctdb = ctdb;
1103 da_ctx->c = talloc_steal(da_ctx, c);
1104 talloc_set_destructor(da_ctx, ctdb_deferred_attach_destructor);
1105 DLIST_ADD(ctdb->deferred_attach, da_ctx);
1107 event_add_timed(ctdb->ev, da_ctx, timeval_current_ofs(ctdb->tunable.deferred_attach_timeout, 0), ctdb_deferred_attach_timeout, da_ctx);
1109 DEBUG(DEBUG_ERR,("DB Attach to database %s deferred for client with pid:%d since node is in recovery mode.\n", db_name, client->pid));
1110 *async_reply = true;
1115 /* the client can optionally pass additional tdb flags, but we
1116 only allow a subset of those on the database in ctdb. Note
1117 that tdb_flags is passed in via the (otherwise unused)
1118 srvid to the attach control */
1119 tdb_flags &= (TDB_NOSYNC|TDB_INCOMPATIBLE_HASH);
1121 /* see if we already have this name */
1122 db = ctdb_db_handle(ctdb, db_name);
1124 if (db->persistent != persistent) {
1125 DEBUG(DEBUG_ERR, ("ERROR: DB Attach %spersistent to %spersistent "
1126 "database %s\n", persistent ? "" : "non-",
1127 db-> persistent ? "" : "non-", db_name));
1130 outdata->dptr = (uint8_t *)&db->db_id;
1131 outdata->dsize = sizeof(db->db_id);
1132 tdb_add_flags(db->ltdb->tdb, tdb_flags);
1136 if (ctdb_local_attach(ctdb, db_name, persistent, NULL, (tdb_flags&TDB_INCOMPATIBLE_HASH)?true:false) != 0) {
1140 db = ctdb_db_handle(ctdb, db_name);
1142 DEBUG(DEBUG_ERR,("Failed to find db handle for name '%s'\n", db_name));
1146 /* remember the flags the client has specified */
1147 tdb_add_flags(db->ltdb->tdb, tdb_flags);
1149 outdata->dptr = (uint8_t *)&db->db_id;
1150 outdata->dsize = sizeof(db->db_id);
1152 /* Try to ensure it's locked in mem */
1153 ctdb_lockdown_memory(ctdb);
1155 /* tell all the other nodes about this database */
1156 ctdb_daemon_send_control(ctdb, CTDB_BROADCAST_ALL, tdb_flags,
1157 persistent?CTDB_CONTROL_DB_ATTACH_PERSISTENT:
1158 CTDB_CONTROL_DB_ATTACH,
1159 0, CTDB_CTRL_FLAG_NOREPLY,
1160 indata, NULL, NULL);
1168 attach to all existing persistent databases
1170 static int ctdb_attach_persistent(struct ctdb_context *ctdb,
1171 const char *unhealthy_reason)
1176 /* open the persistent db directory and scan it for files */
1177 d = opendir(ctdb->db_directory_persistent);
1182 while ((de=readdir(d))) {
1184 size_t len = strlen(de->d_name);
1186 int invalid_name = 0;
1188 s = talloc_strdup(ctdb, de->d_name);
1189 CTDB_NO_MEMORY(ctdb, s);
1191 /* only accept names ending in .tdb */
1192 p = strstr(s, ".tdb.");
1193 if (len < 7 || p == NULL) {
1198 /* only accept names ending with .tdb. and any number of digits */
1200 while (*q != 0 && invalid_name == 0) {
1201 if (!isdigit(*q++)) {
1205 if (invalid_name == 1 || sscanf(p+5, "%u", &node) != 1 || node != ctdb->pnn) {
1206 DEBUG(DEBUG_ERR,("Ignoring persistent database '%s'\n", de->d_name));
1212 if (ctdb_local_attach(ctdb, s, true, unhealthy_reason, 0) != 0) {
1213 DEBUG(DEBUG_ERR,("Failed to attach to persistent database '%s'\n", de->d_name));
1219 DEBUG(DEBUG_INFO,("Attached to persistent database %s\n", s));
1227 int ctdb_attach_databases(struct ctdb_context *ctdb)
1230 char *persistent_health_path = NULL;
1231 char *unhealthy_reason = NULL;
1232 bool first_try = true;
1234 if (ctdb->db_directory == NULL) {
1235 ctdb->db_directory = VARDIR "/ctdb";
1237 if (ctdb->db_directory_persistent == NULL) {
1238 ctdb->db_directory_persistent = VARDIR "/ctdb/persistent";
1240 if (ctdb->db_directory_state == NULL) {
1241 ctdb->db_directory_state = VARDIR "/ctdb/state";
1244 /* make sure the db directory exists */
1245 ret = mkdir(ctdb->db_directory, 0700);
1246 if (ret == -1 && errno != EEXIST) {
1247 DEBUG(DEBUG_CRIT,(__location__ " Unable to create ctdb directory '%s'\n",
1248 ctdb->db_directory));
1252 /* make sure the persistent db directory exists */
1253 ret = mkdir(ctdb->db_directory_persistent, 0700);
1254 if (ret == -1 && errno != EEXIST) {
1255 DEBUG(DEBUG_CRIT,(__location__ " Unable to create ctdb persistent directory '%s'\n",
1256 ctdb->db_directory_persistent));
1260 /* make sure the internal state db directory exists */
1261 ret = mkdir(ctdb->db_directory_state, 0700);
1262 if (ret == -1 && errno != EEXIST) {
1263 DEBUG(DEBUG_CRIT,(__location__ " Unable to create ctdb state directory '%s'\n",
1264 ctdb->db_directory_state));
1268 persistent_health_path = talloc_asprintf(ctdb, "%s/%s.%u",
1269 ctdb->db_directory_state,
1270 PERSISTENT_HEALTH_TDB,
1272 if (persistent_health_path == NULL) {
1273 DEBUG(DEBUG_CRIT,(__location__ " talloc_asprintf() failed\n"));
1279 ctdb->db_persistent_health = tdb_wrap_open(ctdb, persistent_health_path,
1280 0, TDB_DISALLOW_NESTING,
1281 O_CREAT | O_RDWR, 0600);
1282 if (ctdb->db_persistent_health == NULL) {
1283 struct tdb_wrap *tdb;
1286 DEBUG(DEBUG_CRIT,("Failed to open tdb '%s': %d - %s\n",
1287 persistent_health_path,
1290 talloc_free(persistent_health_path);
1291 talloc_free(unhealthy_reason);
1296 unhealthy_reason = talloc_asprintf(ctdb, "WARNING - '%s' %s - %s",
1297 persistent_health_path,
1298 "was cleared after a failure",
1299 "manual verification needed");
1300 if (unhealthy_reason == NULL) {
1301 DEBUG(DEBUG_CRIT,(__location__ " talloc_asprintf() failed\n"));
1302 talloc_free(persistent_health_path);
1306 DEBUG(DEBUG_CRIT,("Failed to open tdb '%s' - retrying after CLEAR_IF_FIRST\n",
1307 persistent_health_path));
1308 tdb = tdb_wrap_open(ctdb, persistent_health_path,
1309 0, TDB_CLEAR_IF_FIRST | TDB_DISALLOW_NESTING,
1310 O_CREAT | O_RDWR, 0600);
1312 DEBUG(DEBUG_CRIT,("Failed to open tdb '%s' - with CLEAR_IF_FIRST: %d - %s\n",
1313 persistent_health_path,
1316 talloc_free(persistent_health_path);
1317 talloc_free(unhealthy_reason);
1324 ret = tdb_check(ctdb->db_persistent_health->tdb, NULL, NULL);
1326 struct tdb_wrap *tdb;
1328 talloc_free(ctdb->db_persistent_health);
1329 ctdb->db_persistent_health = NULL;
1332 DEBUG(DEBUG_CRIT,("tdb_check('%s') failed\n",
1333 persistent_health_path));
1334 talloc_free(persistent_health_path);
1335 talloc_free(unhealthy_reason);
1340 unhealthy_reason = talloc_asprintf(ctdb, "WARNING - '%s' %s - %s",
1341 persistent_health_path,
1342 "was cleared after a failure",
1343 "manual verification needed");
1344 if (unhealthy_reason == NULL) {
1345 DEBUG(DEBUG_CRIT,(__location__ " talloc_asprintf() failed\n"));
1346 talloc_free(persistent_health_path);
1350 DEBUG(DEBUG_CRIT,("tdb_check('%s') failed - retrying after CLEAR_IF_FIRST\n",
1351 persistent_health_path));
1352 tdb = tdb_wrap_open(ctdb, persistent_health_path,
1353 0, TDB_CLEAR_IF_FIRST | TDB_DISALLOW_NESTING,
1354 O_CREAT | O_RDWR, 0600);
1356 DEBUG(DEBUG_CRIT,("Failed to open tdb '%s' - with CLEAR_IF_FIRST: %d - %s\n",
1357 persistent_health_path,
1360 talloc_free(persistent_health_path);
1361 talloc_free(unhealthy_reason);
1368 talloc_free(persistent_health_path);
1370 ret = ctdb_attach_persistent(ctdb, unhealthy_reason);
1371 talloc_free(unhealthy_reason);
1380 called when a broadcast seqnum update comes in
1382 int32_t ctdb_ltdb_update_seqnum(struct ctdb_context *ctdb, uint32_t db_id, uint32_t srcnode)
1384 struct ctdb_db_context *ctdb_db;
1385 if (srcnode == ctdb->pnn) {
1386 /* don't update ourselves! */
1390 ctdb_db = find_ctdb_db(ctdb, db_id);
1392 DEBUG(DEBUG_ERR,("Unknown db_id 0x%x in ctdb_ltdb_update_seqnum\n", db_id));
1396 if (ctdb_db->unhealthy_reason) {
1397 DEBUG(DEBUG_ERR,("db(%s) unhealty in ctdb_ltdb_update_seqnum: %s\n",
1398 ctdb_db->db_name, ctdb_db->unhealthy_reason));
1402 tdb_increment_seqnum_nonblock(ctdb_db->ltdb->tdb);
1403 ctdb_db->seqnum = tdb_get_seqnum(ctdb_db->ltdb->tdb);
1408 timer to check for seqnum changes in a ltdb and propogate them
1410 static void ctdb_ltdb_seqnum_check(struct event_context *ev, struct timed_event *te,
1411 struct timeval t, void *p)
1413 struct ctdb_db_context *ctdb_db = talloc_get_type(p, struct ctdb_db_context);
1414 struct ctdb_context *ctdb = ctdb_db->ctdb;
1415 uint32_t new_seqnum = tdb_get_seqnum(ctdb_db->ltdb->tdb);
1416 if (new_seqnum != ctdb_db->seqnum) {
1417 /* something has changed - propogate it */
1419 data.dptr = (uint8_t *)&ctdb_db->db_id;
1420 data.dsize = sizeof(uint32_t);
1421 ctdb_daemon_send_control(ctdb, CTDB_BROADCAST_VNNMAP, 0,
1422 CTDB_CONTROL_UPDATE_SEQNUM, 0, CTDB_CTRL_FLAG_NOREPLY,
1425 ctdb_db->seqnum = new_seqnum;
1427 /* setup a new timer */
1428 ctdb_db->seqnum_update =
1429 event_add_timed(ctdb->ev, ctdb_db,
1430 timeval_current_ofs(ctdb->tunable.seqnum_interval/1000, (ctdb->tunable.seqnum_interval%1000)*1000),
1431 ctdb_ltdb_seqnum_check, ctdb_db);
1435 enable seqnum handling on this db
1437 int32_t ctdb_ltdb_enable_seqnum(struct ctdb_context *ctdb, uint32_t db_id)
1439 struct ctdb_db_context *ctdb_db;
1440 ctdb_db = find_ctdb_db(ctdb, db_id);
1442 DEBUG(DEBUG_ERR,("Unknown db_id 0x%x in ctdb_ltdb_enable_seqnum\n", db_id));
1446 if (ctdb_db->seqnum_update == NULL) {
1447 ctdb_db->seqnum_update =
1448 event_add_timed(ctdb->ev, ctdb_db,
1449 timeval_current_ofs(ctdb->tunable.seqnum_interval/1000, (ctdb->tunable.seqnum_interval%1000)*1000),
1450 ctdb_ltdb_seqnum_check, ctdb_db);
1453 tdb_enable_seqnum(ctdb_db->ltdb->tdb);
1454 ctdb_db->seqnum = tdb_get_seqnum(ctdb_db->ltdb->tdb);
1458 int32_t ctdb_control_set_db_priority(struct ctdb_context *ctdb, TDB_DATA indata)
1460 struct ctdb_db_priority *db_prio = (struct ctdb_db_priority *)indata.dptr;
1461 struct ctdb_db_context *ctdb_db;
1463 ctdb_db = find_ctdb_db(ctdb, db_prio->db_id);
1465 DEBUG(DEBUG_ERR,("Unknown db_id 0x%x in ctdb_set_db_priority\n", db_prio->db_id));
1469 if ((db_prio->priority<1) || (db_prio->priority>NUM_DB_PRIORITIES)) {
1470 DEBUG(DEBUG_ERR,("Trying to set invalid priority : %u\n", db_prio->priority));
1474 ctdb_db->priority = db_prio->priority;
1475 DEBUG(DEBUG_INFO,("Setting DB priority to %u for db 0x%08x\n", db_prio->priority, db_prio->db_id));
1481 int ctdb_set_db_sticky(struct ctdb_context *ctdb, struct ctdb_db_context *ctdb_db)
1484 DEBUG(DEBUG_NOTICE,("set db sticky %s\n", ctdb_db->db_name));
1486 if (ctdb_db->sticky) {
1490 if (ctdb_db->persistent) {
1491 DEBUG(DEBUG_ERR,("Trying to set persistent database with sticky property\n"));
1495 ctdb_db->sticky_records = trbt_create(ctdb_db, 0);
1497 ctdb_db->sticky = true;
1502 int32_t ctdb_control_get_db_statistics(struct ctdb_context *ctdb,
1506 struct ctdb_db_context *ctdb_db;
1507 struct ctdb_db_statistics_wire *stats;
1512 ctdb_db = find_ctdb_db(ctdb, db_id);
1514 DEBUG(DEBUG_ERR,("Unknown db_id 0x%x in get_db_statistics\n", db_id));
1518 len = offsetof(struct ctdb_db_statistics_wire, hot_keys);
1519 for (i = 0; i < MAX_HOT_KEYS; i++) {
1520 len += 8 + ctdb_db->statistics.hot_keys[i].key.dsize;
1523 stats = talloc_size(outdata, len);
1524 if (stats == NULL) {
1525 DEBUG(DEBUG_ERR,("Failed to allocate db statistics wire structure\n"));
1529 stats->db_ro_delegations = ctdb_db->statistics.db_ro_delegations;
1530 stats->db_ro_revokes = ctdb_db->statistics.db_ro_revokes;
1531 for (i = 0; i < MAX_COUNT_BUCKETS; i++) {
1532 stats->hop_count_bucket[i] = ctdb_db->statistics.hop_count_bucket[i];
1534 stats->num_hot_keys = MAX_HOT_KEYS;
1536 ptr = &stats->hot_keys[0];
1537 for (i = 0; i < MAX_HOT_KEYS; i++) {
1538 *(uint32_t *)ptr = ctdb_db->statistics.hot_keys[i].count;
1541 *(uint32_t *)ptr = ctdb_db->statistics.hot_keys[i].key.dsize;
1544 memcpy(ptr, ctdb_db->statistics.hot_keys[i].key.dptr, ctdb_db->statistics.hot_keys[i].key.dsize);
1545 ptr += ctdb_db->statistics.hot_keys[i].key.dsize;
1548 outdata->dptr = (uint8_t *)stats;
1549 outdata->dsize = len;