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/>.
20 #include "lib/events/events.h"
21 #include "lib/tdb/include/tdb.h"
22 #include "system/network.h"
23 #include "system/filesys.h"
24 #include "system/wait.h"
25 #include "../include/ctdb_private.h"
26 #include "lib/util/dlinklist.h"
29 static bool later_db(const char *name)
31 return (strstr(name, "notify") || strstr(name, "serverid"));
37 static int ctdb_lock_all_databases(struct ctdb_context *ctdb, uint32_t priority)
39 struct ctdb_db_context *ctdb_db;
41 /* This double loop is for backward compatibility and deadlock
42 avoidance for old samba versions that not yet support
44 This code shall be removed later
46 for (ctdb_db=ctdb->db_list;ctdb_db;ctdb_db=ctdb_db->next) {
47 if (ctdb_db->priority != priority) {
50 if (later_db(ctdb_db->db_name)) {
53 DEBUG(DEBUG_INFO,("locking database 0x%08x priority:%u %s\n", ctdb_db->db_id, ctdb_db->priority, ctdb_db->db_name));
54 if (tdb_lockall(ctdb_db->ltdb->tdb) != 0) {
55 DEBUG(DEBUG_ERR,(__location__ " Failed to lock database %s\n", ctdb_db->db_name));
59 for (ctdb_db=ctdb->db_list;ctdb_db;ctdb_db=ctdb_db->next) {
60 if (ctdb_db->priority != priority) {
63 if (!later_db(ctdb_db->db_name)) {
66 DEBUG(DEBUG_INFO,("locking database 0x%08x priority:%u %s\n", ctdb_db->db_id, ctdb_db->priority, ctdb_db->db_name));
67 if (tdb_lockall(ctdb_db->ltdb->tdb) != 0) {
68 DEBUG(DEBUG_ERR,(__location__ " Failed to lock database %s\n", ctdb_db->db_name));
76 a list of control requests waiting for a freeze lock child to get
79 struct ctdb_freeze_waiter {
80 struct ctdb_freeze_waiter *next, *prev;
81 struct ctdb_context *ctdb;
82 struct ctdb_req_control *c;
87 /* a handle to a freeze lock child process */
88 struct ctdb_freeze_handle {
89 struct ctdb_context *ctdb;
93 struct ctdb_freeze_waiter *waiters;
97 destroy a freeze handle
99 static int ctdb_freeze_handle_destructor(struct ctdb_freeze_handle *h)
101 struct ctdb_context *ctdb = h->ctdb;
102 struct ctdb_db_context *ctdb_db;
104 DEBUG(DEBUG_ERR,("Release freeze handler for prio %u\n", h->priority));
106 /* cancel any pending transactions */
107 if (ctdb->freeze_transaction_started) {
108 for (ctdb_db=ctdb->db_list;ctdb_db;ctdb_db=ctdb_db->next) {
109 if (ctdb_db->priority != h->priority) {
112 tdb_add_flags(ctdb_db->ltdb->tdb, TDB_NOLOCK);
113 if (tdb_transaction_cancel(ctdb_db->ltdb->tdb) != 0) {
114 DEBUG(DEBUG_ERR,(__location__ " Failed to cancel transaction for db '%s'\n",
117 tdb_remove_flags(ctdb_db->ltdb->tdb, TDB_NOLOCK);
119 ctdb->freeze_transaction_started = false;
122 ctdb->freeze_mode[h->priority] = CTDB_FREEZE_NONE;
123 ctdb->freeze_handles[h->priority] = NULL;
125 kill(h->child, SIGKILL);
130 called when the child writes its status to us
132 static void ctdb_freeze_lock_handler(struct event_context *ev, struct fd_event *fde,
133 uint16_t flags, void *private_data)
135 struct ctdb_freeze_handle *h = talloc_get_type(private_data, struct ctdb_freeze_handle);
137 struct ctdb_freeze_waiter *w;
139 if (h->ctdb->freeze_mode[h->priority] == CTDB_FREEZE_FROZEN) {
140 DEBUG(DEBUG_INFO,("freeze child died - unfreezing\n"));
145 if (read(h->fd, &status, sizeof(status)) != sizeof(status)) {
146 DEBUG(DEBUG_ERR,("read error from freeze lock child\n"));
151 DEBUG(DEBUG_ERR,("Failed to get locks in ctdb_freeze_child\n"));
152 /* we didn't get the locks - destroy the handle */
157 h->ctdb->freeze_mode[h->priority] = CTDB_FREEZE_FROZEN;
159 /* notify the waiters */
160 if (h != h->ctdb->freeze_handles[h->priority]) {
161 DEBUG(DEBUG_ERR,("lockwait finished but h is not linked\n"));
163 while ((w = h->waiters)) {
165 DLIST_REMOVE(h->waiters, w);
171 create a child which gets locks on all the open databases, then calls the callback telling the parent
174 static struct ctdb_freeze_handle *ctdb_freeze_lock(struct ctdb_context *ctdb, uint32_t priority)
176 struct ctdb_freeze_handle *h;
178 struct fd_event *fde;
180 h = talloc_zero(ctdb, struct ctdb_freeze_handle);
181 CTDB_NO_MEMORY_NULL(ctdb, h);
184 h->priority = priority;
186 if (pipe(fd) == -1) {
187 DEBUG(DEBUG_ERR,("Failed to create pipe for ctdb_freeze_lock\n"));
193 if (h->child == -1) {
194 DEBUG(DEBUG_ERR,("Failed to fork child for ctdb_freeze_lock\n"));
205 debug_extra = talloc_asprintf(NULL, "freeze_lock-%u:", priority);
206 ret = ctdb_lock_all_databases(ctdb, priority);
211 ret = write(fd[1], &ret, sizeof(ret));
212 if (ret != sizeof(ret)) {
213 DEBUG(DEBUG_ERR, (__location__ " Failed to write to socket from freeze child. ret:%d errno:%u\n", ret, errno));
219 if (kill(ctdb->ctdbd_pid, 0) != 0) {
220 DEBUG(DEBUG_ERR,("Parent died. Exiting lock wait child\n"));
227 talloc_set_destructor(h, ctdb_freeze_handle_destructor);
230 set_close_on_exec(fd[0]);
235 fde = event_add_fd(ctdb->ev, h, h->fd, EVENT_FD_READ|EVENT_FD_AUTOCLOSE,
236 ctdb_freeze_lock_handler, h);
238 DEBUG(DEBUG_ERR,("Failed to setup fd event for ctdb_freeze_lock\n"));
248 destroy a waiter for a freeze mode change
250 static int ctdb_freeze_waiter_destructor(struct ctdb_freeze_waiter *w)
252 ctdb_request_control_reply(w->ctdb, w->c, NULL, w->status, NULL);
257 start the freeze process for a certain priority
259 int ctdb_start_freeze(struct ctdb_context *ctdb, uint32_t priority)
262 DEBUG(DEBUG_ERR,("Freeze priority 0 requested, remapping to priority 1\n"));
266 if ((priority < 1) || (priority > NUM_DB_PRIORITIES)) {
267 DEBUG(DEBUG_ERR,(__location__ " Invalid db priority : %u\n", priority));
271 if (ctdb->freeze_mode[priority] == CTDB_FREEZE_FROZEN) {
272 /* we're already frozen */
276 /* if there isn't a freeze lock child then create one */
277 if (ctdb->freeze_handles[priority] == NULL) {
278 ctdb->freeze_handles[priority] = ctdb_freeze_lock(ctdb, priority);
279 CTDB_NO_MEMORY(ctdb, ctdb->freeze_handles[priority]);
280 ctdb->freeze_mode[priority] = CTDB_FREEZE_PENDING;
289 int32_t ctdb_control_freeze(struct ctdb_context *ctdb, struct ctdb_req_control *c, bool *async_reply)
291 struct ctdb_freeze_waiter *w;
294 priority = (uint32_t)c->srvid;
296 DEBUG(DEBUG_ERR, ("Freeze priority %u\n", priority));
299 DEBUG(DEBUG_ERR,("Freeze priority 0 requested, remapping to priority 1\n"));
303 if ((priority < 1) || (priority > NUM_DB_PRIORITIES)) {
304 DEBUG(DEBUG_ERR,(__location__ " Invalid db priority : %u\n", priority));
308 if (ctdb->freeze_mode[priority] == CTDB_FREEZE_FROZEN) {
309 /* we're already frozen */
313 if (ctdb_start_freeze(ctdb, priority) != 0) {
314 DEBUG(DEBUG_ERR,(__location__ " Failed to start freezing databases with priority %u\n", priority));
318 /* add ourselves to list of waiters */
319 if (ctdb->freeze_handles[priority] == NULL) {
320 DEBUG(DEBUG_ERR,("No freeze lock handle when adding a waiter\n"));
324 w = talloc(ctdb->freeze_handles[priority], struct ctdb_freeze_waiter);
325 CTDB_NO_MEMORY(ctdb, w);
327 w->c = talloc_steal(w, c);
328 w->priority = priority;
330 talloc_set_destructor(w, ctdb_freeze_waiter_destructor);
331 DLIST_ADD(ctdb->freeze_handles[priority]->waiters, w);
333 /* we won't reply till later */
340 block until we are frozen, used during daemon startup
342 bool ctdb_blocking_freeze(struct ctdb_context *ctdb)
346 for (i=1; i<=NUM_DB_PRIORITIES; i++) {
347 if (ctdb_start_freeze(ctdb, i)) {
348 DEBUG(DEBUG_ERR,(__location__ " Failed to freeze databases of prio %u\n", i));
352 /* block until frozen */
353 while (ctdb->freeze_mode[i] == CTDB_FREEZE_PENDING) {
354 event_loop_once(ctdb->ev);
362 static void thaw_priority(struct ctdb_context *ctdb, uint32_t priority)
364 DEBUG(DEBUG_ERR,("Thawing priority %u\n", priority));
366 /* cancel any pending transactions */
367 if (ctdb->freeze_transaction_started) {
368 struct ctdb_db_context *ctdb_db;
370 for (ctdb_db=ctdb->db_list;ctdb_db;ctdb_db=ctdb_db->next) {
371 tdb_add_flags(ctdb_db->ltdb->tdb, TDB_NOLOCK);
372 if (tdb_transaction_cancel(ctdb_db->ltdb->tdb) != 0) {
373 DEBUG(DEBUG_ERR,(__location__ " Failed to cancel transaction for db '%s'\n",
376 tdb_remove_flags(ctdb_db->ltdb->tdb, TDB_NOLOCK);
379 ctdb->freeze_transaction_started = false;
382 /* this hack can be used to get a copy of the databases at the end of a recovery */
383 system("mkdir -p /var/ctdb.saved; /usr/bin/rsync --delete -a /var/ctdb/ /var/ctdb.saved/$$ 2>&1 > /dev/null");
387 /* and this one for local testing */
388 system("mkdir -p test.db.saved; /usr/bin/rsync --delete -a test.db/ test.db.saved/$$ 2>&1 > /dev/null");
391 if (ctdb->freeze_handles[priority] != NULL) {
392 talloc_free(ctdb->freeze_handles[priority]);
393 ctdb->freeze_handles[priority] = NULL;
400 int32_t ctdb_control_thaw(struct ctdb_context *ctdb, uint32_t priority)
403 if (priority > NUM_DB_PRIORITIES) {
404 DEBUG(DEBUG_ERR,(__location__ " Invalid db priority : %u\n", priority));
410 for (i=1;i<=NUM_DB_PRIORITIES; i++) {
411 thaw_priority(ctdb, i);
414 thaw_priority(ctdb, priority);
417 ctdb_call_resend_all(ctdb);
423 start a transaction on all databases - used for recovery
425 int32_t ctdb_control_transaction_start(struct ctdb_context *ctdb, uint32_t id)
427 struct ctdb_db_context *ctdb_db;
430 for (i=1;i<=NUM_DB_PRIORITIES; i++) {
431 if (ctdb->freeze_mode[i] != CTDB_FREEZE_FROZEN) {
432 DEBUG(DEBUG_ERR,(__location__ " Failed transaction_start while not frozen\n"));
437 for (ctdb_db=ctdb->db_list;ctdb_db;ctdb_db=ctdb_db->next) {
440 tdb_add_flags(ctdb_db->ltdb->tdb, TDB_NOLOCK);
442 if (ctdb->freeze_transaction_started) {
443 if (tdb_transaction_cancel(ctdb_db->ltdb->tdb) != 0) {
444 DEBUG(DEBUG_ERR,(__location__ " Failed to cancel transaction for db '%s'\n",
446 /* not a fatal error */
450 ret = tdb_transaction_start(ctdb_db->ltdb->tdb);
452 tdb_remove_flags(ctdb_db->ltdb->tdb, TDB_NOLOCK);
455 DEBUG(DEBUG_ERR,(__location__ " Failed to start transaction for db '%s'\n",
461 ctdb->freeze_transaction_started = true;
462 ctdb->freeze_transaction_id = id;
468 cancel a transaction for all databases - used for recovery
470 int32_t ctdb_control_transaction_cancel(struct ctdb_context *ctdb)
472 struct ctdb_db_context *ctdb_db;
474 DEBUG(DEBUG_ERR,(__location__ " recovery transaction cancelled called\n"));
476 for (ctdb_db=ctdb->db_list;ctdb_db;ctdb_db=ctdb_db->next) {
477 tdb_add_flags(ctdb_db->ltdb->tdb, TDB_NOLOCK);
479 if (tdb_transaction_cancel(ctdb_db->ltdb->tdb) != 0) {
480 DEBUG(DEBUG_ERR,(__location__ " Failed to cancel transaction for db '%s'\n", ctdb_db->db_name));
481 /* not a fatal error */
484 tdb_remove_flags(ctdb_db->ltdb->tdb, TDB_NOLOCK);
487 ctdb->freeze_transaction_started = false;
493 commit transactions on all databases
495 int32_t ctdb_control_transaction_commit(struct ctdb_context *ctdb, uint32_t id)
497 struct ctdb_db_context *ctdb_db;
499 int healthy_nodes = 0;
501 for (i=1;i<=NUM_DB_PRIORITIES; i++) {
502 if (ctdb->freeze_mode[i] != CTDB_FREEZE_FROZEN) {
503 DEBUG(DEBUG_ERR,(__location__ " Failed transaction_start while not frozen\n"));
508 if (!ctdb->freeze_transaction_started) {
509 DEBUG(DEBUG_ERR,(__location__ " transaction not started\n"));
513 if (id != ctdb->freeze_transaction_id) {
514 DEBUG(DEBUG_ERR,(__location__ " incorrect transaction id 0x%x in commit\n", id));
518 DEBUG(DEBUG_DEBUG,(__location__ " num_nodes[%d]\n", ctdb->num_nodes));
519 for (i=0; i < ctdb->num_nodes; i++) {
520 DEBUG(DEBUG_DEBUG,(__location__ " node[%d].flags[0x%X]\n",
521 i, ctdb->nodes[i]->flags));
522 if (ctdb->nodes[i]->flags == 0) {
526 DEBUG(DEBUG_INFO,(__location__ " healthy_nodes[%d]\n", healthy_nodes));
528 for (ctdb_db=ctdb->db_list;ctdb_db;ctdb_db=ctdb_db->next) {
531 tdb_add_flags(ctdb_db->ltdb->tdb, TDB_NOLOCK);
532 ret = tdb_transaction_commit(ctdb_db->ltdb->tdb);
534 DEBUG(DEBUG_ERR,(__location__ " Failed to commit transaction for db '%s'. Cancel all transactions and resetting transaction_started to false.\n",
538 tdb_remove_flags(ctdb_db->ltdb->tdb, TDB_NOLOCK);
540 ret = ctdb_update_persistent_health(ctdb, ctdb_db, NULL, healthy_nodes);
542 DEBUG(DEBUG_CRIT,(__location__ " Failed to update persistent health for db '%s'. "
543 "Cancel all remaining transactions and resetting transaction_started to false.\n",
549 ctdb->freeze_transaction_started = false;
550 ctdb->freeze_transaction_id = 0;
555 /* cancel any pending transactions */
556 for (ctdb_db=ctdb->db_list;ctdb_db;ctdb_db=ctdb_db->next) {
557 tdb_add_flags(ctdb_db->ltdb->tdb, TDB_NOLOCK);
558 if (tdb_transaction_cancel(ctdb_db->ltdb->tdb) != 0) {
559 DEBUG(DEBUG_ERR,(__location__ " Failed to cancel transaction for db '%s'\n",
562 tdb_remove_flags(ctdb_db->ltdb->tdb, TDB_NOLOCK);
564 ctdb->freeze_transaction_started = false;
570 wipe a database - only possible when in a frozen transaction
572 int32_t ctdb_control_wipe_database(struct ctdb_context *ctdb, TDB_DATA indata)
574 struct ctdb_control_wipe_database w = *(struct ctdb_control_wipe_database *)indata.dptr;
575 struct ctdb_db_context *ctdb_db;
577 ctdb_db = find_ctdb_db(ctdb, w.db_id);
579 DEBUG(DEBUG_ERR,(__location__ " Unknown db 0x%x\n", w.db_id));
583 if (ctdb->freeze_mode[ctdb_db->priority] != CTDB_FREEZE_FROZEN) {
584 DEBUG(DEBUG_ERR,(__location__ " Failed transaction_start while not frozen\n"));
588 if (!ctdb->freeze_transaction_started) {
589 DEBUG(DEBUG_ERR,(__location__ " transaction not started\n"));
593 if (w.transaction_id != ctdb->freeze_transaction_id) {
594 DEBUG(DEBUG_ERR,(__location__ " incorrect transaction id 0x%x in commit\n", w.transaction_id));
598 if (tdb_wipe_all(ctdb_db->ltdb->tdb) != 0) {
599 DEBUG(DEBUG_ERR,(__location__ " Failed to wipe database for db '%s'\n",