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"
33 static int ctdb_lock_all_databases(struct ctdb_context *ctdb, uint32_t priority)
35 struct ctdb_db_context *ctdb_db;
37 /* This double loop is for backward compatibility and deadlock
38 avoidance for old samba versions that not yet support
40 This code shall be removed later
42 for (ctdb_db=ctdb->db_list;ctdb_db;ctdb_db=ctdb_db->next) {
43 if (ctdb_db->priority != priority) {
46 if (strstr(ctdb_db->db_name, "notify") != NULL) {
49 DEBUG(DEBUG_INFO,("locking database 0x%08x priority:%u %s\n", ctdb_db->db_id, ctdb_db->priority, ctdb_db->db_name));
50 if (tdb_lockall(ctdb_db->ltdb->tdb) != 0) {
54 for (ctdb_db=ctdb->db_list;ctdb_db;ctdb_db=ctdb_db->next) {
55 if (ctdb_db->priority != priority) {
58 if (strstr(ctdb_db->db_name, "notify") == NULL) {
61 DEBUG(DEBUG_INFO,("locking database 0x%08x priority:%u %s\n", ctdb_db->db_id, ctdb_db->priority, ctdb_db->db_name));
62 if (tdb_lockall(ctdb_db->ltdb->tdb) != 0) {
70 a list of control requests waiting for a freeze lock child to get
73 struct ctdb_freeze_waiter {
74 struct ctdb_freeze_waiter *next, *prev;
75 struct ctdb_context *ctdb;
76 struct ctdb_req_control *c;
81 /* a handle to a freeze lock child process */
82 struct ctdb_freeze_handle {
83 struct ctdb_context *ctdb;
87 struct ctdb_freeze_waiter *waiters;
91 destroy a freeze handle
93 static int ctdb_freeze_handle_destructor(struct ctdb_freeze_handle *h)
95 struct ctdb_context *ctdb = h->ctdb;
96 struct ctdb_db_context *ctdb_db;
98 DEBUG(DEBUG_ERR,("Release freeze handler for prio %u\n", h->priority));
100 /* cancel any pending transactions */
101 if (ctdb->freeze_transaction_started) {
102 for (ctdb_db=ctdb->db_list;ctdb_db;ctdb_db=ctdb_db->next) {
103 if (ctdb_db->priority != h->priority) {
106 tdb_add_flags(ctdb_db->ltdb->tdb, TDB_NOLOCK);
107 if (tdb_transaction_cancel(ctdb_db->ltdb->tdb) != 0) {
108 DEBUG(DEBUG_ERR,(__location__ " Failed to cancel transaction for db '%s'\n",
111 tdb_remove_flags(ctdb_db->ltdb->tdb, TDB_NOLOCK);
113 ctdb->freeze_transaction_started = false;
116 ctdb->freeze_mode[h->priority] = CTDB_FREEZE_NONE;
117 ctdb->freeze_handles[h->priority] = NULL;
119 kill(h->child, SIGKILL);
124 called when the child writes its status to us
126 static void ctdb_freeze_lock_handler(struct event_context *ev, struct fd_event *fde,
127 uint16_t flags, void *private_data)
129 struct ctdb_freeze_handle *h = talloc_get_type(private_data, struct ctdb_freeze_handle);
131 struct ctdb_freeze_waiter *w;
133 if (h->ctdb->freeze_mode[h->priority] == CTDB_FREEZE_FROZEN) {
134 DEBUG(DEBUG_INFO,("freeze child died - unfreezing\n"));
139 if (read(h->fd, &status, sizeof(status)) != sizeof(status)) {
140 DEBUG(DEBUG_ERR,("read error from freeze lock child\n"));
145 DEBUG(DEBUG_ERR,("Failed to get locks in ctdb_freeze_child\n"));
146 /* we didn't get the locks - destroy the handle */
151 h->ctdb->freeze_mode[h->priority] = CTDB_FREEZE_FROZEN;
153 /* notify the waiters */
154 if (h != h->ctdb->freeze_handles[h->priority]) {
155 DEBUG(DEBUG_ERR,("lockwait finished but h is not linked\n"));
157 while ((w = h->waiters)) {
159 DLIST_REMOVE(h->waiters, w);
165 create a child which gets locks on all the open databases, then calls the callback telling the parent
168 static struct ctdb_freeze_handle *ctdb_freeze_lock(struct ctdb_context *ctdb, uint32_t priority)
170 struct ctdb_freeze_handle *h;
172 struct fd_event *fde;
174 h = talloc_zero(ctdb, struct ctdb_freeze_handle);
175 CTDB_NO_MEMORY_NULL(ctdb, h);
178 h->priority = priority;
180 if (pipe(fd) == -1) {
181 DEBUG(DEBUG_ERR,("Failed to create pipe for ctdb_freeze_lock\n"));
187 if (h->child == -1) {
188 DEBUG(DEBUG_ERR,("Failed to fork child for ctdb_freeze_lock\n"));
199 ret = ctdb_lock_all_databases(ctdb, priority);
204 ret = write(fd[1], &ret, sizeof(ret));
205 if (ret != sizeof(ret)) {
206 DEBUG(DEBUG_ERR, (__location__ " Failed to write to socket from freeze child. ret:%d errno:%u\n", ret, errno));
212 if (kill(ctdb->ctdbd_pid, 0) != 0) {
213 DEBUG(DEBUG_ERR,("Parent died. Exiting lock wait child\n"));
220 talloc_set_destructor(h, ctdb_freeze_handle_destructor);
223 set_close_on_exec(fd[0]);
228 fde = event_add_fd(ctdb->ev, h, h->fd, EVENT_FD_READ|EVENT_FD_AUTOCLOSE,
229 ctdb_freeze_lock_handler, h);
231 DEBUG(DEBUG_ERR,("Failed to setup fd event for ctdb_freeze_lock\n"));
241 destroy a waiter for a freeze mode change
243 static int ctdb_freeze_waiter_destructor(struct ctdb_freeze_waiter *w)
245 ctdb_request_control_reply(w->ctdb, w->c, NULL, w->status, NULL);
250 start the freeze process for a certain priority
252 int ctdb_start_freeze(struct ctdb_context *ctdb, uint32_t priority)
255 DEBUG(DEBUG_ERR,("Freeze priority 0 requested, remapping to priority 1\n"));
259 if ((priority < 1) || (priority > NUM_DB_PRIORITIES)) {
260 DEBUG(DEBUG_ERR,(__location__ " Invalid db priority : %u\n", priority));
264 if (ctdb->freeze_mode[priority] == CTDB_FREEZE_FROZEN) {
265 /* we're already frozen */
269 /* if there isn't a freeze lock child then create one */
270 if (ctdb->freeze_handles[priority] == NULL) {
271 ctdb->freeze_handles[priority] = ctdb_freeze_lock(ctdb, priority);
272 CTDB_NO_MEMORY(ctdb, ctdb->freeze_handles[priority]);
273 ctdb->freeze_mode[priority] = CTDB_FREEZE_PENDING;
282 int32_t ctdb_control_freeze(struct ctdb_context *ctdb, struct ctdb_req_control *c, bool *async_reply)
284 struct ctdb_freeze_waiter *w;
287 priority = (uint32_t)c->srvid;
289 DEBUG(DEBUG_ERR, ("Freeze priority %u\n", priority));
292 DEBUG(DEBUG_ERR,("Freeze priority 0 requested, remapping to priority 1\n"));
296 if ((priority < 1) || (priority > NUM_DB_PRIORITIES)) {
297 DEBUG(DEBUG_ERR,(__location__ " Invalid db priority : %u\n", priority));
301 if (ctdb->freeze_mode[priority] == CTDB_FREEZE_FROZEN) {
302 /* we're already frozen */
306 if (ctdb_start_freeze(ctdb, priority) != 0) {
307 DEBUG(DEBUG_ERR,(__location__ " Failed to start freezing databases with priority %u\n", priority));
311 /* add ourselves to list of waiters */
312 if (ctdb->freeze_handles[priority] == NULL) {
313 DEBUG(DEBUG_ERR,("No freeze lock handle when adding a waiter\n"));
317 w = talloc(ctdb->freeze_handles[priority], struct ctdb_freeze_waiter);
318 CTDB_NO_MEMORY(ctdb, w);
320 w->c = talloc_steal(w, c);
321 w->priority = priority;
323 talloc_set_destructor(w, ctdb_freeze_waiter_destructor);
324 DLIST_ADD(ctdb->freeze_handles[priority]->waiters, w);
326 /* we won't reply till later */
333 block until we are frozen, used during daemon startup
335 bool ctdb_blocking_freeze(struct ctdb_context *ctdb)
339 for (i=1; i<=NUM_DB_PRIORITIES; i++) {
340 if (ctdb_start_freeze(ctdb, i)) {
341 DEBUG(DEBUG_ERR,(__location__ " Failed to freeze databases of prio %u\n", i));
345 /* block until frozen */
346 while (ctdb->freeze_mode[i] == CTDB_FREEZE_PENDING) {
347 event_loop_once(ctdb->ev);
355 static void thaw_priority(struct ctdb_context *ctdb, uint32_t priority)
357 DEBUG(DEBUG_ERR,("Thawing priority %u\n", priority));
359 /* cancel any pending transactions */
360 if (ctdb->freeze_transaction_started) {
361 struct ctdb_db_context *ctdb_db;
363 for (ctdb_db=ctdb->db_list;ctdb_db;ctdb_db=ctdb_db->next) {
364 tdb_add_flags(ctdb_db->ltdb->tdb, TDB_NOLOCK);
365 if (tdb_transaction_cancel(ctdb_db->ltdb->tdb) != 0) {
366 DEBUG(DEBUG_ERR,(__location__ " Failed to cancel transaction for db '%s'\n",
369 tdb_remove_flags(ctdb_db->ltdb->tdb, TDB_NOLOCK);
372 ctdb->freeze_transaction_started = false;
375 /* this hack can be used to get a copy of the databases at the end of a recovery */
376 system("mkdir -p /var/ctdb.saved; /usr/bin/rsync --delete -a /var/ctdb/ /var/ctdb.saved/$$ 2>&1 > /dev/null");
380 /* and this one for local testing */
381 system("mkdir -p test.db.saved; /usr/bin/rsync --delete -a test.db/ test.db.saved/$$ 2>&1 > /dev/null");
384 if (ctdb->freeze_handles[priority] != NULL) {
385 talloc_free(ctdb->freeze_handles[priority]);
386 ctdb->freeze_handles[priority] = NULL;
393 int32_t ctdb_control_thaw(struct ctdb_context *ctdb, uint32_t priority)
396 if (priority > NUM_DB_PRIORITIES) {
397 DEBUG(DEBUG_ERR,(__location__ " Invalid db priority : %u\n", priority));
403 for (i=1;i<=NUM_DB_PRIORITIES; i++) {
404 thaw_priority(ctdb, i);
407 thaw_priority(ctdb, priority);
410 ctdb_call_resend_all(ctdb);
416 start a transaction on all databases - used for recovery
418 int32_t ctdb_control_transaction_start(struct ctdb_context *ctdb, uint32_t id)
420 struct ctdb_db_context *ctdb_db;
423 for (i=1;i<=NUM_DB_PRIORITIES; i++) {
424 if (ctdb->freeze_mode[i] != CTDB_FREEZE_FROZEN) {
425 DEBUG(DEBUG_ERR,(__location__ " Failed transaction_start while not frozen\n"));
430 for (ctdb_db=ctdb->db_list;ctdb_db;ctdb_db=ctdb_db->next) {
433 tdb_add_flags(ctdb_db->ltdb->tdb, TDB_NOLOCK);
435 if (ctdb->freeze_transaction_started) {
436 if (tdb_transaction_cancel(ctdb_db->ltdb->tdb) != 0) {
437 DEBUG(DEBUG_ERR,(__location__ " Failed to cancel transaction for db '%s'\n",
439 /* not a fatal error */
443 ret = tdb_transaction_start(ctdb_db->ltdb->tdb);
445 tdb_remove_flags(ctdb_db->ltdb->tdb, TDB_NOLOCK);
448 DEBUG(DEBUG_ERR,(__location__ " Failed to start transaction for db '%s'\n",
454 ctdb->freeze_transaction_started = true;
455 ctdb->freeze_transaction_id = id;
461 cancel a transaction for all databases - used for recovery
463 int32_t ctdb_control_transaction_cancel(struct ctdb_context *ctdb)
465 struct ctdb_db_context *ctdb_db;
467 DEBUG(DEBUG_ERR,(__location__ " recovery transaction cancelled called\n"));
469 for (ctdb_db=ctdb->db_list;ctdb_db;ctdb_db=ctdb_db->next) {
470 tdb_add_flags(ctdb_db->ltdb->tdb, TDB_NOLOCK);
472 if (tdb_transaction_cancel(ctdb_db->ltdb->tdb) != 0) {
473 DEBUG(DEBUG_ERR,(__location__ " Failed to cancel transaction for db '%s'\n", ctdb_db->db_name));
474 /* not a fatal error */
477 tdb_remove_flags(ctdb_db->ltdb->tdb, TDB_NOLOCK);
480 ctdb->freeze_transaction_started = false;
486 commit transactions on all databases
488 int32_t ctdb_control_transaction_commit(struct ctdb_context *ctdb, uint32_t id)
490 struct ctdb_db_context *ctdb_db;
492 int healthy_nodes = 0;
494 for (i=1;i<=NUM_DB_PRIORITIES; i++) {
495 if (ctdb->freeze_mode[i] != CTDB_FREEZE_FROZEN) {
496 DEBUG(DEBUG_ERR,(__location__ " Failed transaction_start while not frozen\n"));
501 if (!ctdb->freeze_transaction_started) {
502 DEBUG(DEBUG_ERR,(__location__ " transaction not started\n"));
506 if (id != ctdb->freeze_transaction_id) {
507 DEBUG(DEBUG_ERR,(__location__ " incorrect transaction id 0x%x in commit\n", id));
511 DEBUG(DEBUG_DEBUG,(__location__ " num_nodes[%d]\n", ctdb->num_nodes));
512 for (i=0; i < ctdb->num_nodes; i++) {
513 DEBUG(DEBUG_DEBUG,(__location__ " node[%d].flags[0x%X]\n",
514 i, ctdb->nodes[i]->flags));
515 if (ctdb->nodes[i]->flags == 0) {
519 DEBUG(DEBUG_INFO,(__location__ " healthy_nodes[%d]\n", healthy_nodes));
521 for (ctdb_db=ctdb->db_list;ctdb_db;ctdb_db=ctdb_db->next) {
524 tdb_add_flags(ctdb_db->ltdb->tdb, TDB_NOLOCK);
525 ret = tdb_transaction_commit(ctdb_db->ltdb->tdb);
527 DEBUG(DEBUG_ERR,(__location__ " Failed to commit transaction for db '%s'. Cancel all transactions and resetting transaction_started to false.\n",
531 tdb_remove_flags(ctdb_db->ltdb->tdb, TDB_NOLOCK);
533 ret = ctdb_update_persistent_health(ctdb, ctdb_db, NULL, healthy_nodes);
535 DEBUG(DEBUG_CRIT,(__location__ " Failed to update persistent health for db '%s'. "
536 "Cancel all remaining transactions and resetting transaction_started to false.\n",
542 ctdb->freeze_transaction_started = false;
543 ctdb->freeze_transaction_id = 0;
548 /* cancel any pending transactions */
549 for (ctdb_db=ctdb->db_list;ctdb_db;ctdb_db=ctdb_db->next) {
550 tdb_add_flags(ctdb_db->ltdb->tdb, TDB_NOLOCK);
551 if (tdb_transaction_cancel(ctdb_db->ltdb->tdb) != 0) {
552 DEBUG(DEBUG_ERR,(__location__ " Failed to cancel transaction for db '%s'\n",
555 tdb_remove_flags(ctdb_db->ltdb->tdb, TDB_NOLOCK);
557 ctdb->freeze_transaction_started = false;
563 wipe a database - only possible when in a frozen transaction
565 int32_t ctdb_control_wipe_database(struct ctdb_context *ctdb, TDB_DATA indata)
567 struct ctdb_control_wipe_database w = *(struct ctdb_control_wipe_database *)indata.dptr;
568 struct ctdb_db_context *ctdb_db;
570 ctdb_db = find_ctdb_db(ctdb, w.db_id);
572 DEBUG(DEBUG_ERR,(__location__ " Unknown db 0x%x\n", w.db_id));
576 if (ctdb->freeze_mode[ctdb_db->priority] != CTDB_FREEZE_FROZEN) {
577 DEBUG(DEBUG_ERR,(__location__ " Failed transaction_start while not frozen\n"));
581 if (!ctdb->freeze_transaction_started) {
582 DEBUG(DEBUG_ERR,(__location__ " transaction not started\n"));
586 if (w.transaction_id != ctdb->freeze_transaction_id) {
587 DEBUG(DEBUG_ERR,(__location__ " incorrect transaction id 0x%x in commit\n", w.transaction_id));
591 if (tdb_wipe_all(ctdb_db->ltdb->tdb) != 0) {
592 DEBUG(DEBUG_ERR,(__location__ " Failed to wipe database for db '%s'\n",