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_protocol.h"
26 #include "include/ctdb_private.h"
27 #include "lib/util/dlinklist.h"
34 static int ctdb_lock_all_databases(struct ctdb_context *ctdb, uint32_t priority)
36 struct ctdb_db_context *ctdb_db;
38 /* This double loop is for backward compatibility and deadlock
39 avoidance for old samba versions that not yet support
41 This code shall be removed later
43 for (ctdb_db=ctdb->db_list;ctdb_db;ctdb_db=ctdb_db->next) {
44 if (ctdb_db->priority != priority) {
47 if (strstr(ctdb_db->db_name, "notify") != NULL) {
50 DEBUG(DEBUG_INFO,("locking database 0x%08x priority:%u %s\n", ctdb_db->db_id, ctdb_db->priority, ctdb_db->db_name));
51 if (tdb_lockall(ctdb_db->ltdb->tdb) != 0) {
55 for (ctdb_db=ctdb->db_list;ctdb_db;ctdb_db=ctdb_db->next) {
56 if (ctdb_db->priority != priority) {
59 if (strstr(ctdb_db->db_name, "notify") == NULL) {
62 DEBUG(DEBUG_INFO,("locking database 0x%08x priority:%u %s\n", ctdb_db->db_id, ctdb_db->priority, ctdb_db->db_name));
63 if (tdb_lockall(ctdb_db->ltdb->tdb) != 0) {
71 a list of control requests waiting for a freeze lock child to get
74 struct ctdb_freeze_waiter {
75 struct ctdb_freeze_waiter *next, *prev;
76 struct ctdb_context *ctdb;
77 struct ctdb_req_control *c;
82 /* a handle to a freeze lock child process */
83 struct ctdb_freeze_handle {
84 struct ctdb_context *ctdb;
88 struct ctdb_freeze_waiter *waiters;
92 destroy a freeze handle
94 static int ctdb_freeze_handle_destructor(struct ctdb_freeze_handle *h)
96 struct ctdb_context *ctdb = h->ctdb;
97 struct ctdb_db_context *ctdb_db;
99 DEBUG(DEBUG_ERR,("Release freeze handler for prio %u\n", h->priority));
101 /* cancel any pending transactions */
102 if (ctdb->freeze_transaction_started) {
103 for (ctdb_db=ctdb->db_list;ctdb_db;ctdb_db=ctdb_db->next) {
104 if (ctdb_db->priority != h->priority) {
107 tdb_add_flags(ctdb_db->ltdb->tdb, TDB_NOLOCK);
108 if (tdb_transaction_cancel(ctdb_db->ltdb->tdb) != 0) {
109 DEBUG(DEBUG_ERR,(__location__ " Failed to cancel transaction for db '%s'\n",
112 tdb_remove_flags(ctdb_db->ltdb->tdb, TDB_NOLOCK);
114 ctdb->freeze_transaction_started = false;
117 ctdb->freeze_mode[h->priority] = CTDB_FREEZE_NONE;
118 ctdb->freeze_handles[h->priority] = NULL;
120 kill(h->child, SIGKILL);
125 called when the child writes its status to us
127 static void ctdb_freeze_lock_handler(struct event_context *ev, struct fd_event *fde,
128 uint16_t flags, void *private_data)
130 struct ctdb_freeze_handle *h = talloc_get_type(private_data, struct ctdb_freeze_handle);
132 struct ctdb_freeze_waiter *w;
134 if (h->ctdb->freeze_mode[h->priority] == CTDB_FREEZE_FROZEN) {
135 DEBUG(DEBUG_INFO,("freeze child died - unfreezing\n"));
140 if (read(h->fd, &status, sizeof(status)) != sizeof(status)) {
141 DEBUG(DEBUG_ERR,("read error from freeze lock child\n"));
146 DEBUG(DEBUG_ERR,("Failed to get locks in ctdb_freeze_child\n"));
147 /* we didn't get the locks - destroy the handle */
152 h->ctdb->freeze_mode[h->priority] = CTDB_FREEZE_FROZEN;
154 /* notify the waiters */
155 if (h != h->ctdb->freeze_handles[h->priority]) {
156 DEBUG(DEBUG_ERR,("lockwait finished but h is not linked\n"));
158 while ((w = h->waiters)) {
160 DLIST_REMOVE(h->waiters, w);
166 create a child which gets locks on all the open databases, then calls the callback telling the parent
169 static struct ctdb_freeze_handle *ctdb_freeze_lock(struct ctdb_context *ctdb, uint32_t priority)
171 struct ctdb_freeze_handle *h;
173 struct fd_event *fde;
175 h = talloc_zero(ctdb, struct ctdb_freeze_handle);
176 CTDB_NO_MEMORY_NULL(ctdb, h);
179 h->priority = priority;
181 if (pipe(fd) == -1) {
182 DEBUG(DEBUG_ERR,("Failed to create pipe for ctdb_freeze_lock\n"));
188 if (h->child == -1) {
189 DEBUG(DEBUG_ERR,("Failed to fork child for ctdb_freeze_lock\n"));
200 ret = ctdb_lock_all_databases(ctdb, priority);
205 ret = write(fd[1], &ret, sizeof(ret));
206 if (ret != sizeof(ret)) {
207 DEBUG(DEBUG_ERR, (__location__ " Failed to write to socket from freeze child. ret:%d errno:%u\n", ret, errno));
213 if (kill(ctdb->ctdbd_pid, 0) != 0) {
214 DEBUG(DEBUG_ERR,("Parent died. Exiting lock wait child\n"));
221 talloc_set_destructor(h, ctdb_freeze_handle_destructor);
224 set_close_on_exec(fd[0]);
229 fde = event_add_fd(ctdb->ev, h, h->fd, EVENT_FD_READ|EVENT_FD_AUTOCLOSE,
230 ctdb_freeze_lock_handler, h);
232 DEBUG(DEBUG_ERR,("Failed to setup fd event for ctdb_freeze_lock\n"));
242 destroy a waiter for a freeze mode change
244 static int ctdb_freeze_waiter_destructor(struct ctdb_freeze_waiter *w)
246 ctdb_request_control_reply(w->ctdb, w->c, NULL, w->status, NULL);
251 start the freeze process for a certain priority
253 int ctdb_start_freeze(struct ctdb_context *ctdb, uint32_t priority)
256 DEBUG(DEBUG_ERR,("Freeze priority 0 requested, remapping to priority 1\n"));
260 if ((priority < 1) || (priority > NUM_DB_PRIORITIES)) {
261 DEBUG(DEBUG_ERR,(__location__ " Invalid db priority : %u\n", priority));
265 if (ctdb->freeze_mode[priority] == CTDB_FREEZE_FROZEN) {
266 /* we're already frozen */
270 /* if there isn't a freeze lock child then create one */
271 if (ctdb->freeze_handles[priority] == NULL) {
272 ctdb->freeze_handles[priority] = ctdb_freeze_lock(ctdb, priority);
273 CTDB_NO_MEMORY(ctdb, ctdb->freeze_handles[priority]);
274 ctdb->freeze_mode[priority] = CTDB_FREEZE_PENDING;
283 int32_t ctdb_control_freeze(struct ctdb_context *ctdb, struct ctdb_req_control *c, bool *async_reply)
285 struct ctdb_freeze_waiter *w;
288 priority = (uint32_t)c->srvid;
290 DEBUG(DEBUG_ERR, ("Freeze priority %u\n", priority));
293 DEBUG(DEBUG_ERR,("Freeze priority 0 requested, remapping to priority 1\n"));
297 if ((priority < 1) || (priority > NUM_DB_PRIORITIES)) {
298 DEBUG(DEBUG_ERR,(__location__ " Invalid db priority : %u\n", priority));
302 if (ctdb->freeze_mode[priority] == CTDB_FREEZE_FROZEN) {
303 /* we're already frozen */
307 if (ctdb_start_freeze(ctdb, priority) != 0) {
308 DEBUG(DEBUG_ERR,(__location__ " Failed to start freezing databases with priority %u\n", priority));
312 /* add ourselves to list of waiters */
313 if (ctdb->freeze_handles[priority] == NULL) {
314 DEBUG(DEBUG_ERR,("No freeze lock handle when adding a waiter\n"));
318 w = talloc(ctdb->freeze_handles[priority], struct ctdb_freeze_waiter);
319 CTDB_NO_MEMORY(ctdb, w);
321 w->c = talloc_steal(w, c);
322 w->priority = priority;
324 talloc_set_destructor(w, ctdb_freeze_waiter_destructor);
325 DLIST_ADD(ctdb->freeze_handles[priority]->waiters, w);
327 /* we won't reply till later */
334 block until we are frozen, used during daemon startup
336 bool ctdb_blocking_freeze(struct ctdb_context *ctdb)
340 for (i=1; i<=NUM_DB_PRIORITIES; i++) {
341 if (ctdb_start_freeze(ctdb, i)) {
342 DEBUG(DEBUG_ERR,(__location__ " Failed to freeze databases of prio %u\n", i));
346 /* block until frozen */
347 while (ctdb->freeze_mode[i] == CTDB_FREEZE_PENDING) {
348 event_loop_once(ctdb->ev);
356 static void thaw_priority(struct ctdb_context *ctdb, uint32_t priority)
358 DEBUG(DEBUG_ERR,("Thawing priority %u\n", priority));
360 /* cancel any pending transactions */
361 if (ctdb->freeze_transaction_started) {
362 struct ctdb_db_context *ctdb_db;
364 for (ctdb_db=ctdb->db_list;ctdb_db;ctdb_db=ctdb_db->next) {
365 tdb_add_flags(ctdb_db->ltdb->tdb, TDB_NOLOCK);
366 if (tdb_transaction_cancel(ctdb_db->ltdb->tdb) != 0) {
367 DEBUG(DEBUG_ERR,(__location__ " Failed to cancel transaction for db '%s'\n",
370 tdb_remove_flags(ctdb_db->ltdb->tdb, TDB_NOLOCK);
373 ctdb->freeze_transaction_started = false;
376 /* this hack can be used to get a copy of the databases at the end of a recovery */
377 system("mkdir -p /var/ctdb.saved; /usr/bin/rsync --delete -a /var/ctdb/ /var/ctdb.saved/$$ 2>&1 > /dev/null");
381 /* and this one for local testing */
382 system("mkdir -p test.db.saved; /usr/bin/rsync --delete -a test.db/ test.db.saved/$$ 2>&1 > /dev/null");
385 if (ctdb->freeze_handles[priority] != NULL) {
386 talloc_free(ctdb->freeze_handles[priority]);
387 ctdb->freeze_handles[priority] = NULL;
394 int32_t ctdb_control_thaw(struct ctdb_context *ctdb, uint32_t priority)
397 if (priority > NUM_DB_PRIORITIES) {
398 DEBUG(DEBUG_ERR,(__location__ " Invalid db priority : %u\n", priority));
404 for (i=1;i<=NUM_DB_PRIORITIES; i++) {
405 thaw_priority(ctdb, i);
408 thaw_priority(ctdb, priority);
411 ctdb_call_resend_all(ctdb);
417 start a transaction on all databases - used for recovery
419 int32_t ctdb_control_transaction_start(struct ctdb_context *ctdb, uint32_t id)
421 struct ctdb_db_context *ctdb_db;
424 for (i=1;i<=NUM_DB_PRIORITIES; i++) {
425 if (ctdb->freeze_mode[i] != CTDB_FREEZE_FROZEN) {
426 DEBUG(DEBUG_ERR,(__location__ " Failed transaction_start while not frozen\n"));
431 for (ctdb_db=ctdb->db_list;ctdb_db;ctdb_db=ctdb_db->next) {
434 tdb_add_flags(ctdb_db->ltdb->tdb, TDB_NOLOCK);
436 if (ctdb->freeze_transaction_started) {
437 if (tdb_transaction_cancel(ctdb_db->ltdb->tdb) != 0) {
438 DEBUG(DEBUG_ERR,(__location__ " Failed to cancel transaction for db '%s'\n",
440 /* not a fatal error */
444 ret = tdb_transaction_start(ctdb_db->ltdb->tdb);
446 tdb_remove_flags(ctdb_db->ltdb->tdb, TDB_NOLOCK);
449 DEBUG(DEBUG_ERR,(__location__ " Failed to start transaction for db '%s'\n",
455 ctdb->freeze_transaction_started = true;
456 ctdb->freeze_transaction_id = id;
462 cancel a transaction for all databases - used for recovery
464 int32_t ctdb_control_transaction_cancel(struct ctdb_context *ctdb)
466 struct ctdb_db_context *ctdb_db;
468 DEBUG(DEBUG_ERR,(__location__ " recovery transaction cancelled called\n"));
470 for (ctdb_db=ctdb->db_list;ctdb_db;ctdb_db=ctdb_db->next) {
471 tdb_add_flags(ctdb_db->ltdb->tdb, TDB_NOLOCK);
473 if (tdb_transaction_cancel(ctdb_db->ltdb->tdb) != 0) {
474 DEBUG(DEBUG_ERR,(__location__ " Failed to cancel transaction for db '%s'\n", ctdb_db->db_name));
475 /* not a fatal error */
478 tdb_remove_flags(ctdb_db->ltdb->tdb, TDB_NOLOCK);
481 ctdb->freeze_transaction_started = false;
487 commit transactions on all databases
489 int32_t ctdb_control_transaction_commit(struct ctdb_context *ctdb, uint32_t id)
491 struct ctdb_db_context *ctdb_db;
493 int healthy_nodes = 0;
495 for (i=1;i<=NUM_DB_PRIORITIES; i++) {
496 if (ctdb->freeze_mode[i] != CTDB_FREEZE_FROZEN) {
497 DEBUG(DEBUG_ERR,(__location__ " Failed transaction_start while not frozen\n"));
502 if (!ctdb->freeze_transaction_started) {
503 DEBUG(DEBUG_ERR,(__location__ " transaction not started\n"));
507 if (id != ctdb->freeze_transaction_id) {
508 DEBUG(DEBUG_ERR,(__location__ " incorrect transaction id 0x%x in commit\n", id));
512 DEBUG(DEBUG_DEBUG,(__location__ " num_nodes[%d]\n", ctdb->num_nodes));
513 for (i=0; i < ctdb->num_nodes; i++) {
514 DEBUG(DEBUG_DEBUG,(__location__ " node[%d].flags[0x%X]\n",
515 i, ctdb->nodes[i]->flags));
516 if (ctdb->nodes[i]->flags == 0) {
520 DEBUG(DEBUG_INFO,(__location__ " healthy_nodes[%d]\n", healthy_nodes));
522 for (ctdb_db=ctdb->db_list;ctdb_db;ctdb_db=ctdb_db->next) {
525 tdb_add_flags(ctdb_db->ltdb->tdb, TDB_NOLOCK);
526 ret = tdb_transaction_commit(ctdb_db->ltdb->tdb);
528 DEBUG(DEBUG_ERR,(__location__ " Failed to commit transaction for db '%s'. Cancel all transactions and resetting transaction_started to false.\n",
532 tdb_remove_flags(ctdb_db->ltdb->tdb, TDB_NOLOCK);
534 ret = ctdb_update_persistent_health(ctdb, ctdb_db, NULL, healthy_nodes);
536 DEBUG(DEBUG_CRIT,(__location__ " Failed to update persistent health for db '%s'. "
537 "Cancel all remaining transactions and resetting transaction_started to false.\n",
543 ctdb->freeze_transaction_started = false;
544 ctdb->freeze_transaction_id = 0;
549 /* cancel any pending transactions */
550 for (ctdb_db=ctdb->db_list;ctdb_db;ctdb_db=ctdb_db->next) {
551 tdb_add_flags(ctdb_db->ltdb->tdb, TDB_NOLOCK);
552 if (tdb_transaction_cancel(ctdb_db->ltdb->tdb) != 0) {
553 DEBUG(DEBUG_ERR,(__location__ " Failed to cancel transaction for db '%s'\n",
556 tdb_remove_flags(ctdb_db->ltdb->tdb, TDB_NOLOCK);
558 ctdb->freeze_transaction_started = false;
564 wipe a database - only possible when in a frozen transaction
566 int32_t ctdb_control_wipe_database(struct ctdb_context *ctdb, TDB_DATA indata)
568 struct ctdb_control_wipe_database w = *(struct ctdb_control_wipe_database *)indata.dptr;
569 struct ctdb_db_context *ctdb_db;
571 ctdb_db = find_ctdb_db(ctdb, w.db_id);
573 DEBUG(DEBUG_ERR,(__location__ " Unknown db 0x%x\n", w.db_id));
577 if (ctdb->freeze_mode[ctdb_db->priority] != CTDB_FREEZE_FROZEN) {
578 DEBUG(DEBUG_ERR,(__location__ " Failed transaction_start while not frozen\n"));
582 if (!ctdb->freeze_transaction_started) {
583 DEBUG(DEBUG_ERR,(__location__ " transaction not started\n"));
587 if (w.transaction_id != ctdb->freeze_transaction_id) {
588 DEBUG(DEBUG_ERR,(__location__ " incorrect transaction id 0x%x in commit\n", w.transaction_id));
592 if (tdb_wipe_all(ctdb_db->ltdb->tdb) != 0) {
593 DEBUG(DEBUG_ERR,(__location__ " Failed to wipe database for db '%s'\n",