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)
35 struct ctdb_db_context *ctdb_db;
36 for (ctdb_db=ctdb->db_list;ctdb_db;ctdb_db=ctdb_db->next) {
37 if (tdb_lockall(ctdb_db->ltdb->tdb) != 0) {
45 a list of control requests waiting for a freeze lock child to get
48 struct ctdb_freeze_waiter {
49 struct ctdb_freeze_waiter *next, *prev;
50 struct ctdb_context *ctdb;
51 struct ctdb_req_control *c;
55 /* a handle to a freeze lock child process */
56 struct ctdb_freeze_handle {
57 struct ctdb_context *ctdb;
60 struct ctdb_freeze_waiter *waiters;
61 bool transaction_started;
62 uint32_t transaction_id;
66 destroy a freeze handle
68 static int ctdb_freeze_handle_destructor(struct ctdb_freeze_handle *h)
70 h->ctdb->freeze_mode = CTDB_FREEZE_NONE;
71 kill(h->child, SIGKILL);
72 waitpid(h->child, NULL, 0);
77 called when the child writes its status to us
79 static void ctdb_freeze_lock_handler(struct event_context *ev, struct fd_event *fde,
80 uint16_t flags, void *private_data)
82 struct ctdb_freeze_handle *h = talloc_get_type(private_data, struct ctdb_freeze_handle);
84 struct ctdb_freeze_waiter *w;
86 if (h->ctdb->freeze_mode == CTDB_FREEZE_FROZEN) {
87 DEBUG(DEBUG_INFO,("freeze child died - unfreezing\n"));
92 if (read(h->fd, &status, sizeof(status)) != sizeof(status)) {
93 DEBUG(DEBUG_ERR,("read error from freeze lock child\n"));
98 DEBUG(DEBUG_ERR,("Failed to get locks in ctdb_freeze_child\n"));
99 /* we didn't get the locks - destroy the handle */
104 h->ctdb->freeze_mode = CTDB_FREEZE_FROZEN;
106 /* notify the waiters */
107 while ((w = h->ctdb->freeze_handle->waiters)) {
109 DLIST_REMOVE(h->ctdb->freeze_handle->waiters, w);
115 create a child which gets locks on all the open databases, then calls the callback telling the parent
118 static struct ctdb_freeze_handle *ctdb_freeze_lock(struct ctdb_context *ctdb)
120 struct ctdb_freeze_handle *h;
122 struct fd_event *fde;
124 h = talloc_zero(ctdb, struct ctdb_freeze_handle);
125 CTDB_NO_MEMORY_VOID(ctdb, h);
129 /* use socketpair() instead of pipe() so we have bi-directional fds */
130 if (socketpair(AF_UNIX, SOCK_STREAM, 0, fd) != 0) {
131 DEBUG(DEBUG_ERR,("Failed to create pipe for ctdb_freeze_lock\n"));
137 if (h->child == -1) {
138 DEBUG(DEBUG_ERR,("Failed to fork child for ctdb_freeze_lock\n"));
147 ret = ctdb_lock_all_databases(ctdb);
151 write(fd[1], &ret, sizeof(ret));
152 /* the read here means we will die if the parent exits */
153 read(fd[1], &ret, sizeof(ret));
157 talloc_set_destructor(h, ctdb_freeze_handle_destructor);
163 fde = event_add_fd(ctdb->ev, h, h->fd, EVENT_FD_READ|EVENT_FD_AUTOCLOSE,
164 ctdb_freeze_lock_handler, h);
166 DEBUG(DEBUG_ERR,("Failed to setup fd event for ctdb_freeze_lock\n"));
176 destroy a waiter for a freeze mode change
178 static int ctdb_freeze_waiter_destructor(struct ctdb_freeze_waiter *w)
180 DLIST_REMOVE(w->ctdb->freeze_handle->waiters, w);
181 ctdb_request_control_reply(w->ctdb, w->c, NULL, w->status, NULL);
186 start the freeze process
188 void ctdb_start_freeze(struct ctdb_context *ctdb)
190 if (ctdb->freeze_mode == CTDB_FREEZE_FROZEN) {
191 /* we're already frozen */
195 /* if there isn't a freeze lock child then create one */
196 if (!ctdb->freeze_handle) {
197 ctdb->freeze_handle = ctdb_freeze_lock(ctdb);
198 CTDB_NO_MEMORY_VOID(ctdb, ctdb->freeze_handle);
199 ctdb->freeze_mode = CTDB_FREEZE_PENDING;
206 int32_t ctdb_control_freeze(struct ctdb_context *ctdb, struct ctdb_req_control *c, bool *async_reply)
208 struct ctdb_freeze_waiter *w;
210 if (ctdb->freeze_mode == CTDB_FREEZE_FROZEN) {
211 /* we're already frozen */
215 ctdb_start_freeze(ctdb);
217 /* add ourselves to list of waiters */
218 w = talloc(ctdb->freeze_handle, struct ctdb_freeze_waiter);
219 CTDB_NO_MEMORY(ctdb, w);
221 w->c = talloc_steal(w, c);
223 talloc_set_destructor(w, ctdb_freeze_waiter_destructor);
224 DLIST_ADD(ctdb->freeze_handle->waiters, w);
226 /* we won't reply till later */
233 block until we are frozen, used during daemon startup
235 bool ctdb_blocking_freeze(struct ctdb_context *ctdb)
237 ctdb_start_freeze(ctdb);
239 /* block until frozen */
240 while (ctdb->freeze_mode == CTDB_FREEZE_PENDING) {
241 event_loop_once(ctdb->ev);
244 return ctdb->freeze_mode == CTDB_FREEZE_FROZEN;
252 int32_t ctdb_control_thaw(struct ctdb_context *ctdb)
254 /* cancel any pending transactions */
255 if (ctdb->freeze_handle && ctdb->freeze_handle->transaction_started) {
256 struct ctdb_db_context *ctdb_db;
258 for (ctdb_db=ctdb->db_list;ctdb_db;ctdb_db=ctdb_db->next) {
259 tdb_add_flags(ctdb_db->ltdb->tdb, TDB_NOLOCK);
260 if (tdb_transaction_cancel(ctdb_db->ltdb->tdb) != 0) {
261 DEBUG(DEBUG_ERR,(__location__ " Failed to cancel transaction for db '%s'\n",
264 tdb_remove_flags(ctdb_db->ltdb->tdb, TDB_NOLOCK);
269 /* this hack can be used to get a copy of the databases at the end of a recovery */
270 system("mkdir -p /var/ctdb.saved; /usr/bin/rsync --delete -a /var/ctdb/ /var/ctdb.saved/$$ 2>&1 > /dev/null");
274 /* and this one for local testing */
275 system("mkdir -p test.db.saved; /usr/bin/rsync --delete -a test.db/ test.db.saved/$$ 2>&1 > /dev/null");
279 talloc_free(ctdb->freeze_handle);
280 ctdb->freeze_handle = NULL;
281 ctdb_call_resend_all(ctdb);
287 start a transaction on all databases - used for recovery
289 int32_t ctdb_control_transaction_start(struct ctdb_context *ctdb, uint32_t id)
291 struct ctdb_db_context *ctdb_db;
293 if (ctdb->freeze_mode != CTDB_FREEZE_FROZEN) {
294 DEBUG(DEBUG_ERR,(__location__ " Failed transaction_start while not frozen\n"));
299 for (ctdb_db=ctdb->db_list;ctdb_db;ctdb_db=ctdb_db->next) {
302 tdb_add_flags(ctdb_db->ltdb->tdb, TDB_NOLOCK);
304 if (ctdb->freeze_handle->transaction_started) {
305 if (tdb_transaction_cancel(ctdb_db->ltdb->tdb) != 0) {
306 DEBUG(DEBUG_ERR,(__location__ " Failed to cancel transaction for db '%s'\n",
308 /* not a fatal error */
312 ret = tdb_transaction_start(ctdb_db->ltdb->tdb);
314 tdb_remove_flags(ctdb_db->ltdb->tdb, TDB_NOLOCK);
317 DEBUG(DEBUG_ERR,(__location__ " Failed to start transaction for db '%s'\n",
323 ctdb->freeze_handle->transaction_started = true;
324 ctdb->freeze_handle->transaction_id = id;
330 commit transactions on all databases
332 int32_t ctdb_control_transaction_commit(struct ctdb_context *ctdb, uint32_t id)
334 struct ctdb_db_context *ctdb_db;
336 if (ctdb->freeze_mode != CTDB_FREEZE_FROZEN) {
337 DEBUG(DEBUG_ERR,(__location__ " Failed transaction_start while not frozen\n"));
341 if (!ctdb->freeze_handle->transaction_started) {
342 DEBUG(DEBUG_ERR,(__location__ " transaction not started\n"));
346 if (id != ctdb->freeze_handle->transaction_id) {
347 DEBUG(DEBUG_ERR,(__location__ " incorrect transaction id 0x%x in commit\n", id));
351 for (ctdb_db=ctdb->db_list;ctdb_db;ctdb_db=ctdb_db->next) {
352 tdb_add_flags(ctdb_db->ltdb->tdb, TDB_NOLOCK);
353 if (tdb_transaction_commit(ctdb_db->ltdb->tdb) != 0) {
354 DEBUG(DEBUG_ERR,(__location__ " Failed to commit transaction for db '%s'\n",
356 /* this has to be fatal to maintain integrity - it should only
357 happen if we run out of disk space */
358 ctdb_fatal(ctdb, "Unable to commit transactions\n");
361 tdb_remove_flags(ctdb_db->ltdb->tdb, TDB_NOLOCK);
364 ctdb->freeze_handle->transaction_started = false;
365 ctdb->freeze_handle->transaction_id = 0;
371 wipe a database - only possible when in a frozen transaction
373 int32_t ctdb_control_wipe_database(struct ctdb_context *ctdb, TDB_DATA indata)
375 struct ctdb_control_wipe_database w = *(struct ctdb_control_wipe_database *)indata.dptr;
376 struct ctdb_db_context *ctdb_db;
378 if (ctdb->freeze_mode != CTDB_FREEZE_FROZEN) {
379 DEBUG(DEBUG_ERR,(__location__ " Failed transaction_start while not frozen\n"));
383 if (!ctdb->freeze_handle->transaction_started) {
384 DEBUG(DEBUG_ERR,(__location__ " transaction not started\n"));
388 if (w.transaction_id != ctdb->freeze_handle->transaction_id) {
389 DEBUG(DEBUG_ERR,(__location__ " incorrect transaction id 0x%x in commit\n", w.transaction_id));
393 ctdb_db = find_ctdb_db(ctdb, w.db_id);
395 DEBUG(DEBUG_ERR,(__location__ " Unknown db 0x%x\n", w.db_id));
399 if (tdb_wipe_all(ctdb_db->ltdb->tdb) != 0) {
400 DEBUG(DEBUG_ERR,(__location__ " Failed to wipe database for db '%s'\n",