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"));
88 if (h->ctdb->freeze_handle == h) {
89 h->ctdb->freeze_handle = NULL;
95 if (read(h->fd, &status, sizeof(status)) != sizeof(status)) {
96 DEBUG(DEBUG_ERR,("read error from freeze lock child\n"));
101 DEBUG(DEBUG_ERR,("Failed to get locks in ctdb_freeze_child\n"));
102 /* we didn't get the locks - destroy the handle */
107 h->ctdb->freeze_mode = CTDB_FREEZE_FROZEN;
109 /* notify the waiters */
110 while ((w = h->ctdb->freeze_handle->waiters)) {
112 DLIST_REMOVE(h->ctdb->freeze_handle->waiters, w);
118 create a child which gets locks on all the open databases, then calls the callback telling the parent
121 static struct ctdb_freeze_handle *ctdb_freeze_lock(struct ctdb_context *ctdb)
123 struct ctdb_freeze_handle *h;
125 struct fd_event *fde;
127 h = talloc_zero(ctdb, struct ctdb_freeze_handle);
128 CTDB_NO_MEMORY_NULL(ctdb, h);
132 /* use socketpair() instead of pipe() so we have bi-directional fds */
133 if (socketpair(AF_UNIX, SOCK_STREAM, 0, fd) != 0) {
134 DEBUG(DEBUG_ERR,("Failed to create pipe for ctdb_freeze_lock\n"));
140 if (h->child == -1) {
141 DEBUG(DEBUG_ERR,("Failed to fork child for ctdb_freeze_lock\n"));
151 ret = ctdb_lock_all_databases(ctdb);
158 while (count++ < 30) {
159 ret = write(fd[1], &ret, sizeof(ret));
160 if (ret == sizeof(ret)) {
163 DEBUG(DEBUG_ERR, (__location__ " Failed to write to socket from freeze child. ret:%d errno:%u\n", ret, errno));
167 DEBUG(DEBUG_ERR, (__location__ " Failed to write to socket from freeze child. Aborting freeze child\n"));
171 /* the read here means we will die if the parent exits */
172 read(fd[1], &ret, sizeof(ret));
176 talloc_set_destructor(h, ctdb_freeze_handle_destructor);
182 fde = event_add_fd(ctdb->ev, h, h->fd, EVENT_FD_READ|EVENT_FD_AUTOCLOSE,
183 ctdb_freeze_lock_handler, h);
185 DEBUG(DEBUG_ERR,("Failed to setup fd event for ctdb_freeze_lock\n"));
195 destroy a waiter for a freeze mode change
197 static int ctdb_freeze_waiter_destructor(struct ctdb_freeze_waiter *w)
199 DLIST_REMOVE(w->ctdb->freeze_handle->waiters, w);
200 ctdb_request_control_reply(w->ctdb, w->c, NULL, w->status, NULL);
205 start the freeze process
207 void ctdb_start_freeze(struct ctdb_context *ctdb)
209 if (ctdb->freeze_mode == CTDB_FREEZE_FROZEN) {
210 /* we're already frozen */
214 /* if there isn't a freeze lock child then create one */
215 if (!ctdb->freeze_handle) {
216 ctdb->freeze_handle = ctdb_freeze_lock(ctdb);
217 CTDB_NO_MEMORY_VOID(ctdb, ctdb->freeze_handle);
218 ctdb->freeze_mode = CTDB_FREEZE_PENDING;
225 int32_t ctdb_control_freeze(struct ctdb_context *ctdb, struct ctdb_req_control *c, bool *async_reply)
227 struct ctdb_freeze_waiter *w;
229 if (ctdb->freeze_mode == CTDB_FREEZE_FROZEN) {
230 /* we're already frozen */
234 ctdb_start_freeze(ctdb);
236 /* add ourselves to list of waiters */
237 w = talloc(ctdb->freeze_handle, struct ctdb_freeze_waiter);
238 CTDB_NO_MEMORY(ctdb, w);
240 w->c = talloc_steal(w, c);
242 talloc_set_destructor(w, ctdb_freeze_waiter_destructor);
243 DLIST_ADD(ctdb->freeze_handle->waiters, w);
245 /* we won't reply till later */
252 block until we are frozen, used during daemon startup
254 bool ctdb_blocking_freeze(struct ctdb_context *ctdb)
256 ctdb_start_freeze(ctdb);
258 /* block until frozen */
259 while (ctdb->freeze_mode == CTDB_FREEZE_PENDING) {
260 event_loop_once(ctdb->ev);
263 return ctdb->freeze_mode == CTDB_FREEZE_FROZEN;
271 int32_t ctdb_control_thaw(struct ctdb_context *ctdb)
273 /* cancel any pending transactions */
274 if (ctdb->freeze_handle && ctdb->freeze_handle->transaction_started) {
275 struct ctdb_db_context *ctdb_db;
277 for (ctdb_db=ctdb->db_list;ctdb_db;ctdb_db=ctdb_db->next) {
278 tdb_add_flags(ctdb_db->ltdb->tdb, TDB_NOLOCK);
279 if (tdb_transaction_cancel(ctdb_db->ltdb->tdb) != 0) {
280 DEBUG(DEBUG_ERR,(__location__ " Failed to cancel transaction for db '%s'\n",
283 tdb_remove_flags(ctdb_db->ltdb->tdb, TDB_NOLOCK);
288 /* this hack can be used to get a copy of the databases at the end of a recovery */
289 system("mkdir -p /var/ctdb.saved; /usr/bin/rsync --delete -a /var/ctdb/ /var/ctdb.saved/$$ 2>&1 > /dev/null");
293 /* and this one for local testing */
294 system("mkdir -p test.db.saved; /usr/bin/rsync --delete -a test.db/ test.db.saved/$$ 2>&1 > /dev/null");
298 talloc_free(ctdb->freeze_handle);
299 ctdb->freeze_handle = NULL;
300 ctdb_call_resend_all(ctdb);
306 start a transaction on all databases - used for recovery
308 int32_t ctdb_control_transaction_start(struct ctdb_context *ctdb, uint32_t id)
310 struct ctdb_db_context *ctdb_db;
312 if (ctdb->freeze_mode != CTDB_FREEZE_FROZEN) {
313 DEBUG(DEBUG_ERR,(__location__ " Failed transaction_start while not frozen\n"));
318 for (ctdb_db=ctdb->db_list;ctdb_db;ctdb_db=ctdb_db->next) {
321 tdb_add_flags(ctdb_db->ltdb->tdb, TDB_NOLOCK);
323 if (ctdb->freeze_handle->transaction_started) {
324 if (tdb_transaction_cancel(ctdb_db->ltdb->tdb) != 0) {
325 DEBUG(DEBUG_ERR,(__location__ " Failed to cancel transaction for db '%s'\n",
327 /* not a fatal error */
331 ret = tdb_transaction_start(ctdb_db->ltdb->tdb);
333 tdb_remove_flags(ctdb_db->ltdb->tdb, TDB_NOLOCK);
336 DEBUG(DEBUG_ERR,(__location__ " Failed to start transaction for db '%s'\n",
342 ctdb->freeze_handle->transaction_started = true;
343 ctdb->freeze_handle->transaction_id = id;
349 commit transactions on all databases
351 int32_t ctdb_control_transaction_commit(struct ctdb_context *ctdb, uint32_t id)
353 struct ctdb_db_context *ctdb_db;
355 if (ctdb->freeze_mode != CTDB_FREEZE_FROZEN) {
356 DEBUG(DEBUG_ERR,(__location__ " Failed transaction_start while not frozen\n"));
360 if (!ctdb->freeze_handle->transaction_started) {
361 DEBUG(DEBUG_ERR,(__location__ " transaction not started\n"));
365 if (id != ctdb->freeze_handle->transaction_id) {
366 DEBUG(DEBUG_ERR,(__location__ " incorrect transaction id 0x%x in commit\n", id));
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_commit(ctdb_db->ltdb->tdb) != 0) {
373 DEBUG(DEBUG_ERR,(__location__ " Failed to commit transaction for db '%s'\n",
375 /* this has to be fatal to maintain integrity - it should only
376 happen if we run out of disk space */
377 ctdb_fatal(ctdb, "Unable to commit transactions\n");
380 tdb_remove_flags(ctdb_db->ltdb->tdb, TDB_NOLOCK);
383 ctdb->freeze_handle->transaction_started = false;
384 ctdb->freeze_handle->transaction_id = 0;
390 wipe a database - only possible when in a frozen transaction
392 int32_t ctdb_control_wipe_database(struct ctdb_context *ctdb, TDB_DATA indata)
394 struct ctdb_control_wipe_database w = *(struct ctdb_control_wipe_database *)indata.dptr;
395 struct ctdb_db_context *ctdb_db;
397 if (ctdb->freeze_mode != CTDB_FREEZE_FROZEN) {
398 DEBUG(DEBUG_ERR,(__location__ " Failed transaction_start while not frozen\n"));
402 if (!ctdb->freeze_handle->transaction_started) {
403 DEBUG(DEBUG_ERR,(__location__ " transaction not started\n"));
407 if (w.transaction_id != ctdb->freeze_handle->transaction_id) {
408 DEBUG(DEBUG_ERR,(__location__ " incorrect transaction id 0x%x in commit\n", w.transaction_id));
412 ctdb_db = find_ctdb_db(ctdb, w.db_id);
414 DEBUG(DEBUG_ERR,(__location__ " Unknown db 0x%x\n", w.db_id));
418 if (tdb_wipe_all(ctdb_db->ltdb->tdb) != 0) {
419 DEBUG(DEBUG_ERR,(__location__ " Failed to wipe database for db '%s'\n",