2 ctdb ltdb code - server side
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/>.
21 #include "lib/events/events.h"
22 #include "lib/tdb/include/tdb.h"
23 #include "system/network.h"
24 #include "system/filesys.h"
25 #include "system/dir.h"
26 #include "../include/ctdb_private.h"
28 #include "lib/util/dlinklist.h"
31 this is the dummy null procedure that all databases support
33 static int ctdb_null_func(struct ctdb_call_info *call)
39 this is a plain fetch procedure that all databases support
41 static int ctdb_fetch_func(struct ctdb_call_info *call)
43 call->reply_data = &call->record_data;
49 struct lock_fetch_state {
50 struct ctdb_context *ctdb;
51 void (*recv_pkt)(void *, struct ctdb_req_header *);
53 struct ctdb_req_header *hdr;
55 bool ignore_generation;
59 called when we should retry the operation
61 static void lock_fetch_callback(void *p)
63 struct lock_fetch_state *state = talloc_get_type(p, struct lock_fetch_state);
64 if (!state->ignore_generation &&
65 state->generation != state->ctdb->vnn_map->generation) {
66 DEBUG(0,("Discarding previous generation lockwait packet\n"));
67 talloc_free(state->hdr);
70 state->recv_pkt(state->recv_context, state->hdr);
71 DEBUG(2,(__location__ " PACKET REQUEUED\n"));
76 do a non-blocking ltdb_lock, deferring this ctdb request until we
79 It does the following:
81 1) tries to get the chainlock. If it succeeds, then it returns 0
83 2) if it fails to get a chainlock immediately then it sets up a
84 non-blocking chainlock via ctdb_lockwait, and when it gets the
85 chainlock it re-submits this ctdb request to the main packet
88 This effectively queues all ctdb requests that cannot be
89 immediately satisfied until it can get the lock. This means that
90 the main ctdb daemon will not block waiting for a chainlock held by
93 There are 3 possible return values:
95 0: means that it got the lock immediately.
96 -1: means that it failed to get the lock, and won't retry
97 -2: means that it failed to get the lock immediately, but will retry
99 int ctdb_ltdb_lock_requeue(struct ctdb_db_context *ctdb_db,
100 TDB_DATA key, struct ctdb_req_header *hdr,
101 void (*recv_pkt)(void *, struct ctdb_req_header *),
102 void *recv_context, bool ignore_generation)
105 struct tdb_context *tdb = ctdb_db->ltdb->tdb;
106 struct lockwait_handle *h;
107 struct lock_fetch_state *state;
109 ret = tdb_chainlock_nonblock(tdb, key);
112 !(errno == EACCES || errno == EAGAIN || errno == EDEADLK)) {
113 /* a hard failure - don't try again */
117 /* when torturing, ensure we test the contended path */
118 if ((ctdb_db->ctdb->flags & CTDB_FLAG_TORTURE) &&
121 tdb_chainunlock(tdb, key);
124 /* first the non-contended path */
129 state = talloc(hdr, struct lock_fetch_state);
130 state->ctdb = ctdb_db->ctdb;
132 state->recv_pkt = recv_pkt;
133 state->recv_context = recv_context;
134 state->generation = ctdb_db->ctdb->vnn_map->generation;
135 state->ignore_generation = ignore_generation;
137 /* now the contended path */
138 h = ctdb_lockwait(ctdb_db, key, lock_fetch_callback, state);
140 tdb_chainunlock(tdb, key);
144 /* we need to move the packet off the temporary context in ctdb_input_pkt(),
145 so it won't be freed yet */
146 talloc_steal(state, hdr);
147 talloc_steal(state, h);
149 /* now tell the caller than we will retry asynchronously */
154 a varient of ctdb_ltdb_lock_requeue that also fetches the record
156 int ctdb_ltdb_lock_fetch_requeue(struct ctdb_db_context *ctdb_db,
157 TDB_DATA key, struct ctdb_ltdb_header *header,
158 struct ctdb_req_header *hdr, TDB_DATA *data,
159 void (*recv_pkt)(void *, struct ctdb_req_header *),
160 void *recv_context, bool ignore_generation)
164 ret = ctdb_ltdb_lock_requeue(ctdb_db, key, hdr, recv_pkt,
165 recv_context, ignore_generation);
167 ret = ctdb_ltdb_fetch(ctdb_db, key, header, hdr, data);
169 ctdb_ltdb_unlock(ctdb_db, key);
177 paraoid check to see if the db is empty
179 static void ctdb_check_db_empty(struct ctdb_db_context *ctdb_db)
181 struct tdb_context *tdb = ctdb_db->ltdb->tdb;
182 int count = tdb_traverse_read(tdb, NULL, NULL);
184 DEBUG(0,(__location__ " tdb '%s' not empty on attach! aborting\n",
186 ctdb_fatal(ctdb_db->ctdb, "database not empty on attach");
192 attach to a database, handling both persistent and non-persistent databases
193 return 0 on success, -1 on failure
195 static int ctdb_local_attach(struct ctdb_context *ctdb, const char *db_name, bool persistent)
197 struct ctdb_db_context *ctdb_db, *tmp_db;
201 ctdb_db = talloc_zero(ctdb, struct ctdb_db_context);
202 CTDB_NO_MEMORY(ctdb, ctdb_db);
204 ctdb_db->ctdb = ctdb;
205 ctdb_db->db_name = talloc_strdup(ctdb_db, db_name);
206 CTDB_NO_MEMORY(ctdb, ctdb_db->db_name);
208 key.dsize = strlen(db_name)+1;
209 key.dptr = discard_const(db_name);
210 ctdb_db->db_id = ctdb_hash(&key);
211 ctdb_db->persistent = persistent;
213 /* check for hash collisions */
214 for (tmp_db=ctdb->db_list;tmp_db;tmp_db=tmp_db->next) {
215 if (tmp_db->db_id == ctdb_db->db_id) {
216 DEBUG(0,("db_id 0x%x hash collision. name1='%s' name2='%s'\n",
217 tmp_db->db_id, db_name, tmp_db->db_name));
218 talloc_free(ctdb_db);
223 if (ctdb->db_directory == NULL) {
224 ctdb->db_directory = VARDIR "/ctdb";
227 /* make sure the db directory exists */
228 if (mkdir(ctdb->db_directory, 0700) == -1 && errno != EEXIST) {
229 DEBUG(0,(__location__ " Unable to create ctdb directory '%s'\n",
230 ctdb->db_directory));
231 talloc_free(ctdb_db);
235 if (persistent && mkdir(ctdb->db_directory_persistent, 0700) == -1 && errno != EEXIST) {
236 DEBUG(0,(__location__ " Unable to create ctdb persistent directory '%s'\n",
237 ctdb->db_directory_persistent));
238 talloc_free(ctdb_db);
242 /* open the database */
243 ctdb_db->db_path = talloc_asprintf(ctdb_db, "%s/%s.%u",
244 persistent?ctdb->db_directory_persistent:ctdb->db_directory,
247 ctdb_db->ltdb = tdb_wrap_open(ctdb, ctdb_db->db_path,
248 ctdb->tunable.database_hash_size,
249 persistent?TDB_DEFAULT:TDB_CLEAR_IF_FIRST,
250 O_CREAT|O_RDWR, 0666);
251 if (ctdb_db->ltdb == NULL) {
252 DEBUG(0,("Failed to open tdb '%s'\n", ctdb_db->db_path));
253 talloc_free(ctdb_db);
258 ctdb_check_db_empty(ctdb_db);
261 DLIST_ADD(ctdb->db_list, ctdb_db);
264 all databases support the "null" function. we need this in
265 order to do forced migration of records
267 ret = ctdb_daemon_set_call(ctdb, ctdb_db->db_id, ctdb_null_func, CTDB_NULL_FUNC);
269 DEBUG(0,("Failed to setup null function for '%s'\n", ctdb_db->db_name));
270 talloc_free(ctdb_db);
275 all databases support the "fetch" function. we need this
276 for efficient Samba3 ctdb fetch
278 ret = ctdb_daemon_set_call(ctdb, ctdb_db->db_id, ctdb_fetch_func, CTDB_FETCH_FUNC);
280 DEBUG(0,("Failed to setup fetch function for '%s'\n", ctdb_db->db_name));
281 talloc_free(ctdb_db);
285 DEBUG(1,("Attached to database '%s'\n", ctdb_db->db_path));
293 a client has asked to attach a new database
295 int32_t ctdb_control_db_attach(struct ctdb_context *ctdb, TDB_DATA indata,
296 TDB_DATA *outdata, bool persistent)
298 const char *db_name = (const char *)indata.dptr;
299 struct ctdb_db_context *db;
300 struct ctdb_node *node = ctdb->nodes[ctdb->pnn];
302 /* If the node is inactive it is not part of the cluster
303 and we should not allow clients to attach to any
306 if (node->flags & NODE_FLAGS_INACTIVE) {
307 DEBUG(0,("DB Attach to database %s refused since node is inactive (disconnected or banned)\n", db_name));
312 /* see if we already have this name */
313 db = ctdb_db_handle(ctdb, db_name);
315 outdata->dptr = (uint8_t *)&db->db_id;
316 outdata->dsize = sizeof(db->db_id);
320 if (ctdb_local_attach(ctdb, db_name, persistent) != 0) {
324 db = ctdb_db_handle(ctdb, db_name);
326 DEBUG(0,("Failed to find db handle for name '%s'\n", db_name));
330 outdata->dptr = (uint8_t *)&db->db_id;
331 outdata->dsize = sizeof(db->db_id);
333 /* tell all the other nodes about this database */
334 ctdb_daemon_send_control(ctdb, CTDB_BROADCAST_ALL, 0,
335 CTDB_CONTROL_DB_ATTACH, 0, CTDB_CTRL_FLAG_NOREPLY,
344 attach to all existing persistent databases
346 int ctdb_attach_persistent(struct ctdb_context *ctdb)
351 /* open the persistent db directory and scan it for files */
352 d = opendir(ctdb->db_directory_persistent);
357 while ((de=readdir(d))) {
359 size_t len = strlen(de->d_name);
362 s = talloc_strdup(ctdb, de->d_name);
363 CTDB_NO_MEMORY(ctdb, s);
365 /* only accept names ending in .tdb */
366 p = strstr(s, ".tdb.");
367 if (len < 7 || p == NULL) {
371 if (sscanf(p+5, "%u", &node) != 1 || node != ctdb->pnn) {
377 if (ctdb_local_attach(ctdb, s, true) != 0) {
378 DEBUG(0,("Failed to attach to persistent database '%s'\n", de->d_name));
383 DEBUG(0,("Attached to persistent database %s\n", s));
392 called when a broadcast seqnum update comes in
394 int32_t ctdb_ltdb_update_seqnum(struct ctdb_context *ctdb, uint32_t db_id, uint32_t srcnode)
396 struct ctdb_db_context *ctdb_db;
397 if (srcnode == ctdb->pnn) {
398 /* don't update ourselves! */
402 ctdb_db = find_ctdb_db(ctdb, db_id);
404 DEBUG(0,("Unknown db_id 0x%x in ctdb_ltdb_update_seqnum\n", db_id));
408 tdb_increment_seqnum_nonblock(ctdb_db->ltdb->tdb);
409 ctdb_db->seqnum = tdb_get_seqnum(ctdb_db->ltdb->tdb);
414 timer to check for seqnum changes in a ltdb and propogate them
416 static void ctdb_ltdb_seqnum_check(struct event_context *ev, struct timed_event *te,
417 struct timeval t, void *p)
419 struct ctdb_db_context *ctdb_db = talloc_get_type(p, struct ctdb_db_context);
420 struct ctdb_context *ctdb = ctdb_db->ctdb;
421 uint32_t new_seqnum = tdb_get_seqnum(ctdb_db->ltdb->tdb);
422 if (new_seqnum != ctdb_db->seqnum) {
423 /* something has changed - propogate it */
425 data.dptr = (uint8_t *)&ctdb_db->db_id;
426 data.dsize = sizeof(uint32_t);
427 ctdb_daemon_send_control(ctdb, CTDB_BROADCAST_VNNMAP, 0,
428 CTDB_CONTROL_UPDATE_SEQNUM, 0, CTDB_CTRL_FLAG_NOREPLY,
431 ctdb_db->seqnum = new_seqnum;
433 /* setup a new timer */
435 event_add_timed(ctdb->ev, ctdb_db,
436 timeval_current_ofs(ctdb->tunable.seqnum_frequency, 0),
437 ctdb_ltdb_seqnum_check, ctdb_db);
441 enable seqnum handling on this db
443 int32_t ctdb_ltdb_enable_seqnum(struct ctdb_context *ctdb, uint32_t db_id)
445 struct ctdb_db_context *ctdb_db;
446 ctdb_db = find_ctdb_db(ctdb, db_id);
448 DEBUG(0,("Unknown db_id 0x%x in ctdb_ltdb_enable_seqnum\n", db_id));
452 if (ctdb_db->te == NULL) {
454 event_add_timed(ctdb->ev, ctdb_db,
455 timeval_current_ofs(ctdb->tunable.seqnum_frequency, 0),
456 ctdb_ltdb_seqnum_check, ctdb_db);
459 tdb_enable_seqnum(ctdb_db->ltdb->tdb);
460 ctdb_db->seqnum = tdb_get_seqnum(ctdb_db->ltdb->tdb);