2 Unix SMB/CIFS implementation.
3 Database interface wrapper around ctdbd
4 Copyright (C) Volker Lendecke 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 #ifdef CLUSTER_SUPPORT
23 #include "ctdb_private.h"
24 #include "ctdbd_conn.h"
27 struct tdb_wrap *wtdb;
32 struct db_ctdb_ctx *ctdb_ctx;
33 struct ctdb_ltdb_header header;
36 static struct db_record *fetch_locked_internal(struct db_ctdb_ctx *ctx,
41 static NTSTATUS tdb_error_to_ntstatus(struct tdb_context *tdb)
44 enum TDB_ERROR tret = tdb_error(tdb);
48 status = NT_STATUS_OBJECT_NAME_COLLISION;
51 status = NT_STATUS_OBJECT_NAME_NOT_FOUND;
54 status = NT_STATUS_INTERNAL_DB_CORRUPTION;
61 static NTSTATUS db_ctdb_store(struct db_record *rec, TDB_DATA data, int flag)
63 struct db_ctdb_rec *crec = talloc_get_type_abort(
64 rec->private_data, struct db_ctdb_rec);
68 cdata.dsize = sizeof(crec->header) + data.dsize;
70 if (!(cdata.dptr = SMB_MALLOC_ARRAY(uint8, cdata.dsize))) {
71 return NT_STATUS_NO_MEMORY;
74 memcpy(cdata.dptr, &crec->header, sizeof(crec->header));
75 memcpy(cdata.dptr + sizeof(crec->header), data.dptr, data.dsize);
77 ret = tdb_store(crec->ctdb_ctx->wtdb->tdb, rec->key, cdata, TDB_REPLACE);
79 SAFE_FREE(cdata.dptr);
81 return (ret == 0) ? NT_STATUS_OK
82 : tdb_error_to_ntstatus(crec->ctdb_ctx->wtdb->tdb);
86 /* for persistent databases the store is a bit different. We have to
87 ask the ctdb daemon to push the record to all nodes after the
89 static NTSTATUS db_ctdb_store_persistent(struct db_record *rec, TDB_DATA data, int flag)
91 struct db_ctdb_rec *crec;
92 struct db_record *record;
97 int max_retries = lp_parm_int(-1, "dbwrap ctdb", "max store retries", 5);
99 for (count = 0, status = NT_STATUS_UNSUCCESSFUL, record = rec;
100 (count < max_retries) && !NT_STATUS_IS_OK(status);
106 * There is a hack here: We use rec as a memory
107 * context and re-use it as the record struct ptr.
108 * We don't free the record data allocated
109 * in each turn. So all gets freed when the caller
110 * releases the original record. This is because
111 * we don't get the record passed in by reference
112 * in the first place and the caller relies on
113 * having to free the record himself.
115 record = fetch_locked_internal(crec->ctdb_ctx,
118 true /* persistent */);
119 if (record == NULL) {
120 DEBUG(5, ("fetch_locked_internal failed.\n"));
121 status = NT_STATUS_NO_MEMORY;
126 crec = talloc_get_type_abort(record->private_data,
129 cdata.dsize = sizeof(crec->header) + data.dsize;
131 if (!(cdata.dptr = SMB_MALLOC_ARRAY(uint8, cdata.dsize))) {
132 return NT_STATUS_NO_MEMORY;
137 memcpy(cdata.dptr, &crec->header, sizeof(crec->header));
138 memcpy(cdata.dptr + sizeof(crec->header), data.dptr, data.dsize);
140 status = ctdbd_start_persistent_update(
141 messaging_ctdbd_connection(),
142 crec->ctdb_ctx->db_id,
146 if (NT_STATUS_IS_OK(status)) {
147 ret = tdb_store(crec->ctdb_ctx->wtdb->tdb, rec->key,
151 : tdb_error_to_ntstatus(
152 crec->ctdb_ctx->wtdb->tdb);
156 * release the lock *now* in order to prevent deadlocks.
158 * There is a tradeoff: Usually, the record is still locked
159 * after db->store operation. This lock is usually released
160 * via the talloc destructor with the TALLOC_FREE to
161 * the record. So we have two choices:
163 * - Either re-lock the record after the call to persistent_store
164 * or cancel_persistent update and this way not changing any
165 * assumptions callers may have about the state, but possibly
166 * introducing new race conditions.
168 * - Or don't lock the record again but just remove the
169 * talloc_destructor. This is less racy but assumes that
170 * the lock is always released via TALLOC_FREE of the record.
172 * I choose the first variant for now since it seems less racy.
173 * We can't guarantee that we succeed in getting the lock
174 * anyways. The only real danger here is that a caller
175 * performs multiple store operations after a fetch_locked()
176 * which is currently not the case.
178 tdb_chainunlock(crec->ctdb_ctx->wtdb->tdb, rec->key);
179 talloc_set_destructor(record, NULL);
181 /* now tell ctdbd to update this record on all other nodes */
182 if (NT_STATUS_IS_OK(status)) {
183 status = ctdbd_persistent_store(
184 messaging_ctdbd_connection(),
185 crec->ctdb_ctx->db_id,
189 ctdbd_cancel_persistent_update(
190 messaging_ctdbd_connection(),
191 crec->ctdb_ctx->db_id,
197 SAFE_FREE(cdata.dptr);
200 if (!NT_STATUS_IS_OK(status)) {
201 DEBUG(5, ("ctdbd_persistent_store still failed after "
202 "%d retries with error %s - giving up.\n",
203 count, nt_errstr(status)));
206 SAFE_FREE(cdata.dptr);
211 static NTSTATUS db_ctdb_delete(struct db_record *rec)
216 * We have to store the header with empty data. TODO: Fix the
222 return db_ctdb_store(rec, data, 0);
226 static NTSTATUS db_ctdb_delete_persistent(struct db_record *rec)
231 * We have to store the header with empty data. TODO: Fix the
237 return db_ctdb_store_persistent(rec, data, 0);
241 static int db_ctdb_record_destr(struct db_record* data)
243 struct db_ctdb_rec *crec = talloc_get_type_abort(
244 data->private_data, struct db_ctdb_rec);
246 DEBUG(10, (DEBUGLEVEL > 10
247 ? "Unlocking db %u key %s\n"
248 : "Unlocking db %u key %.20s\n",
249 (int)crec->ctdb_ctx->db_id,
250 hex_encode(data, (unsigned char *)data->key.dptr,
253 if (tdb_chainunlock(crec->ctdb_ctx->wtdb->tdb, data->key) != 0) {
254 DEBUG(0, ("tdb_chainunlock failed\n"));
261 static struct db_record *fetch_locked_internal(struct db_ctdb_ctx *ctx,
266 struct db_record *result;
267 struct db_ctdb_rec *crec;
270 int migrate_attempts = 0;
272 if (!(result = talloc(mem_ctx, struct db_record))) {
273 DEBUG(0, ("talloc failed\n"));
277 if (!(crec = TALLOC_ZERO_P(result, struct db_ctdb_rec))) {
278 DEBUG(0, ("talloc failed\n"));
283 result->private_data = (void *)crec;
284 crec->ctdb_ctx = ctx;
286 result->key.dsize = key.dsize;
287 result->key.dptr = (uint8 *)talloc_memdup(result, key.dptr, key.dsize);
288 if (result->key.dptr == NULL) {
289 DEBUG(0, ("talloc failed\n"));
295 * Do a blocking lock on the record
299 if (DEBUGLEVEL >= 10) {
300 char *keystr = hex_encode(result, key.dptr, key.dsize);
301 DEBUG(10, (DEBUGLEVEL > 10
302 ? "Locking db %u key %s\n"
303 : "Locking db %u key %.20s\n",
304 (int)crec->ctdb_ctx->db_id, keystr));
308 if (tdb_chainlock(ctx->wtdb->tdb, key) != 0) {
309 DEBUG(3, ("tdb_chainlock failed\n"));
315 result->store = db_ctdb_store_persistent;
316 result->delete_rec = db_ctdb_delete_persistent;
318 result->store = db_ctdb_store;
319 result->delete_rec = db_ctdb_delete;
321 talloc_set_destructor(result, db_ctdb_record_destr);
323 ctdb_data = tdb_fetch(ctx->wtdb->tdb, key);
326 * See if we have a valid record and we are the dmaster. If so, we can
327 * take the shortcut and just return it.
330 if ((ctdb_data.dptr == NULL) ||
331 (ctdb_data.dsize < sizeof(struct ctdb_ltdb_header)) ||
332 ((struct ctdb_ltdb_header *)ctdb_data.dptr)->dmaster != get_my_vnn()
334 || (random() % 2 != 0)
337 SAFE_FREE(ctdb_data.dptr);
338 tdb_chainunlock(ctx->wtdb->tdb, key);
339 talloc_set_destructor(result, NULL);
341 migrate_attempts += 1;
343 DEBUG(10, ("ctdb_data.dptr = %p, dmaster = %u (%u)\n",
344 ctdb_data.dptr, ctdb_data.dptr ?
345 ((struct ctdb_ltdb_header *)ctdb_data.dptr)->dmaster : -1,
348 status = ctdbd_migrate(messaging_ctdbd_connection(),ctx->db_id, key);
349 if (!NT_STATUS_IS_OK(status)) {
350 DEBUG(5, ("ctdb_migrate failed: %s\n",
355 /* now its migrated, try again */
359 if (migrate_attempts > 10) {
360 DEBUG(0, ("db_ctdb_fetch_locked needed %d attempts\n",
364 memcpy(&crec->header, ctdb_data.dptr, sizeof(crec->header));
366 result->value.dsize = ctdb_data.dsize - sizeof(crec->header);
367 result->value.dptr = NULL;
369 if ((result->value.dsize != 0)
370 && !(result->value.dptr = (uint8 *)talloc_memdup(
371 result, ctdb_data.dptr + sizeof(crec->header),
372 result->value.dsize))) {
373 DEBUG(0, ("talloc failed\n"));
377 SAFE_FREE(ctdb_data.dptr);
382 static struct db_record *db_ctdb_fetch_locked(struct db_context *db,
386 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
389 return fetch_locked_internal(ctx, mem_ctx, key, db->persistent);
393 fetch (unlocked, no migration) operation on ctdb
395 static int db_ctdb_fetch(struct db_context *db, TALLOC_CTX *mem_ctx,
396 TDB_DATA key, TDB_DATA *data)
398 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
403 /* try a direct fetch */
404 ctdb_data = tdb_fetch(ctx->wtdb->tdb, key);
407 * See if we have a valid record and we are the dmaster. If so, we can
408 * take the shortcut and just return it.
409 * we bypass the dmaster check for persistent databases
411 if ((ctdb_data.dptr != NULL) &&
412 (ctdb_data.dsize >= sizeof(struct ctdb_ltdb_header)) &&
414 ((struct ctdb_ltdb_header *)ctdb_data.dptr)->dmaster == get_my_vnn())) {
415 /* we are the dmaster - avoid the ctdb protocol op */
417 data->dsize = ctdb_data.dsize - sizeof(struct ctdb_ltdb_header);
418 if (data->dsize == 0) {
419 SAFE_FREE(ctdb_data.dptr);
424 data->dptr = (uint8 *)talloc_memdup(
425 mem_ctx, ctdb_data.dptr+sizeof(struct ctdb_ltdb_header),
428 SAFE_FREE(ctdb_data.dptr);
430 if (data->dptr == NULL) {
436 SAFE_FREE(ctdb_data.dptr);
438 /* we weren't able to get it locally - ask ctdb to fetch it for us */
439 status = ctdbd_fetch(messaging_ctdbd_connection(),ctx->db_id, key, mem_ctx, data);
440 if (!NT_STATUS_IS_OK(status)) {
441 DEBUG(5, ("ctdbd_fetch failed: %s\n", nt_errstr(status)));
448 struct traverse_state {
449 struct db_context *db;
450 int (*fn)(struct db_record *rec, void *private_data);
454 static void traverse_callback(TDB_DATA key, TDB_DATA data, void *private_data)
456 struct traverse_state *state = (struct traverse_state *)private_data;
457 struct db_record *rec;
458 TALLOC_CTX *tmp_ctx = talloc_new(state->db);
459 /* we have to give them a locked record to prevent races */
460 rec = db_ctdb_fetch_locked(state->db, tmp_ctx, key);
461 if (rec && rec->value.dsize > 0) {
462 state->fn(rec, state->private_data);
464 talloc_free(tmp_ctx);
467 static int traverse_persistent_callback(TDB_CONTEXT *tdb, TDB_DATA kbuf, TDB_DATA dbuf,
470 struct traverse_state *state = (struct traverse_state *)private_data;
471 struct db_record *rec;
472 TALLOC_CTX *tmp_ctx = talloc_new(state->db);
474 /* we have to give them a locked record to prevent races */
475 rec = db_ctdb_fetch_locked(state->db, tmp_ctx, kbuf);
476 if (rec && rec->value.dsize > 0) {
477 ret = state->fn(rec, state->private_data);
479 talloc_free(tmp_ctx);
483 static int db_ctdb_traverse(struct db_context *db,
484 int (*fn)(struct db_record *rec,
488 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
490 struct traverse_state state;
494 state.private_data = private_data;
496 if (db->persistent) {
497 /* for persistent databases we don't need to do a ctdb traverse,
498 we can do a faster local traverse */
499 return tdb_traverse(ctx->wtdb->tdb, traverse_persistent_callback, &state);
503 ctdbd_traverse(ctx->db_id, traverse_callback, &state);
507 static NTSTATUS db_ctdb_store_deny(struct db_record *rec, TDB_DATA data, int flag)
509 return NT_STATUS_MEDIA_WRITE_PROTECTED;
512 static NTSTATUS db_ctdb_delete_deny(struct db_record *rec)
514 return NT_STATUS_MEDIA_WRITE_PROTECTED;
517 static void traverse_read_callback(TDB_DATA key, TDB_DATA data, void *private_data)
519 struct traverse_state *state = (struct traverse_state *)private_data;
520 struct db_record rec;
523 rec.store = db_ctdb_store_deny;
524 rec.delete_rec = db_ctdb_delete_deny;
525 rec.private_data = state->db;
526 state->fn(&rec, state->private_data);
529 static int traverse_persistent_callback_read(TDB_CONTEXT *tdb, TDB_DATA kbuf, TDB_DATA dbuf,
532 struct traverse_state *state = (struct traverse_state *)private_data;
533 struct db_record rec;
536 rec.store = db_ctdb_store_deny;
537 rec.delete_rec = db_ctdb_delete_deny;
538 rec.private_data = state->db;
540 if (rec.value.dsize <= sizeof(struct ctdb_ltdb_header)) {
541 /* a deleted record */
544 rec.value.dsize -= sizeof(struct ctdb_ltdb_header);
545 rec.value.dptr += sizeof(struct ctdb_ltdb_header);
547 return state->fn(&rec, state->private_data);
550 static int db_ctdb_traverse_read(struct db_context *db,
551 int (*fn)(struct db_record *rec,
555 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
557 struct traverse_state state;
561 state.private_data = private_data;
563 if (db->persistent) {
564 /* for persistent databases we don't need to do a ctdb traverse,
565 we can do a faster local traverse */
566 return tdb_traverse_read(ctx->wtdb->tdb, traverse_persistent_callback_read, &state);
569 ctdbd_traverse(ctx->db_id, traverse_read_callback, &state);
573 static int db_ctdb_get_seqnum(struct db_context *db)
575 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
577 return tdb_get_seqnum(ctx->wtdb->tdb);
580 static int db_ctdb_trans_dummy(struct db_context *db)
583 * Not implemented yet, just return ok
588 struct db_context *db_open_ctdb(TALLOC_CTX *mem_ctx,
590 int hash_size, int tdb_flags,
591 int open_flags, mode_t mode)
593 struct db_context *result;
594 struct db_ctdb_ctx *db_ctdb;
597 if (!lp_clustering()) {
598 DEBUG(10, ("Clustering disabled -- no ctdb\n"));
602 if (!(result = TALLOC_ZERO_P(mem_ctx, struct db_context))) {
603 DEBUG(0, ("talloc failed\n"));
608 if (!(db_ctdb = TALLOC_P(result, struct db_ctdb_ctx))) {
609 DEBUG(0, ("talloc failed\n"));
614 if (!NT_STATUS_IS_OK(ctdbd_db_attach(messaging_ctdbd_connection(),name, &db_ctdb->db_id, tdb_flags))) {
615 DEBUG(0, ("ctdbd_db_attach failed for %s\n", name));
620 db_path = ctdbd_dbpath(messaging_ctdbd_connection(), db_ctdb, db_ctdb->db_id);
622 result->persistent = ((tdb_flags & TDB_CLEAR_IF_FIRST) == 0);
624 /* only pass through specific flags */
625 tdb_flags &= TDB_SEQNUM;
627 /* honor permissions if user has specified O_CREAT */
628 if (open_flags & O_CREAT) {
629 chmod(db_path, mode);
632 db_ctdb->wtdb = tdb_wrap_open(db_ctdb, db_path, hash_size, tdb_flags, O_RDWR, 0);
633 if (db_ctdb->wtdb == NULL) {
634 DEBUG(0, ("Could not open tdb %s: %s\n", db_path, strerror(errno)));
638 talloc_free(db_path);
640 result->private_data = (void *)db_ctdb;
641 result->fetch_locked = db_ctdb_fetch_locked;
642 result->fetch = db_ctdb_fetch;
643 result->traverse = db_ctdb_traverse;
644 result->traverse_read = db_ctdb_traverse_read;
645 result->get_seqnum = db_ctdb_get_seqnum;
646 result->transaction_start = db_ctdb_trans_dummy;
647 result->transaction_commit = db_ctdb_trans_dummy;
648 result->transaction_cancel = db_ctdb_trans_dummy;
650 DEBUG(3,("db_open_ctdb: opened database '%s' with dbid 0x%x\n",
651 name, db_ctdb->db_id));