2 Unix SMB/CIFS implementation.
3 Database interface wrapper around ctdbd
4 Copyright (C) Volker Lendecke 2007-2009
5 Copyright (C) Michael Adam 2009
7 This program is free software; you can redistribute it and/or modify
8 it under the terms of the GNU General Public License as published by
9 the Free Software Foundation; either version 3 of the License, or
10 (at your option) any later version.
12 This program is distributed in the hope that it will be useful,
13 but WITHOUT ANY WARRANTY; without even the implied warranty of
14 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 GNU General Public License for more details.
17 You should have received a copy of the GNU General Public License
18 along with this program. If not, see <http://www.gnu.org/licenses/>.
22 #include "system/filesys.h"
23 #ifdef CLUSTER_SUPPORT
25 #include "ctdb_private.h"
26 #include "ctdbd_conn.h"
30 struct db_ctdb_transaction_handle {
31 struct db_ctdb_ctx *ctx;
33 * we store the reads and writes done under a transaction:
34 * - one list stores both reads and writes (m_all),
35 * - the other just writes (m_write)
37 struct ctdb_marshall_buffer *m_all;
38 struct ctdb_marshall_buffer *m_write;
45 struct db_context *db;
46 struct tdb_wrap *wtdb;
48 struct db_ctdb_transaction_handle *transaction;
49 struct g_lock_ctx *lock_ctx;
53 struct db_ctdb_ctx *ctdb_ctx;
54 struct ctdb_ltdb_header header;
55 struct timeval lock_time;
58 static NTSTATUS tdb_error_to_ntstatus(struct tdb_context *tdb)
61 enum TDB_ERROR tret = tdb_error(tdb);
65 status = NT_STATUS_OBJECT_NAME_COLLISION;
68 status = NT_STATUS_OBJECT_NAME_NOT_FOUND;
71 status = NT_STATUS_INTERNAL_DB_CORRUPTION;
80 * fetch a record from the tdb, separating out the header
81 * information and returning the body of the record.
83 static NTSTATUS db_ctdb_ltdb_fetch(struct db_ctdb_ctx *db,
85 struct ctdb_ltdb_header *header,
92 rec = tdb_fetch(db->wtdb->tdb, key);
93 if (rec.dsize < sizeof(struct ctdb_ltdb_header)) {
94 status = NT_STATUS_NOT_FOUND;
99 header->dmaster = (uint32_t)-1;
106 *header = *(struct ctdb_ltdb_header *)rec.dptr;
110 data->dsize = rec.dsize - sizeof(struct ctdb_ltdb_header);
111 if (data->dsize == 0) {
114 data->dptr = (unsigned char *)talloc_memdup(mem_ctx,
116 + sizeof(struct ctdb_ltdb_header),
118 if (data->dptr == NULL) {
119 status = NT_STATUS_NO_MEMORY;
125 status = NT_STATUS_OK;
133 * Store a record together with the ctdb record header
134 * in the local copy of the database.
136 static NTSTATUS db_ctdb_ltdb_store(struct db_ctdb_ctx *db,
138 struct ctdb_ltdb_header *header,
141 TALLOC_CTX *tmp_ctx = talloc_stackframe();
145 rec.dsize = data.dsize + sizeof(struct ctdb_ltdb_header);
146 rec.dptr = (uint8_t *)talloc_size(tmp_ctx, rec.dsize);
148 if (rec.dptr == NULL) {
149 talloc_free(tmp_ctx);
150 return NT_STATUS_NO_MEMORY;
153 memcpy(rec.dptr, header, sizeof(struct ctdb_ltdb_header));
154 memcpy(sizeof(struct ctdb_ltdb_header) + (uint8_t *)rec.dptr, data.dptr, data.dsize);
156 ret = tdb_store(db->wtdb->tdb, key, rec, TDB_REPLACE);
158 talloc_free(tmp_ctx);
160 return (ret == 0) ? NT_STATUS_OK
161 : tdb_error_to_ntstatus(db->wtdb->tdb);
166 form a ctdb_rec_data record from a key/data pair
168 note that header may be NULL. If not NULL then it is included in the data portion
171 static struct ctdb_rec_data *db_ctdb_marshall_record(TALLOC_CTX *mem_ctx, uint32_t reqid,
173 struct ctdb_ltdb_header *header,
177 struct ctdb_rec_data *d;
179 length = offsetof(struct ctdb_rec_data, data) + key.dsize +
180 data.dsize + (header?sizeof(*header):0);
181 d = (struct ctdb_rec_data *)talloc_size(mem_ctx, length);
187 d->keylen = key.dsize;
188 memcpy(&d->data[0], key.dptr, key.dsize);
190 d->datalen = data.dsize + sizeof(*header);
191 memcpy(&d->data[key.dsize], header, sizeof(*header));
192 memcpy(&d->data[key.dsize+sizeof(*header)], data.dptr, data.dsize);
194 d->datalen = data.dsize;
195 memcpy(&d->data[key.dsize], data.dptr, data.dsize);
201 /* helper function for marshalling multiple records */
202 static struct ctdb_marshall_buffer *db_ctdb_marshall_add(TALLOC_CTX *mem_ctx,
203 struct ctdb_marshall_buffer *m,
207 struct ctdb_ltdb_header *header,
210 struct ctdb_rec_data *r;
211 size_t m_size, r_size;
212 struct ctdb_marshall_buffer *m2 = NULL;
214 r = db_ctdb_marshall_record(talloc_tos(), reqid, key, header, data);
221 m = (struct ctdb_marshall_buffer *)talloc_zero_size(
222 mem_ctx, offsetof(struct ctdb_marshall_buffer, data));
229 m_size = talloc_get_size(m);
230 r_size = talloc_get_size(r);
232 m2 = (struct ctdb_marshall_buffer *)talloc_realloc_size(
233 mem_ctx, m, m_size + r_size);
239 memcpy(m_size + (uint8_t *)m2, r, r_size);
248 /* we've finished marshalling, return a data blob with the marshalled records */
249 static TDB_DATA db_ctdb_marshall_finish(struct ctdb_marshall_buffer *m)
252 data.dptr = (uint8_t *)m;
253 data.dsize = talloc_get_size(m);
258 loop over a marshalling buffer
260 - pass r==NULL to start
261 - loop the number of times indicated by m->count
263 static struct ctdb_rec_data *db_ctdb_marshall_loop_next(struct ctdb_marshall_buffer *m, struct ctdb_rec_data *r,
265 struct ctdb_ltdb_header *header,
266 TDB_DATA *key, TDB_DATA *data)
269 r = (struct ctdb_rec_data *)&m->data[0];
271 r = (struct ctdb_rec_data *)(r->length + (uint8_t *)r);
279 key->dptr = &r->data[0];
280 key->dsize = r->keylen;
283 data->dptr = &r->data[r->keylen];
284 data->dsize = r->datalen;
285 if (header != NULL) {
286 data->dptr += sizeof(*header);
287 data->dsize -= sizeof(*header);
291 if (header != NULL) {
292 if (r->datalen < sizeof(*header)) {
295 *header = *(struct ctdb_ltdb_header *)&r->data[r->keylen];
302 * CTDB transaction destructor
304 static int db_ctdb_transaction_destructor(struct db_ctdb_transaction_handle *h)
308 status = g_lock_unlock(h->ctx->lock_ctx, h->lock_name);
309 if (!NT_STATUS_IS_OK(status)) {
310 DEBUG(0, ("g_lock_unlock failed: %s\n", nt_errstr(status)));
317 * CTDB dbwrap API: transaction_start function
318 * starts a transaction on a persistent database
320 static int db_ctdb_transaction_start(struct db_context *db)
322 struct db_ctdb_transaction_handle *h;
324 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
327 if (!db->persistent) {
328 DEBUG(0,("transactions not supported on non-persistent database 0x%08x\n",
333 if (ctx->transaction) {
334 ctx->transaction->nesting++;
338 h = talloc_zero(db, struct db_ctdb_transaction_handle);
340 DEBUG(0,(__location__ " oom for transaction handle\n"));
346 h->lock_name = talloc_asprintf(h, "transaction_db_0x%08x",
347 (unsigned int)ctx->db_id);
348 if (h->lock_name == NULL) {
349 DEBUG(0, ("talloc_asprintf failed\n"));
355 * Wait a day, i.e. forever...
357 status = g_lock_lock(ctx->lock_ctx, h->lock_name, G_LOCK_WRITE,
358 timeval_set(86400, 0));
359 if (!NT_STATUS_IS_OK(status)) {
360 DEBUG(0, ("g_lock_lock failed: %s\n", nt_errstr(status)));
365 talloc_set_destructor(h, db_ctdb_transaction_destructor);
367 ctx->transaction = h;
369 DEBUG(5,(__location__ " Started transaction on db 0x%08x\n", ctx->db_id));
374 static bool pull_newest_from_marshall_buffer(struct ctdb_marshall_buffer *buf,
376 struct ctdb_ltdb_header *pheader,
380 struct ctdb_rec_data *rec = NULL;
381 struct ctdb_ltdb_header h;
394 * Walk the list of records written during this
395 * transaction. If we want to read one we have already
396 * written, return the last written sample. Thus we do not do
397 * a "break;" for the first hit, this record might have been
401 for (i=0; i<buf->count; i++) {
402 TDB_DATA tkey, tdata;
404 struct ctdb_ltdb_header hdr;
408 rec = db_ctdb_marshall_loop_next(buf, rec, &reqid, &hdr, &tkey,
414 if (tdb_data_equal(key, tkey)) {
426 data.dptr = (uint8_t *)talloc_memdup(mem_ctx, data.dptr,
428 if ((data.dsize != 0) && (data.dptr == NULL)) {
434 if (pheader != NULL) {
442 fetch a record inside a transaction
444 static int db_ctdb_transaction_fetch(struct db_ctdb_ctx *db,
446 TDB_DATA key, TDB_DATA *data)
448 struct db_ctdb_transaction_handle *h = db->transaction;
452 found = pull_newest_from_marshall_buffer(h->m_write, key, NULL,
458 status = db_ctdb_ltdb_fetch(h->ctx, key, NULL, mem_ctx, data);
460 if (NT_STATUS_EQUAL(status, NT_STATUS_NOT_FOUND)) {
462 } else if (!NT_STATUS_IS_OK(status)) {
466 h->m_all = db_ctdb_marshall_add(h, h->m_all, h->ctx->db_id, 1, key,
468 if (h->m_all == NULL) {
469 DEBUG(0,(__location__ " Failed to add to marshalling "
472 talloc_free(data->dptr);
480 * Fetch a record from a persistent database
481 * without record locking and without an active transaction.
483 * This just fetches from the local database copy.
484 * Since the databases are kept in syc cluster-wide,
485 * there is no point in doing a ctdb call to fetch the
486 * record from the lmaster. It does even harm since migration
487 * of records bump their RSN and hence render the persistent
488 * database inconsistent.
490 static int db_ctdb_fetch_persistent(struct db_ctdb_ctx *db,
492 TDB_DATA key, TDB_DATA *data)
497 status = db_ctdb_ltdb_fetch(db, key, NULL, mem_ctx, data);
499 if (NT_STATUS_EQUAL(status, NT_STATUS_NOT_FOUND)) {
501 } else if (!NT_STATUS_IS_OK(status)) {
508 static NTSTATUS db_ctdb_store_transaction(struct db_record *rec, TDB_DATA data, int flag);
509 static NTSTATUS db_ctdb_delete_transaction(struct db_record *rec);
511 static struct db_record *db_ctdb_fetch_locked_transaction(struct db_ctdb_ctx *ctx,
515 struct db_record *result;
518 if (!(result = talloc(mem_ctx, struct db_record))) {
519 DEBUG(0, ("talloc failed\n"));
523 result->private_data = ctx->transaction;
525 result->key.dsize = key.dsize;
526 result->key.dptr = (uint8 *)talloc_memdup(result, key.dptr, key.dsize);
527 if (result->key.dptr == NULL) {
528 DEBUG(0, ("talloc failed\n"));
533 result->store = db_ctdb_store_transaction;
534 result->delete_rec = db_ctdb_delete_transaction;
536 if (pull_newest_from_marshall_buffer(ctx->transaction->m_write, key,
537 NULL, result, &result->value)) {
541 ctdb_data = tdb_fetch(ctx->wtdb->tdb, key);
542 if (ctdb_data.dptr == NULL) {
543 /* create the record */
544 result->value = tdb_null;
548 result->value.dsize = ctdb_data.dsize - sizeof(struct ctdb_ltdb_header);
549 result->value.dptr = NULL;
551 if ((result->value.dsize != 0)
552 && !(result->value.dptr = (uint8 *)talloc_memdup(
553 result, ctdb_data.dptr + sizeof(struct ctdb_ltdb_header),
554 result->value.dsize))) {
555 DEBUG(0, ("talloc failed\n"));
559 SAFE_FREE(ctdb_data.dptr);
564 static int db_ctdb_record_destructor(struct db_record **recp)
566 struct db_record *rec = talloc_get_type_abort(*recp, struct db_record);
567 struct db_ctdb_transaction_handle *h = talloc_get_type_abort(
568 rec->private_data, struct db_ctdb_transaction_handle);
569 int ret = h->ctx->db->transaction_commit(h->ctx->db);
571 DEBUG(0,(__location__ " transaction_commit failed\n"));
577 auto-create a transaction for persistent databases
579 static struct db_record *db_ctdb_fetch_locked_persistent(struct db_ctdb_ctx *ctx,
584 struct db_record *rec, **recp;
586 res = db_ctdb_transaction_start(ctx->db);
591 rec = db_ctdb_fetch_locked_transaction(ctx, mem_ctx, key);
593 ctx->db->transaction_cancel(ctx->db);
597 /* destroy this transaction when we release the lock */
598 recp = talloc(rec, struct db_record *);
600 ctx->db->transaction_cancel(ctx->db);
605 talloc_set_destructor(recp, db_ctdb_record_destructor);
611 stores a record inside a transaction
613 static NTSTATUS db_ctdb_transaction_store(struct db_ctdb_transaction_handle *h,
614 TDB_DATA key, TDB_DATA data)
616 TALLOC_CTX *tmp_ctx = talloc_new(h);
618 struct ctdb_ltdb_header header;
622 /* we need the header so we can update the RSN */
624 if (!pull_newest_from_marshall_buffer(h->m_write, key, &header,
627 rec = tdb_fetch(h->ctx->wtdb->tdb, key);
629 if (rec.dptr != NULL) {
630 memcpy(&header, rec.dptr,
631 sizeof(struct ctdb_ltdb_header));
632 rec.dsize -= sizeof(struct ctdb_ltdb_header);
635 * a special case, we are writing the same
636 * data that is there now
638 if (data.dsize == rec.dsize &&
640 rec.dptr + sizeof(struct ctdb_ltdb_header),
643 talloc_free(tmp_ctx);
650 header.dmaster = get_my_vnn();
653 h->m_all = db_ctdb_marshall_add(h, h->m_all, h->ctx->db_id, 0, key,
655 if (h->m_all == NULL) {
656 DEBUG(0,(__location__ " Failed to add to marshalling "
658 talloc_free(tmp_ctx);
659 return NT_STATUS_NO_MEMORY;
662 h->m_write = db_ctdb_marshall_add(h, h->m_write, h->ctx->db_id, 0, key, &header, data);
663 if (h->m_write == NULL) {
664 DEBUG(0,(__location__ " Failed to add to marshalling record\n"));
665 talloc_free(tmp_ctx);
666 return NT_STATUS_NO_MEMORY;
669 talloc_free(tmp_ctx);
675 a record store inside a transaction
677 static NTSTATUS db_ctdb_store_transaction(struct db_record *rec, TDB_DATA data, int flag)
679 struct db_ctdb_transaction_handle *h = talloc_get_type_abort(
680 rec->private_data, struct db_ctdb_transaction_handle);
683 status = db_ctdb_transaction_store(h, rec->key, data);
688 a record delete inside a transaction
690 static NTSTATUS db_ctdb_delete_transaction(struct db_record *rec)
692 struct db_ctdb_transaction_handle *h = talloc_get_type_abort(
693 rec->private_data, struct db_ctdb_transaction_handle);
696 status = db_ctdb_transaction_store(h, rec->key, tdb_null);
701 * Fetch the db sequence number of a persistent db directly from the db.
703 static NTSTATUS db_ctdb_fetch_db_seqnum_from_db(struct db_ctdb_ctx *db,
707 const char *keyname = CTDB_DB_SEQNUM_KEY;
710 struct ctdb_ltdb_header header;
711 TALLOC_CTX *mem_ctx = talloc_stackframe();
713 if (seqnum == NULL) {
714 return NT_STATUS_INVALID_PARAMETER;
717 key = string_term_tdb_data(keyname);
719 status = db_ctdb_ltdb_fetch(db, key, &header, mem_ctx, &data);
720 if (!NT_STATUS_IS_OK(status) &&
721 !NT_STATUS_EQUAL(status, NT_STATUS_NOT_FOUND))
726 status = NT_STATUS_OK;
728 if (data.dsize != sizeof(uint64_t)) {
733 *seqnum = *(uint64_t *)data.dptr;
736 TALLOC_FREE(mem_ctx);
741 * Store the database sequence number inside a transaction.
743 static NTSTATUS db_ctdb_store_db_seqnum(struct db_ctdb_transaction_handle *h,
747 const char *keyname = CTDB_DB_SEQNUM_KEY;
751 key = string_term_tdb_data(keyname);
753 data.dptr = (uint8_t *)&seqnum;
754 data.dsize = sizeof(uint64_t);
756 status = db_ctdb_transaction_store(h, key, data);
764 static int db_ctdb_transaction_commit(struct db_context *db)
766 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
770 struct db_ctdb_transaction_handle *h = ctx->transaction;
771 uint64_t old_seqnum, new_seqnum;
775 DEBUG(0,(__location__ " transaction commit with no open transaction on db 0x%08x\n", ctx->db_id));
779 if (h->nested_cancel) {
780 db->transaction_cancel(db);
781 DEBUG(5,(__location__ " Failed transaction commit after nested cancel\n"));
785 if (h->nesting != 0) {
790 if (h->m_write == NULL) {
792 * No changes were made, so don't change the seqnum,
793 * don't push to other node, just exit with success.
799 DEBUG(5,(__location__ " Commit transaction on db 0x%08x\n", ctx->db_id));
802 * As the last db action before committing, bump the database sequence
803 * number. Note that this undoes all changes to the seqnum records
804 * performed under the transaction. This record is not meant to be
805 * modified by user interaction. It is for internal use only...
807 rets = db_ctdb_fetch_db_seqnum_from_db(ctx, &old_seqnum);
808 if (!NT_STATUS_IS_OK(rets)) {
809 DEBUG(1, (__location__ " failed to fetch the db sequence number "
810 "in transaction commit on db 0x%08x\n", ctx->db_id));
815 new_seqnum = old_seqnum + 1;
817 rets = db_ctdb_store_db_seqnum(h, new_seqnum);
818 if (!NT_STATUS_IS_OK(rets)) {
819 DEBUG(1, (__location__ "failed to store the db sequence number "
820 " in transaction commit on db 0x%08x\n", ctx->db_id));
826 /* tell ctdbd to commit to the other nodes */
827 rets = ctdbd_control_local(messaging_ctdbd_connection(),
828 CTDB_CONTROL_TRANS3_COMMIT,
830 db_ctdb_marshall_finish(h->m_write),
831 NULL, NULL, &status);
832 if (!NT_STATUS_IS_OK(rets) || status != 0) {
834 * The TRANS3_COMMIT control should only possibly fail when a
835 * recovery has been running concurrently. In any case, the db
836 * will be the same on all nodes, either the new copy or the
837 * old copy. This can be detected by comparing the old and new
838 * local sequence numbers.
840 rets = db_ctdb_fetch_db_seqnum_from_db(ctx, &new_seqnum);
841 if (!NT_STATUS_IS_OK(rets)) {
842 DEBUG(1, (__location__ " failed to refetch db sequence "
843 "number after failed TRANS3_COMMIT\n"));
848 if (new_seqnum == old_seqnum) {
849 /* Recovery prevented all our changes: retry. */
851 } else if (new_seqnum != (old_seqnum + 1)) {
852 DEBUG(0, (__location__ " ERROR: new_seqnum[%lu] != "
853 "old_seqnum[%lu] + (0 or 1) after failed "
854 "TRANS3_COMMIT - this should not happen!\n",
855 (unsigned long)new_seqnum,
856 (unsigned long)old_seqnum));
861 * Recovery propagated our changes to all nodes, completing
862 * our commit for us - succeed.
869 h->ctx->transaction = NULL;
878 static int db_ctdb_transaction_cancel(struct db_context *db)
880 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
882 struct db_ctdb_transaction_handle *h = ctx->transaction;
885 DEBUG(0,(__location__ " transaction cancel with no open transaction on db 0x%08x\n", ctx->db_id));
889 if (h->nesting != 0) {
891 h->nested_cancel = true;
895 DEBUG(5,(__location__ " Cancel transaction on db 0x%08x\n", ctx->db_id));
897 ctx->transaction = NULL;
903 static NTSTATUS db_ctdb_store(struct db_record *rec, TDB_DATA data, int flag)
905 struct db_ctdb_rec *crec = talloc_get_type_abort(
906 rec->private_data, struct db_ctdb_rec);
908 return db_ctdb_ltdb_store(crec->ctdb_ctx, rec->key, &(crec->header), data);
913 static NTSTATUS db_ctdb_delete(struct db_record *rec)
918 * We have to store the header with empty data. TODO: Fix the
924 return db_ctdb_store(rec, data, 0);
928 static int db_ctdb_record_destr(struct db_record* data)
930 struct db_ctdb_rec *crec = talloc_get_type_abort(
931 data->private_data, struct db_ctdb_rec);
934 DEBUG(10, (DEBUGLEVEL > 10
935 ? "Unlocking db %u key %s\n"
936 : "Unlocking db %u key %.20s\n",
937 (int)crec->ctdb_ctx->db_id,
938 hex_encode_talloc(data, (unsigned char *)data->key.dptr,
941 if (tdb_chainunlock(crec->ctdb_ctx->wtdb->tdb, data->key) != 0) {
942 DEBUG(0, ("tdb_chainunlock failed\n"));
946 threshold = lp_ctdb_locktime_warn_threshold();
947 if (threshold != 0) {
948 double timediff = timeval_elapsed(&crec->lock_time);
949 if ((timediff * 1000) > threshold) {
950 DEBUG(0, ("Held tdb lock %f seconds\n", timediff));
957 static struct db_record *fetch_locked_internal(struct db_ctdb_ctx *ctx,
961 struct db_record *result;
962 struct db_ctdb_rec *crec;
965 int migrate_attempts = 0;
967 if (!(result = talloc(mem_ctx, struct db_record))) {
968 DEBUG(0, ("talloc failed\n"));
972 if (!(crec = TALLOC_ZERO_P(result, struct db_ctdb_rec))) {
973 DEBUG(0, ("talloc failed\n"));
978 result->private_data = (void *)crec;
979 crec->ctdb_ctx = ctx;
981 result->key.dsize = key.dsize;
982 result->key.dptr = (uint8 *)talloc_memdup(result, key.dptr, key.dsize);
983 if (result->key.dptr == NULL) {
984 DEBUG(0, ("talloc failed\n"));
990 * Do a blocking lock on the record
994 if (DEBUGLEVEL >= 10) {
995 char *keystr = hex_encode_talloc(result, key.dptr, key.dsize);
996 DEBUG(10, (DEBUGLEVEL > 10
997 ? "Locking db %u key %s\n"
998 : "Locking db %u key %.20s\n",
999 (int)crec->ctdb_ctx->db_id, keystr));
1000 TALLOC_FREE(keystr);
1003 if (tdb_chainlock(ctx->wtdb->tdb, key) != 0) {
1004 DEBUG(3, ("tdb_chainlock failed\n"));
1005 TALLOC_FREE(result);
1009 result->store = db_ctdb_store;
1010 result->delete_rec = db_ctdb_delete;
1011 talloc_set_destructor(result, db_ctdb_record_destr);
1013 ctdb_data = tdb_fetch(ctx->wtdb->tdb, key);
1016 * See if we have a valid record and we are the dmaster. If so, we can
1017 * take the shortcut and just return it.
1020 if ((ctdb_data.dptr == NULL) ||
1021 (ctdb_data.dsize < sizeof(struct ctdb_ltdb_header)) ||
1022 ((struct ctdb_ltdb_header *)ctdb_data.dptr)->dmaster != get_my_vnn()
1024 || (random() % 2 != 0)
1027 SAFE_FREE(ctdb_data.dptr);
1028 tdb_chainunlock(ctx->wtdb->tdb, key);
1029 talloc_set_destructor(result, NULL);
1031 migrate_attempts += 1;
1033 DEBUG(10, ("ctdb_data.dptr = %p, dmaster = %u (%u)\n",
1034 ctdb_data.dptr, ctdb_data.dptr ?
1035 ((struct ctdb_ltdb_header *)ctdb_data.dptr)->dmaster : -1,
1038 status = ctdbd_migrate(messaging_ctdbd_connection(), ctx->db_id,
1040 if (!NT_STATUS_IS_OK(status)) {
1041 DEBUG(5, ("ctdb_migrate failed: %s\n",
1042 nt_errstr(status)));
1043 TALLOC_FREE(result);
1046 /* now its migrated, try again */
1050 if (migrate_attempts > 10) {
1051 DEBUG(0, ("db_ctdb_fetch_locked needed %d attempts\n",
1055 GetTimeOfDay(&crec->lock_time);
1057 memcpy(&crec->header, ctdb_data.dptr, sizeof(crec->header));
1059 result->value.dsize = ctdb_data.dsize - sizeof(crec->header);
1060 result->value.dptr = NULL;
1062 if ((result->value.dsize != 0)
1063 && !(result->value.dptr = (uint8 *)talloc_memdup(
1064 result, ctdb_data.dptr + sizeof(crec->header),
1065 result->value.dsize))) {
1066 DEBUG(0, ("talloc failed\n"));
1067 TALLOC_FREE(result);
1070 SAFE_FREE(ctdb_data.dptr);
1075 static struct db_record *db_ctdb_fetch_locked(struct db_context *db,
1076 TALLOC_CTX *mem_ctx,
1079 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
1080 struct db_ctdb_ctx);
1082 if (ctx->transaction != NULL) {
1083 return db_ctdb_fetch_locked_transaction(ctx, mem_ctx, key);
1086 if (db->persistent) {
1087 return db_ctdb_fetch_locked_persistent(ctx, mem_ctx, key);
1090 return fetch_locked_internal(ctx, mem_ctx, key);
1094 fetch (unlocked, no migration) operation on ctdb
1096 static int db_ctdb_fetch(struct db_context *db, TALLOC_CTX *mem_ctx,
1097 TDB_DATA key, TDB_DATA *data)
1099 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
1100 struct db_ctdb_ctx);
1104 if (ctx->transaction) {
1105 return db_ctdb_transaction_fetch(ctx, mem_ctx, key, data);
1108 if (db->persistent) {
1109 return db_ctdb_fetch_persistent(ctx, mem_ctx, key, data);
1112 /* try a direct fetch */
1113 ctdb_data = tdb_fetch(ctx->wtdb->tdb, key);
1116 * See if we have a valid record and we are the dmaster. If so, we can
1117 * take the shortcut and just return it.
1118 * we bypass the dmaster check for persistent databases
1120 if ((ctdb_data.dptr != NULL) &&
1121 (ctdb_data.dsize >= sizeof(struct ctdb_ltdb_header)) &&
1122 ((struct ctdb_ltdb_header *)ctdb_data.dptr)->dmaster == get_my_vnn())
1124 /* we are the dmaster - avoid the ctdb protocol op */
1126 data->dsize = ctdb_data.dsize - sizeof(struct ctdb_ltdb_header);
1127 if (data->dsize == 0) {
1128 SAFE_FREE(ctdb_data.dptr);
1133 data->dptr = (uint8 *)talloc_memdup(
1134 mem_ctx, ctdb_data.dptr+sizeof(struct ctdb_ltdb_header),
1137 SAFE_FREE(ctdb_data.dptr);
1139 if (data->dptr == NULL) {
1145 SAFE_FREE(ctdb_data.dptr);
1147 /* we weren't able to get it locally - ask ctdb to fetch it for us */
1148 status = ctdbd_fetch(messaging_ctdbd_connection(), ctx->db_id, key,
1150 if (!NT_STATUS_IS_OK(status)) {
1151 DEBUG(5, ("ctdbd_fetch failed: %s\n", nt_errstr(status)));
1158 struct traverse_state {
1159 struct db_context *db;
1160 int (*fn)(struct db_record *rec, void *private_data);
1164 static void traverse_callback(TDB_DATA key, TDB_DATA data, void *private_data)
1166 struct traverse_state *state = (struct traverse_state *)private_data;
1167 struct db_record *rec;
1168 TALLOC_CTX *tmp_ctx = talloc_new(state->db);
1169 /* we have to give them a locked record to prevent races */
1170 rec = db_ctdb_fetch_locked(state->db, tmp_ctx, key);
1171 if (rec && rec->value.dsize > 0) {
1172 state->fn(rec, state->private_data);
1174 talloc_free(tmp_ctx);
1177 static int traverse_persistent_callback(TDB_CONTEXT *tdb, TDB_DATA kbuf, TDB_DATA dbuf,
1180 struct traverse_state *state = (struct traverse_state *)private_data;
1181 struct db_record *rec;
1182 TALLOC_CTX *tmp_ctx = talloc_new(state->db);
1184 /* we have to give them a locked record to prevent races */
1185 rec = db_ctdb_fetch_locked(state->db, tmp_ctx, kbuf);
1186 if (rec && rec->value.dsize > 0) {
1187 ret = state->fn(rec, state->private_data);
1189 talloc_free(tmp_ctx);
1193 static int db_ctdb_traverse(struct db_context *db,
1194 int (*fn)(struct db_record *rec,
1195 void *private_data),
1198 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
1199 struct db_ctdb_ctx);
1200 struct traverse_state state;
1204 state.private_data = private_data;
1206 if (db->persistent) {
1207 /* for persistent databases we don't need to do a ctdb traverse,
1208 we can do a faster local traverse */
1209 return tdb_traverse(ctx->wtdb->tdb, traverse_persistent_callback, &state);
1213 ctdbd_traverse(ctx->db_id, traverse_callback, &state);
1217 static NTSTATUS db_ctdb_store_deny(struct db_record *rec, TDB_DATA data, int flag)
1219 return NT_STATUS_MEDIA_WRITE_PROTECTED;
1222 static NTSTATUS db_ctdb_delete_deny(struct db_record *rec)
1224 return NT_STATUS_MEDIA_WRITE_PROTECTED;
1227 static void traverse_read_callback(TDB_DATA key, TDB_DATA data, void *private_data)
1229 struct traverse_state *state = (struct traverse_state *)private_data;
1230 struct db_record rec;
1233 rec.store = db_ctdb_store_deny;
1234 rec.delete_rec = db_ctdb_delete_deny;
1235 rec.private_data = state->db;
1236 state->fn(&rec, state->private_data);
1239 static int traverse_persistent_callback_read(TDB_CONTEXT *tdb, TDB_DATA kbuf, TDB_DATA dbuf,
1242 struct traverse_state *state = (struct traverse_state *)private_data;
1243 struct db_record rec;
1246 rec.store = db_ctdb_store_deny;
1247 rec.delete_rec = db_ctdb_delete_deny;
1248 rec.private_data = state->db;
1250 if (rec.value.dsize <= sizeof(struct ctdb_ltdb_header)) {
1251 /* a deleted record */
1254 rec.value.dsize -= sizeof(struct ctdb_ltdb_header);
1255 rec.value.dptr += sizeof(struct ctdb_ltdb_header);
1257 return state->fn(&rec, state->private_data);
1260 static int db_ctdb_traverse_read(struct db_context *db,
1261 int (*fn)(struct db_record *rec,
1262 void *private_data),
1265 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
1266 struct db_ctdb_ctx);
1267 struct traverse_state state;
1271 state.private_data = private_data;
1273 if (db->persistent) {
1274 /* for persistent databases we don't need to do a ctdb traverse,
1275 we can do a faster local traverse */
1276 return tdb_traverse_read(ctx->wtdb->tdb, traverse_persistent_callback_read, &state);
1279 ctdbd_traverse(ctx->db_id, traverse_read_callback, &state);
1283 static int db_ctdb_get_seqnum(struct db_context *db)
1285 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
1286 struct db_ctdb_ctx);
1287 return tdb_get_seqnum(ctx->wtdb->tdb);
1290 static int db_ctdb_get_flags(struct db_context *db)
1292 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
1293 struct db_ctdb_ctx);
1294 return tdb_get_flags(ctx->wtdb->tdb);
1297 struct db_context *db_open_ctdb(TALLOC_CTX *mem_ctx,
1299 int hash_size, int tdb_flags,
1300 int open_flags, mode_t mode)
1302 struct db_context *result;
1303 struct db_ctdb_ctx *db_ctdb;
1305 struct ctdbd_connection *conn;
1307 if (!lp_clustering()) {
1308 DEBUG(10, ("Clustering disabled -- no ctdb\n"));
1312 if (!(result = TALLOC_ZERO_P(mem_ctx, struct db_context))) {
1313 DEBUG(0, ("talloc failed\n"));
1314 TALLOC_FREE(result);
1318 if (!(db_ctdb = TALLOC_P(result, struct db_ctdb_ctx))) {
1319 DEBUG(0, ("talloc failed\n"));
1320 TALLOC_FREE(result);
1324 db_ctdb->transaction = NULL;
1325 db_ctdb->db = result;
1327 conn = messaging_ctdbd_connection();
1329 DEBUG(1, ("Could not connect to ctdb\n"));
1330 TALLOC_FREE(result);
1334 if (!NT_STATUS_IS_OK(ctdbd_db_attach(conn, name, &db_ctdb->db_id, tdb_flags))) {
1335 DEBUG(0, ("ctdbd_db_attach failed for %s\n", name));
1336 TALLOC_FREE(result);
1340 db_path = ctdbd_dbpath(conn, db_ctdb, db_ctdb->db_id);
1342 result->persistent = ((tdb_flags & TDB_CLEAR_IF_FIRST) == 0);
1344 /* only pass through specific flags */
1345 tdb_flags &= TDB_SEQNUM;
1347 /* honor permissions if user has specified O_CREAT */
1348 if (open_flags & O_CREAT) {
1349 chmod(db_path, mode);
1352 db_ctdb->wtdb = tdb_wrap_open(db_ctdb, db_path, hash_size, tdb_flags, O_RDWR, 0);
1353 if (db_ctdb->wtdb == NULL) {
1354 DEBUG(0, ("Could not open tdb %s: %s\n", db_path, strerror(errno)));
1355 TALLOC_FREE(result);
1358 talloc_free(db_path);
1360 if (result->persistent) {
1361 db_ctdb->lock_ctx = g_lock_ctx_init(db_ctdb,
1362 ctdb_conn_msg_ctx(conn));
1363 if (db_ctdb->lock_ctx == NULL) {
1364 DEBUG(0, ("g_lock_ctx_init failed\n"));
1365 TALLOC_FREE(result);
1370 result->private_data = (void *)db_ctdb;
1371 result->fetch_locked = db_ctdb_fetch_locked;
1372 result->fetch = db_ctdb_fetch;
1373 result->traverse = db_ctdb_traverse;
1374 result->traverse_read = db_ctdb_traverse_read;
1375 result->get_seqnum = db_ctdb_get_seqnum;
1376 result->get_flags = db_ctdb_get_flags;
1377 result->transaction_start = db_ctdb_transaction_start;
1378 result->transaction_commit = db_ctdb_transaction_commit;
1379 result->transaction_cancel = db_ctdb_transaction_cancel;
1381 DEBUG(3,("db_open_ctdb: opened database '%s' with dbid 0x%x\n",
1382 name, db_ctdb->db_id));