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 #include "lib/util/tdb_wrap.h"
25 #ifdef CLUSTER_SUPPORT
27 #include "ctdb_private.h"
28 #include "ctdbd_conn.h"
32 struct db_ctdb_transaction_handle {
33 struct db_ctdb_ctx *ctx;
35 * we store the reads and writes done under a transaction:
36 * - one list stores both reads and writes (m_all),
37 * - the other just writes (m_write)
39 struct ctdb_marshall_buffer *m_all;
40 struct ctdb_marshall_buffer *m_write;
47 struct db_context *db;
48 struct tdb_wrap *wtdb;
50 struct db_ctdb_transaction_handle *transaction;
51 struct g_lock_ctx *lock_ctx;
55 struct db_ctdb_ctx *ctdb_ctx;
56 struct ctdb_ltdb_header header;
57 struct timeval lock_time;
60 static NTSTATUS tdb_error_to_ntstatus(struct tdb_context *tdb)
63 enum TDB_ERROR tret = tdb_error(tdb);
67 status = NT_STATUS_OBJECT_NAME_COLLISION;
70 status = NT_STATUS_OBJECT_NAME_NOT_FOUND;
73 status = NT_STATUS_INTERNAL_DB_CORRUPTION;
82 * fetch a record from the tdb, separating out the header
83 * information and returning the body of the record.
85 static NTSTATUS db_ctdb_ltdb_fetch(struct db_ctdb_ctx *db,
87 struct ctdb_ltdb_header *header,
94 rec = tdb_fetch(db->wtdb->tdb, key);
95 if (rec.dsize < sizeof(struct ctdb_ltdb_header)) {
96 status = NT_STATUS_NOT_FOUND;
101 header->dmaster = (uint32_t)-1;
108 *header = *(struct ctdb_ltdb_header *)rec.dptr;
112 data->dsize = rec.dsize - sizeof(struct ctdb_ltdb_header);
113 if (data->dsize == 0) {
116 data->dptr = (unsigned char *)talloc_memdup(mem_ctx,
118 + sizeof(struct ctdb_ltdb_header),
120 if (data->dptr == NULL) {
121 status = NT_STATUS_NO_MEMORY;
127 status = NT_STATUS_OK;
135 * Store a record together with the ctdb record header
136 * in the local copy of the database.
138 static NTSTATUS db_ctdb_ltdb_store(struct db_ctdb_ctx *db,
140 struct ctdb_ltdb_header *header,
143 TALLOC_CTX *tmp_ctx = talloc_stackframe();
147 rec.dsize = data.dsize + sizeof(struct ctdb_ltdb_header);
148 rec.dptr = (uint8_t *)talloc_size(tmp_ctx, rec.dsize);
150 if (rec.dptr == NULL) {
151 talloc_free(tmp_ctx);
152 return NT_STATUS_NO_MEMORY;
155 memcpy(rec.dptr, header, sizeof(struct ctdb_ltdb_header));
156 memcpy(sizeof(struct ctdb_ltdb_header) + (uint8_t *)rec.dptr, data.dptr, data.dsize);
158 ret = tdb_store(db->wtdb->tdb, key, rec, TDB_REPLACE);
160 talloc_free(tmp_ctx);
162 return (ret == 0) ? NT_STATUS_OK
163 : tdb_error_to_ntstatus(db->wtdb->tdb);
168 form a ctdb_rec_data record from a key/data pair
170 note that header may be NULL. If not NULL then it is included in the data portion
173 static struct ctdb_rec_data *db_ctdb_marshall_record(TALLOC_CTX *mem_ctx, uint32_t reqid,
175 struct ctdb_ltdb_header *header,
179 struct ctdb_rec_data *d;
181 length = offsetof(struct ctdb_rec_data, data) + key.dsize +
182 data.dsize + (header?sizeof(*header):0);
183 d = (struct ctdb_rec_data *)talloc_size(mem_ctx, length);
189 d->keylen = key.dsize;
190 memcpy(&d->data[0], key.dptr, key.dsize);
192 d->datalen = data.dsize + sizeof(*header);
193 memcpy(&d->data[key.dsize], header, sizeof(*header));
194 memcpy(&d->data[key.dsize+sizeof(*header)], data.dptr, data.dsize);
196 d->datalen = data.dsize;
197 memcpy(&d->data[key.dsize], data.dptr, data.dsize);
203 /* helper function for marshalling multiple records */
204 static struct ctdb_marshall_buffer *db_ctdb_marshall_add(TALLOC_CTX *mem_ctx,
205 struct ctdb_marshall_buffer *m,
209 struct ctdb_ltdb_header *header,
212 struct ctdb_rec_data *r;
213 size_t m_size, r_size;
214 struct ctdb_marshall_buffer *m2 = NULL;
216 r = db_ctdb_marshall_record(talloc_tos(), reqid, key, header, data);
223 m = (struct ctdb_marshall_buffer *)talloc_zero_size(
224 mem_ctx, offsetof(struct ctdb_marshall_buffer, data));
231 m_size = talloc_get_size(m);
232 r_size = talloc_get_size(r);
234 m2 = (struct ctdb_marshall_buffer *)talloc_realloc_size(
235 mem_ctx, m, m_size + r_size);
241 memcpy(m_size + (uint8_t *)m2, r, r_size);
250 /* we've finished marshalling, return a data blob with the marshalled records */
251 static TDB_DATA db_ctdb_marshall_finish(struct ctdb_marshall_buffer *m)
254 data.dptr = (uint8_t *)m;
255 data.dsize = talloc_get_size(m);
260 loop over a marshalling buffer
262 - pass r==NULL to start
263 - loop the number of times indicated by m->count
265 static struct ctdb_rec_data *db_ctdb_marshall_loop_next(struct ctdb_marshall_buffer *m, struct ctdb_rec_data *r,
267 struct ctdb_ltdb_header *header,
268 TDB_DATA *key, TDB_DATA *data)
271 r = (struct ctdb_rec_data *)&m->data[0];
273 r = (struct ctdb_rec_data *)(r->length + (uint8_t *)r);
281 key->dptr = &r->data[0];
282 key->dsize = r->keylen;
285 data->dptr = &r->data[r->keylen];
286 data->dsize = r->datalen;
287 if (header != NULL) {
288 data->dptr += sizeof(*header);
289 data->dsize -= sizeof(*header);
293 if (header != NULL) {
294 if (r->datalen < sizeof(*header)) {
297 *header = *(struct ctdb_ltdb_header *)&r->data[r->keylen];
304 * CTDB transaction destructor
306 static int db_ctdb_transaction_destructor(struct db_ctdb_transaction_handle *h)
310 status = g_lock_unlock(h->ctx->lock_ctx, h->lock_name);
311 if (!NT_STATUS_IS_OK(status)) {
312 DEBUG(0, ("g_lock_unlock failed: %s\n", nt_errstr(status)));
319 * CTDB dbwrap API: transaction_start function
320 * starts a transaction on a persistent database
322 static int db_ctdb_transaction_start(struct db_context *db)
324 struct db_ctdb_transaction_handle *h;
326 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
329 if (!db->persistent) {
330 DEBUG(0,("transactions not supported on non-persistent database 0x%08x\n",
335 if (ctx->transaction) {
336 ctx->transaction->nesting++;
340 h = talloc_zero(db, struct db_ctdb_transaction_handle);
342 DEBUG(0,(__location__ " oom for transaction handle\n"));
348 h->lock_name = talloc_asprintf(h, "transaction_db_0x%08x",
349 (unsigned int)ctx->db_id);
350 if (h->lock_name == NULL) {
351 DEBUG(0, ("talloc_asprintf failed\n"));
357 * Wait a day, i.e. forever...
359 status = g_lock_lock(ctx->lock_ctx, h->lock_name, G_LOCK_WRITE,
360 timeval_set(86400, 0));
361 if (!NT_STATUS_IS_OK(status)) {
362 DEBUG(0, ("g_lock_lock failed: %s\n", nt_errstr(status)));
367 talloc_set_destructor(h, db_ctdb_transaction_destructor);
369 ctx->transaction = h;
371 DEBUG(5,(__location__ " Started transaction on db 0x%08x\n", ctx->db_id));
376 static bool pull_newest_from_marshall_buffer(struct ctdb_marshall_buffer *buf,
378 struct ctdb_ltdb_header *pheader,
382 struct ctdb_rec_data *rec = NULL;
383 struct ctdb_ltdb_header h;
396 * Walk the list of records written during this
397 * transaction. If we want to read one we have already
398 * written, return the last written sample. Thus we do not do
399 * a "break;" for the first hit, this record might have been
403 for (i=0; i<buf->count; i++) {
404 TDB_DATA tkey, tdata;
406 struct ctdb_ltdb_header hdr;
410 rec = db_ctdb_marshall_loop_next(buf, rec, &reqid, &hdr, &tkey,
416 if (tdb_data_equal(key, tkey)) {
428 data.dptr = (uint8_t *)talloc_memdup(mem_ctx, data.dptr,
430 if ((data.dsize != 0) && (data.dptr == NULL)) {
436 if (pheader != NULL) {
444 fetch a record inside a transaction
446 static int db_ctdb_transaction_fetch(struct db_ctdb_ctx *db,
448 TDB_DATA key, TDB_DATA *data)
450 struct db_ctdb_transaction_handle *h = db->transaction;
454 found = pull_newest_from_marshall_buffer(h->m_write, key, NULL,
460 status = db_ctdb_ltdb_fetch(h->ctx, key, NULL, mem_ctx, data);
462 if (NT_STATUS_EQUAL(status, NT_STATUS_NOT_FOUND)) {
464 } else if (!NT_STATUS_IS_OK(status)) {
468 h->m_all = db_ctdb_marshall_add(h, h->m_all, h->ctx->db_id, 1, key,
470 if (h->m_all == NULL) {
471 DEBUG(0,(__location__ " Failed to add to marshalling "
474 talloc_free(data->dptr);
482 * Fetch a record from a persistent database
483 * without record locking and without an active transaction.
485 * This just fetches from the local database copy.
486 * Since the databases are kept in syc cluster-wide,
487 * there is no point in doing a ctdb call to fetch the
488 * record from the lmaster. It does even harm since migration
489 * of records bump their RSN and hence render the persistent
490 * database inconsistent.
492 static int db_ctdb_fetch_persistent(struct db_ctdb_ctx *db,
494 TDB_DATA key, TDB_DATA *data)
498 status = db_ctdb_ltdb_fetch(db, key, NULL, mem_ctx, data);
500 if (NT_STATUS_EQUAL(status, NT_STATUS_NOT_FOUND)) {
502 } else if (!NT_STATUS_IS_OK(status)) {
509 static NTSTATUS db_ctdb_store_transaction(struct db_record *rec, TDB_DATA data, int flag);
510 static NTSTATUS db_ctdb_delete_transaction(struct db_record *rec);
512 static struct db_record *db_ctdb_fetch_locked_transaction(struct db_ctdb_ctx *ctx,
516 struct db_record *result;
519 if (!(result = talloc(mem_ctx, struct db_record))) {
520 DEBUG(0, ("talloc failed\n"));
524 result->private_data = ctx->transaction;
526 result->key.dsize = key.dsize;
527 result->key.dptr = (uint8 *)talloc_memdup(result, key.dptr, key.dsize);
528 if (result->key.dptr == NULL) {
529 DEBUG(0, ("talloc failed\n"));
534 result->store = db_ctdb_store_transaction;
535 result->delete_rec = db_ctdb_delete_transaction;
537 if (pull_newest_from_marshall_buffer(ctx->transaction->m_write, key,
538 NULL, result, &result->value)) {
542 ctdb_data = tdb_fetch(ctx->wtdb->tdb, key);
543 if (ctdb_data.dptr == NULL) {
544 /* create the record */
545 result->value = tdb_null;
549 result->value.dsize = ctdb_data.dsize - sizeof(struct ctdb_ltdb_header);
550 result->value.dptr = NULL;
552 if ((result->value.dsize != 0)
553 && !(result->value.dptr = (uint8 *)talloc_memdup(
554 result, ctdb_data.dptr + sizeof(struct ctdb_ltdb_header),
555 result->value.dsize))) {
556 DEBUG(0, ("talloc failed\n"));
560 SAFE_FREE(ctdb_data.dptr);
565 static int db_ctdb_record_destructor(struct db_record **recp)
567 struct db_record *rec = talloc_get_type_abort(*recp, struct db_record);
568 struct db_ctdb_transaction_handle *h = talloc_get_type_abort(
569 rec->private_data, struct db_ctdb_transaction_handle);
570 int ret = h->ctx->db->transaction_commit(h->ctx->db);
572 DEBUG(0,(__location__ " transaction_commit failed\n"));
578 auto-create a transaction for persistent databases
580 static struct db_record *db_ctdb_fetch_locked_persistent(struct db_ctdb_ctx *ctx,
585 struct db_record *rec, **recp;
587 res = db_ctdb_transaction_start(ctx->db);
592 rec = db_ctdb_fetch_locked_transaction(ctx, mem_ctx, key);
594 ctx->db->transaction_cancel(ctx->db);
598 /* destroy this transaction when we release the lock */
599 recp = talloc(rec, struct db_record *);
601 ctx->db->transaction_cancel(ctx->db);
606 talloc_set_destructor(recp, db_ctdb_record_destructor);
612 stores a record inside a transaction
614 static NTSTATUS db_ctdb_transaction_store(struct db_ctdb_transaction_handle *h,
615 TDB_DATA key, TDB_DATA data)
617 TALLOC_CTX *tmp_ctx = talloc_new(h);
619 struct ctdb_ltdb_header header;
623 /* we need the header so we can update the RSN */
625 if (!pull_newest_from_marshall_buffer(h->m_write, key, &header,
628 rec = tdb_fetch(h->ctx->wtdb->tdb, key);
630 if (rec.dptr != NULL) {
631 memcpy(&header, rec.dptr,
632 sizeof(struct ctdb_ltdb_header));
633 rec.dsize -= sizeof(struct ctdb_ltdb_header);
636 * a special case, we are writing the same
637 * data that is there now
639 if (data.dsize == rec.dsize &&
641 rec.dptr + sizeof(struct ctdb_ltdb_header),
644 talloc_free(tmp_ctx);
651 header.dmaster = get_my_vnn();
654 h->m_all = db_ctdb_marshall_add(h, h->m_all, h->ctx->db_id, 0, key,
656 if (h->m_all == NULL) {
657 DEBUG(0,(__location__ " Failed to add to marshalling "
659 talloc_free(tmp_ctx);
660 return NT_STATUS_NO_MEMORY;
663 h->m_write = db_ctdb_marshall_add(h, h->m_write, h->ctx->db_id, 0, key, &header, data);
664 if (h->m_write == NULL) {
665 DEBUG(0,(__location__ " Failed to add to marshalling record\n"));
666 talloc_free(tmp_ctx);
667 return NT_STATUS_NO_MEMORY;
670 talloc_free(tmp_ctx);
676 a record store inside a transaction
678 static NTSTATUS db_ctdb_store_transaction(struct db_record *rec, TDB_DATA data, int flag)
680 struct db_ctdb_transaction_handle *h = talloc_get_type_abort(
681 rec->private_data, struct db_ctdb_transaction_handle);
684 status = db_ctdb_transaction_store(h, rec->key, data);
689 a record delete inside a transaction
691 static NTSTATUS db_ctdb_delete_transaction(struct db_record *rec)
693 struct db_ctdb_transaction_handle *h = talloc_get_type_abort(
694 rec->private_data, struct db_ctdb_transaction_handle);
697 status = db_ctdb_transaction_store(h, rec->key, tdb_null);
702 * Fetch the db sequence number of a persistent db directly from the db.
704 static NTSTATUS db_ctdb_fetch_db_seqnum_from_db(struct db_ctdb_ctx *db,
708 const char *keyname = CTDB_DB_SEQNUM_KEY;
711 struct ctdb_ltdb_header header;
712 TALLOC_CTX *mem_ctx = talloc_stackframe();
714 if (seqnum == NULL) {
715 return NT_STATUS_INVALID_PARAMETER;
718 key = string_term_tdb_data(keyname);
720 status = db_ctdb_ltdb_fetch(db, key, &header, mem_ctx, &data);
721 if (!NT_STATUS_IS_OK(status) &&
722 !NT_STATUS_EQUAL(status, NT_STATUS_NOT_FOUND))
727 status = NT_STATUS_OK;
729 if (data.dsize != sizeof(uint64_t)) {
734 *seqnum = *(uint64_t *)data.dptr;
737 TALLOC_FREE(mem_ctx);
742 * Store the database sequence number inside a transaction.
744 static NTSTATUS db_ctdb_store_db_seqnum(struct db_ctdb_transaction_handle *h,
748 const char *keyname = CTDB_DB_SEQNUM_KEY;
752 key = string_term_tdb_data(keyname);
754 data.dptr = (uint8_t *)&seqnum;
755 data.dsize = sizeof(uint64_t);
757 status = db_ctdb_transaction_store(h, key, data);
765 static int db_ctdb_transaction_commit(struct db_context *db)
767 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
771 struct db_ctdb_transaction_handle *h = ctx->transaction;
772 uint64_t old_seqnum, new_seqnum;
776 DEBUG(0,(__location__ " transaction commit with no open transaction on db 0x%08x\n", ctx->db_id));
780 if (h->nested_cancel) {
781 db->transaction_cancel(db);
782 DEBUG(5,(__location__ " Failed transaction commit after nested cancel\n"));
786 if (h->nesting != 0) {
791 if (h->m_write == NULL) {
793 * No changes were made, so don't change the seqnum,
794 * don't push to other node, just exit with success.
800 DEBUG(5,(__location__ " Commit transaction on db 0x%08x\n", ctx->db_id));
803 * As the last db action before committing, bump the database sequence
804 * number. Note that this undoes all changes to the seqnum records
805 * performed under the transaction. This record is not meant to be
806 * modified by user interaction. It is for internal use only...
808 rets = db_ctdb_fetch_db_seqnum_from_db(ctx, &old_seqnum);
809 if (!NT_STATUS_IS_OK(rets)) {
810 DEBUG(1, (__location__ " failed to fetch the db sequence number "
811 "in transaction commit on db 0x%08x\n", ctx->db_id));
816 new_seqnum = old_seqnum + 1;
818 rets = db_ctdb_store_db_seqnum(h, new_seqnum);
819 if (!NT_STATUS_IS_OK(rets)) {
820 DEBUG(1, (__location__ "failed to store the db sequence number "
821 " in transaction commit on db 0x%08x\n", ctx->db_id));
827 /* tell ctdbd to commit to the other nodes */
828 rets = ctdbd_control_local(messaging_ctdbd_connection(),
829 CTDB_CONTROL_TRANS3_COMMIT,
831 db_ctdb_marshall_finish(h->m_write),
832 NULL, NULL, &status);
833 if (!NT_STATUS_IS_OK(rets) || status != 0) {
835 * The TRANS3_COMMIT control should only possibly fail when a
836 * recovery has been running concurrently. In any case, the db
837 * will be the same on all nodes, either the new copy or the
838 * old copy. This can be detected by comparing the old and new
839 * local sequence numbers.
841 rets = db_ctdb_fetch_db_seqnum_from_db(ctx, &new_seqnum);
842 if (!NT_STATUS_IS_OK(rets)) {
843 DEBUG(1, (__location__ " failed to refetch db sequence "
844 "number after failed TRANS3_COMMIT\n"));
849 if (new_seqnum == old_seqnum) {
850 /* Recovery prevented all our changes: retry. */
852 } else if (new_seqnum != (old_seqnum + 1)) {
853 DEBUG(0, (__location__ " ERROR: new_seqnum[%lu] != "
854 "old_seqnum[%lu] + (0 or 1) after failed "
855 "TRANS3_COMMIT - this should not happen!\n",
856 (unsigned long)new_seqnum,
857 (unsigned long)old_seqnum));
862 * Recovery propagated our changes to all nodes, completing
863 * our commit for us - succeed.
870 h->ctx->transaction = NULL;
879 static int db_ctdb_transaction_cancel(struct db_context *db)
881 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
883 struct db_ctdb_transaction_handle *h = ctx->transaction;
886 DEBUG(0,(__location__ " transaction cancel with no open transaction on db 0x%08x\n", ctx->db_id));
890 if (h->nesting != 0) {
892 h->nested_cancel = true;
896 DEBUG(5,(__location__ " Cancel transaction on db 0x%08x\n", ctx->db_id));
898 ctx->transaction = NULL;
904 static NTSTATUS db_ctdb_store(struct db_record *rec, TDB_DATA data, int flag)
906 struct db_ctdb_rec *crec = talloc_get_type_abort(
907 rec->private_data, struct db_ctdb_rec);
909 return db_ctdb_ltdb_store(crec->ctdb_ctx, rec->key, &(crec->header), data);
914 #ifdef CTDB_CONTROL_SCHEDULE_FOR_DELETION
915 static NTSTATUS db_ctdb_send_schedule_for_deletion(struct db_record *rec)
918 struct ctdb_control_schedule_for_deletion *dd;
921 struct db_ctdb_rec *crec = talloc_get_type_abort(
922 rec->private_data, struct db_ctdb_rec);
924 indata.dsize = offsetof(struct ctdb_control_schedule_for_deletion, key) + rec->key.dsize;
925 indata.dptr = talloc_zero_array(crec, uint8_t, indata.dsize);
926 if (indata.dptr == NULL) {
927 DEBUG(0, (__location__ " talloc failed!\n"));
928 return NT_STATUS_NO_MEMORY;
931 dd = (struct ctdb_control_schedule_for_deletion *)(void *)indata.dptr;
932 dd->db_id = crec->ctdb_ctx->db_id;
933 dd->hdr = crec->header;
934 dd->keylen = rec->key.dsize;
935 memcpy(dd->key, rec->key.dptr, rec->key.dsize);
937 status = ctdbd_control_local(messaging_ctdbd_connection(),
938 CTDB_CONTROL_SCHEDULE_FOR_DELETION,
939 crec->ctdb_ctx->db_id,
940 CTDB_CTRL_FLAG_NOREPLY, /* flags */
945 talloc_free(indata.dptr);
947 if (!NT_STATUS_IS_OK(status) || cstatus != 0) {
948 DEBUG(1, (__location__ " Error sending local control "
949 "SCHEDULE_FOR_DELETION: %s, cstatus = %d\n",
950 nt_errstr(status), cstatus));
951 if (NT_STATUS_IS_OK(status)) {
952 status = NT_STATUS_UNSUCCESSFUL;
960 static NTSTATUS db_ctdb_delete(struct db_record *rec)
966 * We have to store the header with empty data. TODO: Fix the
972 status = db_ctdb_store(rec, data, 0);
973 if (!NT_STATUS_IS_OK(status)) {
977 #ifdef CTDB_CONTROL_SCHEDULE_FOR_DELETION
978 status = db_ctdb_send_schedule_for_deletion(rec);
984 static int db_ctdb_record_destr(struct db_record* data)
986 struct db_ctdb_rec *crec = talloc_get_type_abort(
987 data->private_data, struct db_ctdb_rec);
990 DEBUG(10, (DEBUGLEVEL > 10
991 ? "Unlocking db %u key %s\n"
992 : "Unlocking db %u key %.20s\n",
993 (int)crec->ctdb_ctx->db_id,
994 hex_encode_talloc(data, (unsigned char *)data->key.dptr,
997 if (tdb_chainunlock(crec->ctdb_ctx->wtdb->tdb, data->key) != 0) {
998 DEBUG(0, ("tdb_chainunlock failed\n"));
1002 threshold = lp_ctdb_locktime_warn_threshold();
1003 if (threshold != 0) {
1004 double timediff = timeval_elapsed(&crec->lock_time);
1005 if ((timediff * 1000) > threshold) {
1006 DEBUG(0, ("Held tdb lock %f seconds\n", timediff));
1013 static struct db_record *fetch_locked_internal(struct db_ctdb_ctx *ctx,
1014 TALLOC_CTX *mem_ctx,
1017 struct db_record *result;
1018 struct db_ctdb_rec *crec;
1021 int migrate_attempts = 0;
1023 if (!(result = talloc(mem_ctx, struct db_record))) {
1024 DEBUG(0, ("talloc failed\n"));
1028 if (!(crec = TALLOC_ZERO_P(result, struct db_ctdb_rec))) {
1029 DEBUG(0, ("talloc failed\n"));
1030 TALLOC_FREE(result);
1034 result->private_data = (void *)crec;
1035 crec->ctdb_ctx = ctx;
1037 result->key.dsize = key.dsize;
1038 result->key.dptr = (uint8 *)talloc_memdup(result, key.dptr, key.dsize);
1039 if (result->key.dptr == NULL) {
1040 DEBUG(0, ("talloc failed\n"));
1041 TALLOC_FREE(result);
1046 * Do a blocking lock on the record
1050 if (DEBUGLEVEL >= 10) {
1051 char *keystr = hex_encode_talloc(result, key.dptr, key.dsize);
1052 DEBUG(10, (DEBUGLEVEL > 10
1053 ? "Locking db %u key %s\n"
1054 : "Locking db %u key %.20s\n",
1055 (int)crec->ctdb_ctx->db_id, keystr));
1056 TALLOC_FREE(keystr);
1059 if (tdb_chainlock(ctx->wtdb->tdb, key) != 0) {
1060 DEBUG(3, ("tdb_chainlock failed\n"));
1061 TALLOC_FREE(result);
1065 result->store = db_ctdb_store;
1066 result->delete_rec = db_ctdb_delete;
1067 talloc_set_destructor(result, db_ctdb_record_destr);
1069 ctdb_data = tdb_fetch(ctx->wtdb->tdb, key);
1072 * See if we have a valid record and we are the dmaster. If so, we can
1073 * take the shortcut and just return it.
1076 if ((ctdb_data.dptr == NULL) ||
1077 (ctdb_data.dsize < sizeof(struct ctdb_ltdb_header)) ||
1078 ((struct ctdb_ltdb_header *)ctdb_data.dptr)->dmaster != get_my_vnn()
1080 || (random() % 2 != 0)
1083 SAFE_FREE(ctdb_data.dptr);
1084 tdb_chainunlock(ctx->wtdb->tdb, key);
1085 talloc_set_destructor(result, NULL);
1087 migrate_attempts += 1;
1089 DEBUG(10, ("ctdb_data.dptr = %p, dmaster = %u (%u)\n",
1090 ctdb_data.dptr, ctdb_data.dptr ?
1091 ((struct ctdb_ltdb_header *)ctdb_data.dptr)->dmaster : -1,
1094 status = ctdbd_migrate(messaging_ctdbd_connection(), ctx->db_id,
1096 if (!NT_STATUS_IS_OK(status)) {
1097 DEBUG(5, ("ctdb_migrate failed: %s\n",
1098 nt_errstr(status)));
1099 TALLOC_FREE(result);
1102 /* now its migrated, try again */
1106 if (migrate_attempts > 10) {
1107 DEBUG(0, ("db_ctdb_fetch_locked needed %d attempts\n",
1111 GetTimeOfDay(&crec->lock_time);
1113 memcpy(&crec->header, ctdb_data.dptr, sizeof(crec->header));
1115 result->value.dsize = ctdb_data.dsize - sizeof(crec->header);
1116 result->value.dptr = NULL;
1118 if ((result->value.dsize != 0)
1119 && !(result->value.dptr = (uint8 *)talloc_memdup(
1120 result, ctdb_data.dptr + sizeof(crec->header),
1121 result->value.dsize))) {
1122 DEBUG(0, ("talloc failed\n"));
1123 TALLOC_FREE(result);
1126 SAFE_FREE(ctdb_data.dptr);
1131 static struct db_record *db_ctdb_fetch_locked(struct db_context *db,
1132 TALLOC_CTX *mem_ctx,
1135 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
1136 struct db_ctdb_ctx);
1138 if (ctx->transaction != NULL) {
1139 return db_ctdb_fetch_locked_transaction(ctx, mem_ctx, key);
1142 if (db->persistent) {
1143 return db_ctdb_fetch_locked_persistent(ctx, mem_ctx, key);
1146 return fetch_locked_internal(ctx, mem_ctx, key);
1150 fetch (unlocked, no migration) operation on ctdb
1152 static int db_ctdb_fetch(struct db_context *db, TALLOC_CTX *mem_ctx,
1153 TDB_DATA key, TDB_DATA *data)
1155 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
1156 struct db_ctdb_ctx);
1160 if (ctx->transaction) {
1161 return db_ctdb_transaction_fetch(ctx, mem_ctx, key, data);
1164 if (db->persistent) {
1165 return db_ctdb_fetch_persistent(ctx, mem_ctx, key, data);
1168 /* try a direct fetch */
1169 ctdb_data = tdb_fetch(ctx->wtdb->tdb, key);
1172 * See if we have a valid record and we are the dmaster. If so, we can
1173 * take the shortcut and just return it.
1174 * we bypass the dmaster check for persistent databases
1176 if ((ctdb_data.dptr != NULL) &&
1177 (ctdb_data.dsize >= sizeof(struct ctdb_ltdb_header)) &&
1178 ((struct ctdb_ltdb_header *)ctdb_data.dptr)->dmaster == get_my_vnn())
1180 /* we are the dmaster - avoid the ctdb protocol op */
1182 data->dsize = ctdb_data.dsize - sizeof(struct ctdb_ltdb_header);
1183 if (data->dsize == 0) {
1184 SAFE_FREE(ctdb_data.dptr);
1189 data->dptr = (uint8 *)talloc_memdup(
1190 mem_ctx, ctdb_data.dptr+sizeof(struct ctdb_ltdb_header),
1193 SAFE_FREE(ctdb_data.dptr);
1195 if (data->dptr == NULL) {
1201 SAFE_FREE(ctdb_data.dptr);
1203 /* we weren't able to get it locally - ask ctdb to fetch it for us */
1204 status = ctdbd_fetch(messaging_ctdbd_connection(), ctx->db_id, key,
1206 if (!NT_STATUS_IS_OK(status)) {
1207 DEBUG(5, ("ctdbd_fetch failed: %s\n", nt_errstr(status)));
1214 struct traverse_state {
1215 struct db_context *db;
1216 int (*fn)(struct db_record *rec, void *private_data);
1220 static void traverse_callback(TDB_DATA key, TDB_DATA data, void *private_data)
1222 struct traverse_state *state = (struct traverse_state *)private_data;
1223 struct db_record *rec;
1224 TALLOC_CTX *tmp_ctx = talloc_new(state->db);
1225 /* we have to give them a locked record to prevent races */
1226 rec = db_ctdb_fetch_locked(state->db, tmp_ctx, key);
1227 if (rec && rec->value.dsize > 0) {
1228 state->fn(rec, state->private_data);
1230 talloc_free(tmp_ctx);
1233 static int traverse_persistent_callback(TDB_CONTEXT *tdb, TDB_DATA kbuf, TDB_DATA dbuf,
1236 struct traverse_state *state = (struct traverse_state *)private_data;
1237 struct db_record *rec;
1238 TALLOC_CTX *tmp_ctx = talloc_new(state->db);
1240 /* we have to give them a locked record to prevent races */
1241 rec = db_ctdb_fetch_locked(state->db, tmp_ctx, kbuf);
1242 if (rec && rec->value.dsize > 0) {
1243 ret = state->fn(rec, state->private_data);
1245 talloc_free(tmp_ctx);
1249 static int db_ctdb_traverse(struct db_context *db,
1250 int (*fn)(struct db_record *rec,
1251 void *private_data),
1254 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
1255 struct db_ctdb_ctx);
1256 struct traverse_state state;
1260 state.private_data = private_data;
1262 if (db->persistent) {
1263 /* for persistent databases we don't need to do a ctdb traverse,
1264 we can do a faster local traverse */
1265 return tdb_traverse(ctx->wtdb->tdb, traverse_persistent_callback, &state);
1269 ctdbd_traverse(ctx->db_id, traverse_callback, &state);
1273 static NTSTATUS db_ctdb_store_deny(struct db_record *rec, TDB_DATA data, int flag)
1275 return NT_STATUS_MEDIA_WRITE_PROTECTED;
1278 static NTSTATUS db_ctdb_delete_deny(struct db_record *rec)
1280 return NT_STATUS_MEDIA_WRITE_PROTECTED;
1283 static void traverse_read_callback(TDB_DATA key, TDB_DATA data, void *private_data)
1285 struct traverse_state *state = (struct traverse_state *)private_data;
1286 struct db_record rec;
1289 rec.store = db_ctdb_store_deny;
1290 rec.delete_rec = db_ctdb_delete_deny;
1291 rec.private_data = state->db;
1292 state->fn(&rec, state->private_data);
1295 static int traverse_persistent_callback_read(TDB_CONTEXT *tdb, TDB_DATA kbuf, TDB_DATA dbuf,
1298 struct traverse_state *state = (struct traverse_state *)private_data;
1299 struct db_record rec;
1302 rec.store = db_ctdb_store_deny;
1303 rec.delete_rec = db_ctdb_delete_deny;
1304 rec.private_data = state->db;
1306 if (rec.value.dsize <= sizeof(struct ctdb_ltdb_header)) {
1307 /* a deleted record */
1310 rec.value.dsize -= sizeof(struct ctdb_ltdb_header);
1311 rec.value.dptr += sizeof(struct ctdb_ltdb_header);
1313 return state->fn(&rec, state->private_data);
1316 static int db_ctdb_traverse_read(struct db_context *db,
1317 int (*fn)(struct db_record *rec,
1318 void *private_data),
1321 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
1322 struct db_ctdb_ctx);
1323 struct traverse_state state;
1327 state.private_data = private_data;
1329 if (db->persistent) {
1330 /* for persistent databases we don't need to do a ctdb traverse,
1331 we can do a faster local traverse */
1332 return tdb_traverse_read(ctx->wtdb->tdb, traverse_persistent_callback_read, &state);
1335 ctdbd_traverse(ctx->db_id, traverse_read_callback, &state);
1339 static int db_ctdb_get_seqnum(struct db_context *db)
1341 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
1342 struct db_ctdb_ctx);
1343 return tdb_get_seqnum(ctx->wtdb->tdb);
1346 static int db_ctdb_get_flags(struct db_context *db)
1348 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
1349 struct db_ctdb_ctx);
1350 return tdb_get_flags(ctx->wtdb->tdb);
1353 struct db_context *db_open_ctdb(TALLOC_CTX *mem_ctx,
1355 int hash_size, int tdb_flags,
1356 int open_flags, mode_t mode)
1358 struct db_context *result;
1359 struct db_ctdb_ctx *db_ctdb;
1361 struct ctdbd_connection *conn;
1363 if (!lp_clustering()) {
1364 DEBUG(10, ("Clustering disabled -- no ctdb\n"));
1368 if (!(result = TALLOC_ZERO_P(mem_ctx, struct db_context))) {
1369 DEBUG(0, ("talloc failed\n"));
1370 TALLOC_FREE(result);
1374 if (!(db_ctdb = talloc(result, struct db_ctdb_ctx))) {
1375 DEBUG(0, ("talloc failed\n"));
1376 TALLOC_FREE(result);
1380 db_ctdb->transaction = NULL;
1381 db_ctdb->db = result;
1383 conn = messaging_ctdbd_connection();
1385 DEBUG(1, ("Could not connect to ctdb\n"));
1386 TALLOC_FREE(result);
1390 if (!NT_STATUS_IS_OK(ctdbd_db_attach(conn, name, &db_ctdb->db_id, tdb_flags))) {
1391 DEBUG(0, ("ctdbd_db_attach failed for %s\n", name));
1392 TALLOC_FREE(result);
1396 db_path = ctdbd_dbpath(conn, db_ctdb, db_ctdb->db_id);
1398 result->persistent = ((tdb_flags & TDB_CLEAR_IF_FIRST) == 0);
1400 /* only pass through specific flags */
1401 tdb_flags &= TDB_SEQNUM;
1403 /* honor permissions if user has specified O_CREAT */
1404 if (open_flags & O_CREAT) {
1405 chmod(db_path, mode);
1408 db_ctdb->wtdb = tdb_wrap_open(db_ctdb, db_path, hash_size, tdb_flags, O_RDWR, 0);
1409 if (db_ctdb->wtdb == NULL) {
1410 DEBUG(0, ("Could not open tdb %s: %s\n", db_path, strerror(errno)));
1411 TALLOC_FREE(result);
1414 talloc_free(db_path);
1416 if (result->persistent) {
1417 db_ctdb->lock_ctx = g_lock_ctx_init(db_ctdb,
1418 ctdb_conn_msg_ctx(conn));
1419 if (db_ctdb->lock_ctx == NULL) {
1420 DEBUG(0, ("g_lock_ctx_init failed\n"));
1421 TALLOC_FREE(result);
1426 result->private_data = (void *)db_ctdb;
1427 result->fetch_locked = db_ctdb_fetch_locked;
1428 result->fetch = db_ctdb_fetch;
1429 result->traverse = db_ctdb_traverse;
1430 result->traverse_read = db_ctdb_traverse_read;
1431 result->get_seqnum = db_ctdb_get_seqnum;
1432 result->get_flags = db_ctdb_get_flags;
1433 result->transaction_start = db_ctdb_transaction_start;
1434 result->transaction_commit = db_ctdb_transaction_commit;
1435 result->transaction_cancel = db_ctdb_transaction_cancel;
1437 DEBUG(3,("db_open_ctdb: opened database '%s' with dbid 0x%x\n",
1438 name, db_ctdb->db_id));