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 #ifdef CLUSTER_SUPPORT
24 #include "ctdb_private.h"
25 #include "ctdbd_conn.h"
28 struct db_ctdb_transaction_handle {
29 struct db_ctdb_ctx *ctx;
31 * we store the reads and writes done under a transaction:
32 * - one list stores both reads and writes (m_all),
33 * - the other just writes (m_write)
35 struct ctdb_marshall_buffer *m_all;
36 struct ctdb_marshall_buffer *m_write;
43 struct db_context *db;
44 struct tdb_wrap *wtdb;
46 struct db_ctdb_transaction_handle *transaction;
47 struct g_lock_ctx *lock_ctx;
51 struct db_ctdb_ctx *ctdb_ctx;
52 struct ctdb_ltdb_header header;
53 struct timeval lock_time;
56 static struct db_record *fetch_locked_internal(struct db_ctdb_ctx *ctx,
61 static NTSTATUS tdb_error_to_ntstatus(struct tdb_context *tdb)
64 enum TDB_ERROR tret = tdb_error(tdb);
68 status = NT_STATUS_OBJECT_NAME_COLLISION;
71 status = NT_STATUS_OBJECT_NAME_NOT_FOUND;
74 status = NT_STATUS_INTERNAL_DB_CORRUPTION;
83 * fetch a record from the tdb, separating out the header
84 * information and returning the body of the record.
86 static NTSTATUS db_ctdb_ltdb_fetch(struct db_ctdb_ctx *db,
88 struct ctdb_ltdb_header *header,
95 rec = tdb_fetch(db->wtdb->tdb, key);
96 if (rec.dsize < sizeof(struct ctdb_ltdb_header)) {
97 status = NT_STATUS_NOT_FOUND;
102 header->dmaster = (uint32_t)-1;
109 *header = *(struct ctdb_ltdb_header *)rec.dptr;
113 data->dsize = rec.dsize - sizeof(struct ctdb_ltdb_header);
114 if (data->dsize == 0) {
117 data->dptr = (unsigned char *)talloc_memdup(mem_ctx,
119 + sizeof(struct ctdb_ltdb_header),
121 if (data->dptr == NULL) {
122 status = NT_STATUS_NO_MEMORY;
128 status = NT_STATUS_OK;
136 * Store a record together with the ctdb record header
137 * in the local copy of the database.
139 static NTSTATUS db_ctdb_ltdb_store(struct db_ctdb_ctx *db,
141 struct ctdb_ltdb_header *header,
144 TALLOC_CTX *tmp_ctx = talloc_stackframe();
148 rec.dsize = data.dsize + sizeof(struct ctdb_ltdb_header);
149 rec.dptr = (uint8_t *)talloc_size(tmp_ctx, rec.dsize);
151 if (rec.dptr == NULL) {
152 talloc_free(tmp_ctx);
153 return NT_STATUS_NO_MEMORY;
156 memcpy(rec.dptr, header, sizeof(struct ctdb_ltdb_header));
157 memcpy(sizeof(struct ctdb_ltdb_header) + (uint8_t *)rec.dptr, data.dptr, data.dsize);
159 ret = tdb_store(db->wtdb->tdb, key, rec, TDB_REPLACE);
161 talloc_free(tmp_ctx);
163 return (ret == 0) ? NT_STATUS_OK
164 : tdb_error_to_ntstatus(db->wtdb->tdb);
169 form a ctdb_rec_data record from a key/data pair
171 note that header may be NULL. If not NULL then it is included in the data portion
174 static struct ctdb_rec_data *db_ctdb_marshall_record(TALLOC_CTX *mem_ctx, uint32_t reqid,
176 struct ctdb_ltdb_header *header,
180 struct ctdb_rec_data *d;
182 length = offsetof(struct ctdb_rec_data, data) + key.dsize +
183 data.dsize + (header?sizeof(*header):0);
184 d = (struct ctdb_rec_data *)talloc_size(mem_ctx, length);
190 d->keylen = key.dsize;
191 memcpy(&d->data[0], key.dptr, key.dsize);
193 d->datalen = data.dsize + sizeof(*header);
194 memcpy(&d->data[key.dsize], header, sizeof(*header));
195 memcpy(&d->data[key.dsize+sizeof(*header)], data.dptr, data.dsize);
197 d->datalen = data.dsize;
198 memcpy(&d->data[key.dsize], data.dptr, data.dsize);
204 /* helper function for marshalling multiple records */
205 static struct ctdb_marshall_buffer *db_ctdb_marshall_add(TALLOC_CTX *mem_ctx,
206 struct ctdb_marshall_buffer *m,
210 struct ctdb_ltdb_header *header,
213 struct ctdb_rec_data *r;
214 size_t m_size, r_size;
215 struct ctdb_marshall_buffer *m2 = NULL;
217 r = db_ctdb_marshall_record(talloc_tos(), reqid, key, header, data);
224 m = (struct ctdb_marshall_buffer *)talloc_zero_size(
225 mem_ctx, offsetof(struct ctdb_marshall_buffer, data));
232 m_size = talloc_get_size(m);
233 r_size = talloc_get_size(r);
235 m2 = (struct ctdb_marshall_buffer *)talloc_realloc_size(
236 mem_ctx, m, m_size + r_size);
242 memcpy(m_size + (uint8_t *)m2, r, r_size);
251 /* we've finished marshalling, return a data blob with the marshalled records */
252 static TDB_DATA db_ctdb_marshall_finish(struct ctdb_marshall_buffer *m)
255 data.dptr = (uint8_t *)m;
256 data.dsize = talloc_get_size(m);
261 loop over a marshalling buffer
263 - pass r==NULL to start
264 - loop the number of times indicated by m->count
266 static struct ctdb_rec_data *db_ctdb_marshall_loop_next(struct ctdb_marshall_buffer *m, struct ctdb_rec_data *r,
268 struct ctdb_ltdb_header *header,
269 TDB_DATA *key, TDB_DATA *data)
272 r = (struct ctdb_rec_data *)&m->data[0];
274 r = (struct ctdb_rec_data *)(r->length + (uint8_t *)r);
282 key->dptr = &r->data[0];
283 key->dsize = r->keylen;
286 data->dptr = &r->data[r->keylen];
287 data->dsize = r->datalen;
288 if (header != NULL) {
289 data->dptr += sizeof(*header);
290 data->dsize -= sizeof(*header);
294 if (header != NULL) {
295 if (r->datalen < sizeof(*header)) {
298 *header = *(struct ctdb_ltdb_header *)&r->data[r->keylen];
305 * CTDB transaction destructor
307 static int db_ctdb_transaction_destructor(struct db_ctdb_transaction_handle *h)
311 status = g_lock_unlock(h->ctx->lock_ctx, h->lock_name);
312 if (!NT_STATUS_IS_OK(status)) {
313 DEBUG(0, ("g_lock_unlock failed: %s\n", nt_errstr(status)));
320 * CTDB dbwrap API: transaction_start function
321 * starts a transaction on a persistent database
323 static int db_ctdb_transaction_start(struct db_context *db)
325 struct db_ctdb_transaction_handle *h;
327 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
330 if (!db->persistent) {
331 DEBUG(0,("transactions not supported on non-persistent database 0x%08x\n",
336 if (ctx->transaction) {
337 ctx->transaction->nesting++;
341 h = talloc_zero(db, struct db_ctdb_transaction_handle);
343 DEBUG(0,(__location__ " oom for transaction handle\n"));
349 h->lock_name = talloc_asprintf(h, "transaction_db_0x%08x",
350 (unsigned int)ctx->db_id);
351 if (h->lock_name == NULL) {
352 DEBUG(0, ("talloc_asprintf failed\n"));
358 * Wait a day, i.e. forever...
360 status = g_lock_lock(ctx->lock_ctx, h->lock_name, G_LOCK_WRITE,
361 timeval_set(86400, 0));
362 if (!NT_STATUS_IS_OK(status)) {
363 DEBUG(0, ("g_lock_lock failed: %s\n", nt_errstr(status)));
368 talloc_set_destructor(h, db_ctdb_transaction_destructor);
370 ctx->transaction = h;
372 DEBUG(5,(__location__ " Started transaction on db 0x%08x\n", ctx->db_id));
377 static bool pull_newest_from_marshall_buffer(struct ctdb_marshall_buffer *buf,
379 struct ctdb_ltdb_header *pheader,
383 struct ctdb_rec_data *rec = NULL;
384 struct ctdb_ltdb_header h;
397 * Walk the list of records written during this
398 * transaction. If we want to read one we have already
399 * written, return the last written sample. Thus we do not do
400 * a "break;" for the first hit, this record might have been
404 for (i=0; i<buf->count; i++) {
405 TDB_DATA tkey, tdata;
407 struct ctdb_ltdb_header hdr;
411 rec = db_ctdb_marshall_loop_next(buf, rec, &reqid, &hdr, &tkey,
417 if (tdb_data_equal(key, tkey)) {
429 data.dptr = (uint8_t *)talloc_memdup(mem_ctx, data.dptr,
431 if ((data.dsize != 0) && (data.dptr == NULL)) {
437 if (pheader != NULL) {
445 fetch a record inside a transaction
447 static int db_ctdb_transaction_fetch(struct db_ctdb_ctx *db,
449 TDB_DATA key, TDB_DATA *data)
451 struct db_ctdb_transaction_handle *h = db->transaction;
455 found = pull_newest_from_marshall_buffer(h->m_write, key, NULL,
461 status = db_ctdb_ltdb_fetch(h->ctx, key, NULL, mem_ctx, data);
463 if (NT_STATUS_EQUAL(status, NT_STATUS_NOT_FOUND)) {
465 } else if (!NT_STATUS_IS_OK(status)) {
469 h->m_all = db_ctdb_marshall_add(h, h->m_all, h->ctx->db_id, 1, key,
471 if (h->m_all == NULL) {
472 DEBUG(0,(__location__ " Failed to add to marshalling "
475 talloc_free(data->dptr);
483 static NTSTATUS db_ctdb_store_transaction(struct db_record *rec, TDB_DATA data, int flag);
484 static NTSTATUS db_ctdb_delete_transaction(struct db_record *rec);
486 static struct db_record *db_ctdb_fetch_locked_transaction(struct db_ctdb_ctx *ctx,
490 struct db_record *result;
493 if (!(result = talloc(mem_ctx, struct db_record))) {
494 DEBUG(0, ("talloc failed\n"));
498 result->private_data = ctx->transaction;
500 result->key.dsize = key.dsize;
501 result->key.dptr = (uint8 *)talloc_memdup(result, key.dptr, key.dsize);
502 if (result->key.dptr == NULL) {
503 DEBUG(0, ("talloc failed\n"));
508 result->store = db_ctdb_store_transaction;
509 result->delete_rec = db_ctdb_delete_transaction;
511 if (pull_newest_from_marshall_buffer(ctx->transaction->m_write, key,
512 NULL, result, &result->value)) {
516 ctdb_data = tdb_fetch(ctx->wtdb->tdb, key);
517 if (ctdb_data.dptr == NULL) {
518 /* create the record */
519 result->value = tdb_null;
523 result->value.dsize = ctdb_data.dsize - sizeof(struct ctdb_ltdb_header);
524 result->value.dptr = NULL;
526 if ((result->value.dsize != 0)
527 && !(result->value.dptr = (uint8 *)talloc_memdup(
528 result, ctdb_data.dptr + sizeof(struct ctdb_ltdb_header),
529 result->value.dsize))) {
530 DEBUG(0, ("talloc failed\n"));
534 SAFE_FREE(ctdb_data.dptr);
539 static int db_ctdb_record_destructor(struct db_record **recp)
541 struct db_record *rec = talloc_get_type_abort(*recp, struct db_record);
542 struct db_ctdb_transaction_handle *h = talloc_get_type_abort(
543 rec->private_data, struct db_ctdb_transaction_handle);
544 int ret = h->ctx->db->transaction_commit(h->ctx->db);
546 DEBUG(0,(__location__ " transaction_commit failed\n"));
552 auto-create a transaction for persistent databases
554 static struct db_record *db_ctdb_fetch_locked_persistent(struct db_ctdb_ctx *ctx,
559 struct db_record *rec, **recp;
561 res = db_ctdb_transaction_start(ctx->db);
566 rec = db_ctdb_fetch_locked_transaction(ctx, mem_ctx, key);
568 ctx->db->transaction_cancel(ctx->db);
572 /* destroy this transaction when we release the lock */
573 recp = talloc(rec, struct db_record *);
575 ctx->db->transaction_cancel(ctx->db);
580 talloc_set_destructor(recp, db_ctdb_record_destructor);
586 stores a record inside a transaction
588 static NTSTATUS db_ctdb_transaction_store(struct db_ctdb_transaction_handle *h,
589 TDB_DATA key, TDB_DATA data)
591 TALLOC_CTX *tmp_ctx = talloc_new(h);
593 struct ctdb_ltdb_header header;
597 /* we need the header so we can update the RSN */
599 if (!pull_newest_from_marshall_buffer(h->m_write, key, &header,
602 rec = tdb_fetch(h->ctx->wtdb->tdb, key);
604 if (rec.dptr != NULL) {
605 memcpy(&header, rec.dptr,
606 sizeof(struct ctdb_ltdb_header));
607 rec.dsize -= sizeof(struct ctdb_ltdb_header);
610 * a special case, we are writing the same
611 * data that is there now
613 if (data.dsize == rec.dsize &&
615 rec.dptr + sizeof(struct ctdb_ltdb_header),
618 talloc_free(tmp_ctx);
625 header.dmaster = get_my_vnn();
628 h->m_all = db_ctdb_marshall_add(h, h->m_all, h->ctx->db_id, 0, key,
630 if (h->m_all == NULL) {
631 DEBUG(0,(__location__ " Failed to add to marshalling "
633 talloc_free(tmp_ctx);
634 return NT_STATUS_NO_MEMORY;
637 h->m_write = db_ctdb_marshall_add(h, h->m_write, h->ctx->db_id, 0, key, &header, data);
638 if (h->m_write == NULL) {
639 DEBUG(0,(__location__ " Failed to add to marshalling record\n"));
640 talloc_free(tmp_ctx);
641 return NT_STATUS_NO_MEMORY;
644 talloc_free(tmp_ctx);
650 a record store inside a transaction
652 static NTSTATUS db_ctdb_store_transaction(struct db_record *rec, TDB_DATA data, int flag)
654 struct db_ctdb_transaction_handle *h = talloc_get_type_abort(
655 rec->private_data, struct db_ctdb_transaction_handle);
658 status = db_ctdb_transaction_store(h, rec->key, data);
663 a record delete inside a transaction
665 static NTSTATUS db_ctdb_delete_transaction(struct db_record *rec)
667 struct db_ctdb_transaction_handle *h = talloc_get_type_abort(
668 rec->private_data, struct db_ctdb_transaction_handle);
671 status = db_ctdb_transaction_store(h, rec->key, tdb_null);
676 * Fetch the db sequence number of a persistent db directly from the db.
678 static NTSTATUS db_ctdb_fetch_db_seqnum_from_db(struct db_ctdb_ctx *db,
682 const char *keyname = CTDB_DB_SEQNUM_KEY;
685 struct ctdb_ltdb_header header;
686 TALLOC_CTX *mem_ctx = talloc_stackframe();
688 if (seqnum == NULL) {
689 return NT_STATUS_INVALID_PARAMETER;
692 key = string_term_tdb_data(keyname);
694 status = db_ctdb_ltdb_fetch(db, key, &header, mem_ctx, &data);
695 if (!NT_STATUS_IS_OK(status) &&
696 !NT_STATUS_EQUAL(status, NT_STATUS_NOT_FOUND))
701 status = NT_STATUS_OK;
703 if (data.dsize != sizeof(uint64_t)) {
708 *seqnum = *(uint64_t *)data.dptr;
711 TALLOC_FREE(mem_ctx);
716 * Store the database sequence number inside a transaction.
718 static NTSTATUS db_ctdb_store_db_seqnum(struct db_ctdb_transaction_handle *h,
722 const char *keyname = CTDB_DB_SEQNUM_KEY;
726 key = string_term_tdb_data(keyname);
728 data.dptr = (uint8_t *)&seqnum;
729 data.dsize = sizeof(uint64_t);
731 status = db_ctdb_transaction_store(h, key, data);
739 static int db_ctdb_transaction_commit(struct db_context *db)
741 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
745 struct db_ctdb_transaction_handle *h = ctx->transaction;
746 uint64_t old_seqnum, new_seqnum;
750 DEBUG(0,(__location__ " transaction commit with no open transaction on db 0x%08x\n", ctx->db_id));
754 if (h->nested_cancel) {
755 db->transaction_cancel(db);
756 DEBUG(5,(__location__ " Failed transaction commit after nested cancel\n"));
760 if (h->nesting != 0) {
765 if (h->m_write == NULL) {
767 * No changes were made, so don't change the seqnum,
768 * don't push to other node, just exit with success.
774 DEBUG(5,(__location__ " Commit transaction on db 0x%08x\n", ctx->db_id));
777 * As the last db action before committing, bump the database sequence
778 * number. Note that this undoes all changes to the seqnum records
779 * performed under the transaction. This record is not meant to be
780 * modified by user interaction. It is for internal use only...
782 rets = db_ctdb_fetch_db_seqnum_from_db(ctx, &old_seqnum);
783 if (!NT_STATUS_IS_OK(rets)) {
784 DEBUG(1, (__location__ " failed to fetch the db sequence number "
785 "in transaction commit on db 0x%08x\n", ctx->db_id));
790 new_seqnum = old_seqnum + 1;
792 rets = db_ctdb_store_db_seqnum(h, new_seqnum);
793 if (!NT_STATUS_IS_OK(rets)) {
794 DEBUG(1, (__location__ "failed to store the db sequence number "
795 " in transaction commit on db 0x%08x\n", ctx->db_id));
801 /* tell ctdbd to commit to the other nodes */
802 rets = ctdbd_control_local(messaging_ctdbd_connection(),
803 CTDB_CONTROL_TRANS3_COMMIT,
805 db_ctdb_marshall_finish(h->m_write),
806 NULL, NULL, &status);
807 if (!NT_STATUS_IS_OK(rets) || status != 0) {
809 * The TRANS3_COMMIT control should only possibly fail when a
810 * recovery has been running concurrently. In any case, the db
811 * will be the same on all nodes, either the new copy or the
812 * old copy. This can be detected by comparing the old and new
813 * local sequence numbers.
815 rets = db_ctdb_fetch_db_seqnum_from_db(ctx, &new_seqnum);
816 if (!NT_STATUS_IS_OK(rets)) {
817 DEBUG(1, (__location__ " failed to refetch db sequence "
818 "number after failed TRANS3_COMMIT\n"));
823 if (new_seqnum == old_seqnum) {
824 /* Recovery prevented all our changes: retry. */
826 } else if (new_seqnum != (old_seqnum + 1)) {
827 DEBUG(0, (__location__ " ERROR: new_seqnum[%lu] != "
828 "old_seqnum[%lu] + (0 or 1) after failed "
829 "TRANS3_COMMIT - this should not happen!\n",
830 (unsigned long)new_seqnum,
831 (unsigned long)old_seqnum));
836 * Recovery propagated our changes to all nodes, completing
837 * our commit for us - succeed.
844 h->ctx->transaction = NULL;
853 static int db_ctdb_transaction_cancel(struct db_context *db)
855 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
857 struct db_ctdb_transaction_handle *h = ctx->transaction;
860 DEBUG(0,(__location__ " transaction cancel with no open transaction on db 0x%08x\n", ctx->db_id));
864 if (h->nesting != 0) {
866 h->nested_cancel = true;
870 DEBUG(5,(__location__ " Cancel transaction on db 0x%08x\n", ctx->db_id));
872 ctx->transaction = NULL;
878 static NTSTATUS db_ctdb_store(struct db_record *rec, TDB_DATA data, int flag)
880 struct db_ctdb_rec *crec = talloc_get_type_abort(
881 rec->private_data, struct db_ctdb_rec);
883 return db_ctdb_ltdb_store(crec->ctdb_ctx, rec->key, &(crec->header), data);
888 static NTSTATUS db_ctdb_delete(struct db_record *rec)
892 struct ctdb_control_schedule_for_deletion *dd;
895 struct db_ctdb_rec *crec = talloc_get_type_abort(
896 rec->private_data, struct db_ctdb_rec);
899 * We have to store the header with empty data. TODO: Fix the
905 status = db_ctdb_store(rec, data, 0);
906 if (!NT_STATUS_IS_OK(status)) {
910 indata.dsize = offsetof(struct ctdb_control_schedule_for_deletion, key) + rec->key.dsize;
911 indata.dptr = talloc_zero_array(crec, uint8_t, indata.dsize);
912 if (indata.dptr == NULL) {
913 printf("out of memory\n");
916 dd = (struct ctdb_control_schedule_for_deletion *)(void *)indata.dptr;
917 dd->db_id = crec->ctdb_ctx->db_id;
918 dd->hdr = crec->header;
919 dd->keylen = rec->key.dsize;
920 memcpy(dd->key, rec->key.dptr, rec->key.dsize);
922 status = ctdbd_control_local(messaging_ctdbd_connection(),
923 CTDB_CONTROL_SCHEDULE_FOR_DELETION,
924 crec->ctdb_ctx->db_id,
930 talloc_free(indata.dptr);
932 if (!NT_STATUS_IS_OK(status) || cstatus != 0) {
933 DEBUG(1, (__location__ " Error sending local control "
934 "SCHEDULE_FOR_DELETION: %s, cstatus = %d\n",
935 nt_errstr(status), cstatus));
936 if (NT_STATUS_IS_OK(status)) {
937 status = NT_STATUS_UNSUCCESSFUL;
944 static int db_ctdb_record_destr(struct db_record* data)
946 struct db_ctdb_rec *crec = talloc_get_type_abort(
947 data->private_data, struct db_ctdb_rec);
950 DEBUG(10, (DEBUGLEVEL > 10
951 ? "Unlocking db %u key %s\n"
952 : "Unlocking db %u key %.20s\n",
953 (int)crec->ctdb_ctx->db_id,
954 hex_encode_talloc(data, (unsigned char *)data->key.dptr,
957 if (tdb_chainunlock(crec->ctdb_ctx->wtdb->tdb, data->key) != 0) {
958 DEBUG(0, ("tdb_chainunlock failed\n"));
962 threshold = lp_ctdb_locktime_warn_threshold();
963 if (threshold != 0) {
964 double timediff = timeval_elapsed(&crec->lock_time);
965 if ((timediff * 1000) > threshold) {
966 DEBUG(0, ("Held tdb lock %f seconds\n", timediff));
973 static struct db_record *fetch_locked_internal(struct db_ctdb_ctx *ctx,
978 struct db_record *result;
979 struct db_ctdb_rec *crec;
982 int migrate_attempts = 0;
984 if (!(result = talloc(mem_ctx, struct db_record))) {
985 DEBUG(0, ("talloc failed\n"));
989 if (!(crec = TALLOC_ZERO_P(result, struct db_ctdb_rec))) {
990 DEBUG(0, ("talloc failed\n"));
995 result->private_data = (void *)crec;
996 crec->ctdb_ctx = ctx;
998 result->key.dsize = key.dsize;
999 result->key.dptr = (uint8 *)talloc_memdup(result, key.dptr, key.dsize);
1000 if (result->key.dptr == NULL) {
1001 DEBUG(0, ("talloc failed\n"));
1002 TALLOC_FREE(result);
1007 * Do a blocking lock on the record
1011 if (DEBUGLEVEL >= 10) {
1012 char *keystr = hex_encode_talloc(result, key.dptr, key.dsize);
1013 DEBUG(10, (DEBUGLEVEL > 10
1014 ? "Locking db %u key %s\n"
1015 : "Locking db %u key %.20s\n",
1016 (int)crec->ctdb_ctx->db_id, keystr));
1017 TALLOC_FREE(keystr);
1020 if (tdb_chainlock(ctx->wtdb->tdb, key) != 0) {
1021 DEBUG(3, ("tdb_chainlock failed\n"));
1022 TALLOC_FREE(result);
1026 result->store = db_ctdb_store;
1027 result->delete_rec = db_ctdb_delete;
1028 talloc_set_destructor(result, db_ctdb_record_destr);
1030 ctdb_data = tdb_fetch(ctx->wtdb->tdb, key);
1033 * See if we have a valid record and we are the dmaster. If so, we can
1034 * take the shortcut and just return it.
1037 if ((ctdb_data.dptr == NULL) ||
1038 (ctdb_data.dsize < sizeof(struct ctdb_ltdb_header)) ||
1039 ((struct ctdb_ltdb_header *)ctdb_data.dptr)->dmaster != get_my_vnn()
1041 || (random() % 2 != 0)
1044 SAFE_FREE(ctdb_data.dptr);
1045 tdb_chainunlock(ctx->wtdb->tdb, key);
1046 talloc_set_destructor(result, NULL);
1048 migrate_attempts += 1;
1050 DEBUG(10, ("ctdb_data.dptr = %p, dmaster = %u (%u)\n",
1051 ctdb_data.dptr, ctdb_data.dptr ?
1052 ((struct ctdb_ltdb_header *)ctdb_data.dptr)->dmaster : -1,
1055 status = ctdbd_migrate(messaging_ctdbd_connection(),ctx->db_id, key);
1056 if (!NT_STATUS_IS_OK(status)) {
1057 DEBUG(5, ("ctdb_migrate failed: %s\n",
1058 nt_errstr(status)));
1059 TALLOC_FREE(result);
1062 /* now its migrated, try again */
1066 if (migrate_attempts > 10) {
1067 DEBUG(0, ("db_ctdb_fetch_locked needed %d attempts\n",
1071 GetTimeOfDay(&crec->lock_time);
1073 memcpy(&crec->header, ctdb_data.dptr, sizeof(crec->header));
1075 result->value.dsize = ctdb_data.dsize - sizeof(crec->header);
1076 result->value.dptr = NULL;
1078 if ((result->value.dsize != 0)
1079 && !(result->value.dptr = (uint8 *)talloc_memdup(
1080 result, ctdb_data.dptr + sizeof(crec->header),
1081 result->value.dsize))) {
1082 DEBUG(0, ("talloc failed\n"));
1083 TALLOC_FREE(result);
1086 SAFE_FREE(ctdb_data.dptr);
1091 static struct db_record *db_ctdb_fetch_locked(struct db_context *db,
1092 TALLOC_CTX *mem_ctx,
1095 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
1096 struct db_ctdb_ctx);
1098 if (ctx->transaction != NULL) {
1099 return db_ctdb_fetch_locked_transaction(ctx, mem_ctx, key);
1102 if (db->persistent) {
1103 return db_ctdb_fetch_locked_persistent(ctx, mem_ctx, key);
1106 return fetch_locked_internal(ctx, mem_ctx, key, db->persistent);
1110 fetch (unlocked, no migration) operation on ctdb
1112 static int db_ctdb_fetch(struct db_context *db, TALLOC_CTX *mem_ctx,
1113 TDB_DATA key, TDB_DATA *data)
1115 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
1116 struct db_ctdb_ctx);
1120 if (ctx->transaction) {
1121 return db_ctdb_transaction_fetch(ctx, mem_ctx, key, data);
1124 /* try a direct fetch */
1125 ctdb_data = tdb_fetch(ctx->wtdb->tdb, key);
1128 * See if we have a valid record and we are the dmaster. If so, we can
1129 * take the shortcut and just return it.
1130 * we bypass the dmaster check for persistent databases
1132 if ((ctdb_data.dptr != NULL) &&
1133 (ctdb_data.dsize >= sizeof(struct ctdb_ltdb_header)) &&
1135 ((struct ctdb_ltdb_header *)ctdb_data.dptr)->dmaster == get_my_vnn())) {
1136 /* we are the dmaster - avoid the ctdb protocol op */
1138 data->dsize = ctdb_data.dsize - sizeof(struct ctdb_ltdb_header);
1139 if (data->dsize == 0) {
1140 SAFE_FREE(ctdb_data.dptr);
1145 data->dptr = (uint8 *)talloc_memdup(
1146 mem_ctx, ctdb_data.dptr+sizeof(struct ctdb_ltdb_header),
1149 SAFE_FREE(ctdb_data.dptr);
1151 if (data->dptr == NULL) {
1157 SAFE_FREE(ctdb_data.dptr);
1159 /* we weren't able to get it locally - ask ctdb to fetch it for us */
1160 status = ctdbd_fetch(messaging_ctdbd_connection(),ctx->db_id, key, mem_ctx, data);
1161 if (!NT_STATUS_IS_OK(status)) {
1162 DEBUG(5, ("ctdbd_fetch failed: %s\n", nt_errstr(status)));
1169 struct traverse_state {
1170 struct db_context *db;
1171 int (*fn)(struct db_record *rec, void *private_data);
1175 static void traverse_callback(TDB_DATA key, TDB_DATA data, void *private_data)
1177 struct traverse_state *state = (struct traverse_state *)private_data;
1178 struct db_record *rec;
1179 TALLOC_CTX *tmp_ctx = talloc_new(state->db);
1180 /* we have to give them a locked record to prevent races */
1181 rec = db_ctdb_fetch_locked(state->db, tmp_ctx, key);
1182 if (rec && rec->value.dsize > 0) {
1183 state->fn(rec, state->private_data);
1185 talloc_free(tmp_ctx);
1188 static int traverse_persistent_callback(TDB_CONTEXT *tdb, TDB_DATA kbuf, TDB_DATA dbuf,
1191 struct traverse_state *state = (struct traverse_state *)private_data;
1192 struct db_record *rec;
1193 TALLOC_CTX *tmp_ctx = talloc_new(state->db);
1195 /* we have to give them a locked record to prevent races */
1196 rec = db_ctdb_fetch_locked(state->db, tmp_ctx, kbuf);
1197 if (rec && rec->value.dsize > 0) {
1198 ret = state->fn(rec, state->private_data);
1200 talloc_free(tmp_ctx);
1204 static int db_ctdb_traverse(struct db_context *db,
1205 int (*fn)(struct db_record *rec,
1206 void *private_data),
1209 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
1210 struct db_ctdb_ctx);
1211 struct traverse_state state;
1215 state.private_data = private_data;
1217 if (db->persistent) {
1218 /* for persistent databases we don't need to do a ctdb traverse,
1219 we can do a faster local traverse */
1220 return tdb_traverse(ctx->wtdb->tdb, traverse_persistent_callback, &state);
1224 ctdbd_traverse(ctx->db_id, traverse_callback, &state);
1228 static NTSTATUS db_ctdb_store_deny(struct db_record *rec, TDB_DATA data, int flag)
1230 return NT_STATUS_MEDIA_WRITE_PROTECTED;
1233 static NTSTATUS db_ctdb_delete_deny(struct db_record *rec)
1235 return NT_STATUS_MEDIA_WRITE_PROTECTED;
1238 static void traverse_read_callback(TDB_DATA key, TDB_DATA data, void *private_data)
1240 struct traverse_state *state = (struct traverse_state *)private_data;
1241 struct db_record rec;
1244 rec.store = db_ctdb_store_deny;
1245 rec.delete_rec = db_ctdb_delete_deny;
1246 rec.private_data = state->db;
1247 state->fn(&rec, state->private_data);
1250 static int traverse_persistent_callback_read(TDB_CONTEXT *tdb, TDB_DATA kbuf, TDB_DATA dbuf,
1253 struct traverse_state *state = (struct traverse_state *)private_data;
1254 struct db_record rec;
1257 rec.store = db_ctdb_store_deny;
1258 rec.delete_rec = db_ctdb_delete_deny;
1259 rec.private_data = state->db;
1261 if (rec.value.dsize <= sizeof(struct ctdb_ltdb_header)) {
1262 /* a deleted record */
1265 rec.value.dsize -= sizeof(struct ctdb_ltdb_header);
1266 rec.value.dptr += sizeof(struct ctdb_ltdb_header);
1268 return state->fn(&rec, state->private_data);
1271 static int db_ctdb_traverse_read(struct db_context *db,
1272 int (*fn)(struct db_record *rec,
1273 void *private_data),
1276 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
1277 struct db_ctdb_ctx);
1278 struct traverse_state state;
1282 state.private_data = private_data;
1284 if (db->persistent) {
1285 /* for persistent databases we don't need to do a ctdb traverse,
1286 we can do a faster local traverse */
1287 return tdb_traverse_read(ctx->wtdb->tdb, traverse_persistent_callback_read, &state);
1290 ctdbd_traverse(ctx->db_id, traverse_read_callback, &state);
1294 static int db_ctdb_get_seqnum(struct db_context *db)
1296 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
1297 struct db_ctdb_ctx);
1298 return tdb_get_seqnum(ctx->wtdb->tdb);
1301 static int db_ctdb_get_flags(struct db_context *db)
1303 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
1304 struct db_ctdb_ctx);
1305 return tdb_get_flags(ctx->wtdb->tdb);
1308 struct db_context *db_open_ctdb(TALLOC_CTX *mem_ctx,
1310 int hash_size, int tdb_flags,
1311 int open_flags, mode_t mode)
1313 struct db_context *result;
1314 struct db_ctdb_ctx *db_ctdb;
1316 struct ctdbd_connection *conn;
1318 if (!lp_clustering()) {
1319 DEBUG(10, ("Clustering disabled -- no ctdb\n"));
1323 if (!(result = TALLOC_ZERO_P(mem_ctx, struct db_context))) {
1324 DEBUG(0, ("talloc failed\n"));
1325 TALLOC_FREE(result);
1329 if (!(db_ctdb = TALLOC_P(result, struct db_ctdb_ctx))) {
1330 DEBUG(0, ("talloc failed\n"));
1331 TALLOC_FREE(result);
1335 db_ctdb->transaction = NULL;
1336 db_ctdb->db = result;
1338 conn = messaging_ctdbd_connection();
1340 DEBUG(1, ("Could not connect to ctdb\n"));
1341 TALLOC_FREE(result);
1345 if (!NT_STATUS_IS_OK(ctdbd_db_attach(conn, name, &db_ctdb->db_id, tdb_flags))) {
1346 DEBUG(0, ("ctdbd_db_attach failed for %s\n", name));
1347 TALLOC_FREE(result);
1351 db_path = ctdbd_dbpath(conn, db_ctdb, db_ctdb->db_id);
1353 result->persistent = ((tdb_flags & TDB_CLEAR_IF_FIRST) == 0);
1355 /* only pass through specific flags */
1356 tdb_flags &= TDB_SEQNUM;
1358 /* honor permissions if user has specified O_CREAT */
1359 if (open_flags & O_CREAT) {
1360 chmod(db_path, mode);
1363 db_ctdb->wtdb = tdb_wrap_open(db_ctdb, db_path, hash_size, tdb_flags, O_RDWR, 0);
1364 if (db_ctdb->wtdb == NULL) {
1365 DEBUG(0, ("Could not open tdb %s: %s\n", db_path, strerror(errno)));
1366 TALLOC_FREE(result);
1369 talloc_free(db_path);
1371 if (result->persistent) {
1372 db_ctdb->lock_ctx = g_lock_ctx_init(db_ctdb,
1373 ctdb_conn_msg_ctx(conn));
1374 if (db_ctdb->lock_ctx == NULL) {
1375 DEBUG(0, ("g_lock_ctx_init failed\n"));
1376 TALLOC_FREE(result);
1381 result->private_data = (void *)db_ctdb;
1382 result->fetch_locked = db_ctdb_fetch_locked;
1383 result->fetch = db_ctdb_fetch;
1384 result->traverse = db_ctdb_traverse;
1385 result->traverse_read = db_ctdb_traverse_read;
1386 result->get_seqnum = db_ctdb_get_seqnum;
1387 result->get_flags = db_ctdb_get_flags;
1388 result->transaction_start = db_ctdb_transaction_start;
1389 result->transaction_commit = db_ctdb_transaction_commit;
1390 result->transaction_cancel = db_ctdb_transaction_cancel;
1392 DEBUG(3,("db_open_ctdb: opened database '%s' with dbid 0x%x\n",
1393 name, db_ctdb->db_id));