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/tdb_wrap/tdb_wrap.h"
25 #include "dbwrap/dbwrap.h"
26 #include "dbwrap/dbwrap_ctdb.h"
27 #include "dbwrap/dbwrap_rbt.h"
28 #include "lib/param/param.h"
30 #include "ctdb/include/ctdb_protocol.h"
31 #include "ctdbd_conn.h"
32 #include "dbwrap/dbwrap.h"
33 #include "dbwrap/dbwrap_private.h"
34 #include "dbwrap/dbwrap_ctdb.h"
37 #include "messages_ctdb.h"
38 #include "lib/cluster_support.h"
39 #include "lib/util/tevent_ntstatus.h"
41 struct db_ctdb_transaction_handle {
42 struct db_ctdb_ctx *ctx;
44 * we store the writes done under a transaction:
46 struct ctdb_marshall_buffer *m_write;
53 struct db_context *db;
54 struct tdb_wrap *wtdb;
56 struct db_ctdb_transaction_handle *transaction;
57 struct g_lock_ctx *lock_ctx;
59 /* thresholds for warning messages */
60 int warn_unlock_msecs;
61 int warn_migrate_msecs;
62 int warn_migrate_attempts;
63 int warn_locktime_msecs;
67 struct db_ctdb_ctx *ctdb_ctx;
68 struct ctdb_ltdb_header header;
69 struct timeval lock_time;
72 struct ctdb_async_ctx {
74 struct ctdbd_connection *async_conn;
77 static struct ctdb_async_ctx ctdb_async_ctx;
79 static int ctdb_async_ctx_init_internal(TALLOC_CTX *mem_ctx,
80 struct tevent_context *ev,
86 TALLOC_FREE(ctdb_async_ctx.async_conn);
87 ctdb_async_ctx.initialized = false;
90 if (ctdb_async_ctx.initialized) {
95 ret = ctdbd_init_connection(mem_ctx,
98 &ctdb_async_ctx.async_conn);
101 if (ret != 0 || ctdb_async_ctx.async_conn == NULL) {
102 DBG_ERR("ctdbd_init_connection failed\n");
106 ret = ctdbd_setup_fde(ctdb_async_ctx.async_conn, ev);
108 DBG_ERR("ctdbd_setup_fde failed\n");
109 TALLOC_FREE(ctdb_async_ctx.async_conn);
113 ctdb_async_ctx.initialized = true;
117 static int ctdb_async_ctx_init(TALLOC_CTX *mem_ctx, struct tevent_context *ev)
119 return ctdb_async_ctx_init_internal(mem_ctx, ev, false);
122 int ctdb_async_ctx_reinit(TALLOC_CTX *mem_ctx, struct tevent_context *ev)
124 return ctdb_async_ctx_init_internal(mem_ctx, ev, true);
127 static NTSTATUS tdb_error_to_ntstatus(struct tdb_context *tdb)
129 enum TDB_ERROR tret = tdb_error(tdb);
131 return map_nt_error_from_tdb(tret);
134 struct db_ctdb_ltdb_parse_state {
135 void (*parser)(TDB_DATA key, struct ctdb_ltdb_header *header,
136 TDB_DATA data, void *private_data);
140 static int db_ctdb_ltdb_parser(TDB_DATA key, TDB_DATA data,
143 struct db_ctdb_ltdb_parse_state *state =
144 (struct db_ctdb_ltdb_parse_state *)private_data;
146 if (data.dsize < sizeof(struct ctdb_ltdb_header)) {
151 key, (struct ctdb_ltdb_header *)data.dptr,
152 make_tdb_data(data.dptr + sizeof(struct ctdb_ltdb_header),
153 data.dsize - sizeof(struct ctdb_ltdb_header)),
154 state->private_data);
158 static NTSTATUS db_ctdb_ltdb_parse(
159 struct db_ctdb_ctx *db, TDB_DATA key,
160 void (*parser)(TDB_DATA key, struct ctdb_ltdb_header *header,
161 TDB_DATA data, void *private_data),
164 struct db_ctdb_ltdb_parse_state state;
167 state.parser = parser;
168 state.private_data = private_data;
170 ret = tdb_parse_record(db->wtdb->tdb, key, db_ctdb_ltdb_parser,
173 return NT_STATUS_NOT_FOUND;
179 * Store a record together with the ctdb record header
180 * in the local copy of the database.
182 static NTSTATUS db_ctdb_ltdb_store(struct db_ctdb_ctx *db,
184 struct ctdb_ltdb_header *header,
185 const TDB_DATA *dbufs, int num_dbufs)
187 TDB_DATA recs[num_dbufs+1];
190 recs[0] = (TDB_DATA) { .dptr = (uint8_t *)header,
191 .dsize = sizeof(struct ctdb_ltdb_header) };
192 memcpy(&recs[1], dbufs, sizeof(TDB_DATA) * num_dbufs);
194 ret = tdb_storev(db->wtdb->tdb, key, recs, num_dbufs + 1, TDB_REPLACE);
196 return (ret == 0) ? NT_STATUS_OK
197 : tdb_error_to_ntstatus(db->wtdb->tdb);
202 form a ctdb_rec_data record from a key/data pair
204 static struct ctdb_rec_data_old *db_ctdb_marshall_record(TALLOC_CTX *mem_ctx, uint32_t reqid,
206 struct ctdb_ltdb_header *header,
210 struct ctdb_rec_data_old *d;
212 length = offsetof(struct ctdb_rec_data_old, data) + key.dsize +
213 data.dsize + sizeof(*header);
214 d = (struct ctdb_rec_data_old *)talloc_size(mem_ctx, length);
220 d->keylen = key.dsize;
221 memcpy(&d->data[0], key.dptr, key.dsize);
223 d->datalen = data.dsize + sizeof(*header);
224 memcpy(&d->data[key.dsize], header, sizeof(*header));
225 memcpy(&d->data[key.dsize+sizeof(*header)], data.dptr, data.dsize);
230 /* helper function for marshalling multiple records */
231 static struct ctdb_marshall_buffer *db_ctdb_marshall_add(TALLOC_CTX *mem_ctx,
232 struct ctdb_marshall_buffer *m,
236 struct ctdb_ltdb_header *header,
239 struct ctdb_rec_data_old *r;
240 size_t m_size, r_size;
241 struct ctdb_marshall_buffer *m2 = NULL;
243 r = db_ctdb_marshall_record(talloc_tos(), reqid, key, header, data);
250 m = (struct ctdb_marshall_buffer *)talloc_zero_size(
251 mem_ctx, offsetof(struct ctdb_marshall_buffer, data));
258 m_size = talloc_get_size(m);
259 r_size = talloc_get_size(r);
261 m2 = (struct ctdb_marshall_buffer *)talloc_realloc_size(
262 mem_ctx, m, m_size + r_size);
268 memcpy(m_size + (uint8_t *)m2, r, r_size);
277 /* we've finished marshalling, return a data blob with the marshalled records */
278 static TDB_DATA db_ctdb_marshall_finish(struct ctdb_marshall_buffer *m)
281 data.dptr = (uint8_t *)m;
282 data.dsize = talloc_get_size(m);
287 loop over a marshalling buffer
289 - pass r==NULL to start
290 - loop the number of times indicated by m->count
292 static struct ctdb_rec_data_old *db_ctdb_marshall_loop_next_key(
293 struct ctdb_marshall_buffer *m, struct ctdb_rec_data_old *r, TDB_DATA *key)
296 r = (struct ctdb_rec_data_old *)&m->data[0];
298 r = (struct ctdb_rec_data_old *)(r->length + (uint8_t *)r);
301 key->dptr = &r->data[0];
302 key->dsize = r->keylen;
306 static bool db_ctdb_marshall_buf_parse(
307 struct ctdb_rec_data_old *r, uint32_t *reqid,
308 struct ctdb_ltdb_header **header, TDB_DATA *data)
310 if (r->datalen < sizeof(struct ctdb_ltdb_header)) {
316 data->dptr = &r->data[r->keylen] + sizeof(struct ctdb_ltdb_header);
317 data->dsize = r->datalen - sizeof(struct ctdb_ltdb_header);
319 *header = (struct ctdb_ltdb_header *)&r->data[r->keylen];
325 * CTDB transaction destructor
327 static int db_ctdb_transaction_destructor(struct db_ctdb_transaction_handle *h)
331 status = g_lock_unlock(h->ctx->lock_ctx,
332 string_term_tdb_data(h->lock_name));
333 if (!NT_STATUS_IS_OK(status)) {
334 DEBUG(0, ("g_lock_unlock failed for %s: %s\n", h->lock_name,
342 * CTDB dbwrap API: transaction_start function
343 * starts a transaction on a persistent database
345 static int db_ctdb_transaction_start(struct db_context *db)
347 struct db_ctdb_transaction_handle *h;
349 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
352 if (!db->persistent) {
353 DEBUG(0,("transactions not supported on non-persistent database 0x%08x\n",
358 if (ctx->transaction) {
359 ctx->transaction->nesting++;
360 DEBUG(5, (__location__ " transaction start on db 0x%08x: nesting %d -> %d\n",
361 ctx->db_id, ctx->transaction->nesting - 1, ctx->transaction->nesting));
365 h = talloc_zero(db, struct db_ctdb_transaction_handle);
367 DEBUG(0,(__location__ " oom for transaction handle\n"));
373 h->lock_name = talloc_asprintf(h, "transaction_db_0x%08x",
374 (unsigned int)ctx->db_id);
375 if (h->lock_name == NULL) {
376 DEBUG(0, ("talloc_asprintf failed\n"));
382 * Wait a day, i.e. forever...
384 status = g_lock_lock(ctx->lock_ctx, string_term_tdb_data(h->lock_name),
385 G_LOCK_WRITE, timeval_set(86400, 0));
386 if (!NT_STATUS_IS_OK(status)) {
387 DEBUG(0, ("g_lock_lock failed: %s\n", nt_errstr(status)));
392 talloc_set_destructor(h, db_ctdb_transaction_destructor);
394 ctx->transaction = h;
396 DEBUG(5,(__location__ " transaction started on db 0x%08x\n", ctx->db_id));
401 static bool parse_newest_in_marshall_buffer(
402 struct ctdb_marshall_buffer *buf, TDB_DATA key,
403 void (*parser)(TDB_DATA key, struct ctdb_ltdb_header *header,
404 TDB_DATA data, void *private_data),
407 struct ctdb_rec_data_old *rec = NULL;
408 struct ctdb_ltdb_header *h = NULL;
417 * Walk the list of records written during this
418 * transaction. If we want to read one we have already
419 * written, return the last written sample. Thus we do not do
420 * a "break;" for the first hit, this record might have been
424 for (i=0; i<buf->count; i++) {
428 rec = db_ctdb_marshall_loop_next_key(buf, rec, &tkey);
433 if (!tdb_data_equal(key, tkey)) {
437 if (!db_ctdb_marshall_buf_parse(rec, &reqid, &h, &data)) {
446 parser(key, h, data, private_data);
451 struct pull_newest_from_marshall_buffer_state {
452 struct ctdb_ltdb_header *pheader;
457 static void pull_newest_from_marshall_buffer_parser(
458 TDB_DATA key, struct ctdb_ltdb_header *header,
459 TDB_DATA data, void *private_data)
461 struct pull_newest_from_marshall_buffer_state *state =
462 (struct pull_newest_from_marshall_buffer_state *)private_data;
464 if (state->pheader != NULL) {
465 memcpy(state->pheader, header, sizeof(*state->pheader));
467 if (state->pdata != NULL) {
468 state->pdata->dsize = data.dsize;
469 state->pdata->dptr = (uint8_t *)talloc_memdup(
470 state->mem_ctx, data.dptr, data.dsize);
474 static bool pull_newest_from_marshall_buffer(struct ctdb_marshall_buffer *buf,
476 struct ctdb_ltdb_header *pheader,
480 struct pull_newest_from_marshall_buffer_state state;
482 state.pheader = pheader;
483 state.mem_ctx = mem_ctx;
486 if (!parse_newest_in_marshall_buffer(
487 buf, key, pull_newest_from_marshall_buffer_parser,
491 if ((pdata != NULL) && (pdata->dsize != 0) && (pdata->dptr == NULL)) {
498 static NTSTATUS db_ctdb_storev_transaction(struct db_record *rec,
499 const TDB_DATA *dbufs, int num_dbufs,
501 static NTSTATUS db_ctdb_delete_transaction(struct db_record *rec);
503 static struct db_record *db_ctdb_fetch_locked_transaction(struct db_ctdb_ctx *ctx,
507 struct db_record *result;
510 if (!(result = talloc(mem_ctx, struct db_record))) {
511 DEBUG(0, ("talloc failed\n"));
515 result->db = ctx->db;
516 result->private_data = ctx->transaction;
518 result->key.dsize = key.dsize;
519 result->key.dptr = (uint8_t *)talloc_memdup(result, key.dptr,
521 if (result->key.dptr == NULL) {
522 DEBUG(0, ("talloc failed\n"));
527 result->storev = db_ctdb_storev_transaction;
528 result->delete_rec = db_ctdb_delete_transaction;
530 if (ctx->transaction == NULL) {
531 DEBUG(0, ("no transaction available\n"));
535 if (pull_newest_from_marshall_buffer(ctx->transaction->m_write, key,
536 NULL, result, &result->value)) {
537 result->value_valid = true;
541 ctdb_data = tdb_fetch(ctx->wtdb->tdb, key);
542 if (ctdb_data.dptr == NULL) {
543 /* create the record */
544 result->value = tdb_null;
545 result->value_valid = true;
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_t *)talloc_memdup(
554 result, ctdb_data.dptr + sizeof(struct ctdb_ltdb_header),
555 result->value.dsize))) {
556 DEBUG(0, ("talloc failed\n"));
560 result->value_valid = true;
562 SAFE_FREE(ctdb_data.dptr);
567 static int db_ctdb_record_destructor(struct db_record **recp)
569 struct db_record *rec = talloc_get_type_abort(*recp, struct db_record);
570 struct db_ctdb_transaction_handle *h = talloc_get_type_abort(
571 rec->private_data, struct db_ctdb_transaction_handle);
572 int ret = h->ctx->db->transaction_commit(h->ctx->db);
574 DEBUG(0,(__location__ " transaction_commit failed\n"));
580 auto-create a transaction for persistent databases
582 static struct db_record *db_ctdb_fetch_locked_persistent(struct db_ctdb_ctx *ctx,
587 struct db_record *rec, **recp;
589 res = db_ctdb_transaction_start(ctx->db);
594 rec = db_ctdb_fetch_locked_transaction(ctx, mem_ctx, key);
596 ctx->db->transaction_cancel(ctx->db);
600 /* destroy this transaction when we release the lock */
601 recp = talloc(rec, struct db_record *);
603 ctx->db->transaction_cancel(ctx->db);
608 talloc_set_destructor(recp, db_ctdb_record_destructor);
614 stores a record inside a transaction
616 static NTSTATUS db_ctdb_transaction_store(struct db_ctdb_transaction_handle *h,
617 TDB_DATA key, TDB_DATA data)
619 TALLOC_CTX *tmp_ctx = talloc_new(h);
621 struct ctdb_ltdb_header header;
625 /* we need the header so we can update the RSN */
627 if (!pull_newest_from_marshall_buffer(h->m_write, key, &header,
630 rec = tdb_fetch(h->ctx->wtdb->tdb, key);
632 if (rec.dptr != NULL) {
633 memcpy(&header, rec.dptr,
634 sizeof(struct ctdb_ltdb_header));
635 rec.dsize -= sizeof(struct ctdb_ltdb_header);
638 * a special case, we are writing the same
639 * data that is there now
641 if (data.dsize == rec.dsize &&
643 rec.dptr + sizeof(struct ctdb_ltdb_header),
646 talloc_free(tmp_ctx);
653 header.dmaster = get_my_vnn();
656 h->m_write = db_ctdb_marshall_add(h, h->m_write, h->ctx->db_id, 0, key, &header, data);
657 if (h->m_write == NULL) {
658 DEBUG(0,(__location__ " Failed to add to marshalling record\n"));
659 talloc_free(tmp_ctx);
660 return NT_STATUS_NO_MEMORY;
663 talloc_free(tmp_ctx);
669 a record store inside a transaction
671 static NTSTATUS db_ctdb_storev_transaction(
672 struct db_record *rec, const TDB_DATA *dbufs, int num_dbufs, int flag)
674 struct db_ctdb_transaction_handle *h = talloc_get_type_abort(
675 rec->private_data, struct db_ctdb_transaction_handle);
679 data = dbwrap_merge_dbufs(rec, dbufs, num_dbufs);
680 if (data.dptr == NULL) {
681 return NT_STATUS_NO_MEMORY;
684 status = db_ctdb_transaction_store(h, rec->key, data);
686 TALLOC_FREE(data.dptr);
692 a record delete inside a transaction
694 static NTSTATUS db_ctdb_delete_transaction(struct db_record *rec)
696 struct db_ctdb_transaction_handle *h = talloc_get_type_abort(
697 rec->private_data, struct db_ctdb_transaction_handle);
700 status = db_ctdb_transaction_store(h, rec->key, tdb_null);
704 static void db_ctdb_fetch_db_seqnum_parser(
705 TDB_DATA key, struct ctdb_ltdb_header *header,
706 TDB_DATA data, void *private_data)
708 uint64_t *seqnum = (uint64_t *)private_data;
710 if (data.dsize != sizeof(uint64_t)) {
714 memcpy(seqnum, data.dptr, sizeof(*seqnum));
718 * Fetch the db sequence number of a persistent db directly from the db.
720 static NTSTATUS db_ctdb_fetch_db_seqnum_from_db(struct db_ctdb_ctx *db,
726 if (seqnum == NULL) {
727 return NT_STATUS_INVALID_PARAMETER;
730 key = string_term_tdb_data(CTDB_DB_SEQNUM_KEY);
732 status = db_ctdb_ltdb_parse(
733 db, key, db_ctdb_fetch_db_seqnum_parser, seqnum);
735 if (NT_STATUS_IS_OK(status)) {
738 if (NT_STATUS_EQUAL(status, NT_STATUS_NOT_FOUND)) {
746 * Store the database sequence number inside a transaction.
748 static NTSTATUS db_ctdb_store_db_seqnum(struct db_ctdb_transaction_handle *h,
752 TDB_DATA key = string_term_tdb_data(CTDB_DB_SEQNUM_KEY);
753 TDB_DATA data = { .dptr=(uint8_t *)&seqnum, .dsize=sizeof(seqnum) };
755 status = db_ctdb_transaction_store(h, key, data);
763 static int db_ctdb_transaction_commit(struct db_context *db)
765 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
769 struct db_ctdb_transaction_handle *h = ctx->transaction;
770 uint64_t old_seqnum, new_seqnum;
774 DEBUG(0,(__location__ " transaction commit with no open transaction on db 0x%08x\n", ctx->db_id));
778 if (h->nested_cancel) {
779 db->transaction_cancel(db);
780 DEBUG(5,(__location__ " Failed transaction commit after nested cancel\n"));
784 if (h->nesting != 0) {
786 DEBUG(5, (__location__ " transaction commit on db 0x%08x: nesting %d -> %d\n",
787 ctx->db_id, ctx->transaction->nesting + 1, ctx->transaction->nesting));
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__ " transaction commit 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 ret = ctdbd_control_local(messaging_ctdb_connection(),
829 CTDB_CONTROL_TRANS3_COMMIT,
831 db_ctdb_marshall_finish(h->m_write),
832 NULL, NULL, &status);
833 if ((ret != 0) || 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. */
853 if (new_seqnum != (old_seqnum + 1)) {
854 DEBUG(0, (__location__ " ERROR: new_seqnum[%lu] != "
855 "old_seqnum[%lu] + (0 or 1) after failed "
856 "TRANS3_COMMIT - this should not happen!\n",
857 (unsigned long)new_seqnum,
858 (unsigned long)old_seqnum));
863 * Recovery propagated our changes to all nodes, completing
864 * our commit for us - succeed.
871 h->ctx->transaction = NULL;
880 static int db_ctdb_transaction_cancel(struct db_context *db)
882 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
884 struct db_ctdb_transaction_handle *h = ctx->transaction;
887 DEBUG(0,(__location__ " transaction cancel with no open transaction on db 0x%08x\n", ctx->db_id));
891 if (h->nesting != 0) {
893 h->nested_cancel = true;
894 DEBUG(5, (__location__ " transaction cancel on db 0x%08x: nesting %d -> %d\n",
895 ctx->db_id, ctx->transaction->nesting + 1, ctx->transaction->nesting));
899 DEBUG(5,(__location__ " Cancel transaction on db 0x%08x\n", ctx->db_id));
901 ctx->transaction = NULL;
907 static NTSTATUS db_ctdb_storev(struct db_record *rec,
908 const TDB_DATA *dbufs, int num_dbufs, int flag)
910 struct db_ctdb_rec *crec = talloc_get_type_abort(
911 rec->private_data, struct db_ctdb_rec);
914 status = db_ctdb_ltdb_store(crec->ctdb_ctx, rec->key, &(crec->header),
921 static NTSTATUS db_ctdb_send_schedule_for_deletion(struct db_record *rec)
923 NTSTATUS status = NT_STATUS_OK;
925 struct ctdb_control_schedule_for_deletion *dd;
928 struct db_ctdb_rec *crec = talloc_get_type_abort(
929 rec->private_data, struct db_ctdb_rec);
930 struct db_ctdb_ctx *ctx = crec->ctdb_ctx;
932 indata.dsize = offsetof(struct ctdb_control_schedule_for_deletion, key) + rec->key.dsize;
933 indata.dptr = talloc_zero_array(crec, uint8_t, indata.dsize);
934 if (indata.dptr == NULL) {
935 DEBUG(0, (__location__ " talloc failed!\n"));
936 return NT_STATUS_NO_MEMORY;
939 dd = (struct ctdb_control_schedule_for_deletion *)(void *)indata.dptr;
940 dd->db_id = ctx->db_id;
941 dd->hdr = crec->header;
942 dd->keylen = rec->key.dsize;
943 memcpy(dd->key, rec->key.dptr, rec->key.dsize);
945 ret = ctdbd_control_local(messaging_ctdb_connection(),
946 CTDB_CONTROL_SCHEDULE_FOR_DELETION,
947 crec->ctdb_ctx->db_id,
948 CTDB_CTRL_FLAG_NOREPLY, /* flags */
953 talloc_free(indata.dptr);
955 if ((ret != 0) || cstatus != 0) {
956 DEBUG(1, (__location__ " Error sending local control "
957 "SCHEDULE_FOR_DELETION: %s, cstatus = %"PRIi32"\n",
958 strerror(ret), cstatus));
960 status = map_nt_error_from_unix(ret);
962 status = NT_STATUS_UNSUCCESSFUL;
969 static NTSTATUS db_ctdb_delete(struct db_record *rec)
974 * We have to store the header with empty data. TODO: Fix the
978 status = db_ctdb_storev(rec, &tdb_null, 1, 0);
979 if (!NT_STATUS_IS_OK(status)) {
983 status = db_ctdb_send_schedule_for_deletion(rec);
987 static int db_ctdb_record_destr(struct db_record* data)
989 struct db_ctdb_rec *crec = talloc_get_type_abort(
990 data->private_data, struct db_ctdb_rec);
993 struct timeval before;
996 DEBUG(10, (DEBUGLEVEL > 10
997 ? "Unlocking db %u key %s\n"
998 : "Unlocking db %u key %.20s\n",
999 (int)crec->ctdb_ctx->db_id,
1000 hex_encode_talloc(data, (unsigned char *)data->key.dptr,
1003 before = timeval_current();
1005 ret = tdb_chainunlock(crec->ctdb_ctx->wtdb->tdb, data->key);
1007 timediff = timeval_elapsed(&before);
1008 timediff *= 1000; /* get us milliseconds */
1010 if (timediff > crec->ctdb_ctx->warn_unlock_msecs) {
1012 key = hex_encode_talloc(talloc_tos(),
1013 (unsigned char *)data->key.dptr,
1015 DEBUG(0, ("tdb_chainunlock on db %s, key %s took %f milliseconds\n",
1016 tdb_name(crec->ctdb_ctx->wtdb->tdb), key,
1022 DEBUG(0, ("tdb_chainunlock failed\n"));
1026 threshold = crec->ctdb_ctx->warn_locktime_msecs;
1027 if (threshold != 0) {
1028 timediff = timeval_elapsed(&crec->lock_time) * 1000;
1029 if (timediff > threshold) {
1032 key = hex_encode_talloc(data,
1033 (unsigned char *)data->key.dptr,
1035 DEBUG(0, ("Held tdb lock on db %s, key %s "
1036 "%f milliseconds\n",
1037 tdb_name(crec->ctdb_ctx->wtdb->tdb),
1046 * Check whether we have a valid local copy of the given record,
1047 * either for reading or for writing.
1049 static bool db_ctdb_can_use_local_hdr(const struct ctdb_ltdb_header *hdr,
1050 uint32_t my_vnn, bool read_only)
1052 if (hdr->dmaster != my_vnn) {
1053 /* If we're not dmaster, it must be r/o copy. */
1054 return read_only && (hdr->flags & CTDB_REC_RO_HAVE_READONLY);
1058 * If we want write access, no one may have r/o copies.
1060 return read_only || !(hdr->flags & CTDB_REC_RO_HAVE_DELEGATIONS);
1063 static bool db_ctdb_can_use_local_copy(TDB_DATA ctdb_data, uint32_t my_vnn,
1066 if (ctdb_data.dptr == NULL) {
1070 if (ctdb_data.dsize < sizeof(struct ctdb_ltdb_header)) {
1074 return db_ctdb_can_use_local_hdr(
1075 (struct ctdb_ltdb_header *)ctdb_data.dptr, my_vnn, read_only);
1078 static struct db_record *fetch_locked_internal(struct db_ctdb_ctx *ctx,
1079 TALLOC_CTX *mem_ctx,
1083 struct db_record *result;
1084 struct db_ctdb_rec *crec;
1086 int migrate_attempts;
1087 struct timeval migrate_start;
1088 struct timeval chainlock_start;
1089 struct timeval ctdb_start_time;
1090 double chainlock_time = 0;
1091 double ctdb_time = 0;
1096 if (!(result = talloc(mem_ctx, struct db_record))) {
1097 DEBUG(0, ("talloc failed\n"));
1101 if (!(crec = talloc_zero(result, struct db_ctdb_rec))) {
1102 DEBUG(0, ("talloc failed\n"));
1103 TALLOC_FREE(result);
1107 result->db = ctx->db;
1108 result->private_data = (void *)crec;
1109 crec->ctdb_ctx = ctx;
1111 result->key.dsize = key.dsize;
1112 result->key.dptr = (uint8_t *)talloc_memdup(result, key.dptr,
1114 if (result->key.dptr == NULL) {
1115 DEBUG(0, ("talloc failed\n"));
1116 TALLOC_FREE(result);
1120 migrate_attempts = 0;
1121 GetTimeOfDay(&migrate_start);
1124 * Do a blocking lock on the record
1128 if (DEBUGLEVEL >= 10) {
1129 char *keystr = hex_encode_talloc(result, key.dptr, key.dsize);
1130 DEBUG(10, (DEBUGLEVEL > 10
1131 ? "Locking db %u key %s\n"
1132 : "Locking db %u key %.20s\n",
1133 (int)crec->ctdb_ctx->db_id, keystr));
1134 TALLOC_FREE(keystr);
1137 GetTimeOfDay(&chainlock_start);
1139 ? tdb_chainlock_nonblock(ctx->wtdb->tdb, key)
1140 : tdb_chainlock(ctx->wtdb->tdb, key);
1141 chainlock_time += timeval_elapsed(&chainlock_start);
1144 DEBUG(3, ("tdb_chainlock failed\n"));
1145 TALLOC_FREE(result);
1149 result->storev = db_ctdb_storev;
1150 result->delete_rec = db_ctdb_delete;
1151 talloc_set_destructor(result, db_ctdb_record_destr);
1153 ctdb_data = tdb_fetch(ctx->wtdb->tdb, key);
1156 * See if we have a valid record and we are the dmaster. If so, we can
1157 * take the shortcut and just return it.
1160 if (!db_ctdb_can_use_local_copy(ctdb_data, get_my_vnn(), false)) {
1161 SAFE_FREE(ctdb_data.dptr);
1162 tdb_chainunlock(ctx->wtdb->tdb, key);
1163 talloc_set_destructor(result, NULL);
1165 if (tryonly && (migrate_attempts != 0)) {
1166 DEBUG(5, ("record migrated away again\n"));
1167 TALLOC_FREE(result);
1171 migrate_attempts += 1;
1173 DEBUG(10, ("ctdb_data.dptr = %p, dmaster = %"PRIu32" "
1174 "(%"PRIu32") %"PRIu32"\n",
1175 ctdb_data.dptr, ctdb_data.dptr ?
1176 ((struct ctdb_ltdb_header *)ctdb_data.dptr)->dmaster :
1180 ((struct ctdb_ltdb_header *)ctdb_data.dptr)->flags : 0));
1182 GetTimeOfDay(&ctdb_start_time);
1183 ret = ctdbd_migrate(messaging_ctdb_connection(), ctx->db_id,
1185 ctdb_time += timeval_elapsed(&ctdb_start_time);
1188 DEBUG(5, ("ctdbd_migrate failed: %s\n",
1190 TALLOC_FREE(result);
1193 /* now its migrated, try again */
1199 duration = timeval_elapsed(&migrate_start);
1202 * Convert the duration to milliseconds to avoid a
1203 * floating-point division of
1204 * lp_parm_int("migrate_duration") by 1000.
1206 duration_msecs = duration * 1000;
1209 if ((migrate_attempts > ctx->warn_migrate_attempts) ||
1210 (duration_msecs > ctx->warn_migrate_msecs)) {
1213 if (tdb_get_flags(ctx->wtdb->tdb) & TDB_INCOMPATIBLE_HASH) {
1214 chain = tdb_jenkins_hash(&key) %
1215 tdb_hash_size(ctx->wtdb->tdb);
1218 DEBUG(0, ("db_ctdb_fetch_locked for %s key %s, chain %d "
1219 "needed %d attempts, %d milliseconds, "
1220 "chainlock: %f ms, CTDB %f ms\n",
1221 tdb_name(ctx->wtdb->tdb),
1222 hex_encode_talloc(talloc_tos(),
1223 (unsigned char *)key.dptr,
1226 migrate_attempts, duration_msecs,
1227 chainlock_time * 1000.0,
1228 ctdb_time * 1000.0));
1231 GetTimeOfDay(&crec->lock_time);
1233 memcpy(&crec->header, ctdb_data.dptr, sizeof(crec->header));
1235 result->value.dsize = ctdb_data.dsize - sizeof(crec->header);
1236 result->value.dptr = NULL;
1238 if (result->value.dsize != 0) {
1239 result->value.dptr = talloc_memdup(
1240 result, ctdb_data.dptr + sizeof(crec->header),
1241 result->value.dsize);
1242 if (result->value.dptr == NULL) {
1243 DBG_ERR("talloc failed\n");
1244 TALLOC_FREE(result);
1248 result->value_valid = true;
1250 SAFE_FREE(ctdb_data.dptr);
1255 static struct db_record *db_ctdb_fetch_locked(struct db_context *db,
1256 TALLOC_CTX *mem_ctx,
1259 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
1260 struct db_ctdb_ctx);
1262 if (ctx->transaction != NULL) {
1263 return db_ctdb_fetch_locked_transaction(ctx, mem_ctx, key);
1266 if (db->persistent) {
1267 return db_ctdb_fetch_locked_persistent(ctx, mem_ctx, key);
1270 return fetch_locked_internal(ctx, mem_ctx, key, false);
1273 static struct db_record *db_ctdb_try_fetch_locked(struct db_context *db,
1274 TALLOC_CTX *mem_ctx,
1277 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
1278 struct db_ctdb_ctx);
1280 if (ctx->transaction != NULL) {
1281 return db_ctdb_fetch_locked_transaction(ctx, mem_ctx, key);
1284 if (db->persistent) {
1285 return db_ctdb_fetch_locked_persistent(ctx, mem_ctx, key);
1288 return fetch_locked_internal(ctx, mem_ctx, key, true);
1291 struct db_ctdb_parse_record_state {
1292 void (*parser)(TDB_DATA key, TDB_DATA data, void *private_data);
1295 bool ask_for_readonly_copy;
1300 static void db_ctdb_parse_record_parser(
1301 TDB_DATA key, struct ctdb_ltdb_header *header,
1302 TDB_DATA data, void *private_data)
1304 struct db_ctdb_parse_record_state *state =
1305 (struct db_ctdb_parse_record_state *)private_data;
1306 state->parser(key, data, state->private_data);
1309 static void db_ctdb_parse_record_parser_nonpersistent(
1310 TDB_DATA key, struct ctdb_ltdb_header *header,
1311 TDB_DATA data, void *private_data)
1313 struct db_ctdb_parse_record_state *state =
1314 (struct db_ctdb_parse_record_state *)private_data;
1316 if (db_ctdb_can_use_local_hdr(header, state->my_vnn, true)) {
1318 * A record consisting only of the ctdb header can be
1319 * a validly created empty record or a tombstone
1320 * record of a deleted record (not vacuumed yet). Mark
1323 state->empty_record = (data.dsize == 0);
1324 if (!state->empty_record) {
1325 state->parser(key, data, state->private_data);
1330 * We found something in the db, so it seems that this record,
1331 * while not usable locally right now, is popular. Ask for a
1334 state->ask_for_readonly_copy = true;
1338 static NTSTATUS db_ctdb_try_parse_local_record(struct db_ctdb_ctx *ctx,
1340 struct db_ctdb_parse_record_state *state)
1344 if (ctx->transaction != NULL) {
1345 struct db_ctdb_transaction_handle *h = ctx->transaction;
1349 * Transactions only happen for persistent db's.
1352 found = parse_newest_in_marshall_buffer(
1353 h->m_write, key, db_ctdb_parse_record_parser, state);
1356 return NT_STATUS_OK;
1360 if (ctx->db->persistent) {
1362 * Persistent db, but not found in the transaction buffer
1364 return db_ctdb_ltdb_parse(
1365 ctx, key, db_ctdb_parse_record_parser, state);
1368 state->done = false;
1369 state->ask_for_readonly_copy = false;
1371 status = db_ctdb_ltdb_parse(
1372 ctx, key, db_ctdb_parse_record_parser_nonpersistent, state);
1373 if (NT_STATUS_IS_OK(status) && state->done) {
1374 if (state->empty_record) {
1376 * We know authoritatively, that this is an empty
1377 * record. Since ctdb does not distinguish between empty
1378 * and deleted records, this can be a record stored as
1379 * empty or a not-yet-vacuumed tombstone record of a
1380 * deleted record. Now Samba right now can live without
1381 * empty records, so we can safely report this record
1384 * See bugs 10008 and 12005.
1386 return NT_STATUS_NOT_FOUND;
1388 return NT_STATUS_OK;
1391 return NT_STATUS_MORE_PROCESSING_REQUIRED;
1394 static NTSTATUS db_ctdb_parse_record(struct db_context *db, TDB_DATA key,
1395 void (*parser)(TDB_DATA key,
1397 void *private_data),
1400 struct db_ctdb_ctx *ctx = talloc_get_type_abort(
1401 db->private_data, struct db_ctdb_ctx);
1402 struct db_ctdb_parse_record_state state;
1406 state.parser = parser;
1407 state.private_data = private_data;
1408 state.my_vnn = get_my_vnn();
1409 state.empty_record = false;
1411 status = db_ctdb_try_parse_local_record(ctx, key, &state);
1412 if (!NT_STATUS_EQUAL(status, NT_STATUS_MORE_PROCESSING_REQUIRED)) {
1416 ret = ctdbd_parse(messaging_ctdb_connection(), ctx->db_id, key,
1417 state.ask_for_readonly_copy, parser, private_data);
1419 if (ret == ENOENT) {
1422 * NT_STATUS_OBJECT_NAME_NOT_FOUND. Our upper
1423 * layers expect NT_STATUS_NOT_FOUND for "no
1424 * record around". We need to convert dbwrap
1425 * to 0/errno away from NTSTATUS ... :-)
1427 return NT_STATUS_NOT_FOUND;
1429 return map_nt_error_from_unix(ret);
1431 return NT_STATUS_OK;
1434 static void db_ctdb_parse_record_done(struct tevent_req *subreq);
1436 static struct tevent_req *db_ctdb_parse_record_send(
1437 TALLOC_CTX *mem_ctx,
1438 struct tevent_context *ev,
1439 struct db_context *db,
1441 void (*parser)(TDB_DATA key,
1443 void *private_data),
1445 enum dbwrap_req_state *req_state)
1447 struct db_ctdb_ctx *ctx = talloc_get_type_abort(
1448 db->private_data, struct db_ctdb_ctx);
1449 struct tevent_req *req = NULL;
1450 struct tevent_req *subreq = NULL;
1451 struct db_ctdb_parse_record_state *state = NULL;
1454 req = tevent_req_create(mem_ctx, &state,
1455 struct db_ctdb_parse_record_state);
1457 *req_state = DBWRAP_REQ_ERROR;
1462 *state = (struct db_ctdb_parse_record_state) {
1464 .private_data = private_data,
1465 .my_vnn = get_my_vnn(),
1466 .empty_record = false,
1469 status = db_ctdb_try_parse_local_record(ctx, key, state);
1470 if (!NT_STATUS_EQUAL(status, NT_STATUS_MORE_PROCESSING_REQUIRED)) {
1471 if (tevent_req_nterror(req, status)) {
1472 *req_state = DBWRAP_REQ_ERROR;
1473 return tevent_req_post(req, ev);
1475 *req_state = DBWRAP_REQ_DONE;
1476 tevent_req_done(req);
1477 return tevent_req_post(req, ev);
1480 subreq = ctdbd_parse_send(state,
1482 ctdb_async_ctx.async_conn,
1485 state->ask_for_readonly_copy,
1489 if (tevent_req_nomem(subreq, req)) {
1490 *req_state = DBWRAP_REQ_ERROR;
1491 return tevent_req_post(req, ev);
1493 tevent_req_set_callback(subreq, db_ctdb_parse_record_done, req);
1498 static void db_ctdb_parse_record_done(struct tevent_req *subreq)
1500 struct tevent_req *req = tevent_req_callback_data(
1501 subreq, struct tevent_req);
1504 ret = ctdbd_parse_recv(subreq);
1505 TALLOC_FREE(subreq);
1507 if (ret == ENOENT) {
1509 * This maps to NT_STATUS_OBJECT_NAME_NOT_FOUND. Our
1510 * upper layers expect NT_STATUS_NOT_FOUND for "no
1511 * record around". We need to convert dbwrap to 0/errno
1512 * away from NTSTATUS ... :-)
1514 tevent_req_nterror(req, NT_STATUS_NOT_FOUND);
1517 tevent_req_nterror(req, map_nt_error_from_unix(ret));
1521 tevent_req_done(req);
1525 static NTSTATUS db_ctdb_parse_record_recv(struct tevent_req *req)
1527 return tevent_req_simple_recv_ntstatus(req);
1530 struct traverse_state {
1531 struct db_context *db;
1532 int (*fn)(struct db_record *rec, void *private_data);
1537 static void traverse_callback(TDB_DATA key, TDB_DATA data, void *private_data)
1539 struct traverse_state *state = (struct traverse_state *)private_data;
1540 struct db_record *rec = NULL;
1541 TALLOC_CTX *tmp_ctx = NULL;
1543 tmp_ctx = talloc_new(state->db);
1544 if (tmp_ctx == NULL) {
1545 DBG_ERR("talloc_new failed\n");
1549 /* we have to give them a locked record to prevent races */
1550 rec = db_ctdb_fetch_locked(state->db, tmp_ctx, key);
1551 if (rec != NULL && rec->value.dsize > 0) {
1552 state->fn(rec, state->private_data);
1555 talloc_free(tmp_ctx);
1558 static int traverse_persistent_callback(TDB_CONTEXT *tdb, TDB_DATA kbuf, TDB_DATA dbuf,
1561 struct traverse_state *state = (struct traverse_state *)private_data;
1562 struct db_record *rec;
1563 TALLOC_CTX *tmp_ctx = talloc_new(state->db);
1567 * Skip the __db_sequence_number__ key:
1568 * This is used for persistent transactions internally.
1570 if (kbuf.dsize == strlen(CTDB_DB_SEQNUM_KEY) + 1 &&
1571 strcmp((const char*)kbuf.dptr, CTDB_DB_SEQNUM_KEY) == 0)
1576 /* we have to give them a locked record to prevent races */
1577 rec = db_ctdb_fetch_locked(state->db, tmp_ctx, kbuf);
1578 if (rec && rec->value.dsize > 0) {
1579 ret = state->fn(rec, state->private_data);
1583 talloc_free(tmp_ctx);
1587 /* wrapper to use traverse_persistent_callback with dbwrap */
1588 static int traverse_persistent_callback_dbwrap(struct db_record *rec, void* data)
1590 return traverse_persistent_callback(NULL, rec->key, rec->value, data);
1593 static int db_ctdbd_traverse(uint32_t db_id,
1594 void (*fn)(TDB_DATA key, TDB_DATA data,
1595 void *private_data),
1598 struct ctdbd_connection *conn;
1602 ret = ctdbd_init_connection(talloc_tos(), lp_ctdbd_socket(),
1603 lp_ctdb_timeout(), &conn);
1606 DBG_WARNING("ctdbd_init_connection failed: %s\n",
1611 ret = ctdbd_traverse(conn, db_id, fn, private_data);
1615 DBG_WARNING("ctdbd_traverse failed: %s\n",
1624 static int db_ctdb_traverse(struct db_context *db,
1625 int (*fn)(struct db_record *rec,
1626 void *private_data),
1630 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
1631 struct db_ctdb_ctx);
1632 struct traverse_state state;
1634 state = (struct traverse_state) {
1637 .private_data = private_data,
1640 if (db->persistent) {
1641 struct tdb_context *ltdb = ctx->wtdb->tdb;
1643 /* for persistent databases we don't need to do a ctdb traverse,
1644 we can do a faster local traverse */
1645 ret = tdb_traverse(ltdb, traverse_persistent_callback, &state);
1649 if (ctx->transaction && ctx->transaction->m_write) {
1651 * we now have to handle keys not yet
1652 * present at transaction start
1654 struct db_context *newkeys = db_open_rbt(talloc_tos());
1655 struct ctdb_marshall_buffer *mbuf = ctx->transaction->m_write;
1656 struct ctdb_rec_data_old *rec=NULL;
1661 if (newkeys == NULL) {
1665 for (i=0; i<mbuf->count; i++) {
1667 rec = db_ctdb_marshall_loop_next_key(
1669 SMB_ASSERT(rec != NULL);
1671 if (!tdb_exists(ltdb, key)) {
1672 dbwrap_store(newkeys, key, tdb_null, 0);
1675 status = dbwrap_traverse(newkeys,
1676 traverse_persistent_callback_dbwrap,
1679 talloc_free(newkeys);
1680 if (!NT_STATUS_IS_OK(status)) {
1688 ret = db_ctdbd_traverse(ctx->db_id, traverse_callback, &state);
1695 static NTSTATUS db_ctdb_storev_deny(struct db_record *rec,
1696 const TDB_DATA *dbufs, int num_dbufs, int flag)
1698 return NT_STATUS_MEDIA_WRITE_PROTECTED;
1701 static NTSTATUS db_ctdb_delete_deny(struct db_record *rec)
1703 return NT_STATUS_MEDIA_WRITE_PROTECTED;
1706 static void traverse_read_callback(TDB_DATA key, TDB_DATA data, void *private_data)
1708 struct traverse_state *state = (struct traverse_state *)private_data;
1709 struct db_record rec;
1715 rec.storev = db_ctdb_storev_deny;
1716 rec.delete_rec = db_ctdb_delete_deny;
1717 rec.private_data = NULL;
1718 rec.value_valid = true;
1719 state->fn(&rec, state->private_data);
1723 static int traverse_persistent_callback_read(TDB_CONTEXT *tdb, TDB_DATA kbuf, TDB_DATA dbuf,
1726 struct traverse_state *state = (struct traverse_state *)private_data;
1727 struct db_record rec;
1730 * Skip the __db_sequence_number__ key:
1731 * This is used for persistent transactions internally.
1733 if (kbuf.dsize == strlen(CTDB_DB_SEQNUM_KEY) + 1 &&
1734 strcmp((const char*)kbuf.dptr, CTDB_DB_SEQNUM_KEY) == 0)
1743 rec.value_valid = true;
1744 rec.storev = db_ctdb_storev_deny;
1745 rec.delete_rec = db_ctdb_delete_deny;
1746 rec.private_data = NULL;
1748 if (rec.value.dsize <= sizeof(struct ctdb_ltdb_header)) {
1749 /* a deleted record */
1752 rec.value.dsize -= sizeof(struct ctdb_ltdb_header);
1753 rec.value.dptr += sizeof(struct ctdb_ltdb_header);
1756 return state->fn(&rec, state->private_data);
1759 static int db_ctdb_traverse_read(struct db_context *db,
1760 int (*fn)(struct db_record *rec,
1761 void *private_data),
1765 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
1766 struct db_ctdb_ctx);
1767 struct traverse_state state;
1769 state = (struct traverse_state) {
1772 .private_data = private_data,
1775 if (db->persistent) {
1776 /* for persistent databases we don't need to do a ctdb traverse,
1777 we can do a faster local traverse */
1780 nrecs = tdb_traverse_read(ctx->wtdb->tdb,
1781 traverse_persistent_callback_read,
1789 ret = db_ctdbd_traverse(ctx->db_id, traverse_read_callback, &state);
1796 static int db_ctdb_get_seqnum(struct db_context *db)
1798 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
1799 struct db_ctdb_ctx);
1800 return tdb_get_seqnum(ctx->wtdb->tdb);
1803 static size_t db_ctdb_id(struct db_context *db, uint8_t *id, size_t idlen)
1805 struct db_ctdb_ctx *ctx = talloc_get_type_abort(
1806 db->private_data, struct db_ctdb_ctx);
1808 if (idlen >= sizeof(ctx->db_id)) {
1809 memcpy(id, &ctx->db_id, sizeof(ctx->db_id));
1812 return sizeof(ctx->db_id);
1815 struct db_context *db_open_ctdb(TALLOC_CTX *mem_ctx,
1816 struct messaging_context *msg_ctx,
1818 int hash_size, int tdb_flags,
1819 int open_flags, mode_t mode,
1820 enum dbwrap_lock_order lock_order,
1821 uint64_t dbwrap_flags)
1823 struct db_context *result;
1824 struct db_ctdb_ctx *db_ctdb;
1826 struct loadparm_context *lp_ctx;
1828 TDB_DATA outdata = {0};
1829 bool persistent = (tdb_flags & TDB_CLEAR_IF_FIRST) == 0;
1833 if (!lp_clustering()) {
1834 DEBUG(10, ("Clustering disabled -- no ctdb\n"));
1838 if (!(result = talloc_zero(mem_ctx, struct db_context))) {
1839 DEBUG(0, ("talloc failed\n"));
1840 TALLOC_FREE(result);
1844 if (!(db_ctdb = talloc(result, struct db_ctdb_ctx))) {
1845 DEBUG(0, ("talloc failed\n"));
1846 TALLOC_FREE(result);
1850 result->name = talloc_strdup(result, name);
1851 if (result->name == NULL) {
1852 DEBUG(0, ("talloc failed\n"));
1853 TALLOC_FREE(result);
1857 db_ctdb->transaction = NULL;
1858 db_ctdb->db = result;
1860 ret = ctdbd_db_attach(messaging_ctdb_connection(), name,
1861 &db_ctdb->db_id, persistent);
1863 DEBUG(0, ("ctdbd_db_attach failed for %s: %s\n", name,
1865 TALLOC_FREE(result);
1869 if (tdb_flags & TDB_SEQNUM) {
1870 data.dptr = (uint8_t *)&db_ctdb->db_id;
1871 data.dsize = sizeof(db_ctdb->db_id);
1873 ret = ctdbd_control_local(messaging_ctdb_connection(),
1874 CTDB_CONTROL_ENABLE_SEQNUM,
1876 NULL, NULL, &cstatus);
1877 if ((ret != 0) || cstatus != 0) {
1878 DBG_ERR("ctdb_control for enable seqnum "
1879 "failed: %s\n", strerror(ret));
1880 TALLOC_FREE(result);
1885 db_path = ctdbd_dbpath(messaging_ctdb_connection(), db_ctdb,
1887 if (db_path == NULL) {
1888 DBG_ERR("ctdbd_dbpath failed\n");
1889 TALLOC_FREE(result);
1893 result->persistent = persistent;
1894 result->lock_order = lock_order;
1896 data.dptr = (uint8_t *)&db_ctdb->db_id;
1897 data.dsize = sizeof(db_ctdb->db_id);
1899 ret = ctdbd_control_local(messaging_ctdb_connection(),
1900 CTDB_CONTROL_DB_OPEN_FLAGS,
1901 0, 0, data, NULL, &outdata, &cstatus);
1903 DBG_ERR(" ctdb control for db_open_flags "
1904 "failed: %s\n", strerror(ret));
1905 TALLOC_FREE(result);
1909 if (cstatus != 0 || outdata.dsize != sizeof(int)) {
1910 DBG_ERR("ctdb_control for db_open_flags failed\n");
1911 TALLOC_FREE(outdata.dptr);
1912 TALLOC_FREE(result);
1916 tdb_flags = *(int *)outdata.dptr;
1917 TALLOC_FREE(outdata.dptr);
1919 if (!result->persistent) {
1920 ret = ctdb_async_ctx_init(NULL, messaging_tevent_context(msg_ctx));
1922 DBG_ERR("ctdb_async_ctx_init failed: %s\n", strerror(ret));
1923 TALLOC_FREE(result);
1928 if (!result->persistent &&
1929 (dbwrap_flags & DBWRAP_FLAG_OPTIMIZE_READONLY_ACCESS))
1933 indata = make_tdb_data((uint8_t *)&db_ctdb->db_id,
1934 sizeof(db_ctdb->db_id));
1936 ret = ctdbd_control_local(
1937 messaging_ctdb_connection(),
1938 CTDB_CONTROL_SET_DB_READONLY, 0, 0,
1939 indata, NULL, NULL, &cstatus);
1940 if ((ret != 0) || (cstatus != 0)) {
1941 DEBUG(1, ("CTDB_CONTROL_SET_DB_READONLY failed: "
1942 "%s, %"PRIi32"\n", strerror(ret), cstatus));
1943 TALLOC_FREE(result);
1948 lp_ctx = loadparm_init_s3(db_path, loadparm_s3_helpers());
1950 if (hash_size == 0) {
1951 hash_size = lpcfg_tdb_hash_size(lp_ctx, db_path);
1954 db_ctdb->wtdb = tdb_wrap_open(db_ctdb, db_path, hash_size,
1955 lpcfg_tdb_flags(lp_ctx, tdb_flags),
1957 talloc_unlink(db_path, lp_ctx);
1958 if (db_ctdb->wtdb == NULL) {
1959 DEBUG(0, ("Could not open tdb %s: %s\n", db_path, strerror(errno)));
1960 TALLOC_FREE(result);
1963 talloc_free(db_path);
1965 /* honor permissions if user has specified O_CREAT */
1966 if (open_flags & O_CREAT) {
1968 fd = tdb_fd(db_ctdb->wtdb->tdb);
1969 ret = fchmod(fd, mode);
1971 DBG_WARNING("fchmod failed: %s\n",
1973 TALLOC_FREE(result);
1978 if (result->persistent) {
1979 db_ctdb->lock_ctx = g_lock_ctx_init(db_ctdb, msg_ctx);
1980 if (db_ctdb->lock_ctx == NULL) {
1981 DEBUG(0, ("g_lock_ctx_init failed\n"));
1982 TALLOC_FREE(result);
1987 db_ctdb->warn_unlock_msecs = lp_parm_int(-1, "ctdb",
1988 "unlock_warn_threshold", 5);
1989 db_ctdb->warn_migrate_attempts = lp_parm_int(-1, "ctdb",
1990 "migrate_attempts", 10);
1991 db_ctdb->warn_migrate_msecs = lp_parm_int(-1, "ctdb",
1992 "migrate_duration", 5000);
1993 db_ctdb->warn_locktime_msecs = lp_ctdb_locktime_warn_threshold();
1995 result->private_data = (void *)db_ctdb;
1996 result->fetch_locked = db_ctdb_fetch_locked;
1997 result->try_fetch_locked = db_ctdb_try_fetch_locked;
1998 result->parse_record = db_ctdb_parse_record;
1999 result->parse_record_send = db_ctdb_parse_record_send;
2000 result->parse_record_recv = db_ctdb_parse_record_recv;
2001 result->traverse = db_ctdb_traverse;
2002 result->traverse_read = db_ctdb_traverse_read;
2003 result->get_seqnum = db_ctdb_get_seqnum;
2004 result->transaction_start = db_ctdb_transaction_start;
2005 result->transaction_commit = db_ctdb_transaction_commit;
2006 result->transaction_cancel = db_ctdb_transaction_cancel;
2007 result->id = db_ctdb_id;
2009 DEBUG(3,("db_open_ctdb: opened database '%s' with dbid 0x%x\n",
2010 name, db_ctdb->db_id));