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;
548 result->value.dsize = ctdb_data.dsize - sizeof(struct ctdb_ltdb_header);
549 result->value.dptr = NULL;
551 if ((result->value.dsize != 0)
552 && !(result->value.dptr = (uint8_t *)talloc_memdup(
553 result, ctdb_data.dptr + sizeof(struct ctdb_ltdb_header),
554 result->value.dsize))) {
555 DEBUG(0, ("talloc failed\n"));
558 result->value_valid = true;
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_write = db_ctdb_marshall_add(h, h->m_write, h->ctx->db_id, 0, key, &header, data);
655 if (h->m_write == NULL) {
656 DEBUG(0,(__location__ " Failed to add to marshalling record\n"));
657 talloc_free(tmp_ctx);
658 return NT_STATUS_NO_MEMORY;
661 talloc_free(tmp_ctx);
667 a record store inside a transaction
669 static NTSTATUS db_ctdb_storev_transaction(
670 struct db_record *rec, const TDB_DATA *dbufs, int num_dbufs, int flag)
672 struct db_ctdb_transaction_handle *h = talloc_get_type_abort(
673 rec->private_data, struct db_ctdb_transaction_handle);
677 data = dbwrap_merge_dbufs(rec, dbufs, num_dbufs);
678 if (data.dptr == NULL) {
679 return NT_STATUS_NO_MEMORY;
682 status = db_ctdb_transaction_store(h, rec->key, data);
684 TALLOC_FREE(data.dptr);
690 a record delete inside a transaction
692 static NTSTATUS db_ctdb_delete_transaction(struct db_record *rec)
694 struct db_ctdb_transaction_handle *h = talloc_get_type_abort(
695 rec->private_data, struct db_ctdb_transaction_handle);
698 status = db_ctdb_transaction_store(h, rec->key, tdb_null);
702 static void db_ctdb_fetch_db_seqnum_parser(
703 TDB_DATA key, struct ctdb_ltdb_header *header,
704 TDB_DATA data, void *private_data)
706 uint64_t *seqnum = (uint64_t *)private_data;
708 if (data.dsize != sizeof(uint64_t)) {
712 memcpy(seqnum, data.dptr, sizeof(*seqnum));
716 * Fetch the db sequence number of a persistent db directly from the db.
718 static NTSTATUS db_ctdb_fetch_db_seqnum_from_db(struct db_ctdb_ctx *db,
724 if (seqnum == NULL) {
725 return NT_STATUS_INVALID_PARAMETER;
728 key = string_term_tdb_data(CTDB_DB_SEQNUM_KEY);
730 status = db_ctdb_ltdb_parse(
731 db, key, db_ctdb_fetch_db_seqnum_parser, seqnum);
733 if (NT_STATUS_IS_OK(status)) {
736 if (NT_STATUS_EQUAL(status, NT_STATUS_NOT_FOUND)) {
744 * Store the database sequence number inside a transaction.
746 static NTSTATUS db_ctdb_store_db_seqnum(struct db_ctdb_transaction_handle *h,
750 const char *keyname = CTDB_DB_SEQNUM_KEY;
754 key = string_term_tdb_data(keyname);
756 data.dptr = (uint8_t *)&seqnum;
757 data.dsize = sizeof(uint64_t);
759 status = db_ctdb_transaction_store(h, key, data);
767 static int db_ctdb_transaction_commit(struct db_context *db)
769 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
773 struct db_ctdb_transaction_handle *h = ctx->transaction;
774 uint64_t old_seqnum, new_seqnum;
778 DEBUG(0,(__location__ " transaction commit with no open transaction on db 0x%08x\n", ctx->db_id));
782 if (h->nested_cancel) {
783 db->transaction_cancel(db);
784 DEBUG(5,(__location__ " Failed transaction commit after nested cancel\n"));
788 if (h->nesting != 0) {
790 DEBUG(5, (__location__ " transaction commit on db 0x%08x: nesting %d -> %d\n",
791 ctx->db_id, ctx->transaction->nesting + 1, ctx->transaction->nesting));
795 if (h->m_write == NULL) {
797 * No changes were made, so don't change the seqnum,
798 * don't push to other node, just exit with success.
804 DEBUG(5,(__location__ " transaction commit on db 0x%08x\n", ctx->db_id));
807 * As the last db action before committing, bump the database sequence
808 * number. Note that this undoes all changes to the seqnum records
809 * performed under the transaction. This record is not meant to be
810 * modified by user interaction. It is for internal use only...
812 rets = db_ctdb_fetch_db_seqnum_from_db(ctx, &old_seqnum);
813 if (!NT_STATUS_IS_OK(rets)) {
814 DEBUG(1, (__location__ " failed to fetch the db sequence number "
815 "in transaction commit on db 0x%08x\n", ctx->db_id));
820 new_seqnum = old_seqnum + 1;
822 rets = db_ctdb_store_db_seqnum(h, new_seqnum);
823 if (!NT_STATUS_IS_OK(rets)) {
824 DEBUG(1, (__location__ "failed to store the db sequence number "
825 " in transaction commit on db 0x%08x\n", ctx->db_id));
831 /* tell ctdbd to commit to the other nodes */
832 ret = ctdbd_control_local(messaging_ctdb_connection(),
833 CTDB_CONTROL_TRANS3_COMMIT,
835 db_ctdb_marshall_finish(h->m_write),
836 NULL, NULL, &status);
837 if ((ret != 0) || status != 0) {
839 * The TRANS3_COMMIT control should only possibly fail when a
840 * recovery has been running concurrently. In any case, the db
841 * will be the same on all nodes, either the new copy or the
842 * old copy. This can be detected by comparing the old and new
843 * local sequence numbers.
845 rets = db_ctdb_fetch_db_seqnum_from_db(ctx, &new_seqnum);
846 if (!NT_STATUS_IS_OK(rets)) {
847 DEBUG(1, (__location__ " failed to refetch db sequence "
848 "number after failed TRANS3_COMMIT\n"));
853 if (new_seqnum == old_seqnum) {
854 /* Recovery prevented all our changes: retry. */
857 if (new_seqnum != (old_seqnum + 1)) {
858 DEBUG(0, (__location__ " ERROR: new_seqnum[%lu] != "
859 "old_seqnum[%lu] + (0 or 1) after failed "
860 "TRANS3_COMMIT - this should not happen!\n",
861 (unsigned long)new_seqnum,
862 (unsigned long)old_seqnum));
867 * Recovery propagated our changes to all nodes, completing
868 * our commit for us - succeed.
875 h->ctx->transaction = NULL;
884 static int db_ctdb_transaction_cancel(struct db_context *db)
886 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
888 struct db_ctdb_transaction_handle *h = ctx->transaction;
891 DEBUG(0,(__location__ " transaction cancel with no open transaction on db 0x%08x\n", ctx->db_id));
895 if (h->nesting != 0) {
897 h->nested_cancel = true;
898 DEBUG(5, (__location__ " transaction cancel on db 0x%08x: nesting %d -> %d\n",
899 ctx->db_id, ctx->transaction->nesting + 1, ctx->transaction->nesting));
903 DEBUG(5,(__location__ " Cancel transaction on db 0x%08x\n", ctx->db_id));
905 ctx->transaction = NULL;
911 static NTSTATUS db_ctdb_storev(struct db_record *rec,
912 const TDB_DATA *dbufs, int num_dbufs, int flag)
914 struct db_ctdb_rec *crec = talloc_get_type_abort(
915 rec->private_data, struct db_ctdb_rec);
918 status = db_ctdb_ltdb_store(crec->ctdb_ctx, rec->key, &(crec->header),
925 static NTSTATUS db_ctdb_send_schedule_for_deletion(struct db_record *rec)
927 NTSTATUS status = NT_STATUS_OK;
929 struct ctdb_control_schedule_for_deletion *dd;
932 struct db_ctdb_rec *crec = talloc_get_type_abort(
933 rec->private_data, struct db_ctdb_rec);
934 struct db_ctdb_ctx *ctx = crec->ctdb_ctx;
936 indata.dsize = offsetof(struct ctdb_control_schedule_for_deletion, key) + rec->key.dsize;
937 indata.dptr = talloc_zero_array(crec, uint8_t, indata.dsize);
938 if (indata.dptr == NULL) {
939 DEBUG(0, (__location__ " talloc failed!\n"));
940 return NT_STATUS_NO_MEMORY;
943 dd = (struct ctdb_control_schedule_for_deletion *)(void *)indata.dptr;
944 dd->db_id = ctx->db_id;
945 dd->hdr = crec->header;
946 dd->keylen = rec->key.dsize;
947 memcpy(dd->key, rec->key.dptr, rec->key.dsize);
949 ret = ctdbd_control_local(messaging_ctdb_connection(),
950 CTDB_CONTROL_SCHEDULE_FOR_DELETION,
951 crec->ctdb_ctx->db_id,
952 CTDB_CTRL_FLAG_NOREPLY, /* flags */
957 talloc_free(indata.dptr);
959 if ((ret != 0) || cstatus != 0) {
960 DEBUG(1, (__location__ " Error sending local control "
961 "SCHEDULE_FOR_DELETION: %s, cstatus = %"PRIi32"\n",
962 strerror(ret), cstatus));
964 status = map_nt_error_from_unix(ret);
966 status = NT_STATUS_UNSUCCESSFUL;
973 static NTSTATUS db_ctdb_delete(struct db_record *rec)
978 * We have to store the header with empty data. TODO: Fix the
982 status = db_ctdb_storev(rec, &tdb_null, 1, 0);
983 if (!NT_STATUS_IS_OK(status)) {
987 status = db_ctdb_send_schedule_for_deletion(rec);
991 static int db_ctdb_record_destr(struct db_record* data)
993 struct db_ctdb_rec *crec = talloc_get_type_abort(
994 data->private_data, struct db_ctdb_rec);
997 struct timeval before;
1000 DEBUG(10, (DEBUGLEVEL > 10
1001 ? "Unlocking db %u key %s\n"
1002 : "Unlocking db %u key %.20s\n",
1003 (int)crec->ctdb_ctx->db_id,
1004 hex_encode_talloc(data, (unsigned char *)data->key.dptr,
1007 before = timeval_current();
1009 ret = tdb_chainunlock(crec->ctdb_ctx->wtdb->tdb, data->key);
1011 timediff = timeval_elapsed(&before);
1012 timediff *= 1000; /* get us milliseconds */
1014 if (timediff > crec->ctdb_ctx->warn_unlock_msecs) {
1016 key = hex_encode_talloc(talloc_tos(),
1017 (unsigned char *)data->key.dptr,
1019 DEBUG(0, ("tdb_chainunlock on db %s, key %s took %f milliseconds\n",
1020 tdb_name(crec->ctdb_ctx->wtdb->tdb), key,
1026 DEBUG(0, ("tdb_chainunlock failed\n"));
1030 threshold = crec->ctdb_ctx->warn_locktime_msecs;
1031 if (threshold != 0) {
1032 timediff = timeval_elapsed(&crec->lock_time) * 1000;
1033 if (timediff > threshold) {
1036 key = hex_encode_talloc(data,
1037 (unsigned char *)data->key.dptr,
1039 DEBUG(0, ("Held tdb lock on db %s, key %s "
1040 "%f milliseconds\n",
1041 tdb_name(crec->ctdb_ctx->wtdb->tdb),
1050 * Check whether we have a valid local copy of the given record,
1051 * either for reading or for writing.
1053 static bool db_ctdb_can_use_local_hdr(const struct ctdb_ltdb_header *hdr,
1054 uint32_t my_vnn, bool read_only)
1056 if (hdr->dmaster != my_vnn) {
1057 /* If we're not dmaster, it must be r/o copy. */
1058 return read_only && (hdr->flags & CTDB_REC_RO_HAVE_READONLY);
1062 * If we want write access, no one may have r/o copies.
1064 return read_only || !(hdr->flags & CTDB_REC_RO_HAVE_DELEGATIONS);
1067 static bool db_ctdb_can_use_local_copy(TDB_DATA ctdb_data, uint32_t my_vnn,
1070 if (ctdb_data.dptr == NULL) {
1074 if (ctdb_data.dsize < sizeof(struct ctdb_ltdb_header)) {
1078 return db_ctdb_can_use_local_hdr(
1079 (struct ctdb_ltdb_header *)ctdb_data.dptr, my_vnn, read_only);
1082 static struct db_record *fetch_locked_internal(struct db_ctdb_ctx *ctx,
1083 TALLOC_CTX *mem_ctx,
1087 struct db_record *result;
1088 struct db_ctdb_rec *crec;
1090 int migrate_attempts;
1091 struct timeval migrate_start;
1092 struct timeval chainlock_start;
1093 struct timeval ctdb_start_time;
1094 double chainlock_time = 0;
1095 double ctdb_time = 0;
1100 if (!(result = talloc(mem_ctx, struct db_record))) {
1101 DEBUG(0, ("talloc failed\n"));
1105 if (!(crec = talloc_zero(result, struct db_ctdb_rec))) {
1106 DEBUG(0, ("talloc failed\n"));
1107 TALLOC_FREE(result);
1111 result->db = ctx->db;
1112 result->private_data = (void *)crec;
1113 crec->ctdb_ctx = ctx;
1115 result->key.dsize = key.dsize;
1116 result->key.dptr = (uint8_t *)talloc_memdup(result, key.dptr,
1118 if (result->key.dptr == NULL) {
1119 DEBUG(0, ("talloc failed\n"));
1120 TALLOC_FREE(result);
1124 migrate_attempts = 0;
1125 GetTimeOfDay(&migrate_start);
1128 * Do a blocking lock on the record
1132 if (DEBUGLEVEL >= 10) {
1133 char *keystr = hex_encode_talloc(result, key.dptr, key.dsize);
1134 DEBUG(10, (DEBUGLEVEL > 10
1135 ? "Locking db %u key %s\n"
1136 : "Locking db %u key %.20s\n",
1137 (int)crec->ctdb_ctx->db_id, keystr));
1138 TALLOC_FREE(keystr);
1141 GetTimeOfDay(&chainlock_start);
1143 ? tdb_chainlock_nonblock(ctx->wtdb->tdb, key)
1144 : tdb_chainlock(ctx->wtdb->tdb, key);
1145 chainlock_time += timeval_elapsed(&chainlock_start);
1148 DEBUG(3, ("tdb_chainlock failed\n"));
1149 TALLOC_FREE(result);
1153 result->storev = db_ctdb_storev;
1154 result->delete_rec = db_ctdb_delete;
1155 talloc_set_destructor(result, db_ctdb_record_destr);
1157 ctdb_data = tdb_fetch(ctx->wtdb->tdb, key);
1160 * See if we have a valid record and we are the dmaster. If so, we can
1161 * take the shortcut and just return it.
1164 if (!db_ctdb_can_use_local_copy(ctdb_data, get_my_vnn(), false)) {
1165 SAFE_FREE(ctdb_data.dptr);
1166 tdb_chainunlock(ctx->wtdb->tdb, key);
1167 talloc_set_destructor(result, NULL);
1169 if (tryonly && (migrate_attempts != 0)) {
1170 DEBUG(5, ("record migrated away again\n"));
1171 TALLOC_FREE(result);
1175 migrate_attempts += 1;
1177 DEBUG(10, ("ctdb_data.dptr = %p, dmaster = %"PRIu32" "
1178 "(%"PRIu32") %"PRIu32"\n",
1179 ctdb_data.dptr, ctdb_data.dptr ?
1180 ((struct ctdb_ltdb_header *)ctdb_data.dptr)->dmaster :
1184 ((struct ctdb_ltdb_header *)ctdb_data.dptr)->flags : 0));
1186 GetTimeOfDay(&ctdb_start_time);
1187 ret = ctdbd_migrate(messaging_ctdb_connection(), ctx->db_id,
1189 ctdb_time += timeval_elapsed(&ctdb_start_time);
1192 DEBUG(5, ("ctdbd_migrate failed: %s\n",
1194 TALLOC_FREE(result);
1197 /* now its migrated, try again */
1203 duration = timeval_elapsed(&migrate_start);
1206 * Convert the duration to milliseconds to avoid a
1207 * floating-point division of
1208 * lp_parm_int("migrate_duration") by 1000.
1210 duration_msecs = duration * 1000;
1213 if ((migrate_attempts > ctx->warn_migrate_attempts) ||
1214 (duration_msecs > ctx->warn_migrate_msecs)) {
1217 if (tdb_get_flags(ctx->wtdb->tdb) & TDB_INCOMPATIBLE_HASH) {
1218 chain = tdb_jenkins_hash(&key) %
1219 tdb_hash_size(ctx->wtdb->tdb);
1222 DEBUG(0, ("db_ctdb_fetch_locked for %s key %s, chain %d "
1223 "needed %d attempts, %d milliseconds, "
1224 "chainlock: %f ms, CTDB %f ms\n",
1225 tdb_name(ctx->wtdb->tdb),
1226 hex_encode_talloc(talloc_tos(),
1227 (unsigned char *)key.dptr,
1230 migrate_attempts, duration_msecs,
1231 chainlock_time * 1000.0,
1232 ctdb_time * 1000.0));
1235 GetTimeOfDay(&crec->lock_time);
1237 memcpy(&crec->header, ctdb_data.dptr, sizeof(crec->header));
1239 result->value.dsize = ctdb_data.dsize - sizeof(crec->header);
1240 result->value.dptr = NULL;
1242 if (result->value.dsize != 0) {
1243 result->value.dptr = talloc_memdup(
1244 result, ctdb_data.dptr + sizeof(crec->header),
1245 result->value.dsize);
1246 if (result->value.dptr == NULL) {
1247 DBG_ERR("talloc failed\n");
1248 TALLOC_FREE(result);
1251 result->value_valid = true;
1253 SAFE_FREE(ctdb_data.dptr);
1258 static struct db_record *db_ctdb_fetch_locked(struct db_context *db,
1259 TALLOC_CTX *mem_ctx,
1262 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
1263 struct db_ctdb_ctx);
1265 if (ctx->transaction != NULL) {
1266 return db_ctdb_fetch_locked_transaction(ctx, mem_ctx, key);
1269 if (db->persistent) {
1270 return db_ctdb_fetch_locked_persistent(ctx, mem_ctx, key);
1273 return fetch_locked_internal(ctx, mem_ctx, key, false);
1276 static struct db_record *db_ctdb_try_fetch_locked(struct db_context *db,
1277 TALLOC_CTX *mem_ctx,
1280 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
1281 struct db_ctdb_ctx);
1283 if (ctx->transaction != NULL) {
1284 return db_ctdb_fetch_locked_transaction(ctx, mem_ctx, key);
1287 if (db->persistent) {
1288 return db_ctdb_fetch_locked_persistent(ctx, mem_ctx, key);
1291 return fetch_locked_internal(ctx, mem_ctx, key, true);
1294 struct db_ctdb_parse_record_state {
1295 void (*parser)(TDB_DATA key, TDB_DATA data, void *private_data);
1298 bool ask_for_readonly_copy;
1303 static void db_ctdb_parse_record_parser(
1304 TDB_DATA key, struct ctdb_ltdb_header *header,
1305 TDB_DATA data, void *private_data)
1307 struct db_ctdb_parse_record_state *state =
1308 (struct db_ctdb_parse_record_state *)private_data;
1309 state->parser(key, data, state->private_data);
1312 static void db_ctdb_parse_record_parser_nonpersistent(
1313 TDB_DATA key, struct ctdb_ltdb_header *header,
1314 TDB_DATA data, void *private_data)
1316 struct db_ctdb_parse_record_state *state =
1317 (struct db_ctdb_parse_record_state *)private_data;
1319 if (db_ctdb_can_use_local_hdr(header, state->my_vnn, true)) {
1321 * A record consisting only of the ctdb header can be
1322 * a validly created empty record or a tombstone
1323 * record of a deleted record (not vacuumed yet). Mark
1326 state->empty_record = (data.dsize == 0);
1327 if (!state->empty_record) {
1328 state->parser(key, data, state->private_data);
1333 * We found something in the db, so it seems that this record,
1334 * while not usable locally right now, is popular. Ask for a
1337 state->ask_for_readonly_copy = true;
1341 static NTSTATUS db_ctdb_try_parse_local_record(struct db_ctdb_ctx *ctx,
1343 struct db_ctdb_parse_record_state *state)
1347 if (ctx->transaction != NULL) {
1348 struct db_ctdb_transaction_handle *h = ctx->transaction;
1352 * Transactions only happen for persistent db's.
1355 found = parse_newest_in_marshall_buffer(
1356 h->m_write, key, db_ctdb_parse_record_parser, state);
1359 return NT_STATUS_OK;
1363 if (ctx->db->persistent) {
1365 * Persistent db, but not found in the transaction buffer
1367 return db_ctdb_ltdb_parse(
1368 ctx, key, db_ctdb_parse_record_parser, state);
1371 state->done = false;
1372 state->ask_for_readonly_copy = false;
1374 status = db_ctdb_ltdb_parse(
1375 ctx, key, db_ctdb_parse_record_parser_nonpersistent, state);
1376 if (NT_STATUS_IS_OK(status) && state->done) {
1377 if (state->empty_record) {
1379 * We know authoritatively, that this is an empty
1380 * record. Since ctdb does not distinguish between empty
1381 * and deleted records, this can be a record stored as
1382 * empty or a not-yet-vacuumed tombstone record of a
1383 * deleted record. Now Samba right now can live without
1384 * empty records, so we can safely report this record
1387 * See bugs 10008 and 12005.
1389 return NT_STATUS_NOT_FOUND;
1391 return NT_STATUS_OK;
1394 return NT_STATUS_MORE_PROCESSING_REQUIRED;
1397 static NTSTATUS db_ctdb_parse_record(struct db_context *db, TDB_DATA key,
1398 void (*parser)(TDB_DATA key,
1400 void *private_data),
1403 struct db_ctdb_ctx *ctx = talloc_get_type_abort(
1404 db->private_data, struct db_ctdb_ctx);
1405 struct db_ctdb_parse_record_state state;
1409 state.parser = parser;
1410 state.private_data = private_data;
1411 state.my_vnn = get_my_vnn();
1412 state.empty_record = false;
1414 status = db_ctdb_try_parse_local_record(ctx, key, &state);
1415 if (!NT_STATUS_EQUAL(status, NT_STATUS_MORE_PROCESSING_REQUIRED)) {
1419 ret = ctdbd_parse(messaging_ctdb_connection(), ctx->db_id, key,
1420 state.ask_for_readonly_copy, parser, private_data);
1422 if (ret == ENOENT) {
1425 * NT_STATUS_OBJECT_NAME_NOT_FOUND. Our upper
1426 * layers expect NT_STATUS_NOT_FOUND for "no
1427 * record around". We need to convert dbwrap
1428 * to 0/errno away from NTSTATUS ... :-)
1430 return NT_STATUS_NOT_FOUND;
1432 return map_nt_error_from_unix(ret);
1434 return NT_STATUS_OK;
1437 static void db_ctdb_parse_record_done(struct tevent_req *subreq);
1439 static struct tevent_req *db_ctdb_parse_record_send(
1440 TALLOC_CTX *mem_ctx,
1441 struct tevent_context *ev,
1442 struct db_context *db,
1444 void (*parser)(TDB_DATA key,
1446 void *private_data),
1448 enum dbwrap_req_state *req_state)
1450 struct db_ctdb_ctx *ctx = talloc_get_type_abort(
1451 db->private_data, struct db_ctdb_ctx);
1452 struct tevent_req *req = NULL;
1453 struct tevent_req *subreq = NULL;
1454 struct db_ctdb_parse_record_state *state = NULL;
1457 req = tevent_req_create(mem_ctx, &state,
1458 struct db_ctdb_parse_record_state);
1460 *req_state = DBWRAP_REQ_ERROR;
1465 *state = (struct db_ctdb_parse_record_state) {
1467 .private_data = private_data,
1468 .my_vnn = get_my_vnn(),
1469 .empty_record = false,
1472 status = db_ctdb_try_parse_local_record(ctx, key, state);
1473 if (!NT_STATUS_EQUAL(status, NT_STATUS_MORE_PROCESSING_REQUIRED)) {
1474 if (tevent_req_nterror(req, status)) {
1475 *req_state = DBWRAP_REQ_ERROR;
1476 return tevent_req_post(req, ev);
1478 *req_state = DBWRAP_REQ_DONE;
1479 tevent_req_done(req);
1480 return tevent_req_post(req, ev);
1483 subreq = ctdbd_parse_send(state,
1485 ctdb_async_ctx.async_conn,
1488 state->ask_for_readonly_copy,
1492 if (tevent_req_nomem(subreq, req)) {
1493 *req_state = DBWRAP_REQ_ERROR;
1494 return tevent_req_post(req, ev);
1496 tevent_req_set_callback(subreq, db_ctdb_parse_record_done, req);
1501 static void db_ctdb_parse_record_done(struct tevent_req *subreq)
1503 struct tevent_req *req = tevent_req_callback_data(
1504 subreq, struct tevent_req);
1507 ret = ctdbd_parse_recv(subreq);
1508 TALLOC_FREE(subreq);
1510 if (ret == ENOENT) {
1512 * This maps to NT_STATUS_OBJECT_NAME_NOT_FOUND. Our
1513 * upper layers expect NT_STATUS_NOT_FOUND for "no
1514 * record around". We need to convert dbwrap to 0/errno
1515 * away from NTSTATUS ... :-)
1517 tevent_req_nterror(req, NT_STATUS_NOT_FOUND);
1520 tevent_req_nterror(req, map_nt_error_from_unix(ret));
1524 tevent_req_done(req);
1528 static NTSTATUS db_ctdb_parse_record_recv(struct tevent_req *req)
1530 return tevent_req_simple_recv_ntstatus(req);
1533 struct traverse_state {
1534 struct db_context *db;
1535 int (*fn)(struct db_record *rec, void *private_data);
1540 static void traverse_callback(TDB_DATA key, TDB_DATA data, void *private_data)
1542 struct traverse_state *state = (struct traverse_state *)private_data;
1543 struct db_record *rec = NULL;
1544 TALLOC_CTX *tmp_ctx = NULL;
1546 tmp_ctx = talloc_new(state->db);
1547 if (tmp_ctx == NULL) {
1548 DBG_ERR("talloc_new failed\n");
1552 /* we have to give them a locked record to prevent races */
1553 rec = db_ctdb_fetch_locked(state->db, tmp_ctx, key);
1554 if (rec != NULL && rec->value.dsize > 0) {
1555 state->fn(rec, state->private_data);
1558 talloc_free(tmp_ctx);
1561 static int traverse_persistent_callback(TDB_CONTEXT *tdb, TDB_DATA kbuf, TDB_DATA dbuf,
1564 struct traverse_state *state = (struct traverse_state *)private_data;
1565 struct db_record *rec;
1566 TALLOC_CTX *tmp_ctx = talloc_new(state->db);
1570 * Skip the __db_sequence_number__ key:
1571 * This is used for persistent transactions internally.
1573 if (kbuf.dsize == strlen(CTDB_DB_SEQNUM_KEY) + 1 &&
1574 strcmp((const char*)kbuf.dptr, CTDB_DB_SEQNUM_KEY) == 0)
1579 /* we have to give them a locked record to prevent races */
1580 rec = db_ctdb_fetch_locked(state->db, tmp_ctx, kbuf);
1581 if (rec && rec->value.dsize > 0) {
1582 ret = state->fn(rec, state->private_data);
1586 talloc_free(tmp_ctx);
1590 /* wrapper to use traverse_persistent_callback with dbwrap */
1591 static int traverse_persistent_callback_dbwrap(struct db_record *rec, void* data)
1593 return traverse_persistent_callback(NULL, rec->key, rec->value, data);
1596 static int db_ctdbd_traverse(uint32_t db_id,
1597 void (*fn)(TDB_DATA key, TDB_DATA data,
1598 void *private_data),
1601 struct ctdbd_connection *conn;
1605 ret = ctdbd_init_connection(talloc_tos(), lp_ctdbd_socket(),
1606 lp_ctdb_timeout(), &conn);
1609 DBG_WARNING("ctdbd_init_connection failed: %s\n",
1614 ret = ctdbd_traverse(conn, db_id, fn, private_data);
1618 DBG_WARNING("ctdbd_traverse failed: %s\n",
1627 static int db_ctdb_traverse(struct db_context *db,
1628 int (*fn)(struct db_record *rec,
1629 void *private_data),
1633 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
1634 struct db_ctdb_ctx);
1635 struct traverse_state state;
1637 state = (struct traverse_state) {
1640 .private_data = private_data,
1643 if (db->persistent) {
1644 struct tdb_context *ltdb = ctx->wtdb->tdb;
1646 /* for persistent databases we don't need to do a ctdb traverse,
1647 we can do a faster local traverse */
1648 ret = tdb_traverse(ltdb, traverse_persistent_callback, &state);
1652 if (ctx->transaction && ctx->transaction->m_write) {
1654 * we now have to handle keys not yet
1655 * present at transaction start
1657 struct db_context *newkeys = db_open_rbt(talloc_tos());
1658 struct ctdb_marshall_buffer *mbuf = ctx->transaction->m_write;
1659 struct ctdb_rec_data_old *rec=NULL;
1664 if (newkeys == NULL) {
1668 for (i=0; i<mbuf->count; i++) {
1670 rec = db_ctdb_marshall_loop_next_key(
1672 SMB_ASSERT(rec != NULL);
1674 if (!tdb_exists(ltdb, key)) {
1675 dbwrap_store(newkeys, key, tdb_null, 0);
1678 status = dbwrap_traverse(newkeys,
1679 traverse_persistent_callback_dbwrap,
1682 talloc_free(newkeys);
1683 if (!NT_STATUS_IS_OK(status)) {
1691 ret = db_ctdbd_traverse(ctx->db_id, traverse_callback, &state);
1698 static NTSTATUS db_ctdb_storev_deny(struct db_record *rec,
1699 const TDB_DATA *dbufs, int num_dbufs, int flag)
1701 return NT_STATUS_MEDIA_WRITE_PROTECTED;
1704 static NTSTATUS db_ctdb_delete_deny(struct db_record *rec)
1706 return NT_STATUS_MEDIA_WRITE_PROTECTED;
1709 static void traverse_read_callback(TDB_DATA key, TDB_DATA data, void *private_data)
1711 struct traverse_state *state = (struct traverse_state *)private_data;
1712 struct db_record rec;
1718 rec.storev = db_ctdb_storev_deny;
1719 rec.delete_rec = db_ctdb_delete_deny;
1720 rec.private_data = NULL;
1721 rec.value_valid = true;
1722 state->fn(&rec, state->private_data);
1726 static int traverse_persistent_callback_read(TDB_CONTEXT *tdb, TDB_DATA kbuf, TDB_DATA dbuf,
1729 struct traverse_state *state = (struct traverse_state *)private_data;
1730 struct db_record rec;
1733 * Skip the __db_sequence_number__ key:
1734 * This is used for persistent transactions internally.
1736 if (kbuf.dsize == strlen(CTDB_DB_SEQNUM_KEY) + 1 &&
1737 strcmp((const char*)kbuf.dptr, CTDB_DB_SEQNUM_KEY) == 0)
1746 rec.value_valid = true;
1747 rec.storev = db_ctdb_storev_deny;
1748 rec.delete_rec = db_ctdb_delete_deny;
1749 rec.private_data = NULL;
1751 if (rec.value.dsize <= sizeof(struct ctdb_ltdb_header)) {
1752 /* a deleted record */
1755 rec.value.dsize -= sizeof(struct ctdb_ltdb_header);
1756 rec.value.dptr += sizeof(struct ctdb_ltdb_header);
1759 return state->fn(&rec, state->private_data);
1762 static int db_ctdb_traverse_read(struct db_context *db,
1763 int (*fn)(struct db_record *rec,
1764 void *private_data),
1768 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
1769 struct db_ctdb_ctx);
1770 struct traverse_state state;
1772 state = (struct traverse_state) {
1775 .private_data = private_data,
1778 if (db->persistent) {
1779 /* for persistent databases we don't need to do a ctdb traverse,
1780 we can do a faster local traverse */
1783 nrecs = tdb_traverse_read(ctx->wtdb->tdb,
1784 traverse_persistent_callback_read,
1792 ret = db_ctdbd_traverse(ctx->db_id, traverse_read_callback, &state);
1799 static int db_ctdb_get_seqnum(struct db_context *db)
1801 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
1802 struct db_ctdb_ctx);
1803 return tdb_get_seqnum(ctx->wtdb->tdb);
1806 static size_t db_ctdb_id(struct db_context *db, uint8_t *id, size_t idlen)
1808 struct db_ctdb_ctx *ctx = talloc_get_type_abort(
1809 db->private_data, struct db_ctdb_ctx);
1811 if (idlen >= sizeof(ctx->db_id)) {
1812 memcpy(id, &ctx->db_id, sizeof(ctx->db_id));
1815 return sizeof(ctx->db_id);
1818 struct db_context *db_open_ctdb(TALLOC_CTX *mem_ctx,
1819 struct messaging_context *msg_ctx,
1821 int hash_size, int tdb_flags,
1822 int open_flags, mode_t mode,
1823 enum dbwrap_lock_order lock_order,
1824 uint64_t dbwrap_flags)
1826 struct db_context *result;
1827 struct db_ctdb_ctx *db_ctdb;
1829 struct loadparm_context *lp_ctx;
1831 bool persistent = (tdb_flags & TDB_CLEAR_IF_FIRST) == 0;
1835 if (!lp_clustering()) {
1836 DEBUG(10, ("Clustering disabled -- no ctdb\n"));
1840 if (!(result = talloc_zero(mem_ctx, struct db_context))) {
1841 DEBUG(0, ("talloc failed\n"));
1842 TALLOC_FREE(result);
1846 if (!(db_ctdb = talloc(result, struct db_ctdb_ctx))) {
1847 DEBUG(0, ("talloc failed\n"));
1848 TALLOC_FREE(result);
1852 result->name = talloc_strdup(result, name);
1853 if (result->name == NULL) {
1854 DEBUG(0, ("talloc failed\n"));
1855 TALLOC_FREE(result);
1859 db_ctdb->transaction = NULL;
1860 db_ctdb->db = result;
1862 ret = ctdbd_db_attach(messaging_ctdb_connection(), name,
1863 &db_ctdb->db_id, persistent);
1865 DEBUG(0, ("ctdbd_db_attach failed for %s: %s\n", name,
1867 TALLOC_FREE(result);
1871 if (tdb_flags & TDB_SEQNUM) {
1872 data.dptr = (uint8_t *)&db_ctdb->db_id;
1873 data.dsize = sizeof(db_ctdb->db_id);
1875 ret = ctdbd_control_local(messaging_ctdb_connection(),
1876 CTDB_CONTROL_ENABLE_SEQNUM,
1878 NULL, NULL, &cstatus);
1879 if ((ret != 0) || cstatus != 0) {
1880 DBG_ERR("ctdb_control for enable seqnum "
1881 "failed: %s\n", strerror(ret));
1882 TALLOC_FREE(result);
1887 db_path = ctdbd_dbpath(messaging_ctdb_connection(), db_ctdb,
1889 if (db_path == NULL) {
1890 DBG_ERR("ctdbd_dbpath failed\n");
1891 TALLOC_FREE(result);
1895 result->persistent = persistent;
1896 result->lock_order = lock_order;
1898 data.dptr = (uint8_t *)&db_ctdb->db_id;
1899 data.dsize = sizeof(db_ctdb->db_id);
1901 ret = ctdbd_control_local(messaging_ctdb_connection(),
1902 CTDB_CONTROL_DB_OPEN_FLAGS,
1903 0, 0, data, NULL, &data, &cstatus);
1905 DBG_ERR(" ctdb control for db_open_flags "
1906 "failed: %s\n", strerror(ret));
1907 TALLOC_FREE(result);
1911 if (cstatus != 0 || data.dsize != sizeof(int)) {
1912 DBG_ERR("ctdb_control for db_open_flags failed\n");
1913 TALLOC_FREE(result);
1917 tdb_flags = *(int *)data.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));