s3:dbwrap_ctdb: exit early when nothing has been written in transaction_commit.
[Samba/bb.git] / source3 / lib / dbwrap_ctdb.c
blob79c4c0ce420f0cdd28085999d99ba9ed2ba654d0
1 /*
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/>.
21 #include "includes.h"
22 #ifdef CLUSTER_SUPPORT
23 #include "ctdb.h"
24 #include "ctdb_private.h"
25 #include "ctdbd_conn.h"
26 #include "g_lock.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;
37 uint32_t nesting;
38 bool nested_cancel;
39 char *lock_name;
42 struct db_ctdb_ctx {
43 struct db_context *db;
44 struct tdb_wrap *wtdb;
45 uint32 db_id;
46 struct db_ctdb_transaction_handle *transaction;
47 struct g_lock_ctx *lock_ctx;
50 struct db_ctdb_rec {
51 struct db_ctdb_ctx *ctdb_ctx;
52 struct ctdb_ltdb_header header;
55 static struct db_record *fetch_locked_internal(struct db_ctdb_ctx *ctx,
56 TALLOC_CTX *mem_ctx,
57 TDB_DATA key,
58 bool persistent);
60 static NTSTATUS tdb_error_to_ntstatus(struct tdb_context *tdb)
62 NTSTATUS status;
63 enum TDB_ERROR tret = tdb_error(tdb);
65 switch (tret) {
66 case TDB_ERR_EXISTS:
67 status = NT_STATUS_OBJECT_NAME_COLLISION;
68 break;
69 case TDB_ERR_NOEXIST:
70 status = NT_STATUS_OBJECT_NAME_NOT_FOUND;
71 break;
72 default:
73 status = NT_STATUS_INTERNAL_DB_CORRUPTION;
74 break;
77 return status;
81 /**
82 * fetch a record from the tdb, separating out the header
83 * information and returning the body of the record.
85 static NTSTATUS db_ctdb_ltdb_fetch(struct db_ctdb_ctx *db,
86 TDB_DATA key,
87 struct ctdb_ltdb_header *header,
88 TALLOC_CTX *mem_ctx,
89 TDB_DATA *data)
91 TDB_DATA rec;
92 NTSTATUS status;
94 rec = tdb_fetch(db->wtdb->tdb, key);
95 if (rec.dsize < sizeof(struct ctdb_ltdb_header)) {
96 status = NT_STATUS_NOT_FOUND;
97 if (data) {
98 ZERO_STRUCTP(data);
100 if (header) {
101 header->dmaster = (uint32_t)-1;
102 header->rsn = 0;
104 goto done;
107 if (header) {
108 *header = *(struct ctdb_ltdb_header *)rec.dptr;
111 if (data) {
112 data->dsize = rec.dsize - sizeof(struct ctdb_ltdb_header);
113 if (data->dsize == 0) {
114 data->dptr = NULL;
115 } else {
116 data->dptr = (unsigned char *)talloc_memdup(mem_ctx,
117 rec.dptr
118 + sizeof(struct ctdb_ltdb_header),
119 data->dsize);
120 if (data->dptr == NULL) {
121 status = NT_STATUS_NO_MEMORY;
122 goto done;
127 status = NT_STATUS_OK;
129 done:
130 SAFE_FREE(rec.dptr);
131 return status;
135 * Store a record together with the ctdb record header
136 * in the local copy of the database.
138 static NTSTATUS db_ctdb_ltdb_store(struct db_ctdb_ctx *db,
139 TDB_DATA key,
140 struct ctdb_ltdb_header *header,
141 TDB_DATA data)
143 TALLOC_CTX *tmp_ctx = talloc_stackframe();
144 TDB_DATA rec;
145 int ret;
147 rec.dsize = data.dsize + sizeof(struct ctdb_ltdb_header);
148 rec.dptr = (uint8_t *)talloc_size(tmp_ctx, rec.dsize);
150 if (rec.dptr == NULL) {
151 talloc_free(tmp_ctx);
152 return NT_STATUS_NO_MEMORY;
155 memcpy(rec.dptr, header, sizeof(struct ctdb_ltdb_header));
156 memcpy(sizeof(struct ctdb_ltdb_header) + (uint8_t *)rec.dptr, data.dptr, data.dsize);
158 ret = tdb_store(db->wtdb->tdb, key, rec, TDB_REPLACE);
160 talloc_free(tmp_ctx);
162 return (ret == 0) ? NT_STATUS_OK
163 : tdb_error_to_ntstatus(db->wtdb->tdb);
168 form a ctdb_rec_data record from a key/data pair
170 note that header may be NULL. If not NULL then it is included in the data portion
171 of the record
173 static struct ctdb_rec_data *db_ctdb_marshall_record(TALLOC_CTX *mem_ctx, uint32_t reqid,
174 TDB_DATA key,
175 struct ctdb_ltdb_header *header,
176 TDB_DATA data)
178 size_t length;
179 struct ctdb_rec_data *d;
181 length = offsetof(struct ctdb_rec_data, data) + key.dsize +
182 data.dsize + (header?sizeof(*header):0);
183 d = (struct ctdb_rec_data *)talloc_size(mem_ctx, length);
184 if (d == NULL) {
185 return NULL;
187 d->length = length;
188 d->reqid = reqid;
189 d->keylen = key.dsize;
190 memcpy(&d->data[0], key.dptr, key.dsize);
191 if (header) {
192 d->datalen = data.dsize + sizeof(*header);
193 memcpy(&d->data[key.dsize], header, sizeof(*header));
194 memcpy(&d->data[key.dsize+sizeof(*header)], data.dptr, data.dsize);
195 } else {
196 d->datalen = data.dsize;
197 memcpy(&d->data[key.dsize], data.dptr, data.dsize);
199 return d;
203 /* helper function for marshalling multiple records */
204 static struct ctdb_marshall_buffer *db_ctdb_marshall_add(TALLOC_CTX *mem_ctx,
205 struct ctdb_marshall_buffer *m,
206 uint64_t db_id,
207 uint32_t reqid,
208 TDB_DATA key,
209 struct ctdb_ltdb_header *header,
210 TDB_DATA data)
212 struct ctdb_rec_data *r;
213 size_t m_size, r_size;
214 struct ctdb_marshall_buffer *m2 = NULL;
216 r = db_ctdb_marshall_record(talloc_tos(), reqid, key, header, data);
217 if (r == NULL) {
218 talloc_free(m);
219 return NULL;
222 if (m == NULL) {
223 m = (struct ctdb_marshall_buffer *)talloc_zero_size(
224 mem_ctx, offsetof(struct ctdb_marshall_buffer, data));
225 if (m == NULL) {
226 goto done;
228 m->db_id = db_id;
231 m_size = talloc_get_size(m);
232 r_size = talloc_get_size(r);
234 m2 = (struct ctdb_marshall_buffer *)talloc_realloc_size(
235 mem_ctx, m, m_size + r_size);
236 if (m2 == NULL) {
237 talloc_free(m);
238 goto done;
241 memcpy(m_size + (uint8_t *)m2, r, r_size);
243 m2->count++;
245 done:
246 talloc_free(r);
247 return m2;
250 /* we've finished marshalling, return a data blob with the marshalled records */
251 static TDB_DATA db_ctdb_marshall_finish(struct ctdb_marshall_buffer *m)
253 TDB_DATA data;
254 data.dptr = (uint8_t *)m;
255 data.dsize = talloc_get_size(m);
256 return data;
260 loop over a marshalling buffer
262 - pass r==NULL to start
263 - loop the number of times indicated by m->count
265 static struct ctdb_rec_data *db_ctdb_marshall_loop_next(struct ctdb_marshall_buffer *m, struct ctdb_rec_data *r,
266 uint32_t *reqid,
267 struct ctdb_ltdb_header *header,
268 TDB_DATA *key, TDB_DATA *data)
270 if (r == NULL) {
271 r = (struct ctdb_rec_data *)&m->data[0];
272 } else {
273 r = (struct ctdb_rec_data *)(r->length + (uint8_t *)r);
276 if (reqid != NULL) {
277 *reqid = r->reqid;
280 if (key != NULL) {
281 key->dptr = &r->data[0];
282 key->dsize = r->keylen;
284 if (data != NULL) {
285 data->dptr = &r->data[r->keylen];
286 data->dsize = r->datalen;
287 if (header != NULL) {
288 data->dptr += sizeof(*header);
289 data->dsize -= sizeof(*header);
293 if (header != NULL) {
294 if (r->datalen < sizeof(*header)) {
295 return NULL;
297 *header = *(struct ctdb_ltdb_header *)&r->data[r->keylen];
300 return r;
304 * CTDB transaction destructor
306 static int db_ctdb_transaction_destructor(struct db_ctdb_transaction_handle *h)
308 NTSTATUS status;
310 status = g_lock_unlock(h->ctx->lock_ctx, h->lock_name);
311 if (!NT_STATUS_IS_OK(status)) {
312 DEBUG(0, ("g_lock_unlock failed: %s\n", nt_errstr(status)));
313 return -1;
315 return 0;
319 * CTDB dbwrap API: transaction_start function
320 * starts a transaction on a persistent database
322 static int db_ctdb_transaction_start(struct db_context *db)
324 struct db_ctdb_transaction_handle *h;
325 NTSTATUS status;
326 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
327 struct db_ctdb_ctx);
329 if (!db->persistent) {
330 DEBUG(0,("transactions not supported on non-persistent database 0x%08x\n",
331 ctx->db_id));
332 return -1;
335 if (ctx->transaction) {
336 ctx->transaction->nesting++;
337 return 0;
340 h = talloc_zero(db, struct db_ctdb_transaction_handle);
341 if (h == NULL) {
342 DEBUG(0,(__location__ " oom for transaction handle\n"));
343 return -1;
346 h->ctx = ctx;
348 h->lock_name = talloc_asprintf(h, "transaction_db_0x%08x",
349 (unsigned int)ctx->db_id);
350 if (h->lock_name == NULL) {
351 DEBUG(0, ("talloc_asprintf failed\n"));
352 TALLOC_FREE(h);
353 return -1;
357 * Wait a day, i.e. forever...
359 status = g_lock_lock(ctx->lock_ctx, h->lock_name, G_LOCK_WRITE,
360 timeval_set(86400, 0));
361 if (!NT_STATUS_IS_OK(status)) {
362 DEBUG(0, ("g_lock_lock failed: %s\n", nt_errstr(status)));
363 TALLOC_FREE(h);
364 return -1;
367 talloc_set_destructor(h, db_ctdb_transaction_destructor);
369 ctx->transaction = h;
371 DEBUG(5,(__location__ " Started transaction on db 0x%08x\n", ctx->db_id));
373 return 0;
376 static bool pull_newest_from_marshall_buffer(struct ctdb_marshall_buffer *buf,
377 TDB_DATA key,
378 struct ctdb_ltdb_header *pheader,
379 TALLOC_CTX *mem_ctx,
380 TDB_DATA *pdata)
382 struct ctdb_rec_data *rec = NULL;
383 struct ctdb_ltdb_header h;
384 bool found = false;
385 TDB_DATA data;
386 int i;
388 if (buf == NULL) {
389 return false;
392 ZERO_STRUCT(h);
393 ZERO_STRUCT(data);
396 * Walk the list of records written during this
397 * transaction. If we want to read one we have already
398 * written, return the last written sample. Thus we do not do
399 * a "break;" for the first hit, this record might have been
400 * overwritten later.
403 for (i=0; i<buf->count; i++) {
404 TDB_DATA tkey, tdata;
405 uint32_t reqid;
406 struct ctdb_ltdb_header hdr;
408 ZERO_STRUCT(hdr);
410 rec = db_ctdb_marshall_loop_next(buf, rec, &reqid, &hdr, &tkey,
411 &tdata);
412 if (rec == NULL) {
413 return false;
416 if (tdb_data_equal(key, tkey)) {
417 found = true;
418 data = tdata;
419 h = hdr;
423 if (!found) {
424 return false;
427 if (pdata != NULL) {
428 data.dptr = (uint8_t *)talloc_memdup(mem_ctx, data.dptr,
429 data.dsize);
430 if ((data.dsize != 0) && (data.dptr == NULL)) {
431 return false;
433 *pdata = data;
436 if (pheader != NULL) {
437 *pheader = h;
440 return true;
444 fetch a record inside a transaction
446 static int db_ctdb_transaction_fetch(struct db_ctdb_ctx *db,
447 TALLOC_CTX *mem_ctx,
448 TDB_DATA key, TDB_DATA *data)
450 struct db_ctdb_transaction_handle *h = db->transaction;
451 NTSTATUS status;
452 bool found;
454 found = pull_newest_from_marshall_buffer(h->m_write, key, NULL,
455 mem_ctx, data);
456 if (found) {
457 return 0;
460 status = db_ctdb_ltdb_fetch(h->ctx, key, NULL, mem_ctx, data);
462 if (NT_STATUS_EQUAL(status, NT_STATUS_NOT_FOUND)) {
463 *data = tdb_null;
464 } else if (!NT_STATUS_IS_OK(status)) {
465 return -1;
468 h->m_all = db_ctdb_marshall_add(h, h->m_all, h->ctx->db_id, 1, key,
469 NULL, *data);
470 if (h->m_all == NULL) {
471 DEBUG(0,(__location__ " Failed to add to marshalling "
472 "record\n"));
473 data->dsize = 0;
474 talloc_free(data->dptr);
475 return -1;
478 return 0;
482 static NTSTATUS db_ctdb_store_transaction(struct db_record *rec, TDB_DATA data, int flag);
483 static NTSTATUS db_ctdb_delete_transaction(struct db_record *rec);
485 static struct db_record *db_ctdb_fetch_locked_transaction(struct db_ctdb_ctx *ctx,
486 TALLOC_CTX *mem_ctx,
487 TDB_DATA key)
489 struct db_record *result;
490 TDB_DATA ctdb_data;
492 if (!(result = talloc(mem_ctx, struct db_record))) {
493 DEBUG(0, ("talloc failed\n"));
494 return NULL;
497 result->private_data = ctx->transaction;
499 result->key.dsize = key.dsize;
500 result->key.dptr = (uint8 *)talloc_memdup(result, key.dptr, key.dsize);
501 if (result->key.dptr == NULL) {
502 DEBUG(0, ("talloc failed\n"));
503 TALLOC_FREE(result);
504 return NULL;
507 result->store = db_ctdb_store_transaction;
508 result->delete_rec = db_ctdb_delete_transaction;
510 if (pull_newest_from_marshall_buffer(ctx->transaction->m_write, key,
511 NULL, result, &result->value)) {
512 return result;
515 ctdb_data = tdb_fetch(ctx->wtdb->tdb, key);
516 if (ctdb_data.dptr == NULL) {
517 /* create the record */
518 result->value = tdb_null;
519 return result;
522 result->value.dsize = ctdb_data.dsize - sizeof(struct ctdb_ltdb_header);
523 result->value.dptr = NULL;
525 if ((result->value.dsize != 0)
526 && !(result->value.dptr = (uint8 *)talloc_memdup(
527 result, ctdb_data.dptr + sizeof(struct ctdb_ltdb_header),
528 result->value.dsize))) {
529 DEBUG(0, ("talloc failed\n"));
530 TALLOC_FREE(result);
533 SAFE_FREE(ctdb_data.dptr);
535 return result;
538 static int db_ctdb_record_destructor(struct db_record **recp)
540 struct db_record *rec = talloc_get_type_abort(*recp, struct db_record);
541 struct db_ctdb_transaction_handle *h = talloc_get_type_abort(
542 rec->private_data, struct db_ctdb_transaction_handle);
543 int ret = h->ctx->db->transaction_commit(h->ctx->db);
544 if (ret != 0) {
545 DEBUG(0,(__location__ " transaction_commit failed\n"));
547 return 0;
551 auto-create a transaction for persistent databases
553 static struct db_record *db_ctdb_fetch_locked_persistent(struct db_ctdb_ctx *ctx,
554 TALLOC_CTX *mem_ctx,
555 TDB_DATA key)
557 int res;
558 struct db_record *rec, **recp;
560 res = db_ctdb_transaction_start(ctx->db);
561 if (res == -1) {
562 return NULL;
565 rec = db_ctdb_fetch_locked_transaction(ctx, mem_ctx, key);
566 if (rec == NULL) {
567 ctx->db->transaction_cancel(ctx->db);
568 return NULL;
571 /* destroy this transaction when we release the lock */
572 recp = talloc(rec, struct db_record *);
573 if (recp == NULL) {
574 ctx->db->transaction_cancel(ctx->db);
575 talloc_free(rec);
576 return NULL;
578 *recp = rec;
579 talloc_set_destructor(recp, db_ctdb_record_destructor);
580 return rec;
585 stores a record inside a transaction
587 static NTSTATUS db_ctdb_transaction_store(struct db_ctdb_transaction_handle *h,
588 TDB_DATA key, TDB_DATA data)
590 TALLOC_CTX *tmp_ctx = talloc_new(h);
591 TDB_DATA rec;
592 struct ctdb_ltdb_header header;
594 ZERO_STRUCT(header);
596 /* we need the header so we can update the RSN */
598 if (!pull_newest_from_marshall_buffer(h->m_write, key, &header,
599 NULL, NULL)) {
601 rec = tdb_fetch(h->ctx->wtdb->tdb, key);
603 if (rec.dptr != NULL) {
604 memcpy(&header, rec.dptr,
605 sizeof(struct ctdb_ltdb_header));
606 rec.dsize -= sizeof(struct ctdb_ltdb_header);
609 * a special case, we are writing the same
610 * data that is there now
612 if (data.dsize == rec.dsize &&
613 memcmp(data.dptr,
614 rec.dptr + sizeof(struct ctdb_ltdb_header),
615 data.dsize) == 0) {
616 SAFE_FREE(rec.dptr);
617 talloc_free(tmp_ctx);
618 return NT_STATUS_OK;
621 SAFE_FREE(rec.dptr);
624 header.dmaster = get_my_vnn();
625 header.rsn++;
627 h->m_all = db_ctdb_marshall_add(h, h->m_all, h->ctx->db_id, 0, key,
628 NULL, data);
629 if (h->m_all == NULL) {
630 DEBUG(0,(__location__ " Failed to add to marshalling "
631 "record\n"));
632 talloc_free(tmp_ctx);
633 return NT_STATUS_NO_MEMORY;
636 h->m_write = db_ctdb_marshall_add(h, h->m_write, h->ctx->db_id, 0, key, &header, data);
637 if (h->m_write == NULL) {
638 DEBUG(0,(__location__ " Failed to add to marshalling record\n"));
639 talloc_free(tmp_ctx);
640 return NT_STATUS_NO_MEMORY;
643 talloc_free(tmp_ctx);
644 return NT_STATUS_OK;
649 a record store inside a transaction
651 static NTSTATUS db_ctdb_store_transaction(struct db_record *rec, TDB_DATA data, int flag)
653 struct db_ctdb_transaction_handle *h = talloc_get_type_abort(
654 rec->private_data, struct db_ctdb_transaction_handle);
655 NTSTATUS status;
657 status = db_ctdb_transaction_store(h, rec->key, data);
658 return status;
662 a record delete inside a transaction
664 static NTSTATUS db_ctdb_delete_transaction(struct db_record *rec)
666 struct db_ctdb_transaction_handle *h = talloc_get_type_abort(
667 rec->private_data, struct db_ctdb_transaction_handle);
668 NTSTATUS status;
670 status = db_ctdb_transaction_store(h, rec->key, tdb_null);
671 return status;
675 * Fetch the db sequence number of a persistent db directly from the db.
677 static NTSTATUS db_ctdb_fetch_db_seqnum_from_db(struct db_ctdb_ctx *db,
678 uint64_t *seqnum)
680 NTSTATUS status;
681 const char *keyname = CTDB_DB_SEQNUM_KEY;
682 TDB_DATA key;
683 TDB_DATA data;
684 struct ctdb_ltdb_header header;
685 TALLOC_CTX *mem_ctx = talloc_stackframe();
687 if (seqnum == NULL) {
688 return NT_STATUS_INVALID_PARAMETER;
691 key.dptr = (uint8_t *)discard_const(keyname);
692 key.dsize = strlen(keyname) + 1;
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))
698 goto done;
701 status = NT_STATUS_OK;
703 if (data.dsize != sizeof(uint64_t)) {
704 *seqnum = 0;
705 goto done;
708 *seqnum = *(uint64_t *)data.dptr;
710 done:
711 TALLOC_FREE(mem_ctx);
712 return status;
716 * Store the database sequence number inside a transaction.
718 static NTSTATUS db_ctdb_store_db_seqnum(struct db_ctdb_transaction_handle *h,
719 uint64_t seqnum)
721 NTSTATUS status;
722 const char *keyname = CTDB_DB_SEQNUM_KEY;
723 TDB_DATA key;
724 TDB_DATA data;
726 key.dptr = (uint8_t *)discard_const(keyname);
727 key.dsize = strlen(keyname);
729 data.dptr = (uint8_t *)&seqnum;
730 data.dsize = sizeof(uint64_t);
732 status = db_ctdb_transaction_store(h, key, data);
734 return status;
738 commit a transaction
740 static int db_ctdb_transaction_commit(struct db_context *db)
742 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
743 struct db_ctdb_ctx);
744 NTSTATUS rets;
745 int status;
746 struct db_ctdb_transaction_handle *h = ctx->transaction;
747 uint64_t old_seqnum, new_seqnum;
748 int ret;
750 if (h == NULL) {
751 DEBUG(0,(__location__ " transaction commit with no open transaction on db 0x%08x\n", ctx->db_id));
752 return -1;
755 if (h->nested_cancel) {
756 db->transaction_cancel(db);
757 DEBUG(5,(__location__ " Failed transaction commit after nested cancel\n"));
758 return -1;
761 if (h->nesting != 0) {
762 h->nesting--;
763 return 0;
766 if (h->m_write == NULL) {
768 * No changes were made, so don't change the seqnum,
769 * don't push to other node, just exit with success.
771 ret = 0;
772 goto done;
775 DEBUG(5,(__location__ " Commit transaction on db 0x%08x\n", ctx->db_id));
778 * As the last db action before committing, bump the database sequence
779 * number. Note that this undoes all changes to the seqnum records
780 * performed under the transaction. This record is not meant to be
781 * modified by user interaction. It is for internal use only...
783 rets = db_ctdb_fetch_db_seqnum_from_db(ctx, &old_seqnum);
784 if (!NT_STATUS_IS_OK(rets)) {
785 DEBUG(1, (__location__ " failed to fetch the db sequence number "
786 "in transaction commit on db 0x%08x\n", ctx->db_id));
787 ret = -1;
788 goto done;
791 new_seqnum = old_seqnum + 1;
793 rets = db_ctdb_store_db_seqnum(h, new_seqnum);
794 if (!NT_STATUS_IS_OK(rets)) {
795 DEBUG(1, (__location__ "failed to store the db sequence number "
796 " in transaction commit on db 0x%08x\n", ctx->db_id));
797 ret = -1;
798 goto done;
801 again:
802 /* tell ctdbd to commit to the other nodes */
803 rets = ctdbd_control_local(messaging_ctdbd_connection(),
804 CTDB_CONTROL_TRANS3_COMMIT,
805 h->ctx->db_id, 0,
806 db_ctdb_marshall_finish(h->m_write),
807 NULL, NULL, &status);
808 if (!NT_STATUS_IS_OK(rets) || status != 0) {
810 * The TRANS3_COMMIT control should only possibly fail when a
811 * recovery has been running concurrently. In any case, the db
812 * will be the same on all nodes, either the new copy or the
813 * old copy. This can be detected by comparing the old and new
814 * local sequence numbers.
816 rets = db_ctdb_fetch_db_seqnum_from_db(ctx, &new_seqnum);
817 if (!NT_STATUS_IS_OK(rets)) {
818 DEBUG(1, (__location__ " failed to refetch db sequence "
819 "number after failed TRANS3_COMMIT\n"));
820 ret = -1;
821 goto done;
824 if (new_seqnum == old_seqnum) {
825 /* Recovery prevented all our changes: retry. */
826 goto again;
827 } else if (new_seqnum != (old_seqnum + 1)) {
828 DEBUG(0, (__location__ " ERROR: new_seqnum[%lu] != "
829 "old_seqnum[%lu] + (0 or 1) after failed "
830 "TRANS3_COMMIT - this should not happen!\n",
831 (unsigned long)new_seqnum,
832 (unsigned long)old_seqnum));
833 ret = -1;
834 goto done;
837 * Recovery propagated our changes to all nodes, completing
838 * our commit for us - succeed.
842 ret = 0;
844 done:
845 h->ctx->transaction = NULL;
846 talloc_free(h);
847 return ret;
852 cancel a transaction
854 static int db_ctdb_transaction_cancel(struct db_context *db)
856 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
857 struct db_ctdb_ctx);
858 struct db_ctdb_transaction_handle *h = ctx->transaction;
860 if (h == NULL) {
861 DEBUG(0,(__location__ " transaction cancel with no open transaction on db 0x%08x\n", ctx->db_id));
862 return -1;
865 if (h->nesting != 0) {
866 h->nesting--;
867 h->nested_cancel = true;
868 return 0;
871 DEBUG(5,(__location__ " Cancel transaction on db 0x%08x\n", ctx->db_id));
873 ctx->transaction = NULL;
874 talloc_free(h);
875 return 0;
879 static NTSTATUS db_ctdb_store(struct db_record *rec, TDB_DATA data, int flag)
881 struct db_ctdb_rec *crec = talloc_get_type_abort(
882 rec->private_data, struct db_ctdb_rec);
884 return db_ctdb_ltdb_store(crec->ctdb_ctx, rec->key, &(crec->header), data);
889 static NTSTATUS db_ctdb_delete(struct db_record *rec)
891 TDB_DATA data;
894 * We have to store the header with empty data. TODO: Fix the
895 * tdb-level cleanup
898 ZERO_STRUCT(data);
900 return db_ctdb_store(rec, data, 0);
904 static int db_ctdb_record_destr(struct db_record* data)
906 struct db_ctdb_rec *crec = talloc_get_type_abort(
907 data->private_data, struct db_ctdb_rec);
909 DEBUG(10, (DEBUGLEVEL > 10
910 ? "Unlocking db %u key %s\n"
911 : "Unlocking db %u key %.20s\n",
912 (int)crec->ctdb_ctx->db_id,
913 hex_encode_talloc(data, (unsigned char *)data->key.dptr,
914 data->key.dsize)));
916 if (tdb_chainunlock(crec->ctdb_ctx->wtdb->tdb, data->key) != 0) {
917 DEBUG(0, ("tdb_chainunlock failed\n"));
918 return -1;
921 return 0;
924 static struct db_record *fetch_locked_internal(struct db_ctdb_ctx *ctx,
925 TALLOC_CTX *mem_ctx,
926 TDB_DATA key,
927 bool persistent)
929 struct db_record *result;
930 struct db_ctdb_rec *crec;
931 NTSTATUS status;
932 TDB_DATA ctdb_data;
933 int migrate_attempts = 0;
935 if (!(result = talloc(mem_ctx, struct db_record))) {
936 DEBUG(0, ("talloc failed\n"));
937 return NULL;
940 if (!(crec = TALLOC_ZERO_P(result, struct db_ctdb_rec))) {
941 DEBUG(0, ("talloc failed\n"));
942 TALLOC_FREE(result);
943 return NULL;
946 result->private_data = (void *)crec;
947 crec->ctdb_ctx = ctx;
949 result->key.dsize = key.dsize;
950 result->key.dptr = (uint8 *)talloc_memdup(result, key.dptr, key.dsize);
951 if (result->key.dptr == NULL) {
952 DEBUG(0, ("talloc failed\n"));
953 TALLOC_FREE(result);
954 return NULL;
958 * Do a blocking lock on the record
960 again:
962 if (DEBUGLEVEL >= 10) {
963 char *keystr = hex_encode_talloc(result, key.dptr, key.dsize);
964 DEBUG(10, (DEBUGLEVEL > 10
965 ? "Locking db %u key %s\n"
966 : "Locking db %u key %.20s\n",
967 (int)crec->ctdb_ctx->db_id, keystr));
968 TALLOC_FREE(keystr);
971 if (tdb_chainlock(ctx->wtdb->tdb, key) != 0) {
972 DEBUG(3, ("tdb_chainlock failed\n"));
973 TALLOC_FREE(result);
974 return NULL;
977 result->store = db_ctdb_store;
978 result->delete_rec = db_ctdb_delete;
979 talloc_set_destructor(result, db_ctdb_record_destr);
981 ctdb_data = tdb_fetch(ctx->wtdb->tdb, key);
984 * See if we have a valid record and we are the dmaster. If so, we can
985 * take the shortcut and just return it.
988 if ((ctdb_data.dptr == NULL) ||
989 (ctdb_data.dsize < sizeof(struct ctdb_ltdb_header)) ||
990 ((struct ctdb_ltdb_header *)ctdb_data.dptr)->dmaster != get_my_vnn()
991 #if 0
992 || (random() % 2 != 0)
993 #endif
995 SAFE_FREE(ctdb_data.dptr);
996 tdb_chainunlock(ctx->wtdb->tdb, key);
997 talloc_set_destructor(result, NULL);
999 migrate_attempts += 1;
1001 DEBUG(10, ("ctdb_data.dptr = %p, dmaster = %u (%u)\n",
1002 ctdb_data.dptr, ctdb_data.dptr ?
1003 ((struct ctdb_ltdb_header *)ctdb_data.dptr)->dmaster : -1,
1004 get_my_vnn()));
1006 status = ctdbd_migrate(messaging_ctdbd_connection(),ctx->db_id, key);
1007 if (!NT_STATUS_IS_OK(status)) {
1008 DEBUG(5, ("ctdb_migrate failed: %s\n",
1009 nt_errstr(status)));
1010 TALLOC_FREE(result);
1011 return NULL;
1013 /* now its migrated, try again */
1014 goto again;
1017 if (migrate_attempts > 10) {
1018 DEBUG(0, ("db_ctdb_fetch_locked needed %d attempts\n",
1019 migrate_attempts));
1022 memcpy(&crec->header, ctdb_data.dptr, sizeof(crec->header));
1024 result->value.dsize = ctdb_data.dsize - sizeof(crec->header);
1025 result->value.dptr = NULL;
1027 if ((result->value.dsize != 0)
1028 && !(result->value.dptr = (uint8 *)talloc_memdup(
1029 result, ctdb_data.dptr + sizeof(crec->header),
1030 result->value.dsize))) {
1031 DEBUG(0, ("talloc failed\n"));
1032 TALLOC_FREE(result);
1035 SAFE_FREE(ctdb_data.dptr);
1037 return result;
1040 static struct db_record *db_ctdb_fetch_locked(struct db_context *db,
1041 TALLOC_CTX *mem_ctx,
1042 TDB_DATA key)
1044 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
1045 struct db_ctdb_ctx);
1047 if (ctx->transaction != NULL) {
1048 return db_ctdb_fetch_locked_transaction(ctx, mem_ctx, key);
1051 if (db->persistent) {
1052 return db_ctdb_fetch_locked_persistent(ctx, mem_ctx, key);
1055 return fetch_locked_internal(ctx, mem_ctx, key, db->persistent);
1059 fetch (unlocked, no migration) operation on ctdb
1061 static int db_ctdb_fetch(struct db_context *db, TALLOC_CTX *mem_ctx,
1062 TDB_DATA key, TDB_DATA *data)
1064 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
1065 struct db_ctdb_ctx);
1066 NTSTATUS status;
1067 TDB_DATA ctdb_data;
1069 if (ctx->transaction) {
1070 return db_ctdb_transaction_fetch(ctx, mem_ctx, key, data);
1073 /* try a direct fetch */
1074 ctdb_data = tdb_fetch(ctx->wtdb->tdb, key);
1077 * See if we have a valid record and we are the dmaster. If so, we can
1078 * take the shortcut and just return it.
1079 * we bypass the dmaster check for persistent databases
1081 if ((ctdb_data.dptr != NULL) &&
1082 (ctdb_data.dsize >= sizeof(struct ctdb_ltdb_header)) &&
1083 (db->persistent ||
1084 ((struct ctdb_ltdb_header *)ctdb_data.dptr)->dmaster == get_my_vnn())) {
1085 /* we are the dmaster - avoid the ctdb protocol op */
1087 data->dsize = ctdb_data.dsize - sizeof(struct ctdb_ltdb_header);
1088 if (data->dsize == 0) {
1089 SAFE_FREE(ctdb_data.dptr);
1090 data->dptr = NULL;
1091 return 0;
1094 data->dptr = (uint8 *)talloc_memdup(
1095 mem_ctx, ctdb_data.dptr+sizeof(struct ctdb_ltdb_header),
1096 data->dsize);
1098 SAFE_FREE(ctdb_data.dptr);
1100 if (data->dptr == NULL) {
1101 return -1;
1103 return 0;
1106 SAFE_FREE(ctdb_data.dptr);
1108 /* we weren't able to get it locally - ask ctdb to fetch it for us */
1109 status = ctdbd_fetch(messaging_ctdbd_connection(),ctx->db_id, key, mem_ctx, data);
1110 if (!NT_STATUS_IS_OK(status)) {
1111 DEBUG(5, ("ctdbd_fetch failed: %s\n", nt_errstr(status)));
1112 return -1;
1115 return 0;
1118 struct traverse_state {
1119 struct db_context *db;
1120 int (*fn)(struct db_record *rec, void *private_data);
1121 void *private_data;
1124 static void traverse_callback(TDB_DATA key, TDB_DATA data, void *private_data)
1126 struct traverse_state *state = (struct traverse_state *)private_data;
1127 struct db_record *rec;
1128 TALLOC_CTX *tmp_ctx = talloc_new(state->db);
1129 /* we have to give them a locked record to prevent races */
1130 rec = db_ctdb_fetch_locked(state->db, tmp_ctx, key);
1131 if (rec && rec->value.dsize > 0) {
1132 state->fn(rec, state->private_data);
1134 talloc_free(tmp_ctx);
1137 static int traverse_persistent_callback(TDB_CONTEXT *tdb, TDB_DATA kbuf, TDB_DATA dbuf,
1138 void *private_data)
1140 struct traverse_state *state = (struct traverse_state *)private_data;
1141 struct db_record *rec;
1142 TALLOC_CTX *tmp_ctx = talloc_new(state->db);
1143 int ret = 0;
1144 /* we have to give them a locked record to prevent races */
1145 rec = db_ctdb_fetch_locked(state->db, tmp_ctx, kbuf);
1146 if (rec && rec->value.dsize > 0) {
1147 ret = state->fn(rec, state->private_data);
1149 talloc_free(tmp_ctx);
1150 return ret;
1153 static int db_ctdb_traverse(struct db_context *db,
1154 int (*fn)(struct db_record *rec,
1155 void *private_data),
1156 void *private_data)
1158 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
1159 struct db_ctdb_ctx);
1160 struct traverse_state state;
1162 state.db = db;
1163 state.fn = fn;
1164 state.private_data = private_data;
1166 if (db->persistent) {
1167 /* for persistent databases we don't need to do a ctdb traverse,
1168 we can do a faster local traverse */
1169 return tdb_traverse(ctx->wtdb->tdb, traverse_persistent_callback, &state);
1173 ctdbd_traverse(ctx->db_id, traverse_callback, &state);
1174 return 0;
1177 static NTSTATUS db_ctdb_store_deny(struct db_record *rec, TDB_DATA data, int flag)
1179 return NT_STATUS_MEDIA_WRITE_PROTECTED;
1182 static NTSTATUS db_ctdb_delete_deny(struct db_record *rec)
1184 return NT_STATUS_MEDIA_WRITE_PROTECTED;
1187 static void traverse_read_callback(TDB_DATA key, TDB_DATA data, void *private_data)
1189 struct traverse_state *state = (struct traverse_state *)private_data;
1190 struct db_record rec;
1191 rec.key = key;
1192 rec.value = data;
1193 rec.store = db_ctdb_store_deny;
1194 rec.delete_rec = db_ctdb_delete_deny;
1195 rec.private_data = state->db;
1196 state->fn(&rec, state->private_data);
1199 static int traverse_persistent_callback_read(TDB_CONTEXT *tdb, TDB_DATA kbuf, TDB_DATA dbuf,
1200 void *private_data)
1202 struct traverse_state *state = (struct traverse_state *)private_data;
1203 struct db_record rec;
1204 rec.key = kbuf;
1205 rec.value = dbuf;
1206 rec.store = db_ctdb_store_deny;
1207 rec.delete_rec = db_ctdb_delete_deny;
1208 rec.private_data = state->db;
1210 if (rec.value.dsize <= sizeof(struct ctdb_ltdb_header)) {
1211 /* a deleted record */
1212 return 0;
1214 rec.value.dsize -= sizeof(struct ctdb_ltdb_header);
1215 rec.value.dptr += sizeof(struct ctdb_ltdb_header);
1217 return state->fn(&rec, state->private_data);
1220 static int db_ctdb_traverse_read(struct db_context *db,
1221 int (*fn)(struct db_record *rec,
1222 void *private_data),
1223 void *private_data)
1225 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
1226 struct db_ctdb_ctx);
1227 struct traverse_state state;
1229 state.db = db;
1230 state.fn = fn;
1231 state.private_data = private_data;
1233 if (db->persistent) {
1234 /* for persistent databases we don't need to do a ctdb traverse,
1235 we can do a faster local traverse */
1236 return tdb_traverse_read(ctx->wtdb->tdb, traverse_persistent_callback_read, &state);
1239 ctdbd_traverse(ctx->db_id, traverse_read_callback, &state);
1240 return 0;
1243 static int db_ctdb_get_seqnum(struct db_context *db)
1245 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
1246 struct db_ctdb_ctx);
1247 return tdb_get_seqnum(ctx->wtdb->tdb);
1250 static int db_ctdb_get_flags(struct db_context *db)
1252 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
1253 struct db_ctdb_ctx);
1254 return tdb_get_flags(ctx->wtdb->tdb);
1257 struct db_context *db_open_ctdb(TALLOC_CTX *mem_ctx,
1258 const char *name,
1259 int hash_size, int tdb_flags,
1260 int open_flags, mode_t mode)
1262 struct db_context *result;
1263 struct db_ctdb_ctx *db_ctdb;
1264 char *db_path;
1265 struct ctdbd_connection *conn;
1267 if (!lp_clustering()) {
1268 DEBUG(10, ("Clustering disabled -- no ctdb\n"));
1269 return NULL;
1272 if (!(result = TALLOC_ZERO_P(mem_ctx, struct db_context))) {
1273 DEBUG(0, ("talloc failed\n"));
1274 TALLOC_FREE(result);
1275 return NULL;
1278 if (!(db_ctdb = TALLOC_P(result, struct db_ctdb_ctx))) {
1279 DEBUG(0, ("talloc failed\n"));
1280 TALLOC_FREE(result);
1281 return NULL;
1284 db_ctdb->transaction = NULL;
1285 db_ctdb->db = result;
1287 conn = messaging_ctdbd_connection();
1289 if (!NT_STATUS_IS_OK(ctdbd_db_attach(conn, name, &db_ctdb->db_id, tdb_flags))) {
1290 DEBUG(0, ("ctdbd_db_attach failed for %s\n", name));
1291 TALLOC_FREE(result);
1292 return NULL;
1295 db_path = ctdbd_dbpath(conn, db_ctdb, db_ctdb->db_id);
1297 result->persistent = ((tdb_flags & TDB_CLEAR_IF_FIRST) == 0);
1299 /* only pass through specific flags */
1300 tdb_flags &= TDB_SEQNUM;
1302 /* honor permissions if user has specified O_CREAT */
1303 if (open_flags & O_CREAT) {
1304 chmod(db_path, mode);
1307 db_ctdb->wtdb = tdb_wrap_open(db_ctdb, db_path, hash_size, tdb_flags, O_RDWR, 0);
1308 if (db_ctdb->wtdb == NULL) {
1309 DEBUG(0, ("Could not open tdb %s: %s\n", db_path, strerror(errno)));
1310 TALLOC_FREE(result);
1311 return NULL;
1313 talloc_free(db_path);
1315 if (result->persistent) {
1316 db_ctdb->lock_ctx = g_lock_ctx_init(db_ctdb,
1317 ctdb_conn_msg_ctx(conn));
1318 if (db_ctdb->lock_ctx == NULL) {
1319 DEBUG(0, ("g_lock_ctx_init failed\n"));
1320 TALLOC_FREE(result);
1321 return NULL;
1325 result->private_data = (void *)db_ctdb;
1326 result->fetch_locked = db_ctdb_fetch_locked;
1327 result->fetch = db_ctdb_fetch;
1328 result->traverse = db_ctdb_traverse;
1329 result->traverse_read = db_ctdb_traverse_read;
1330 result->get_seqnum = db_ctdb_get_seqnum;
1331 result->get_flags = db_ctdb_get_flags;
1332 result->transaction_start = db_ctdb_transaction_start;
1333 result->transaction_commit = db_ctdb_transaction_commit;
1334 result->transaction_cancel = db_ctdb_transaction_cancel;
1336 DEBUG(3,("db_open_ctdb: opened database '%s' with dbid 0x%x\n",
1337 name, db_ctdb->db_id));
1339 return result;
1341 #endif