s3:dbwrap_ctdb: maintain a database sequence number that bumps in transactions
[Samba.git] / source3 / lib / dbwrap_ctdb.c
blobfb99e1d9cf52c5265d6856dd5bd682c02f279a4c
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;
385 TDB_DATA data;
386 int i;
388 if (buf == NULL) {
389 return false;
393 * Walk the list of records written during this
394 * transaction. If we want to read one we have already
395 * written, return the last written sample. Thus we do not do
396 * a "break;" for the first hit, this record might have been
397 * overwritten later.
400 for (i=0; i<buf->count; i++) {
401 TDB_DATA tkey, tdata;
402 uint32_t reqid;
404 rec = db_ctdb_marshall_loop_next(buf, rec, &reqid, &h, &tkey,
405 &tdata);
406 if (rec == NULL) {
407 return false;
410 if (tdb_data_equal(key, tkey)) {
411 found = true;
412 data = tdata;
416 if (!found) {
417 return false;
420 if (pdata != NULL) {
421 data.dptr = (uint8_t *)talloc_memdup(mem_ctx, data.dptr,
422 data.dsize);
423 if ((data.dsize != 0) && (data.dptr == NULL)) {
424 return false;
426 *pdata = data;
429 if (pheader != NULL) {
430 *pheader = h;
433 return true;
437 fetch a record inside a transaction
439 static int db_ctdb_transaction_fetch(struct db_ctdb_ctx *db,
440 TALLOC_CTX *mem_ctx,
441 TDB_DATA key, TDB_DATA *data)
443 struct db_ctdb_transaction_handle *h = db->transaction;
444 NTSTATUS status;
445 bool found;
447 found = pull_newest_from_marshall_buffer(h->m_write, key, NULL,
448 mem_ctx, data);
449 if (found) {
450 return 0;
453 status = db_ctdb_ltdb_fetch(h->ctx, key, NULL, mem_ctx, data);
455 if (NT_STATUS_EQUAL(status, NT_STATUS_NOT_FOUND)) {
456 *data = tdb_null;
457 } else if (!NT_STATUS_IS_OK(status)) {
458 return -1;
461 h->m_all = db_ctdb_marshall_add(h, h->m_all, h->ctx->db_id, 1, key,
462 NULL, *data);
463 if (h->m_all == NULL) {
464 DEBUG(0,(__location__ " Failed to add to marshalling "
465 "record\n"));
466 data->dsize = 0;
467 talloc_free(data->dptr);
468 return -1;
471 return 0;
475 static NTSTATUS db_ctdb_store_transaction(struct db_record *rec, TDB_DATA data, int flag);
476 static NTSTATUS db_ctdb_delete_transaction(struct db_record *rec);
478 static struct db_record *db_ctdb_fetch_locked_transaction(struct db_ctdb_ctx *ctx,
479 TALLOC_CTX *mem_ctx,
480 TDB_DATA key)
482 struct db_record *result;
483 TDB_DATA ctdb_data;
485 if (!(result = talloc(mem_ctx, struct db_record))) {
486 DEBUG(0, ("talloc failed\n"));
487 return NULL;
490 result->private_data = ctx->transaction;
492 result->key.dsize = key.dsize;
493 result->key.dptr = (uint8 *)talloc_memdup(result, key.dptr, key.dsize);
494 if (result->key.dptr == NULL) {
495 DEBUG(0, ("talloc failed\n"));
496 TALLOC_FREE(result);
497 return NULL;
500 result->store = db_ctdb_store_transaction;
501 result->delete_rec = db_ctdb_delete_transaction;
503 if (pull_newest_from_marshall_buffer(ctx->transaction->m_write, key,
504 NULL, result, &result->value)) {
505 return result;
508 ctdb_data = tdb_fetch(ctx->wtdb->tdb, key);
509 if (ctdb_data.dptr == NULL) {
510 /* create the record */
511 result->value = tdb_null;
512 return result;
515 result->value.dsize = ctdb_data.dsize - sizeof(struct ctdb_ltdb_header);
516 result->value.dptr = NULL;
518 if ((result->value.dsize != 0)
519 && !(result->value.dptr = (uint8 *)talloc_memdup(
520 result, ctdb_data.dptr + sizeof(struct ctdb_ltdb_header),
521 result->value.dsize))) {
522 DEBUG(0, ("talloc failed\n"));
523 TALLOC_FREE(result);
526 SAFE_FREE(ctdb_data.dptr);
528 return result;
531 static int db_ctdb_record_destructor(struct db_record **recp)
533 struct db_record *rec = talloc_get_type_abort(*recp, struct db_record);
534 struct db_ctdb_transaction_handle *h = talloc_get_type_abort(
535 rec->private_data, struct db_ctdb_transaction_handle);
536 int ret = h->ctx->db->transaction_commit(h->ctx->db);
537 if (ret != 0) {
538 DEBUG(0,(__location__ " transaction_commit failed\n"));
540 return 0;
544 auto-create a transaction for persistent databases
546 static struct db_record *db_ctdb_fetch_locked_persistent(struct db_ctdb_ctx *ctx,
547 TALLOC_CTX *mem_ctx,
548 TDB_DATA key)
550 int res;
551 struct db_record *rec, **recp;
553 res = db_ctdb_transaction_start(ctx->db);
554 if (res == -1) {
555 return NULL;
558 rec = db_ctdb_fetch_locked_transaction(ctx, mem_ctx, key);
559 if (rec == NULL) {
560 ctx->db->transaction_cancel(ctx->db);
561 return NULL;
564 /* destroy this transaction when we release the lock */
565 recp = talloc(rec, struct db_record *);
566 if (recp == NULL) {
567 ctx->db->transaction_cancel(ctx->db);
568 talloc_free(rec);
569 return NULL;
571 *recp = rec;
572 talloc_set_destructor(recp, db_ctdb_record_destructor);
573 return rec;
578 stores a record inside a transaction
580 static NTSTATUS db_ctdb_transaction_store(struct db_ctdb_transaction_handle *h,
581 TDB_DATA key, TDB_DATA data)
583 TALLOC_CTX *tmp_ctx = talloc_new(h);
584 TDB_DATA rec;
585 struct ctdb_ltdb_header header;
587 ZERO_STRUCT(header);
589 /* we need the header so we can update the RSN */
591 if (!pull_newest_from_marshall_buffer(h->m_write, key, &header,
592 NULL, NULL)) {
594 rec = tdb_fetch(h->ctx->wtdb->tdb, key);
596 if (rec.dptr != NULL) {
597 memcpy(&header, rec.dptr,
598 sizeof(struct ctdb_ltdb_header));
599 rec.dsize -= sizeof(struct ctdb_ltdb_header);
602 * a special case, we are writing the same
603 * data that is there now
605 if (data.dsize == rec.dsize &&
606 memcmp(data.dptr,
607 rec.dptr + sizeof(struct ctdb_ltdb_header),
608 data.dsize) == 0) {
609 SAFE_FREE(rec.dptr);
610 talloc_free(tmp_ctx);
611 return NT_STATUS_OK;
614 SAFE_FREE(rec.dptr);
617 header.dmaster = get_my_vnn();
618 header.rsn++;
620 h->m_all = db_ctdb_marshall_add(h, h->m_all, h->ctx->db_id, 0, key,
621 NULL, data);
622 if (h->m_all == NULL) {
623 DEBUG(0,(__location__ " Failed to add to marshalling "
624 "record\n"));
625 talloc_free(tmp_ctx);
626 return NT_STATUS_NO_MEMORY;
629 h->m_write = db_ctdb_marshall_add(h, h->m_write, h->ctx->db_id, 0, key, &header, data);
630 if (h->m_write == NULL) {
631 DEBUG(0,(__location__ " Failed to add to marshalling record\n"));
632 talloc_free(tmp_ctx);
633 return NT_STATUS_NO_MEMORY;
636 talloc_free(tmp_ctx);
637 return NT_STATUS_OK;
642 a record store inside a transaction
644 static NTSTATUS db_ctdb_store_transaction(struct db_record *rec, TDB_DATA data, int flag)
646 struct db_ctdb_transaction_handle *h = talloc_get_type_abort(
647 rec->private_data, struct db_ctdb_transaction_handle);
648 NTSTATUS status;
650 status = db_ctdb_transaction_store(h, rec->key, data);
651 return status;
655 a record delete inside a transaction
657 static NTSTATUS db_ctdb_delete_transaction(struct db_record *rec)
659 struct db_ctdb_transaction_handle *h = talloc_get_type_abort(
660 rec->private_data, struct db_ctdb_transaction_handle);
661 NTSTATUS status;
663 status = db_ctdb_transaction_store(h, rec->key, tdb_null);
664 return status;
668 * Fetch the db sequence number of a persistent db directly from the db.
670 static NTSTATUS db_ctdb_fetch_db_seqnum_from_db(struct db_ctdb_ctx *db,
671 uint64_t *seqnum)
673 NTSTATUS status;
674 const char *keyname = CTDB_DB_SEQNUM_KEY;
675 TDB_DATA key;
676 TDB_DATA data;
677 struct ctdb_ltdb_header header;
678 TALLOC_CTX *mem_ctx = talloc_stackframe();
680 if (seqnum == NULL) {
681 return NT_STATUS_INVALID_PARAMETER;
684 key.dptr = (uint8_t *)discard_const(keyname);
685 key.dsize = strlen(keyname) + 1;
687 status = db_ctdb_ltdb_fetch(db, key, &header, mem_ctx, &data);
688 if (!NT_STATUS_IS_OK(status)) {
689 goto done;
692 if (data.dsize != sizeof(uint64_t)) {
693 *seqnum = 0;
694 goto done;
697 *seqnum = *(uint64_t *)data.dptr;
699 done:
700 TALLOC_FREE(mem_ctx);
701 return status;
705 * Store the database sequence number inside a transaction.
707 static NTSTATUS db_ctdb_store_db_seqnum(struct db_ctdb_transaction_handle *h,
708 uint64_t seqnum)
710 NTSTATUS status;
711 const char *keyname = CTDB_DB_SEQNUM_KEY;
712 TDB_DATA key;
713 TDB_DATA data;
715 key.dptr = (uint8_t *)discard_const(keyname);
716 key.dsize = strlen(keyname);
718 data.dptr = (uint8_t *)&seqnum;
719 data.dsize = sizeof(uint64_t);
721 status = db_ctdb_transaction_store(h, key, data);
723 return status;
727 commit a transaction
729 static int db_ctdb_transaction_commit(struct db_context *db)
731 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
732 struct db_ctdb_ctx);
733 NTSTATUS rets;
734 int status;
735 struct db_ctdb_transaction_handle *h = ctx->transaction;
736 uint64_t old_seqnum, new_seqnum;
737 int ret;
739 if (h == NULL) {
740 DEBUG(0,(__location__ " transaction commit with no open transaction on db 0x%08x\n", ctx->db_id));
741 return -1;
744 if (h->nested_cancel) {
745 db->transaction_cancel(db);
746 DEBUG(5,(__location__ " Failed transaction commit after nested cancel\n"));
747 return -1;
750 if (h->nesting != 0) {
751 h->nesting--;
752 return 0;
755 DEBUG(5,(__location__ " Commit transaction on db 0x%08x\n", ctx->db_id));
758 * As the last db action before committing, bump the database sequence
759 * number. Note that this undoes all changes to the seqnum records
760 * performed under the transaction. This record is not meant to be
761 * modified by user interaction. It is for internal use only...
763 rets = db_ctdb_fetch_db_seqnum_from_db(ctx, &old_seqnum);
764 if (!NT_STATUS_IS_OK(rets)) {
765 DEBUG(1, (__location__ " failed to fetch the db sequence number "
766 "in transaction commit on db 0x%08x\n", ctx->db_id));
767 ret = -1;
768 goto done;
771 new_seqnum = old_seqnum + 1;
773 rets = db_ctdb_store_db_seqnum(h, new_seqnum);
774 if (!NT_STATUS_IS_OK(rets)) {
775 DEBUG(1, (__location__ "failed to store the db sequence number "
776 " in transaction commit on db 0x%08x\n", ctx->db_id));
777 ret = -1;
778 goto done;
781 again:
782 if (h->m_write == NULL) {
783 /* no changes were made, potentially after a retry */
784 goto done;
787 /* tell ctdbd to commit to the other nodes */
788 rets = ctdbd_control_local(messaging_ctdbd_connection(),
789 CTDB_CONTROL_TRANS3_COMMIT,
790 h->ctx->db_id, 0,
791 db_ctdb_marshall_finish(h->m_write),
792 NULL, NULL, &status);
793 if (!NT_STATUS_IS_OK(rets) || status != 0) {
795 * The TRANS3_COMMIT control should only possibly fail when a
796 * recovery has been running concurrently. In any case, the db
797 * will be the same on all nodes, either the new copy or the
798 * old copy. This can be detected by comparing the old and new
799 * local sequence numbers.
801 rets = db_ctdb_fetch_db_seqnum_from_db(ctx, &new_seqnum);
802 if (!NT_STATUS_IS_OK(rets)) {
803 DEBUG(1, (__location__ " failed to refetch db sequence "
804 "number after failed TRANS3_COMMIT\n"));
805 ret = -1;
806 goto done;
809 if (new_seqnum == old_seqnum) {
810 /* Recovery prevented all our changes: retry. */
811 goto again;
812 } else if (new_seqnum != (old_seqnum + 1)) {
813 DEBUG(0, (__location__ " ERROR: new_seqnum[%lu] != "
814 "old_seqnum[%lu] + (0 or 1) after failed "
815 "TRANS3_COMMIT - this should not happen!\n",
816 (unsigned long)new_seqnum,
817 (unsigned long)old_seqnum));
818 ret = -1;
819 goto done;
822 * Recovery propagated our changes to all nodes, completing
823 * our commit for us - succeed.
827 ret = 0;
829 done:
830 h->ctx->transaction = NULL;
831 talloc_free(h);
832 return 0;
837 cancel a transaction
839 static int db_ctdb_transaction_cancel(struct db_context *db)
841 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
842 struct db_ctdb_ctx);
843 struct db_ctdb_transaction_handle *h = ctx->transaction;
845 if (h == NULL) {
846 DEBUG(0,(__location__ " transaction cancel with no open transaction on db 0x%08x\n", ctx->db_id));
847 return -1;
850 if (h->nesting != 0) {
851 h->nesting--;
852 h->nested_cancel = true;
853 return 0;
856 DEBUG(5,(__location__ " Cancel transaction on db 0x%08x\n", ctx->db_id));
858 ctx->transaction = NULL;
859 talloc_free(h);
860 return 0;
864 static NTSTATUS db_ctdb_store(struct db_record *rec, TDB_DATA data, int flag)
866 struct db_ctdb_rec *crec = talloc_get_type_abort(
867 rec->private_data, struct db_ctdb_rec);
869 return db_ctdb_ltdb_store(crec->ctdb_ctx, rec->key, &(crec->header), data);
874 static NTSTATUS db_ctdb_delete(struct db_record *rec)
876 TDB_DATA data;
879 * We have to store the header with empty data. TODO: Fix the
880 * tdb-level cleanup
883 ZERO_STRUCT(data);
885 return db_ctdb_store(rec, data, 0);
889 static int db_ctdb_record_destr(struct db_record* data)
891 struct db_ctdb_rec *crec = talloc_get_type_abort(
892 data->private_data, struct db_ctdb_rec);
894 DEBUG(10, (DEBUGLEVEL > 10
895 ? "Unlocking db %u key %s\n"
896 : "Unlocking db %u key %.20s\n",
897 (int)crec->ctdb_ctx->db_id,
898 hex_encode_talloc(data, (unsigned char *)data->key.dptr,
899 data->key.dsize)));
901 if (tdb_chainunlock(crec->ctdb_ctx->wtdb->tdb, data->key) != 0) {
902 DEBUG(0, ("tdb_chainunlock failed\n"));
903 return -1;
906 return 0;
909 static struct db_record *fetch_locked_internal(struct db_ctdb_ctx *ctx,
910 TALLOC_CTX *mem_ctx,
911 TDB_DATA key,
912 bool persistent)
914 struct db_record *result;
915 struct db_ctdb_rec *crec;
916 NTSTATUS status;
917 TDB_DATA ctdb_data;
918 int migrate_attempts = 0;
920 if (!(result = talloc(mem_ctx, struct db_record))) {
921 DEBUG(0, ("talloc failed\n"));
922 return NULL;
925 if (!(crec = TALLOC_ZERO_P(result, struct db_ctdb_rec))) {
926 DEBUG(0, ("talloc failed\n"));
927 TALLOC_FREE(result);
928 return NULL;
931 result->private_data = (void *)crec;
932 crec->ctdb_ctx = ctx;
934 result->key.dsize = key.dsize;
935 result->key.dptr = (uint8 *)talloc_memdup(result, key.dptr, key.dsize);
936 if (result->key.dptr == NULL) {
937 DEBUG(0, ("talloc failed\n"));
938 TALLOC_FREE(result);
939 return NULL;
943 * Do a blocking lock on the record
945 again:
947 if (DEBUGLEVEL >= 10) {
948 char *keystr = hex_encode_talloc(result, key.dptr, key.dsize);
949 DEBUG(10, (DEBUGLEVEL > 10
950 ? "Locking db %u key %s\n"
951 : "Locking db %u key %.20s\n",
952 (int)crec->ctdb_ctx->db_id, keystr));
953 TALLOC_FREE(keystr);
956 if (tdb_chainlock(ctx->wtdb->tdb, key) != 0) {
957 DEBUG(3, ("tdb_chainlock failed\n"));
958 TALLOC_FREE(result);
959 return NULL;
962 result->store = db_ctdb_store;
963 result->delete_rec = db_ctdb_delete;
964 talloc_set_destructor(result, db_ctdb_record_destr);
966 ctdb_data = tdb_fetch(ctx->wtdb->tdb, key);
969 * See if we have a valid record and we are the dmaster. If so, we can
970 * take the shortcut and just return it.
973 if ((ctdb_data.dptr == NULL) ||
974 (ctdb_data.dsize < sizeof(struct ctdb_ltdb_header)) ||
975 ((struct ctdb_ltdb_header *)ctdb_data.dptr)->dmaster != get_my_vnn()
976 #if 0
977 || (random() % 2 != 0)
978 #endif
980 SAFE_FREE(ctdb_data.dptr);
981 tdb_chainunlock(ctx->wtdb->tdb, key);
982 talloc_set_destructor(result, NULL);
984 migrate_attempts += 1;
986 DEBUG(10, ("ctdb_data.dptr = %p, dmaster = %u (%u)\n",
987 ctdb_data.dptr, ctdb_data.dptr ?
988 ((struct ctdb_ltdb_header *)ctdb_data.dptr)->dmaster : -1,
989 get_my_vnn()));
991 status = ctdbd_migrate(messaging_ctdbd_connection(),ctx->db_id, key);
992 if (!NT_STATUS_IS_OK(status)) {
993 DEBUG(5, ("ctdb_migrate failed: %s\n",
994 nt_errstr(status)));
995 TALLOC_FREE(result);
996 return NULL;
998 /* now its migrated, try again */
999 goto again;
1002 if (migrate_attempts > 10) {
1003 DEBUG(0, ("db_ctdb_fetch_locked needed %d attempts\n",
1004 migrate_attempts));
1007 memcpy(&crec->header, ctdb_data.dptr, sizeof(crec->header));
1009 result->value.dsize = ctdb_data.dsize - sizeof(crec->header);
1010 result->value.dptr = NULL;
1012 if ((result->value.dsize != 0)
1013 && !(result->value.dptr = (uint8 *)talloc_memdup(
1014 result, ctdb_data.dptr + sizeof(crec->header),
1015 result->value.dsize))) {
1016 DEBUG(0, ("talloc failed\n"));
1017 TALLOC_FREE(result);
1020 SAFE_FREE(ctdb_data.dptr);
1022 return result;
1025 static struct db_record *db_ctdb_fetch_locked(struct db_context *db,
1026 TALLOC_CTX *mem_ctx,
1027 TDB_DATA key)
1029 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
1030 struct db_ctdb_ctx);
1032 if (ctx->transaction != NULL) {
1033 return db_ctdb_fetch_locked_transaction(ctx, mem_ctx, key);
1036 if (db->persistent) {
1037 return db_ctdb_fetch_locked_persistent(ctx, mem_ctx, key);
1040 return fetch_locked_internal(ctx, mem_ctx, key, db->persistent);
1044 fetch (unlocked, no migration) operation on ctdb
1046 static int db_ctdb_fetch(struct db_context *db, TALLOC_CTX *mem_ctx,
1047 TDB_DATA key, TDB_DATA *data)
1049 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
1050 struct db_ctdb_ctx);
1051 NTSTATUS status;
1052 TDB_DATA ctdb_data;
1054 if (ctx->transaction) {
1055 return db_ctdb_transaction_fetch(ctx, mem_ctx, key, data);
1058 /* try a direct fetch */
1059 ctdb_data = tdb_fetch(ctx->wtdb->tdb, key);
1062 * See if we have a valid record and we are the dmaster. If so, we can
1063 * take the shortcut and just return it.
1064 * we bypass the dmaster check for persistent databases
1066 if ((ctdb_data.dptr != NULL) &&
1067 (ctdb_data.dsize >= sizeof(struct ctdb_ltdb_header)) &&
1068 (db->persistent ||
1069 ((struct ctdb_ltdb_header *)ctdb_data.dptr)->dmaster == get_my_vnn())) {
1070 /* we are the dmaster - avoid the ctdb protocol op */
1072 data->dsize = ctdb_data.dsize - sizeof(struct ctdb_ltdb_header);
1073 if (data->dsize == 0) {
1074 SAFE_FREE(ctdb_data.dptr);
1075 data->dptr = NULL;
1076 return 0;
1079 data->dptr = (uint8 *)talloc_memdup(
1080 mem_ctx, ctdb_data.dptr+sizeof(struct ctdb_ltdb_header),
1081 data->dsize);
1083 SAFE_FREE(ctdb_data.dptr);
1085 if (data->dptr == NULL) {
1086 return -1;
1088 return 0;
1091 SAFE_FREE(ctdb_data.dptr);
1093 /* we weren't able to get it locally - ask ctdb to fetch it for us */
1094 status = ctdbd_fetch(messaging_ctdbd_connection(),ctx->db_id, key, mem_ctx, data);
1095 if (!NT_STATUS_IS_OK(status)) {
1096 DEBUG(5, ("ctdbd_fetch failed: %s\n", nt_errstr(status)));
1097 return -1;
1100 return 0;
1103 struct traverse_state {
1104 struct db_context *db;
1105 int (*fn)(struct db_record *rec, void *private_data);
1106 void *private_data;
1109 static void traverse_callback(TDB_DATA key, TDB_DATA data, void *private_data)
1111 struct traverse_state *state = (struct traverse_state *)private_data;
1112 struct db_record *rec;
1113 TALLOC_CTX *tmp_ctx = talloc_new(state->db);
1114 /* we have to give them a locked record to prevent races */
1115 rec = db_ctdb_fetch_locked(state->db, tmp_ctx, key);
1116 if (rec && rec->value.dsize > 0) {
1117 state->fn(rec, state->private_data);
1119 talloc_free(tmp_ctx);
1122 static int traverse_persistent_callback(TDB_CONTEXT *tdb, TDB_DATA kbuf, TDB_DATA dbuf,
1123 void *private_data)
1125 struct traverse_state *state = (struct traverse_state *)private_data;
1126 struct db_record *rec;
1127 TALLOC_CTX *tmp_ctx = talloc_new(state->db);
1128 int ret = 0;
1129 /* we have to give them a locked record to prevent races */
1130 rec = db_ctdb_fetch_locked(state->db, tmp_ctx, kbuf);
1131 if (rec && rec->value.dsize > 0) {
1132 ret = state->fn(rec, state->private_data);
1134 talloc_free(tmp_ctx);
1135 return ret;
1138 static int db_ctdb_traverse(struct db_context *db,
1139 int (*fn)(struct db_record *rec,
1140 void *private_data),
1141 void *private_data)
1143 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
1144 struct db_ctdb_ctx);
1145 struct traverse_state state;
1147 state.db = db;
1148 state.fn = fn;
1149 state.private_data = private_data;
1151 if (db->persistent) {
1152 /* for persistent databases we don't need to do a ctdb traverse,
1153 we can do a faster local traverse */
1154 return tdb_traverse(ctx->wtdb->tdb, traverse_persistent_callback, &state);
1158 ctdbd_traverse(ctx->db_id, traverse_callback, &state);
1159 return 0;
1162 static NTSTATUS db_ctdb_store_deny(struct db_record *rec, TDB_DATA data, int flag)
1164 return NT_STATUS_MEDIA_WRITE_PROTECTED;
1167 static NTSTATUS db_ctdb_delete_deny(struct db_record *rec)
1169 return NT_STATUS_MEDIA_WRITE_PROTECTED;
1172 static void traverse_read_callback(TDB_DATA key, TDB_DATA data, void *private_data)
1174 struct traverse_state *state = (struct traverse_state *)private_data;
1175 struct db_record rec;
1176 rec.key = key;
1177 rec.value = data;
1178 rec.store = db_ctdb_store_deny;
1179 rec.delete_rec = db_ctdb_delete_deny;
1180 rec.private_data = state->db;
1181 state->fn(&rec, state->private_data);
1184 static int traverse_persistent_callback_read(TDB_CONTEXT *tdb, TDB_DATA kbuf, TDB_DATA dbuf,
1185 void *private_data)
1187 struct traverse_state *state = (struct traverse_state *)private_data;
1188 struct db_record rec;
1189 rec.key = kbuf;
1190 rec.value = dbuf;
1191 rec.store = db_ctdb_store_deny;
1192 rec.delete_rec = db_ctdb_delete_deny;
1193 rec.private_data = state->db;
1195 if (rec.value.dsize <= sizeof(struct ctdb_ltdb_header)) {
1196 /* a deleted record */
1197 return 0;
1199 rec.value.dsize -= sizeof(struct ctdb_ltdb_header);
1200 rec.value.dptr += sizeof(struct ctdb_ltdb_header);
1202 return state->fn(&rec, state->private_data);
1205 static int db_ctdb_traverse_read(struct db_context *db,
1206 int (*fn)(struct db_record *rec,
1207 void *private_data),
1208 void *private_data)
1210 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
1211 struct db_ctdb_ctx);
1212 struct traverse_state state;
1214 state.db = db;
1215 state.fn = fn;
1216 state.private_data = private_data;
1218 if (db->persistent) {
1219 /* for persistent databases we don't need to do a ctdb traverse,
1220 we can do a faster local traverse */
1221 return tdb_traverse_read(ctx->wtdb->tdb, traverse_persistent_callback_read, &state);
1224 ctdbd_traverse(ctx->db_id, traverse_read_callback, &state);
1225 return 0;
1228 static int db_ctdb_get_seqnum(struct db_context *db)
1230 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
1231 struct db_ctdb_ctx);
1232 return tdb_get_seqnum(ctx->wtdb->tdb);
1235 static int db_ctdb_get_flags(struct db_context *db)
1237 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
1238 struct db_ctdb_ctx);
1239 return tdb_get_flags(ctx->wtdb->tdb);
1242 struct db_context *db_open_ctdb(TALLOC_CTX *mem_ctx,
1243 const char *name,
1244 int hash_size, int tdb_flags,
1245 int open_flags, mode_t mode)
1247 struct db_context *result;
1248 struct db_ctdb_ctx *db_ctdb;
1249 char *db_path;
1250 struct ctdbd_connection *conn;
1252 if (!lp_clustering()) {
1253 DEBUG(10, ("Clustering disabled -- no ctdb\n"));
1254 return NULL;
1257 if (!(result = TALLOC_ZERO_P(mem_ctx, struct db_context))) {
1258 DEBUG(0, ("talloc failed\n"));
1259 TALLOC_FREE(result);
1260 return NULL;
1263 if (!(db_ctdb = TALLOC_P(result, struct db_ctdb_ctx))) {
1264 DEBUG(0, ("talloc failed\n"));
1265 TALLOC_FREE(result);
1266 return NULL;
1269 db_ctdb->transaction = NULL;
1270 db_ctdb->db = result;
1272 conn = messaging_ctdbd_connection();
1274 if (!NT_STATUS_IS_OK(ctdbd_db_attach(conn, name, &db_ctdb->db_id, tdb_flags))) {
1275 DEBUG(0, ("ctdbd_db_attach failed for %s\n", name));
1276 TALLOC_FREE(result);
1277 return NULL;
1280 db_path = ctdbd_dbpath(conn, db_ctdb, db_ctdb->db_id);
1282 result->persistent = ((tdb_flags & TDB_CLEAR_IF_FIRST) == 0);
1284 /* only pass through specific flags */
1285 tdb_flags &= TDB_SEQNUM;
1287 /* honor permissions if user has specified O_CREAT */
1288 if (open_flags & O_CREAT) {
1289 chmod(db_path, mode);
1292 db_ctdb->wtdb = tdb_wrap_open(db_ctdb, db_path, hash_size, tdb_flags, O_RDWR, 0);
1293 if (db_ctdb->wtdb == NULL) {
1294 DEBUG(0, ("Could not open tdb %s: %s\n", db_path, strerror(errno)));
1295 TALLOC_FREE(result);
1296 return NULL;
1298 talloc_free(db_path);
1300 if (result->persistent) {
1301 db_ctdb->lock_ctx = g_lock_ctx_init(db_ctdb,
1302 ctdb_conn_msg_ctx(conn));
1303 if (db_ctdb->lock_ctx == NULL) {
1304 DEBUG(0, ("g_lock_ctx_init failed\n"));
1305 TALLOC_FREE(result);
1306 return NULL;
1310 result->private_data = (void *)db_ctdb;
1311 result->fetch_locked = db_ctdb_fetch_locked;
1312 result->fetch = db_ctdb_fetch;
1313 result->traverse = db_ctdb_traverse;
1314 result->traverse_read = db_ctdb_traverse_read;
1315 result->get_seqnum = db_ctdb_get_seqnum;
1316 result->get_flags = db_ctdb_get_flags;
1317 result->transaction_start = db_ctdb_transaction_start;
1318 result->transaction_commit = db_ctdb_transaction_commit;
1319 result->transaction_cancel = db_ctdb_transaction_cancel;
1321 DEBUG(3,("db_open_ctdb: opened database '%s' with dbid 0x%x\n",
1322 name, db_ctdb->db_id));
1324 return result;
1326 #endif