?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
s3:dbwrap_ctdb: fix db_ctdb_fetch_db_seqnum_from_db() when NT_STATUS_NOT_FOUND.
[Samba.git] / source3 / lib / dbwrap_ctdb.c
blob4c4486c53971f775dc5a0d46f93e8765f1252b41
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
6
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.
11
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.
16
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/>.
19 */
20
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"
27
28 struct db_ctdb_transaction_handle {
29 struct db_ctdb_ctx *ctx;
30 /*
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)
34 */
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;
40 };
41
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;
48 };
49
50 struct db_ctdb_rec {
51 struct db_ctdb_ctx *ctdb_ctx;
52 struct ctdb_ltdb_header header;
53 };
54
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);
59
60 static NTSTATUS tdb_error_to_ntstatus(struct tdb_context *tdb)
61 {
62 NTSTATUS status;
63 enum TDB_ERROR tret = tdb_error(tdb);
64
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;
75 }
76
77 return status;
78 }
79
80
81 /**
82 * fetch a record from the tdb, separating out the header
83 * information and returning the body of the record.
84 */
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)
90 {
91 TDB_DATA rec;
92 NTSTATUS status;
93
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);
99 }
100 if (header) {
101 header->dmaster = (uint32_t)-1;
102 header->rsn = 0;
103 }
104 goto done;
105 }
106
107 if (header) {
108 *header = *(struct ctdb_ltdb_header *)rec.dptr;
109 }
110
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;
123 }
124 }
125 }
126
127 status = NT_STATUS_OK;
128
129 done:
130 SAFE_FREE(rec.dptr);
131 return status;
132 }
133
134 /*
135 * Store a record together with the ctdb record header
136 * in the local copy of the database.
137 */
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)
142 {
143 TALLOC_CTX *tmp_ctx = talloc_stackframe();
144 TDB_DATA rec;
145 int ret;
146
147 rec.dsize = data.dsize + sizeof(struct ctdb_ltdb_header);
148 rec.dptr = (uint8_t *)talloc_size(tmp_ctx, rec.dsize);
149
150 if (rec.dptr == NULL) {
151 talloc_free(tmp_ctx);
152 return NT_STATUS_NO_MEMORY;
153 }
154
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);
157
158 ret = tdb_store(db->wtdb->tdb, key, rec, TDB_REPLACE);
159
160 talloc_free(tmp_ctx);
161
162 return (ret == 0) ? NT_STATUS_OK
163 : tdb_error_to_ntstatus(db->wtdb->tdb);
164
165 }
166
167 /*
168 form a ctdb_rec_data record from a key/data pair
169
170 note that header may be NULL. If not NULL then it is included in the data portion
171 of the record
172 */
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)
177 {
178 size_t length;
179 struct ctdb_rec_data *d;
180
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;
186 }
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);
198 }
199 return d;
200 }
201
202
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)
211 {
212 struct ctdb_rec_data *r;
213 size_t m_size, r_size;
214 struct ctdb_marshall_buffer *m2 = NULL;
215
216 r = db_ctdb_marshall_record(talloc_tos(), reqid, key, header, data);
217 if (r == NULL) {
218 talloc_free(m);
219 return NULL;
220 }
221
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;
227 }
228 m->db_id = db_id;
229 }
230
231 m_size = talloc_get_size(m);
232 r_size = talloc_get_size(r);
233
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;
239 }
240
241 memcpy(m_size + (uint8_t *)m2, r, r_size);
242
243 m2->count++;
244
245 done:
246 talloc_free(r);
247 return m2;
248 }
249
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)
252 {
253 TDB_DATA data;
254 data.dptr = (uint8_t *)m;
255 data.dsize = talloc_get_size(m);
256 return data;
257 }
258
259 /*
260 loop over a marshalling buffer
261
262 - pass r==NULL to start
263 - loop the number of times indicated by m->count
264 */
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)
269 {
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);
274 }
275
276 if (reqid != NULL) {
277 *reqid = r->reqid;
278 }
279
280 if (key != NULL) {
281 key->dptr = &r->data[0];
282 key->dsize = r->keylen;
283 }
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);
290 }
291 }
292
293 if (header != NULL) {
294 if (r->datalen < sizeof(*header)) {
295 return NULL;
296 }
297 *header = *(struct ctdb_ltdb_header *)&r->data[r->keylen];
298 }
299
300 return r;
301 }
302
303 /**
304 * CTDB transaction destructor
305 */
306 static int db_ctdb_transaction_destructor(struct db_ctdb_transaction_handle *h)
307 {
308 NTSTATUS status;
309
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;
314 }
315 return 0;
316 }
317
318 /**
319 * CTDB dbwrap API: transaction_start function
320 * starts a transaction on a persistent database
321 */
322 static int db_ctdb_transaction_start(struct db_context *db)
323 {
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);
328
329 if (!db->persistent) {
330 DEBUG(0,("transactions not supported on non-persistent database 0x%08x\n",
331 ctx->db_id));
332 return -1;
333 }
334
335 if (ctx->transaction) {
336 ctx->transaction->nesting++;
337 return 0;
338 }
339
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;
344 }
345
346 h->ctx = ctx;
347
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;
354 }
355
356 /*
357 * Wait a day, i.e. forever...
358 */
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;
365 }
366
367 talloc_set_destructor(h, db_ctdb_transaction_destructor);
368
369 ctx->transaction = h;
370
371 DEBUG(5,(__location__ " Started transaction on db 0x%08x\n", ctx->db_id));
372
373 return 0;
374 }
375
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)
381 {
382 struct ctdb_rec_data *rec = NULL;
383 struct ctdb_ltdb_header h;
384 bool found;
385 TDB_DATA data;
386 int i;
387
388 if (buf == NULL) {
389 return false;
390 }
391
392 /*
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.
398 */
399
400 for (i=0; i<buf->count; i++) {
401 TDB_DATA tkey, tdata;
402 uint32_t reqid;
403
404 rec = db_ctdb_marshall_loop_next(buf, rec, &reqid, &h, &tkey,
405 &tdata);
406 if (rec == NULL) {
407 return false;
408 }
409
410 if (tdb_data_equal(key, tkey)) {
411 found = true;
412 data = tdata;
413 }
414 }
415
416 if (!found) {
417 return false;
418 }
419
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;
425 }
426 *pdata = data;
427 }
428
429 if (pheader != NULL) {
430 *pheader = h;
431 }
432
433 return true;
434 }
435
436 /*
437 fetch a record inside a transaction
438 */
439 static int db_ctdb_transaction_fetch(struct db_ctdb_ctx *db,
440 TALLOC_CTX *mem_ctx,
441 TDB_DATA key, TDB_DATA *data)
442 {
443 struct db_ctdb_transaction_handle *h = db->transaction;
444 NTSTATUS status;
445 bool found;
446
447 found = pull_newest_from_marshall_buffer(h->m_write, key, NULL,
448 mem_ctx, data);
449 if (found) {
450 return 0;
451 }
452
453 status = db_ctdb_ltdb_fetch(h->ctx, key, NULL, mem_ctx, data);
454
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;
459 }
460
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;
469 }
470
471 return 0;
472 }
473
474
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);
477
478 static struct db_record *db_ctdb_fetch_locked_transaction(struct db_ctdb_ctx *ctx,
479 TALLOC_CTX *mem_ctx,
480 TDB_DATA key)
481 {
482 struct db_record *result;
483 TDB_DATA ctdb_data;
484
485 if (!(result = talloc(mem_ctx, struct db_record))) {
486 DEBUG(0, ("talloc failed\n"));
487 return NULL;
488 }
489
490 result->private_data = ctx->transaction;
491
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;
498 }
499
500 result->store = db_ctdb_store_transaction;
501 result->delete_rec = db_ctdb_delete_transaction;
502
503 if (pull_newest_from_marshall_buffer(ctx->transaction->m_write, key,
504 NULL, result, &result->value)) {
505 return result;
506 }
507
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;
513 }
514
515 result->value.dsize = ctdb_data.dsize - sizeof(struct ctdb_ltdb_header);
516 result->value.dptr = NULL;
517
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);
524 }
525
526 SAFE_FREE(ctdb_data.dptr);
527
528 return result;
529 }
530
531 static int db_ctdb_record_destructor(struct db_record **recp)
532 {
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"));
539 }
540 return 0;
541 }
542
543 /*
544 auto-create a transaction for persistent databases
545 */
546 static struct db_record *db_ctdb_fetch_locked_persistent(struct db_ctdb_ctx *ctx,
547 TALLOC_CTX *mem_ctx,
548 TDB_DATA key)
549 {
550 int res;
551 struct db_record *rec, **recp;
552
553 res = db_ctdb_transaction_start(ctx->db);
554 if (res == -1) {
555 return NULL;
556 }
557
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;
562 }
563
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;
570 }
571 *recp = rec;
572 talloc_set_destructor(recp, db_ctdb_record_destructor);
573 return rec;
574 }
575
576
577 /*
578 stores a record inside a transaction
579 */
580 static NTSTATUS db_ctdb_transaction_store(struct db_ctdb_transaction_handle *h,
581 TDB_DATA key, TDB_DATA data)
582 {
583 TALLOC_CTX *tmp_ctx = talloc_new(h);
584 TDB_DATA rec;
585 struct ctdb_ltdb_header header;
586
587 ZERO_STRUCT(header);
588
589 /* we need the header so we can update the RSN */
590
591 if (!pull_newest_from_marshall_buffer(h->m_write, key, &header,
592 NULL, NULL)) {
593
594 rec = tdb_fetch(h->ctx->wtdb->tdb, key);
595
596 if (rec.dptr != NULL) {
597 memcpy(&header, rec.dptr,
598 sizeof(struct ctdb_ltdb_header));
599 rec.dsize -= sizeof(struct ctdb_ltdb_header);
600
601 /*
602 * a special case, we are writing the same
603 * data that is there now
604 */
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;
612 }
613 }
614 SAFE_FREE(rec.dptr);
615 }
616
617 header.dmaster = get_my_vnn();
618 header.rsn++;
619
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;
627 }
628
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;
634 }
635
636 talloc_free(tmp_ctx);
637 return NT_STATUS_OK;
638 }
639
640
641 /*
642 a record store inside a transaction
643 */
644 static NTSTATUS db_ctdb_store_transaction(struct db_record *rec, TDB_DATA data, int flag)
645 {
646 struct db_ctdb_transaction_handle *h = talloc_get_type_abort(
647 rec->private_data, struct db_ctdb_transaction_handle);
648 NTSTATUS status;
649
650 status = db_ctdb_transaction_store(h, rec->key, data);
651 return status;
652 }
653
654 /*
655 a record delete inside a transaction
656 */
657 static NTSTATUS db_ctdb_delete_transaction(struct db_record *rec)
658 {
659 struct db_ctdb_transaction_handle *h = talloc_get_type_abort(
660 rec->private_data, struct db_ctdb_transaction_handle);
661 NTSTATUS status;
662
663 status = db_ctdb_transaction_store(h, rec->key, tdb_null);
664 return status;
665 }
666
667 /**
668 * Fetch the db sequence number of a persistent db directly from the db.
669 */
670 static NTSTATUS db_ctdb_fetch_db_seqnum_from_db(struct db_ctdb_ctx *db,
671 uint64_t *seqnum)
672 {
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();
679
680 if (seqnum == NULL) {
681 return NT_STATUS_INVALID_PARAMETER;
682 }
683
684 key.dptr = (uint8_t *)discard_const(keyname);
685 key.dsize = strlen(keyname) + 1;
686
687 status = db_ctdb_ltdb_fetch(db, key, &header, mem_ctx, &data);
688 if (!NT_STATUS_IS_OK(status) &&
689 !NT_STATUS_EQUAL(status, NT_STATUS_NOT_FOUND))
690 {
691 goto done;
692 }
693
694 status = NT_STATUS_OK;
695
696 if (data.dsize != sizeof(uint64_t)) {
697 *seqnum = 0;
698 goto done;
699 }
700
701 *seqnum = *(uint64_t *)data.dptr;
702
703 done:
704 TALLOC_FREE(mem_ctx);
705 return status;
706 }
707
708 /**
709 * Store the database sequence number inside a transaction.
710 */
711 static NTSTATUS db_ctdb_store_db_seqnum(struct db_ctdb_transaction_handle *h,
712 uint64_t seqnum)
713 {
714 NTSTATUS status;
715 const char *keyname = CTDB_DB_SEQNUM_KEY;
716 TDB_DATA key;
717 TDB_DATA data;
718
719 key.dptr = (uint8_t *)discard_const(keyname);
720 key.dsize = strlen(keyname);
721
722 data.dptr = (uint8_t *)&seqnum;
723 data.dsize = sizeof(uint64_t);
724
725 status = db_ctdb_transaction_store(h, key, data);
726
727 return status;
728 }
729
730 /*
731 commit a transaction
732 */
733 static int db_ctdb_transaction_commit(struct db_context *db)
734 {
735 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
736 struct db_ctdb_ctx);
737 NTSTATUS rets;
738 int status;
739 struct db_ctdb_transaction_handle *h = ctx->transaction;
740 uint64_t old_seqnum, new_seqnum;
741 int ret;
742
743 if (h == NULL) {
744 DEBUG(0,(__location__ " transaction commit with no open transaction on db 0x%08x\n", ctx->db_id));
745 return -1;
746 }
747
748 if (h->nested_cancel) {
749 db->transaction_cancel(db);
750 DEBUG(5,(__location__ " Failed transaction commit after nested cancel\n"));
751 return -1;
752 }
753
754 if (h->nesting != 0) {
755 h->nesting--;
756 return 0;
757 }
758
759 DEBUG(5,(__location__ " Commit transaction on db 0x%08x\n", ctx->db_id));
760
761 /*
762 * As the last db action before committing, bump the database sequence
763 * number. Note that this undoes all changes to the seqnum records
764 * performed under the transaction. This record is not meant to be
765 * modified by user interaction. It is for internal use only...
766 */
767 rets = db_ctdb_fetch_db_seqnum_from_db(ctx, &old_seqnum);
768 if (!NT_STATUS_IS_OK(rets)) {
769 DEBUG(1, (__location__ " failed to fetch the db sequence number "
770 "in transaction commit on db 0x%08x\n", ctx->db_id));
771 ret = -1;
772 goto done;
773 }
774
775 new_seqnum = old_seqnum + 1;
776
777 rets = db_ctdb_store_db_seqnum(h, new_seqnum);
778 if (!NT_STATUS_IS_OK(rets)) {
779 DEBUG(1, (__location__ "failed to store the db sequence number "
780 " in transaction commit on db 0x%08x\n", ctx->db_id));
781 ret = -1;
782 goto done;
783 }
784
785 again:
786 if (h->m_write == NULL) {
787 /* no changes were made, potentially after a retry */
788 goto done;
789 }
790
791 /* tell ctdbd to commit to the other nodes */
792 rets = ctdbd_control_local(messaging_ctdbd_connection(),
793 CTDB_CONTROL_TRANS3_COMMIT,
794 h->ctx->db_id, 0,
795 db_ctdb_marshall_finish(h->m_write),
796 NULL, NULL, &status);
797 if (!NT_STATUS_IS_OK(rets) || status != 0) {
798 /*
799 * The TRANS3_COMMIT control should only possibly fail when a
800 * recovery has been running concurrently. In any case, the db
801 * will be the same on all nodes, either the new copy or the
802 * old copy. This can be detected by comparing the old and new
803 * local sequence numbers.
804 */
805 rets = db_ctdb_fetch_db_seqnum_from_db(ctx, &new_seqnum);
806 if (!NT_STATUS_IS_OK(rets)) {
807 DEBUG(1, (__location__ " failed to refetch db sequence "
808 "number after failed TRANS3_COMMIT\n"));
809 ret = -1;
810 goto done;
811 }
812
813 if (new_seqnum == old_seqnum) {
814 /* Recovery prevented all our changes: retry. */
815 goto again;
816 } else if (new_seqnum != (old_seqnum + 1)) {
817 DEBUG(0, (__location__ " ERROR: new_seqnum[%lu] != "
818 "old_seqnum[%lu] + (0 or 1) after failed "
819 "TRANS3_COMMIT - this should not happen!\n",
820 (unsigned long)new_seqnum,
821 (unsigned long)old_seqnum));
822 ret = -1;
823 goto done;
824 }
825 /*
826 * Recovery propagated our changes to all nodes, completing
827 * our commit for us - succeed.
828 */
829 }
830
831 ret = 0;
832
833 done:
834 h->ctx->transaction = NULL;
835 talloc_free(h);
836 return 0;
837 }
838
839
840 /*
841 cancel a transaction
842 */
843 static int db_ctdb_transaction_cancel(struct db_context *db)
844 {
845 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
846 struct db_ctdb_ctx);
847 struct db_ctdb_transaction_handle *h = ctx->transaction;
848
849 if (h == NULL) {
850 DEBUG(0,(__location__ " transaction cancel with no open transaction on db 0x%08x\n", ctx->db_id));
851 return -1;
852 }
853
854 if (h->nesting != 0) {
855 h->nesting--;
856 h->nested_cancel = true;
857 return 0;
858 }
859
860 DEBUG(5,(__location__ " Cancel transaction on db 0x%08x\n", ctx->db_id));
861
862 ctx->transaction = NULL;
863 talloc_free(h);
864 return 0;
865 }
866
867
868 static NTSTATUS db_ctdb_store(struct db_record *rec, TDB_DATA data, int flag)
869 {
870 struct db_ctdb_rec *crec = talloc_get_type_abort(
871 rec->private_data, struct db_ctdb_rec);
872
873 return db_ctdb_ltdb_store(crec->ctdb_ctx, rec->key, &(crec->header), data);
874 }
875
876
877
878 static NTSTATUS db_ctdb_delete(struct db_record *rec)
879 {
880 TDB_DATA data;
881
882 /*
883 * We have to store the header with empty data. TODO: Fix the
884 * tdb-level cleanup
885 */
886
887 ZERO_STRUCT(data);
888
889 return db_ctdb_store(rec, data, 0);
890
891 }
892
893 static int db_ctdb_record_destr(struct db_record* data)
894 {
895 struct db_ctdb_rec *crec = talloc_get_type_abort(
896 data->private_data, struct db_ctdb_rec);
897
898 DEBUG(10, (DEBUGLEVEL > 10
899 ? "Unlocking db %u key %s\n"
900 : "Unlocking db %u key %.20s\n",
901 (int)crec->ctdb_ctx->db_id,
902 hex_encode_talloc(data, (unsigned char *)data->key.dptr,
903 data->key.dsize)));
904
905 if (tdb_chainunlock(crec->ctdb_ctx->wtdb->tdb, data->key) != 0) {
906 DEBUG(0, ("tdb_chainunlock failed\n"));
907 return -1;
908 }
909
910 return 0;
911 }
912
913 static struct db_record *fetch_locked_internal(struct db_ctdb_ctx *ctx,
914 TALLOC_CTX *mem_ctx,
915 TDB_DATA key,
916 bool persistent)
917 {
918 struct db_record *result;
919 struct db_ctdb_rec *crec;
920 NTSTATUS status;
921 TDB_DATA ctdb_data;
922 int migrate_attempts = 0;
923
924 if (!(result = talloc(mem_ctx, struct db_record))) {
925 DEBUG(0, ("talloc failed\n"));
926 return NULL;
927 }
928
929 if (!(crec = TALLOC_ZERO_P(result, struct db_ctdb_rec))) {
930 DEBUG(0, ("talloc failed\n"));
931 TALLOC_FREE(result);
932 return NULL;
933 }
934
935 result->private_data = (void *)crec;
936 crec->ctdb_ctx = ctx;
937
938 result->key.dsize = key.dsize;
939 result->key.dptr = (uint8 *)talloc_memdup(result, key.dptr, key.dsize);
940 if (result->key.dptr == NULL) {
941 DEBUG(0, ("talloc failed\n"));
942 TALLOC_FREE(result);
943 return NULL;
944 }
945
946 /*
947 * Do a blocking lock on the record
948 */
949 again:
950
951 if (DEBUGLEVEL >= 10) {
952 char *keystr = hex_encode_talloc(result, key.dptr, key.dsize);
953 DEBUG(10, (DEBUGLEVEL > 10
954 ? "Locking db %u key %s\n"
955 : "Locking db %u key %.20s\n",
956 (int)crec->ctdb_ctx->db_id, keystr));
957 TALLOC_FREE(keystr);
958 }
959
960 if (tdb_chainlock(ctx->wtdb->tdb, key) != 0) {
961 DEBUG(3, ("tdb_chainlock failed\n"));
962 TALLOC_FREE(result);
963 return NULL;
964 }
965
966 result->store = db_ctdb_store;
967 result->delete_rec = db_ctdb_delete;
968 talloc_set_destructor(result, db_ctdb_record_destr);
969
970 ctdb_data = tdb_fetch(ctx->wtdb->tdb, key);
971
972 /*
973 * See if we have a valid record and we are the dmaster. If so, we can
974 * take the shortcut and just return it.
975 */
976
977 if ((ctdb_data.dptr == NULL) ||
978 (ctdb_data.dsize < sizeof(struct ctdb_ltdb_header)) ||
979 ((struct ctdb_ltdb_header *)ctdb_data.dptr)->dmaster != get_my_vnn()
980 #if 0
981 || (random() % 2 != 0)
982 #endif
983 ) {
984 SAFE_FREE(ctdb_data.dptr);
985 tdb_chainunlock(ctx->wtdb->tdb, key);
986 talloc_set_destructor(result, NULL);
987
988 migrate_attempts += 1;
989
990 DEBUG(10, ("ctdb_data.dptr = %p, dmaster = %u (%u)\n",
991 ctdb_data.dptr, ctdb_data.dptr ?
992 ((struct ctdb_ltdb_header *)ctdb_data.dptr)->dmaster : -1,
993 get_my_vnn()));
994
995 status = ctdbd_migrate(messaging_ctdbd_connection(),ctx->db_id, key);
996 if (!NT_STATUS_IS_OK(status)) {
997 DEBUG(5, ("ctdb_migrate failed: %s\n",
998 nt_errstr(status)));
999 TALLOC_FREE(result);
1000 return NULL;
1001 }
1002 /* now its migrated, try again */
1003 goto again;
1004 }
1005
1006 if (migrate_attempts > 10) {
1007 DEBUG(0, ("db_ctdb_fetch_locked needed %d attempts\n",
1008 migrate_attempts));
1009 }
1010
1011 memcpy(&crec->header, ctdb_data.dptr, sizeof(crec->header));
1012
1013 result->value.dsize = ctdb_data.dsize - sizeof(crec->header);
1014 result->value.dptr = NULL;
1015
1016 if ((result->value.dsize != 0)
1017 && !(result->value.dptr = (uint8 *)talloc_memdup(
1018 result, ctdb_data.dptr + sizeof(crec->header),
1019 result->value.dsize))) {
1020 DEBUG(0, ("talloc failed\n"));
1021 TALLOC_FREE(result);
1022 }
1023
1024 SAFE_FREE(ctdb_data.dptr);
1025
1026 return result;
1027 }
1028
1029 static struct db_record *db_ctdb_fetch_locked(struct db_context *db,
1030 TALLOC_CTX *mem_ctx,
1031 TDB_DATA key)
1032 {
1033 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
1034 struct db_ctdb_ctx);
1035
1036 if (ctx->transaction != NULL) {
1037 return db_ctdb_fetch_locked_transaction(ctx, mem_ctx, key);
1038 }
1039
1040 if (db->persistent) {
1041 return db_ctdb_fetch_locked_persistent(ctx, mem_ctx, key);
1042 }
1043
1044 return fetch_locked_internal(ctx, mem_ctx, key, db->persistent);
1045 }
1046
1047 /*
1048 fetch (unlocked, no migration) operation on ctdb
1049 */
1050 static int db_ctdb_fetch(struct db_context *db, TALLOC_CTX *mem_ctx,
1051 TDB_DATA key, TDB_DATA *data)
1052 {
1053 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
1054 struct db_ctdb_ctx);
1055 NTSTATUS status;
1056 TDB_DATA ctdb_data;
1057
1058 if (ctx->transaction) {
1059 return db_ctdb_transaction_fetch(ctx, mem_ctx, key, data);
1060 }
1061
1062 /* try a direct fetch */
1063 ctdb_data = tdb_fetch(ctx->wtdb->tdb, key);
1064
1065 /*
1066 * See if we have a valid record and we are the dmaster. If so, we can
1067 * take the shortcut and just return it.
1068 * we bypass the dmaster check for persistent databases
1069 */
1070 if ((ctdb_data.dptr != NULL) &&
1071 (ctdb_data.dsize >= sizeof(struct ctdb_ltdb_header)) &&
1072 (db->persistent ||
1073 ((struct ctdb_ltdb_header *)ctdb_data.dptr)->dmaster == get_my_vnn())) {
1074 /* we are the dmaster - avoid the ctdb protocol op */
1075
1076 data->dsize = ctdb_data.dsize - sizeof(struct ctdb_ltdb_header);
1077 if (data->dsize == 0) {
1078 SAFE_FREE(ctdb_data.dptr);
1079 data->dptr = NULL;
1080 return 0;
1081 }
1082
1083 data->dptr = (uint8 *)talloc_memdup(
1084 mem_ctx, ctdb_data.dptr+sizeof(struct ctdb_ltdb_header),
1085 data->dsize);
1086
1087 SAFE_FREE(ctdb_data.dptr);
1088
1089 if (data->dptr == NULL) {
1090 return -1;
1091 }
1092 return 0;
1093 }
1094
1095 SAFE_FREE(ctdb_data.dptr);
1096
1097 /* we weren't able to get it locally - ask ctdb to fetch it for us */
1098 status = ctdbd_fetch(messaging_ctdbd_connection(),ctx->db_id, key, mem_ctx, data);
1099 if (!NT_STATUS_IS_OK(status)) {
1100 DEBUG(5, ("ctdbd_fetch failed: %s\n", nt_errstr(status)));
1101 return -1;
1102 }
1103
1104 return 0;
1105 }
1106
1107 struct traverse_state {
1108 struct db_context *db;
1109 int (*fn)(struct db_record *rec, void *private_data);
1110 void *private_data;
1111 };
1112
1113 static void traverse_callback(TDB_DATA key, TDB_DATA data, void *private_data)
1114 {
1115 struct traverse_state *state = (struct traverse_state *)private_data;
1116 struct db_record *rec;
1117 TALLOC_CTX *tmp_ctx = talloc_new(state->db);
1118 /* we have to give them a locked record to prevent races */
1119 rec = db_ctdb_fetch_locked(state->db, tmp_ctx, key);
1120 if (rec && rec->value.dsize > 0) {
1121 state->fn(rec, state->private_data);
1122 }
1123 talloc_free(tmp_ctx);
1124 }
1125
1126 static int traverse_persistent_callback(TDB_CONTEXT *tdb, TDB_DATA kbuf, TDB_DATA dbuf,
1127 void *private_data)
1128 {
1129 struct traverse_state *state = (struct traverse_state *)private_data;
1130 struct db_record *rec;
1131 TALLOC_CTX *tmp_ctx = talloc_new(state->db);
1132 int ret = 0;
1133 /* we have to give them a locked record to prevent races */
1134 rec = db_ctdb_fetch_locked(state->db, tmp_ctx, kbuf);
1135 if (rec && rec->value.dsize > 0) {
1136 ret = state->fn(rec, state->private_data);
1137 }
1138 talloc_free(tmp_ctx);
1139 return ret;
1140 }
1141
1142 static int db_ctdb_traverse(struct db_context *db,
1143 int (*fn)(struct db_record *rec,
1144 void *private_data),
1145 void *private_data)
1146 {
1147 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
1148 struct db_ctdb_ctx);
1149 struct traverse_state state;
1150
1151 state.db = db;
1152 state.fn = fn;
1153 state.private_data = private_data;
1154
1155 if (db->persistent) {
1156 /* for persistent databases we don't need to do a ctdb traverse,
1157 we can do a faster local traverse */
1158 return tdb_traverse(ctx->wtdb->tdb, traverse_persistent_callback, &state);
1159 }
1160
1161
1162 ctdbd_traverse(ctx->db_id, traverse_callback, &state);
1163 return 0;
1164 }
1165
1166 static NTSTATUS db_ctdb_store_deny(struct db_record *rec, TDB_DATA data, int flag)
1167 {
1168 return NT_STATUS_MEDIA_WRITE_PROTECTED;
1169 }
1170
1171 static NTSTATUS db_ctdb_delete_deny(struct db_record *rec)
1172 {
1173 return NT_STATUS_MEDIA_WRITE_PROTECTED;
1174 }
1175
1176 static void traverse_read_callback(TDB_DATA key, TDB_DATA data, void *private_data)
1177 {
1178 struct traverse_state *state = (struct traverse_state *)private_data;
1179 struct db_record rec;
1180 rec.key = key;
1181 rec.value = data;
1182 rec.store = db_ctdb_store_deny;
1183 rec.delete_rec = db_ctdb_delete_deny;
1184 rec.private_data = state->db;
1185 state->fn(&rec, state->private_data);
1186 }
1187
1188 static int traverse_persistent_callback_read(TDB_CONTEXT *tdb, TDB_DATA kbuf, TDB_DATA dbuf,
1189 void *private_data)
1190 {
1191 struct traverse_state *state = (struct traverse_state *)private_data;
1192 struct db_record rec;
1193 rec.key = kbuf;
1194 rec.value = dbuf;
1195 rec.store = db_ctdb_store_deny;
1196 rec.delete_rec = db_ctdb_delete_deny;
1197 rec.private_data = state->db;
1198
1199 if (rec.value.dsize <= sizeof(struct ctdb_ltdb_header)) {
1200 /* a deleted record */
1201 return 0;
1202 }
1203 rec.value.dsize -= sizeof(struct ctdb_ltdb_header);
1204 rec.value.dptr += sizeof(struct ctdb_ltdb_header);
1205
1206 return state->fn(&rec, state->private_data);
1207 }
1208
1209 static int db_ctdb_traverse_read(struct db_context *db,
1210 int (*fn)(struct db_record *rec,
1211 void *private_data),
1212 void *private_data)
1213 {
1214 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
1215 struct db_ctdb_ctx);
1216 struct traverse_state state;
1217
1218 state.db = db;
1219 state.fn = fn;
1220 state.private_data = private_data;
1221
1222 if (db->persistent) {
1223 /* for persistent databases we don't need to do a ctdb traverse,
1224 we can do a faster local traverse */
1225 return tdb_traverse_read(ctx->wtdb->tdb, traverse_persistent_callback_read, &state);
1226 }
1227
1228 ctdbd_traverse(ctx->db_id, traverse_read_callback, &state);
1229 return 0;
1230 }
1231
1232 static int db_ctdb_get_seqnum(struct db_context *db)
1233 {
1234 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
1235 struct db_ctdb_ctx);
1236 return tdb_get_seqnum(ctx->wtdb->tdb);
1237 }
1238
1239 static int db_ctdb_get_flags(struct db_context *db)
1240 {
1241 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
1242 struct db_ctdb_ctx);
1243 return tdb_get_flags(ctx->wtdb->tdb);
1244 }
1245
1246 struct db_context *db_open_ctdb(TALLOC_CTX *mem_ctx,
1247 const char *name,
1248 int hash_size, int tdb_flags,
1249 int open_flags, mode_t mode)
1250 {
1251 struct db_context *result;
1252 struct db_ctdb_ctx *db_ctdb;
1253 char *db_path;
1254 struct ctdbd_connection *conn;
1255
1256 if (!lp_clustering()) {
1257 DEBUG(10, ("Clustering disabled -- no ctdb\n"));
1258 return NULL;
1259 }
1260
1261 if (!(result = TALLOC_ZERO_P(mem_ctx, struct db_context))) {
1262 DEBUG(0, ("talloc failed\n"));
1263 TALLOC_FREE(result);
1264 return NULL;
1265 }
1266
1267 if (!(db_ctdb = TALLOC_P(result, struct db_ctdb_ctx))) {
1268 DEBUG(0, ("talloc failed\n"));
1269 TALLOC_FREE(result);
1270 return NULL;
1271 }
1272
1273 db_ctdb->transaction = NULL;
1274 db_ctdb->db = result;
1275
1276 conn = messaging_ctdbd_connection();
1277
1278 if (!NT_STATUS_IS_OK(ctdbd_db_attach(conn, name, &db_ctdb->db_id, tdb_flags))) {
1279 DEBUG(0, ("ctdbd_db_attach failed for %s\n", name));
1280 TALLOC_FREE(result);
1281 return NULL;
1282 }
1283
1284 db_path = ctdbd_dbpath(conn, db_ctdb, db_ctdb->db_id);
1285
1286 result->persistent = ((tdb_flags & TDB_CLEAR_IF_FIRST) == 0);
1287
1288 /* only pass through specific flags */
1289 tdb_flags &= TDB_SEQNUM;
1290
1291 /* honor permissions if user has specified O_CREAT */
1292 if (open_flags & O_CREAT) {
1293 chmod(db_path, mode);
1294 }
1295
1296 db_ctdb->wtdb = tdb_wrap_open(db_ctdb, db_path, hash_size, tdb_flags, O_RDWR, 0);
1297 if (db_ctdb->wtdb == NULL) {
1298 DEBUG(0, ("Could not open tdb %s: %s\n", db_path, strerror(errno)));
1299 TALLOC_FREE(result);
1300 return NULL;
1301 }
1302 talloc_free(db_path);
1303
1304 if (result->persistent) {
1305 db_ctdb->lock_ctx = g_lock_ctx_init(db_ctdb,
1306 ctdb_conn_msg_ctx(conn));
1307 if (db_ctdb->lock_ctx == NULL) {
1308 DEBUG(0, ("g_lock_ctx_init failed\n"));
1309 TALLOC_FREE(result);
1310 return NULL;
1311 }
1312 }
1313
1314 result->private_data = (void *)db_ctdb;
1315 result->fetch_locked = db_ctdb_fetch_locked;
1316 result->fetch = db_ctdb_fetch;
1317 result->traverse = db_ctdb_traverse;
1318 result->traverse_read = db_ctdb_traverse_read;
1319 result->get_seqnum = db_ctdb_get_seqnum;
1320 result->get_flags = db_ctdb_get_flags;
1321 result->transaction_start = db_ctdb_transaction_start;
1322 result->transaction_commit = db_ctdb_transaction_commit;
1323 result->transaction_cancel = db_ctdb_transaction_cancel;
1324
1325 DEBUG(3,("db_open_ctdb: opened database '%s' with dbid 0x%x\n",
1326 name, db_ctdb->db_id));
1327
1328 return result;
1329 }
1330 #endif
Samba GIT mirror