?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
s3: db->persistent==true was handled earlier, make this more obvious (cherry picked...
[Samba.git] / source3 / lib / dbwrap_ctdb.c
blob05ac77785883b5e455fe216491df3df1d342bc7e
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 = false;
385 TDB_DATA data;
386 int i;
387
388 if (buf == NULL) {
389 return false;
390 }
391
392 ZERO_STRUCT(h);
393 ZERO_STRUCT(data);
394
395 /*
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.
401 */
402
403 for (i=0; i<buf->count; i++) {
404 TDB_DATA tkey, tdata;
405 uint32_t reqid;
406 struct ctdb_ltdb_header hdr;
407
408 ZERO_STRUCT(hdr);
409
410 rec = db_ctdb_marshall_loop_next(buf, rec, &reqid, &hdr, &tkey,
411 &tdata);
412 if (rec == NULL) {
413 return false;
414 }
415
416 if (tdb_data_equal(key, tkey)) {
417 found = true;
418 data = tdata;
419 h = hdr;
420 }
421 }
422
423 if (!found) {
424 return false;
425 }
426
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;
432 }
433 *pdata = data;
434 }
435
436 if (pheader != NULL) {
437 *pheader = h;
438 }
439
440 return true;
441 }
442
443 /*
444 fetch a record inside a transaction
445 */
446 static int db_ctdb_transaction_fetch(struct db_ctdb_ctx *db,
447 TALLOC_CTX *mem_ctx,
448 TDB_DATA key, TDB_DATA *data)
449 {
450 struct db_ctdb_transaction_handle *h = db->transaction;
451 NTSTATUS status;
452 bool found;
453
454 found = pull_newest_from_marshall_buffer(h->m_write, key, NULL,
455 mem_ctx, data);
456 if (found) {
457 return 0;
458 }
459
460 status = db_ctdb_ltdb_fetch(h->ctx, key, NULL, mem_ctx, data);
461
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;
466 }
467
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;
476 }
477
478 return 0;
479 }
480
481
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);
484
485 static struct db_record *db_ctdb_fetch_locked_transaction(struct db_ctdb_ctx *ctx,
486 TALLOC_CTX *mem_ctx,
487 TDB_DATA key)
488 {
489 struct db_record *result;
490 TDB_DATA ctdb_data;
491
492 if (!(result = talloc(mem_ctx, struct db_record))) {
493 DEBUG(0, ("talloc failed\n"));
494 return NULL;
495 }
496
497 result->private_data = ctx->transaction;
498
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;
505 }
506
507 result->store = db_ctdb_store_transaction;
508 result->delete_rec = db_ctdb_delete_transaction;
509
510 if (pull_newest_from_marshall_buffer(ctx->transaction->m_write, key,
511 NULL, result, &result->value)) {
512 return result;
513 }
514
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;
520 }
521
522 result->value.dsize = ctdb_data.dsize - sizeof(struct ctdb_ltdb_header);
523 result->value.dptr = NULL;
524
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);
531 }
532
533 SAFE_FREE(ctdb_data.dptr);
534
535 return result;
536 }
537
538 static int db_ctdb_record_destructor(struct db_record **recp)
539 {
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"));
546 }
547 return 0;
548 }
549
550 /*
551 auto-create a transaction for persistent databases
552 */
553 static struct db_record *db_ctdb_fetch_locked_persistent(struct db_ctdb_ctx *ctx,
554 TALLOC_CTX *mem_ctx,
555 TDB_DATA key)
556 {
557 int res;
558 struct db_record *rec, **recp;
559
560 res = db_ctdb_transaction_start(ctx->db);
561 if (res == -1) {
562 return NULL;
563 }
564
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;
569 }
570
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;
577 }
578 *recp = rec;
579 talloc_set_destructor(recp, db_ctdb_record_destructor);
580 return rec;
581 }
582
583
584 /*
585 stores a record inside a transaction
586 */
587 static NTSTATUS db_ctdb_transaction_store(struct db_ctdb_transaction_handle *h,
588 TDB_DATA key, TDB_DATA data)
589 {
590 TALLOC_CTX *tmp_ctx = talloc_new(h);
591 TDB_DATA rec;
592 struct ctdb_ltdb_header header;
593
594 ZERO_STRUCT(header);
595
596 /* we need the header so we can update the RSN */
597
598 if (!pull_newest_from_marshall_buffer(h->m_write, key, &header,
599 NULL, NULL)) {
600
601 rec = tdb_fetch(h->ctx->wtdb->tdb, key);
602
603 if (rec.dptr != NULL) {
604 memcpy(&header, rec.dptr,
605 sizeof(struct ctdb_ltdb_header));
606 rec.dsize -= sizeof(struct ctdb_ltdb_header);
607
608 /*
609 * a special case, we are writing the same
610 * data that is there now
611 */
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;
619 }
620 }
621 SAFE_FREE(rec.dptr);
622 }
623
624 header.dmaster = get_my_vnn();
625 header.rsn++;
626
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;
634 }
635
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;
641 }
642
643 talloc_free(tmp_ctx);
644 return NT_STATUS_OK;
645 }
646
647
648 /*
649 a record store inside a transaction
650 */
651 static NTSTATUS db_ctdb_store_transaction(struct db_record *rec, TDB_DATA data, int flag)
652 {
653 struct db_ctdb_transaction_handle *h = talloc_get_type_abort(
654 rec->private_data, struct db_ctdb_transaction_handle);
655 NTSTATUS status;
656
657 status = db_ctdb_transaction_store(h, rec->key, data);
658 return status;
659 }
660
661 /*
662 a record delete inside a transaction
663 */
664 static NTSTATUS db_ctdb_delete_transaction(struct db_record *rec)
665 {
666 struct db_ctdb_transaction_handle *h = talloc_get_type_abort(
667 rec->private_data, struct db_ctdb_transaction_handle);
668 NTSTATUS status;
669
670 status = db_ctdb_transaction_store(h, rec->key, tdb_null);
671 return status;
672 }
673
674 /**
675 * Fetch the db sequence number of a persistent db directly from the db.
676 */
677 static NTSTATUS db_ctdb_fetch_db_seqnum_from_db(struct db_ctdb_ctx *db,
678 uint64_t *seqnum)
679 {
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();
686
687 if (seqnum == NULL) {
688 return NT_STATUS_INVALID_PARAMETER;
689 }
690
691 key = string_term_tdb_data(keyname);
692
693 status = db_ctdb_ltdb_fetch(db, key, &header, mem_ctx, &data);
694 if (!NT_STATUS_IS_OK(status) &&
695 !NT_STATUS_EQUAL(status, NT_STATUS_NOT_FOUND))
696 {
697 goto done;
698 }
699
700 status = NT_STATUS_OK;
701
702 if (data.dsize != sizeof(uint64_t)) {
703 *seqnum = 0;
704 goto done;
705 }
706
707 *seqnum = *(uint64_t *)data.dptr;
708
709 done:
710 TALLOC_FREE(mem_ctx);
711 return status;
712 }
713
714 /**
715 * Store the database sequence number inside a transaction.
716 */
717 static NTSTATUS db_ctdb_store_db_seqnum(struct db_ctdb_transaction_handle *h,
718 uint64_t seqnum)
719 {
720 NTSTATUS status;
721 const char *keyname = CTDB_DB_SEQNUM_KEY;
722 TDB_DATA key;
723 TDB_DATA data;
724
725 key = string_term_tdb_data(keyname);
726
727 data.dptr = (uint8_t *)&seqnum;
728 data.dsize = sizeof(uint64_t);
729
730 status = db_ctdb_transaction_store(h, key, data);
731
732 return status;
733 }
734
735 /*
736 commit a transaction
737 */
738 static int db_ctdb_transaction_commit(struct db_context *db)
739 {
740 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
741 struct db_ctdb_ctx);
742 NTSTATUS rets;
743 int status;
744 struct db_ctdb_transaction_handle *h = ctx->transaction;
745 uint64_t old_seqnum, new_seqnum;
746 int ret;
747
748 if (h == NULL) {
749 DEBUG(0,(__location__ " transaction commit with no open transaction on db 0x%08x\n", ctx->db_id));
750 return -1;
751 }
752
753 if (h->nested_cancel) {
754 db->transaction_cancel(db);
755 DEBUG(5,(__location__ " Failed transaction commit after nested cancel\n"));
756 return -1;
757 }
758
759 if (h->nesting != 0) {
760 h->nesting--;
761 return 0;
762 }
763
764 if (h->m_write == NULL) {
765 /*
766 * No changes were made, so don't change the seqnum,
767 * don't push to other node, just exit with success.
768 */
769 ret = 0;
770 goto done;
771 }
772
773 DEBUG(5,(__location__ " Commit transaction on db 0x%08x\n", ctx->db_id));
774
775 /*
776 * As the last db action before committing, bump the database sequence
777 * number. Note that this undoes all changes to the seqnum records
778 * performed under the transaction. This record is not meant to be
779 * modified by user interaction. It is for internal use only...
780 */
781 rets = db_ctdb_fetch_db_seqnum_from_db(ctx, &old_seqnum);
782 if (!NT_STATUS_IS_OK(rets)) {
783 DEBUG(1, (__location__ " failed to fetch the db sequence number "
784 "in transaction commit on db 0x%08x\n", ctx->db_id));
785 ret = -1;
786 goto done;
787 }
788
789 new_seqnum = old_seqnum + 1;
790
791 rets = db_ctdb_store_db_seqnum(h, new_seqnum);
792 if (!NT_STATUS_IS_OK(rets)) {
793 DEBUG(1, (__location__ "failed to store the db sequence number "
794 " in transaction commit on db 0x%08x\n", ctx->db_id));
795 ret = -1;
796 goto done;
797 }
798
799 again:
800 /* tell ctdbd to commit to the other nodes */
801 rets = ctdbd_control_local(messaging_ctdbd_connection(),
802 CTDB_CONTROL_TRANS3_COMMIT,
803 h->ctx->db_id, 0,
804 db_ctdb_marshall_finish(h->m_write),
805 NULL, NULL, &status);
806 if (!NT_STATUS_IS_OK(rets) || status != 0) {
807 /*
808 * The TRANS3_COMMIT control should only possibly fail when a
809 * recovery has been running concurrently. In any case, the db
810 * will be the same on all nodes, either the new copy or the
811 * old copy. This can be detected by comparing the old and new
812 * local sequence numbers.
813 */
814 rets = db_ctdb_fetch_db_seqnum_from_db(ctx, &new_seqnum);
815 if (!NT_STATUS_IS_OK(rets)) {
816 DEBUG(1, (__location__ " failed to refetch db sequence "
817 "number after failed TRANS3_COMMIT\n"));
818 ret = -1;
819 goto done;
820 }
821
822 if (new_seqnum == old_seqnum) {
823 /* Recovery prevented all our changes: retry. */
824 goto again;
825 } else if (new_seqnum != (old_seqnum + 1)) {
826 DEBUG(0, (__location__ " ERROR: new_seqnum[%lu] != "
827 "old_seqnum[%lu] + (0 or 1) after failed "
828 "TRANS3_COMMIT - this should not happen!\n",
829 (unsigned long)new_seqnum,
830 (unsigned long)old_seqnum));
831 ret = -1;
832 goto done;
833 }
834 /*
835 * Recovery propagated our changes to all nodes, completing
836 * our commit for us - succeed.
837 */
838 }
839
840 ret = 0;
841
842 done:
843 h->ctx->transaction = NULL;
844 talloc_free(h);
845 return ret;
846 }
847
848
849 /*
850 cancel a transaction
851 */
852 static int db_ctdb_transaction_cancel(struct db_context *db)
853 {
854 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
855 struct db_ctdb_ctx);
856 struct db_ctdb_transaction_handle *h = ctx->transaction;
857
858 if (h == NULL) {
859 DEBUG(0,(__location__ " transaction cancel with no open transaction on db 0x%08x\n", ctx->db_id));
860 return -1;
861 }
862
863 if (h->nesting != 0) {
864 h->nesting--;
865 h->nested_cancel = true;
866 return 0;
867 }
868
869 DEBUG(5,(__location__ " Cancel transaction on db 0x%08x\n", ctx->db_id));
870
871 ctx->transaction = NULL;
872 talloc_free(h);
873 return 0;
874 }
875
876
877 static NTSTATUS db_ctdb_store(struct db_record *rec, TDB_DATA data, int flag)
878 {
879 struct db_ctdb_rec *crec = talloc_get_type_abort(
880 rec->private_data, struct db_ctdb_rec);
881
882 return db_ctdb_ltdb_store(crec->ctdb_ctx, rec->key, &(crec->header), data);
883 }
884
885
886
887 static NTSTATUS db_ctdb_delete(struct db_record *rec)
888 {
889 TDB_DATA data;
890
891 /*
892 * We have to store the header with empty data. TODO: Fix the
893 * tdb-level cleanup
894 */
895
896 ZERO_STRUCT(data);
897
898 return db_ctdb_store(rec, data, 0);
899
900 }
901
902 static int db_ctdb_record_destr(struct db_record* data)
903 {
904 struct db_ctdb_rec *crec = talloc_get_type_abort(
905 data->private_data, struct db_ctdb_rec);
906
907 DEBUG(10, (DEBUGLEVEL > 10
908 ? "Unlocking db %u key %s\n"
909 : "Unlocking db %u key %.20s\n",
910 (int)crec->ctdb_ctx->db_id,
911 hex_encode_talloc(data, (unsigned char *)data->key.dptr,
912 data->key.dsize)));
913
914 if (tdb_chainunlock(crec->ctdb_ctx->wtdb->tdb, data->key) != 0) {
915 DEBUG(0, ("tdb_chainunlock failed\n"));
916 return -1;
917 }
918
919 return 0;
920 }
921
922 static struct db_record *fetch_locked_internal(struct db_ctdb_ctx *ctx,
923 TALLOC_CTX *mem_ctx,
924 TDB_DATA key,
925 bool persistent)
926 {
927 struct db_record *result;
928 struct db_ctdb_rec *crec;
929 NTSTATUS status;
930 TDB_DATA ctdb_data;
931 int migrate_attempts = 0;
932
933 if (!(result = talloc(mem_ctx, struct db_record))) {
934 DEBUG(0, ("talloc failed\n"));
935 return NULL;
936 }
937
938 if (!(crec = TALLOC_ZERO_P(result, struct db_ctdb_rec))) {
939 DEBUG(0, ("talloc failed\n"));
940 TALLOC_FREE(result);
941 return NULL;
942 }
943
944 result->private_data = (void *)crec;
945 crec->ctdb_ctx = ctx;
946
947 result->key.dsize = key.dsize;
948 result->key.dptr = (uint8 *)talloc_memdup(result, key.dptr, key.dsize);
949 if (result->key.dptr == NULL) {
950 DEBUG(0, ("talloc failed\n"));
951 TALLOC_FREE(result);
952 return NULL;
953 }
954
955 /*
956 * Do a blocking lock on the record
957 */
958 again:
959
960 if (DEBUGLEVEL >= 10) {
961 char *keystr = hex_encode_talloc(result, key.dptr, key.dsize);
962 DEBUG(10, (DEBUGLEVEL > 10
963 ? "Locking db %u key %s\n"
964 : "Locking db %u key %.20s\n",
965 (int)crec->ctdb_ctx->db_id, keystr));
966 TALLOC_FREE(keystr);
967 }
968
969 if (tdb_chainlock(ctx->wtdb->tdb, key) != 0) {
970 DEBUG(3, ("tdb_chainlock failed\n"));
971 TALLOC_FREE(result);
972 return NULL;
973 }
974
975 result->store = db_ctdb_store;
976 result->delete_rec = db_ctdb_delete;
977 talloc_set_destructor(result, db_ctdb_record_destr);
978
979 ctdb_data = tdb_fetch(ctx->wtdb->tdb, key);
980
981 /*
982 * See if we have a valid record and we are the dmaster. If so, we can
983 * take the shortcut and just return it.
984 */
985
986 if ((ctdb_data.dptr == NULL) ||
987 (ctdb_data.dsize < sizeof(struct ctdb_ltdb_header)) ||
988 ((struct ctdb_ltdb_header *)ctdb_data.dptr)->dmaster != get_my_vnn()
989 #if 0
990 || (random() % 2 != 0)
991 #endif
992 ) {
993 SAFE_FREE(ctdb_data.dptr);
994 tdb_chainunlock(ctx->wtdb->tdb, key);
995 talloc_set_destructor(result, NULL);
996
997 migrate_attempts += 1;
998
999 DEBUG(10, ("ctdb_data.dptr = %p, dmaster = %u (%u)\n",
1000 ctdb_data.dptr, ctdb_data.dptr ?
1001 ((struct ctdb_ltdb_header *)ctdb_data.dptr)->dmaster : -1,
1002 get_my_vnn()));
1003
1004 status = ctdbd_migrate(messaging_ctdbd_connection(),ctx->db_id, key);
1005 if (!NT_STATUS_IS_OK(status)) {
1006 DEBUG(5, ("ctdb_migrate failed: %s\n",
1007 nt_errstr(status)));
1008 TALLOC_FREE(result);
1009 return NULL;
1010 }
1011 /* now its migrated, try again */
1012 goto again;
1013 }
1014
1015 if (migrate_attempts > 10) {
1016 DEBUG(0, ("db_ctdb_fetch_locked needed %d attempts\n",
1017 migrate_attempts));
1018 }
1019
1020 memcpy(&crec->header, ctdb_data.dptr, sizeof(crec->header));
1021
1022 result->value.dsize = ctdb_data.dsize - sizeof(crec->header);
1023 result->value.dptr = NULL;
1024
1025 if ((result->value.dsize != 0)
1026 && !(result->value.dptr = (uint8 *)talloc_memdup(
1027 result, ctdb_data.dptr + sizeof(crec->header),
1028 result->value.dsize))) {
1029 DEBUG(0, ("talloc failed\n"));
1030 TALLOC_FREE(result);
1031 }
1032
1033 SAFE_FREE(ctdb_data.dptr);
1034
1035 return result;
1036 }
1037
1038 static struct db_record *db_ctdb_fetch_locked(struct db_context *db,
1039 TALLOC_CTX *mem_ctx,
1040 TDB_DATA key)
1041 {
1042 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
1043 struct db_ctdb_ctx);
1044
1045 if (ctx->transaction != NULL) {
1046 return db_ctdb_fetch_locked_transaction(ctx, mem_ctx, key);
1047 }
1048
1049 if (db->persistent) {
1050 return db_ctdb_fetch_locked_persistent(ctx, mem_ctx, key);
1051 }
1052
1053 return fetch_locked_internal(ctx, mem_ctx, key, false);
1054 }
1055
1056 /*
1057 fetch (unlocked, no migration) operation on ctdb
1058 */
1059 static int db_ctdb_fetch(struct db_context *db, TALLOC_CTX *mem_ctx,
1060 TDB_DATA key, TDB_DATA *data)
1061 {
1062 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
1063 struct db_ctdb_ctx);
1064 NTSTATUS status;
1065 TDB_DATA ctdb_data;
1066
1067 if (ctx->transaction) {
1068 return db_ctdb_transaction_fetch(ctx, mem_ctx, key, data);
1069 }
1070
1071 /* try a direct fetch */
1072 ctdb_data = tdb_fetch(ctx->wtdb->tdb, key);
1073
1074 /*
1075 * See if we have a valid record and we are the dmaster. If so, we can
1076 * take the shortcut and just return it.
1077 * we bypass the dmaster check for persistent databases
1078 */
1079 if ((ctdb_data.dptr != NULL) &&
1080 (ctdb_data.dsize >= sizeof(struct ctdb_ltdb_header)) &&
1081 (db->persistent ||
1082 ((struct ctdb_ltdb_header *)ctdb_data.dptr)->dmaster == get_my_vnn())) {
1083 /* we are the dmaster - avoid the ctdb protocol op */
1084
1085 data->dsize = ctdb_data.dsize - sizeof(struct ctdb_ltdb_header);
1086 if (data->dsize == 0) {
1087 SAFE_FREE(ctdb_data.dptr);
1088 data->dptr = NULL;
1089 return 0;
1090 }
1091
1092 data->dptr = (uint8 *)talloc_memdup(
1093 mem_ctx, ctdb_data.dptr+sizeof(struct ctdb_ltdb_header),
1094 data->dsize);
1095
1096 SAFE_FREE(ctdb_data.dptr);
1097
1098 if (data->dptr == NULL) {
1099 return -1;
1100 }
1101 return 0;
1102 }
1103
1104 SAFE_FREE(ctdb_data.dptr);
1105
1106 /* we weren't able to get it locally - ask ctdb to fetch it for us */
1107 status = ctdbd_fetch(messaging_ctdbd_connection(),ctx->db_id, key, mem_ctx, data);
1108 if (!NT_STATUS_IS_OK(status)) {
1109 DEBUG(5, ("ctdbd_fetch failed: %s\n", nt_errstr(status)));
1110 return -1;
1111 }
1112
1113 return 0;
1114 }
1115
1116 struct traverse_state {
1117 struct db_context *db;
1118 int (*fn)(struct db_record *rec, void *private_data);
1119 void *private_data;
1120 };
1121
1122 static void traverse_callback(TDB_DATA key, TDB_DATA data, void *private_data)
1123 {
1124 struct traverse_state *state = (struct traverse_state *)private_data;
1125 struct db_record *rec;
1126 TALLOC_CTX *tmp_ctx = talloc_new(state->db);
1127 /* we have to give them a locked record to prevent races */
1128 rec = db_ctdb_fetch_locked(state->db, tmp_ctx, key);
1129 if (rec && rec->value.dsize > 0) {
1130 state->fn(rec, state->private_data);
1131 }
1132 talloc_free(tmp_ctx);
1133 }
1134
1135 static int traverse_persistent_callback(TDB_CONTEXT *tdb, TDB_DATA kbuf, TDB_DATA dbuf,
1136 void *private_data)
1137 {
1138 struct traverse_state *state = (struct traverse_state *)private_data;
1139 struct db_record *rec;
1140 TALLOC_CTX *tmp_ctx = talloc_new(state->db);
1141 int ret = 0;
1142 /* we have to give them a locked record to prevent races */
1143 rec = db_ctdb_fetch_locked(state->db, tmp_ctx, kbuf);
1144 if (rec && rec->value.dsize > 0) {
1145 ret = state->fn(rec, state->private_data);
1146 }
1147 talloc_free(tmp_ctx);
1148 return ret;
1149 }
1150
1151 static int db_ctdb_traverse(struct db_context *db,
1152 int (*fn)(struct db_record *rec,
1153 void *private_data),
1154 void *private_data)
1155 {
1156 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
1157 struct db_ctdb_ctx);
1158 struct traverse_state state;
1159
1160 state.db = db;
1161 state.fn = fn;
1162 state.private_data = private_data;
1163
1164 if (db->persistent) {
1165 /* for persistent databases we don't need to do a ctdb traverse,
1166 we can do a faster local traverse */
1167 return tdb_traverse(ctx->wtdb->tdb, traverse_persistent_callback, &state);
1168 }
1169
1170
1171 ctdbd_traverse(ctx->db_id, traverse_callback, &state);
1172 return 0;
1173 }
1174
1175 static NTSTATUS db_ctdb_store_deny(struct db_record *rec, TDB_DATA data, int flag)
1176 {
1177 return NT_STATUS_MEDIA_WRITE_PROTECTED;
1178 }
1179
1180 static NTSTATUS db_ctdb_delete_deny(struct db_record *rec)
1181 {
1182 return NT_STATUS_MEDIA_WRITE_PROTECTED;
1183 }
1184
1185 static void traverse_read_callback(TDB_DATA key, TDB_DATA data, void *private_data)
1186 {
1187 struct traverse_state *state = (struct traverse_state *)private_data;
1188 struct db_record rec;
1189 rec.key = key;
1190 rec.value = data;
1191 rec.store = db_ctdb_store_deny;
1192 rec.delete_rec = db_ctdb_delete_deny;
1193 rec.private_data = state->db;
1194 state->fn(&rec, state->private_data);
1195 }
1196
1197 static int traverse_persistent_callback_read(TDB_CONTEXT *tdb, TDB_DATA kbuf, TDB_DATA dbuf,
1198 void *private_data)
1199 {
1200 struct traverse_state *state = (struct traverse_state *)private_data;
1201 struct db_record rec;
1202 rec.key = kbuf;
1203 rec.value = dbuf;
1204 rec.store = db_ctdb_store_deny;
1205 rec.delete_rec = db_ctdb_delete_deny;
1206 rec.private_data = state->db;
1207
1208 if (rec.value.dsize <= sizeof(struct ctdb_ltdb_header)) {
1209 /* a deleted record */
1210 return 0;
1211 }
1212 rec.value.dsize -= sizeof(struct ctdb_ltdb_header);
1213 rec.value.dptr += sizeof(struct ctdb_ltdb_header);
1214
1215 return state->fn(&rec, state->private_data);
1216 }
1217
1218 static int db_ctdb_traverse_read(struct db_context *db,
1219 int (*fn)(struct db_record *rec,
1220 void *private_data),
1221 void *private_data)
1222 {
1223 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
1224 struct db_ctdb_ctx);
1225 struct traverse_state state;
1226
1227 state.db = db;
1228 state.fn = fn;
1229 state.private_data = private_data;
1230
1231 if (db->persistent) {
1232 /* for persistent databases we don't need to do a ctdb traverse,
1233 we can do a faster local traverse */
1234 return tdb_traverse_read(ctx->wtdb->tdb, traverse_persistent_callback_read, &state);
1235 }
1236
1237 ctdbd_traverse(ctx->db_id, traverse_read_callback, &state);
1238 return 0;
1239 }
1240
1241 static int db_ctdb_get_seqnum(struct db_context *db)
1242 {
1243 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
1244 struct db_ctdb_ctx);
1245 return tdb_get_seqnum(ctx->wtdb->tdb);
1246 }
1247
1248 static int db_ctdb_get_flags(struct db_context *db)
1249 {
1250 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
1251 struct db_ctdb_ctx);
1252 return tdb_get_flags(ctx->wtdb->tdb);
1253 }
1254
1255 struct db_context *db_open_ctdb(TALLOC_CTX *mem_ctx,
1256 const char *name,
1257 int hash_size, int tdb_flags,
1258 int open_flags, mode_t mode)
1259 {
1260 struct db_context *result;
1261 struct db_ctdb_ctx *db_ctdb;
1262 char *db_path;
1263 struct ctdbd_connection *conn;
1264
1265 if (!lp_clustering()) {
1266 DEBUG(10, ("Clustering disabled -- no ctdb\n"));
1267 return NULL;
1268 }
1269
1270 if (!(result = TALLOC_ZERO_P(mem_ctx, struct db_context))) {
1271 DEBUG(0, ("talloc failed\n"));
1272 TALLOC_FREE(result);
1273 return NULL;
1274 }
1275
1276 if (!(db_ctdb = TALLOC_P(result, struct db_ctdb_ctx))) {
1277 DEBUG(0, ("talloc failed\n"));
1278 TALLOC_FREE(result);
1279 return NULL;
1280 }
1281
1282 db_ctdb->transaction = NULL;
1283 db_ctdb->db = result;
1284
1285 conn = messaging_ctdbd_connection();
1286
1287 if (!NT_STATUS_IS_OK(ctdbd_db_attach(conn, name, &db_ctdb->db_id, tdb_flags))) {
1288 DEBUG(0, ("ctdbd_db_attach failed for %s\n", name));
1289 TALLOC_FREE(result);
1290 return NULL;
1291 }
1292
1293 db_path = ctdbd_dbpath(conn, db_ctdb, db_ctdb->db_id);
1294
1295 result->persistent = ((tdb_flags & TDB_CLEAR_IF_FIRST) == 0);
1296
1297 /* only pass through specific flags */
1298 tdb_flags &= TDB_SEQNUM;
1299
1300 /* honor permissions if user has specified O_CREAT */
1301 if (open_flags & O_CREAT) {
1302 chmod(db_path, mode);
1303 }
1304
1305 db_ctdb->wtdb = tdb_wrap_open(db_ctdb, db_path, hash_size, tdb_flags, O_RDWR, 0);
1306 if (db_ctdb->wtdb == NULL) {
1307 DEBUG(0, ("Could not open tdb %s: %s\n", db_path, strerror(errno)));
1308 TALLOC_FREE(result);
1309 return NULL;
1310 }
1311 talloc_free(db_path);
1312
1313 if (result->persistent) {
1314 db_ctdb->lock_ctx = g_lock_ctx_init(db_ctdb,
1315 ctdb_conn_msg_ctx(conn));
1316 if (db_ctdb->lock_ctx == NULL) {
1317 DEBUG(0, ("g_lock_ctx_init failed\n"));
1318 TALLOC_FREE(result);
1319 return NULL;
1320 }
1321 }
1322
1323 result->private_data = (void *)db_ctdb;
1324 result->fetch_locked = db_ctdb_fetch_locked;
1325 result->fetch = db_ctdb_fetch;
1326 result->traverse = db_ctdb_traverse;
1327 result->traverse_read = db_ctdb_traverse_read;
1328 result->get_seqnum = db_ctdb_get_seqnum;
1329 result->get_flags = db_ctdb_get_flags;
1330 result->transaction_start = db_ctdb_transaction_start;
1331 result->transaction_commit = db_ctdb_transaction_commit;
1332 result->transaction_cancel = db_ctdb_transaction_cancel;
1333
1334 DEBUG(3,("db_open_ctdb: opened database '%s' with dbid 0x%x\n",
1335 name, db_ctdb->db_id));
1336
1337 return result;
1338 }
1339 #endif
Samba GIT mirror