?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
ctdb_conn: Use messaging_ctdb_connection
[Samba.git] / source3 / lib / dbwrap / dbwrap_ctdb.c
blob25db140583fe851bf8da74ca8f78291740aeefc7
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 #include "system/filesys.h"
23 #include "lib/tdb_wrap/tdb_wrap.h"
24 #include "util_tdb.h"
25 #include "dbwrap/dbwrap.h"
26 #include "dbwrap/dbwrap_ctdb.h"
27 #include "dbwrap/dbwrap_rbt.h"
28 #include "lib/param/param.h"
29
30 #include "ctdb/include/ctdb_protocol.h"
31 #include "ctdbd_conn.h"
32 #include "dbwrap/dbwrap.h"
33 #include "dbwrap/dbwrap_private.h"
34 #include "dbwrap/dbwrap_ctdb.h"
35 #include "g_lock.h"
36 #include "messages.h"
37 #include "messages_ctdb.h"
38 #include "lib/cluster_support.h"
39 #include "lib/util/tevent_ntstatus.h"
40
41 struct db_ctdb_transaction_handle {
42 struct db_ctdb_ctx *ctx;
43 /*
44 * we store the writes done under a transaction:
45 */
46 struct ctdb_marshall_buffer *m_write;
47 uint32_t nesting;
48 bool nested_cancel;
49 char *lock_name;
50 };
51
52 struct db_ctdb_ctx {
53 struct db_context *db;
54 struct tdb_wrap *wtdb;
55 uint32_t db_id;
56 struct db_ctdb_transaction_handle *transaction;
57 struct g_lock_ctx *lock_ctx;
58
59 /* thresholds for warning messages */
60 int warn_unlock_msecs;
61 int warn_migrate_msecs;
62 int warn_migrate_attempts;
63 int warn_locktime_msecs;
64 };
65
66 struct db_ctdb_rec {
67 struct db_ctdb_ctx *ctdb_ctx;
68 struct ctdb_ltdb_header header;
69 struct timeval lock_time;
70 };
71
72 struct ctdb_async_ctx {
73 bool initialized;
74 struct ctdbd_connection *async_conn;
75 };
76
77 static struct ctdb_async_ctx ctdb_async_ctx;
78
79 static int ctdb_async_ctx_init_internal(TALLOC_CTX *mem_ctx,
80 struct tevent_context *ev,
81 bool reinit)
82 {
83 int ret;
84
85 if (reinit) {
86 TALLOC_FREE(ctdb_async_ctx.async_conn);
87 ctdb_async_ctx.initialized = false;
88 }
89
90 if (ctdb_async_ctx.initialized) {
91 return 0;
92 }
93
94 become_root();
95 ret = ctdbd_init_connection(mem_ctx,
96 lp_ctdbd_socket(),
97 lp_ctdb_timeout(),
98 &ctdb_async_ctx.async_conn);
99 unbecome_root();
100
101 if (ctdb_async_ctx.async_conn == NULL) {
102 DBG_ERR("ctdbd_init_connection failed\n");
103 return EIO;
104 }
105
106 ret = ctdbd_setup_fde(ctdb_async_ctx.async_conn, ev);
107 if (ret != 0) {
108 DBG_ERR("ctdbd_setup_fde failed\n");
109 TALLOC_FREE(ctdb_async_ctx.async_conn);
110 return ret;
111 }
112
113 ctdb_async_ctx.initialized = true;
114 return 0;
115 }
116
117 static int ctdb_async_ctx_init(TALLOC_CTX *mem_ctx, struct tevent_context *ev)
118 {
119 return ctdb_async_ctx_init_internal(mem_ctx, ev, false);
120 }
121
122 int ctdb_async_ctx_reinit(TALLOC_CTX *mem_ctx, struct tevent_context *ev)
123 {
124 return ctdb_async_ctx_init_internal(mem_ctx, ev, true);
125 }
126
127 static NTSTATUS tdb_error_to_ntstatus(struct tdb_context *tdb)
128 {
129 enum TDB_ERROR tret = tdb_error(tdb);
130
131 return map_nt_error_from_tdb(tret);
132 }
133
134 struct db_ctdb_ltdb_parse_state {
135 void (*parser)(TDB_DATA key, struct ctdb_ltdb_header *header,
136 TDB_DATA data, void *private_data);
137 void *private_data;
138 };
139
140 static int db_ctdb_ltdb_parser(TDB_DATA key, TDB_DATA data,
141 void *private_data)
142 {
143 struct db_ctdb_ltdb_parse_state *state =
144 (struct db_ctdb_ltdb_parse_state *)private_data;
145
146 if (data.dsize < sizeof(struct ctdb_ltdb_header)) {
147 return -1;
148 }
149
150 state->parser(
151 key, (struct ctdb_ltdb_header *)data.dptr,
152 make_tdb_data(data.dptr + sizeof(struct ctdb_ltdb_header),
153 data.dsize - sizeof(struct ctdb_ltdb_header)),
154 state->private_data);
155 return 0;
156 }
157
158 static NTSTATUS db_ctdb_ltdb_parse(
159 struct db_ctdb_ctx *db, TDB_DATA key,
160 void (*parser)(TDB_DATA key, struct ctdb_ltdb_header *header,
161 TDB_DATA data, void *private_data),
162 void *private_data)
163 {
164 struct db_ctdb_ltdb_parse_state state;
165 int ret;
166
167 state.parser = parser;
168 state.private_data = private_data;
169
170 ret = tdb_parse_record(db->wtdb->tdb, key, db_ctdb_ltdb_parser,
171 &state);
172 if (ret == -1) {
173 return NT_STATUS_NOT_FOUND;
174 }
175 return NT_STATUS_OK;
176 }
177
178 /*
179 * Store a record together with the ctdb record header
180 * in the local copy of the database.
181 */
182 static NTSTATUS db_ctdb_ltdb_store(struct db_ctdb_ctx *db,
183 TDB_DATA key,
184 struct ctdb_ltdb_header *header,
185 const TDB_DATA *dbufs, int num_dbufs)
186 {
187 TDB_DATA recs[num_dbufs+1];
188 int ret;
189
190 recs[0] = (TDB_DATA) { .dptr = (uint8_t *)header,
191 .dsize = sizeof(struct ctdb_ltdb_header) };
192 memcpy(&recs[1], dbufs, sizeof(TDB_DATA) * num_dbufs);
193
194 ret = tdb_storev(db->wtdb->tdb, key, recs, num_dbufs + 1, TDB_REPLACE);
195
196 return (ret == 0) ? NT_STATUS_OK
197 : tdb_error_to_ntstatus(db->wtdb->tdb);
198
199 }
200
201 /*
202 form a ctdb_rec_data record from a key/data pair
203 */
204 static struct ctdb_rec_data_old *db_ctdb_marshall_record(TALLOC_CTX *mem_ctx, uint32_t reqid,
205 TDB_DATA key,
206 struct ctdb_ltdb_header *header,
207 TDB_DATA data)
208 {
209 size_t length;
210 struct ctdb_rec_data_old *d;
211
212 length = offsetof(struct ctdb_rec_data_old, data) + key.dsize +
213 data.dsize + sizeof(*header);
214 d = (struct ctdb_rec_data_old *)talloc_size(mem_ctx, length);
215 if (d == NULL) {
216 return NULL;
217 }
218 d->length = length;
219 d->reqid = reqid;
220 d->keylen = key.dsize;
221 memcpy(&d->data[0], key.dptr, key.dsize);
222
223 d->datalen = data.dsize + sizeof(*header);
224 memcpy(&d->data[key.dsize], header, sizeof(*header));
225 memcpy(&d->data[key.dsize+sizeof(*header)], data.dptr, data.dsize);
226 return d;
227 }
228
229
230 /* helper function for marshalling multiple records */
231 static struct ctdb_marshall_buffer *db_ctdb_marshall_add(TALLOC_CTX *mem_ctx,
232 struct ctdb_marshall_buffer *m,
233 uint32_t db_id,
234 uint32_t reqid,
235 TDB_DATA key,
236 struct ctdb_ltdb_header *header,
237 TDB_DATA data)
238 {
239 struct ctdb_rec_data_old *r;
240 size_t m_size, r_size;
241 struct ctdb_marshall_buffer *m2 = NULL;
242
243 r = db_ctdb_marshall_record(talloc_tos(), reqid, key, header, data);
244 if (r == NULL) {
245 talloc_free(m);
246 return NULL;
247 }
248
249 if (m == NULL) {
250 m = (struct ctdb_marshall_buffer *)talloc_zero_size(
251 mem_ctx, offsetof(struct ctdb_marshall_buffer, data));
252 if (m == NULL) {
253 goto done;
254 }
255 m->db_id = db_id;
256 }
257
258 m_size = talloc_get_size(m);
259 r_size = talloc_get_size(r);
260
261 m2 = (struct ctdb_marshall_buffer *)talloc_realloc_size(
262 mem_ctx, m, m_size + r_size);
263 if (m2 == NULL) {
264 talloc_free(m);
265 goto done;
266 }
267
268 memcpy(m_size + (uint8_t *)m2, r, r_size);
269
270 m2->count++;
271
272 done:
273 talloc_free(r);
274 return m2;
275 }
276
277 /* we've finished marshalling, return a data blob with the marshalled records */
278 static TDB_DATA db_ctdb_marshall_finish(struct ctdb_marshall_buffer *m)
279 {
280 TDB_DATA data;
281 data.dptr = (uint8_t *)m;
282 data.dsize = talloc_get_size(m);
283 return data;
284 }
285
286 /*
287 loop over a marshalling buffer
288
289 - pass r==NULL to start
290 - loop the number of times indicated by m->count
291 */
292 static struct ctdb_rec_data_old *db_ctdb_marshall_loop_next_key(
293 struct ctdb_marshall_buffer *m, struct ctdb_rec_data_old *r, TDB_DATA *key)
294 {
295 if (r == NULL) {
296 r = (struct ctdb_rec_data_old *)&m->data[0];
297 } else {
298 r = (struct ctdb_rec_data_old *)(r->length + (uint8_t *)r);
299 }
300
301 key->dptr = &r->data[0];
302 key->dsize = r->keylen;
303 return r;
304 }
305
306 static bool db_ctdb_marshall_buf_parse(
307 struct ctdb_rec_data_old *r, uint32_t *reqid,
308 struct ctdb_ltdb_header **header, TDB_DATA *data)
309 {
310 if (r->datalen < sizeof(struct ctdb_ltdb_header)) {
311 return false;
312 }
313
314 *reqid = r->reqid;
315
316 data->dptr = &r->data[r->keylen] + sizeof(struct ctdb_ltdb_header);
317 data->dsize = r->datalen - sizeof(struct ctdb_ltdb_header);
318
319 *header = (struct ctdb_ltdb_header *)&r->data[r->keylen];
320
321 return true;
322 }
323
324 /**
325 * CTDB transaction destructor
326 */
327 static int db_ctdb_transaction_destructor(struct db_ctdb_transaction_handle *h)
328 {
329 NTSTATUS status;
330
331 status = g_lock_unlock(h->ctx->lock_ctx, h->lock_name);
332 if (!NT_STATUS_IS_OK(status)) {
333 DEBUG(0, ("g_lock_unlock failed for %s: %s\n", h->lock_name,
334 nt_errstr(status)));
335 return -1;
336 }
337 return 0;
338 }
339
340 /**
341 * CTDB dbwrap API: transaction_start function
342 * starts a transaction on a persistent database
343 */
344 static int db_ctdb_transaction_start(struct db_context *db)
345 {
346 struct db_ctdb_transaction_handle *h;
347 NTSTATUS status;
348 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
349 struct db_ctdb_ctx);
350
351 if (!db->persistent) {
352 DEBUG(0,("transactions not supported on non-persistent database 0x%08x\n",
353 ctx->db_id));
354 return -1;
355 }
356
357 if (ctx->transaction) {
358 ctx->transaction->nesting++;
359 DEBUG(5, (__location__ " transaction start on db 0x%08x: nesting %d -> %d\n",
360 ctx->db_id, ctx->transaction->nesting - 1, ctx->transaction->nesting));
361 return 0;
362 }
363
364 h = talloc_zero(db, struct db_ctdb_transaction_handle);
365 if (h == NULL) {
366 DEBUG(0,(__location__ " oom for transaction handle\n"));
367 return -1;
368 }
369
370 h->ctx = ctx;
371
372 h->lock_name = talloc_asprintf(h, "transaction_db_0x%08x",
373 (unsigned int)ctx->db_id);
374 if (h->lock_name == NULL) {
375 DEBUG(0, ("talloc_asprintf failed\n"));
376 TALLOC_FREE(h);
377 return -1;
378 }
379
380 /*
381 * Wait a day, i.e. forever...
382 */
383 status = g_lock_lock(ctx->lock_ctx, h->lock_name, G_LOCK_WRITE,
384 timeval_set(86400, 0));
385 if (!NT_STATUS_IS_OK(status)) {
386 DEBUG(0, ("g_lock_lock failed: %s\n", nt_errstr(status)));
387 TALLOC_FREE(h);
388 return -1;
389 }
390
391 talloc_set_destructor(h, db_ctdb_transaction_destructor);
392
393 ctx->transaction = h;
394
395 DEBUG(5,(__location__ " transaction started on db 0x%08x\n", ctx->db_id));
396
397 return 0;
398 }
399
400 static bool parse_newest_in_marshall_buffer(
401 struct ctdb_marshall_buffer *buf, TDB_DATA key,
402 void (*parser)(TDB_DATA key, struct ctdb_ltdb_header *header,
403 TDB_DATA data, void *private_data),
404 void *private_data)
405 {
406 struct ctdb_rec_data_old *rec = NULL;
407 struct ctdb_ltdb_header *h = NULL;
408 TDB_DATA data;
409 uint32_t i;
410
411 if (buf == NULL) {
412 return false;
413 }
414
415 /*
416 * Walk the list of records written during this
417 * transaction. If we want to read one we have already
418 * written, return the last written sample. Thus we do not do
419 * a "break;" for the first hit, this record might have been
420 * overwritten later.
421 */
422
423 for (i=0; i<buf->count; i++) {
424 TDB_DATA tkey;
425 uint32_t reqid;
426
427 rec = db_ctdb_marshall_loop_next_key(buf, rec, &tkey);
428 if (rec == NULL) {
429 return false;
430 }
431
432 if (!tdb_data_equal(key, tkey)) {
433 continue;
434 }
435
436 if (!db_ctdb_marshall_buf_parse(rec, &reqid, &h, &data)) {
437 return false;
438 }
439 }
440
441 if (h == NULL) {
442 return false;
443 }
444
445 parser(key, h, data, private_data);
446
447 return true;
448 }
449
450 struct pull_newest_from_marshall_buffer_state {
451 struct ctdb_ltdb_header *pheader;
452 TALLOC_CTX *mem_ctx;
453 TDB_DATA *pdata;
454 };
455
456 static void pull_newest_from_marshall_buffer_parser(
457 TDB_DATA key, struct ctdb_ltdb_header *header,
458 TDB_DATA data, void *private_data)
459 {
460 struct pull_newest_from_marshall_buffer_state *state =
461 (struct pull_newest_from_marshall_buffer_state *)private_data;
462
463 if (state->pheader != NULL) {
464 memcpy(state->pheader, header, sizeof(*state->pheader));
465 }
466 if (state->pdata != NULL) {
467 state->pdata->dsize = data.dsize;
468 state->pdata->dptr = (uint8_t *)talloc_memdup(
469 state->mem_ctx, data.dptr, data.dsize);
470 }
471 }
472
473 static bool pull_newest_from_marshall_buffer(struct ctdb_marshall_buffer *buf,
474 TDB_DATA key,
475 struct ctdb_ltdb_header *pheader,
476 TALLOC_CTX *mem_ctx,
477 TDB_DATA *pdata)
478 {
479 struct pull_newest_from_marshall_buffer_state state;
480
481 state.pheader = pheader;
482 state.mem_ctx = mem_ctx;
483 state.pdata = pdata;
484
485 if (!parse_newest_in_marshall_buffer(
486 buf, key, pull_newest_from_marshall_buffer_parser,
487 &state)) {
488 return false;
489 }
490 if ((pdata != NULL) && (pdata->dsize != 0) && (pdata->dptr == NULL)) {
491 /* ENOMEM */
492 return false;
493 }
494 return true;
495 }
496
497 static NTSTATUS db_ctdb_storev_transaction(struct db_record *rec,
498 const TDB_DATA *dbufs, int num_dbufs,
499 int flag);
500 static NTSTATUS db_ctdb_delete_transaction(struct db_record *rec);
501
502 static struct db_record *db_ctdb_fetch_locked_transaction(struct db_ctdb_ctx *ctx,
503 TALLOC_CTX *mem_ctx,
504 TDB_DATA key)
505 {
506 struct db_record *result;
507 TDB_DATA ctdb_data;
508
509 if (!(result = talloc(mem_ctx, struct db_record))) {
510 DEBUG(0, ("talloc failed\n"));
511 return NULL;
512 }
513
514 result->db = ctx->db;
515 result->private_data = ctx->transaction;
516
517 result->key.dsize = key.dsize;
518 result->key.dptr = (uint8_t *)talloc_memdup(result, key.dptr,
519 key.dsize);
520 if (result->key.dptr == NULL) {
521 DEBUG(0, ("talloc failed\n"));
522 TALLOC_FREE(result);
523 return NULL;
524 }
525
526 result->storev = db_ctdb_storev_transaction;
527 result->delete_rec = db_ctdb_delete_transaction;
528
529 if (pull_newest_from_marshall_buffer(ctx->transaction->m_write, key,
530 NULL, result, &result->value)) {
531 return result;
532 }
533
534 ctdb_data = tdb_fetch(ctx->wtdb->tdb, key);
535 if (ctdb_data.dptr == NULL) {
536 /* create the record */
537 result->value = tdb_null;
538 return result;
539 }
540
541 result->value.dsize = ctdb_data.dsize - sizeof(struct ctdb_ltdb_header);
542 result->value.dptr = NULL;
543
544 if ((result->value.dsize != 0)
545 && !(result->value.dptr = (uint8_t *)talloc_memdup(
546 result, ctdb_data.dptr + sizeof(struct ctdb_ltdb_header),
547 result->value.dsize))) {
548 DEBUG(0, ("talloc failed\n"));
549 TALLOC_FREE(result);
550 }
551
552 SAFE_FREE(ctdb_data.dptr);
553
554 return result;
555 }
556
557 static int db_ctdb_record_destructor(struct db_record **recp)
558 {
559 struct db_record *rec = talloc_get_type_abort(*recp, struct db_record);
560 struct db_ctdb_transaction_handle *h = talloc_get_type_abort(
561 rec->private_data, struct db_ctdb_transaction_handle);
562 int ret = h->ctx->db->transaction_commit(h->ctx->db);
563 if (ret != 0) {
564 DEBUG(0,(__location__ " transaction_commit failed\n"));
565 }
566 return 0;
567 }
568
569 /*
570 auto-create a transaction for persistent databases
571 */
572 static struct db_record *db_ctdb_fetch_locked_persistent(struct db_ctdb_ctx *ctx,
573 TALLOC_CTX *mem_ctx,
574 TDB_DATA key)
575 {
576 int res;
577 struct db_record *rec, **recp;
578
579 res = db_ctdb_transaction_start(ctx->db);
580 if (res == -1) {
581 return NULL;
582 }
583
584 rec = db_ctdb_fetch_locked_transaction(ctx, mem_ctx, key);
585 if (rec == NULL) {
586 ctx->db->transaction_cancel(ctx->db);
587 return NULL;
588 }
589
590 /* destroy this transaction when we release the lock */
591 recp = talloc(rec, struct db_record *);
592 if (recp == NULL) {
593 ctx->db->transaction_cancel(ctx->db);
594 talloc_free(rec);
595 return NULL;
596 }
597 *recp = rec;
598 talloc_set_destructor(recp, db_ctdb_record_destructor);
599 return rec;
600 }
601
602
603 /*
604 stores a record inside a transaction
605 */
606 static NTSTATUS db_ctdb_transaction_store(struct db_ctdb_transaction_handle *h,
607 TDB_DATA key, TDB_DATA data)
608 {
609 TALLOC_CTX *tmp_ctx = talloc_new(h);
610 TDB_DATA rec;
611 struct ctdb_ltdb_header header;
612
613 ZERO_STRUCT(header);
614
615 /* we need the header so we can update the RSN */
616
617 if (!pull_newest_from_marshall_buffer(h->m_write, key, &header,
618 NULL, NULL)) {
619
620 rec = tdb_fetch(h->ctx->wtdb->tdb, key);
621
622 if (rec.dptr != NULL) {
623 memcpy(&header, rec.dptr,
624 sizeof(struct ctdb_ltdb_header));
625 rec.dsize -= sizeof(struct ctdb_ltdb_header);
626
627 /*
628 * a special case, we are writing the same
629 * data that is there now
630 */
631 if (data.dsize == rec.dsize &&
632 memcmp(data.dptr,
633 rec.dptr + sizeof(struct ctdb_ltdb_header),
634 data.dsize) == 0) {
635 SAFE_FREE(rec.dptr);
636 talloc_free(tmp_ctx);
637 return NT_STATUS_OK;
638 }
639 }
640 SAFE_FREE(rec.dptr);
641 }
642
643 header.dmaster = get_my_vnn();
644 header.rsn++;
645
646 h->m_write = db_ctdb_marshall_add(h, h->m_write, h->ctx->db_id, 0, key, &header, data);
647 if (h->m_write == NULL) {
648 DEBUG(0,(__location__ " Failed to add to marshalling record\n"));
649 talloc_free(tmp_ctx);
650 return NT_STATUS_NO_MEMORY;
651 }
652
653 talloc_free(tmp_ctx);
654 return NT_STATUS_OK;
655 }
656
657
658 /*
659 a record store inside a transaction
660 */
661 static NTSTATUS db_ctdb_storev_transaction(
662 struct db_record *rec, const TDB_DATA *dbufs, int num_dbufs, int flag)
663 {
664 struct db_ctdb_transaction_handle *h = talloc_get_type_abort(
665 rec->private_data, struct db_ctdb_transaction_handle);
666 NTSTATUS status;
667 TDB_DATA data;
668
669 data = dbwrap_merge_dbufs(rec, dbufs, num_dbufs);
670 if (data.dptr == NULL) {
671 return NT_STATUS_NO_MEMORY;
672 }
673
674 status = db_ctdb_transaction_store(h, rec->key, data);
675
676 TALLOC_FREE(data.dptr);
677
678 return status;
679 }
680
681 /*
682 a record delete inside a transaction
683 */
684 static NTSTATUS db_ctdb_delete_transaction(struct db_record *rec)
685 {
686 struct db_ctdb_transaction_handle *h = talloc_get_type_abort(
687 rec->private_data, struct db_ctdb_transaction_handle);
688 NTSTATUS status;
689
690 status = db_ctdb_transaction_store(h, rec->key, tdb_null);
691 return status;
692 }
693
694 static void db_ctdb_fetch_db_seqnum_parser(
695 TDB_DATA key, struct ctdb_ltdb_header *header,
696 TDB_DATA data, void *private_data)
697 {
698 uint64_t *seqnum = (uint64_t *)private_data;
699
700 if (data.dsize != sizeof(uint64_t)) {
701 *seqnum = 0;
702 return;
703 }
704 memcpy(seqnum, data.dptr, sizeof(*seqnum));
705 }
706
707 /**
708 * Fetch the db sequence number of a persistent db directly from the db.
709 */
710 static NTSTATUS db_ctdb_fetch_db_seqnum_from_db(struct db_ctdb_ctx *db,
711 uint64_t *seqnum)
712 {
713 NTSTATUS status;
714 TDB_DATA key;
715
716 if (seqnum == NULL) {
717 return NT_STATUS_INVALID_PARAMETER;
718 }
719
720 key = string_term_tdb_data(CTDB_DB_SEQNUM_KEY);
721
722 status = db_ctdb_ltdb_parse(
723 db, key, db_ctdb_fetch_db_seqnum_parser, seqnum);
724
725 if (NT_STATUS_IS_OK(status)) {
726 return NT_STATUS_OK;
727 }
728 if (NT_STATUS_EQUAL(status, NT_STATUS_NOT_FOUND)) {
729 *seqnum = 0;
730 return NT_STATUS_OK;
731 }
732 return status;
733 }
734
735 /**
736 * Store the database sequence number inside a transaction.
737 */
738 static NTSTATUS db_ctdb_store_db_seqnum(struct db_ctdb_transaction_handle *h,
739 uint64_t seqnum)
740 {
741 NTSTATUS status;
742 const char *keyname = CTDB_DB_SEQNUM_KEY;
743 TDB_DATA key;
744 TDB_DATA data;
745
746 key = string_term_tdb_data(keyname);
747
748 data.dptr = (uint8_t *)&seqnum;
749 data.dsize = sizeof(uint64_t);
750
751 status = db_ctdb_transaction_store(h, key, data);
752
753 return status;
754 }
755
756 /*
757 commit a transaction
758 */
759 static int db_ctdb_transaction_commit(struct db_context *db)
760 {
761 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
762 struct db_ctdb_ctx);
763 NTSTATUS rets;
764 int32_t status;
765 struct db_ctdb_transaction_handle *h = ctx->transaction;
766 uint64_t old_seqnum, new_seqnum;
767 int ret;
768
769 if (h == NULL) {
770 DEBUG(0,(__location__ " transaction commit with no open transaction on db 0x%08x\n", ctx->db_id));
771 return -1;
772 }
773
774 if (h->nested_cancel) {
775 db->transaction_cancel(db);
776 DEBUG(5,(__location__ " Failed transaction commit after nested cancel\n"));
777 return -1;
778 }
779
780 if (h->nesting != 0) {
781 h->nesting--;
782 DEBUG(5, (__location__ " transaction commit on db 0x%08x: nesting %d -> %d\n",
783 ctx->db_id, ctx->transaction->nesting + 1, ctx->transaction->nesting));
784 return 0;
785 }
786
787 if (h->m_write == NULL) {
788 /*
789 * No changes were made, so don't change the seqnum,
790 * don't push to other node, just exit with success.
791 */
792 ret = 0;
793 goto done;
794 }
795
796 DEBUG(5,(__location__ " transaction commit on db 0x%08x\n", ctx->db_id));
797
798 /*
799 * As the last db action before committing, bump the database sequence
800 * number. Note that this undoes all changes to the seqnum records
801 * performed under the transaction. This record is not meant to be
802 * modified by user interaction. It is for internal use only...
803 */
804 rets = db_ctdb_fetch_db_seqnum_from_db(ctx, &old_seqnum);
805 if (!NT_STATUS_IS_OK(rets)) {
806 DEBUG(1, (__location__ " failed to fetch the db sequence number "
807 "in transaction commit on db 0x%08x\n", ctx->db_id));
808 ret = -1;
809 goto done;
810 }
811
812 new_seqnum = old_seqnum + 1;
813
814 rets = db_ctdb_store_db_seqnum(h, new_seqnum);
815 if (!NT_STATUS_IS_OK(rets)) {
816 DEBUG(1, (__location__ "failed to store the db sequence number "
817 " in transaction commit on db 0x%08x\n", ctx->db_id));
818 ret = -1;
819 goto done;
820 }
821
822 again:
823 /* tell ctdbd to commit to the other nodes */
824 ret = ctdbd_control_local(messaging_ctdb_connection(),
825 CTDB_CONTROL_TRANS3_COMMIT,
826 h->ctx->db_id, 0,
827 db_ctdb_marshall_finish(h->m_write),
828 NULL, NULL, &status);
829 if ((ret != 0) || status != 0) {
830 /*
831 * The TRANS3_COMMIT control should only possibly fail when a
832 * recovery has been running concurrently. In any case, the db
833 * will be the same on all nodes, either the new copy or the
834 * old copy. This can be detected by comparing the old and new
835 * local sequence numbers.
836 */
837 rets = db_ctdb_fetch_db_seqnum_from_db(ctx, &new_seqnum);
838 if (!NT_STATUS_IS_OK(rets)) {
839 DEBUG(1, (__location__ " failed to refetch db sequence "
840 "number after failed TRANS3_COMMIT\n"));
841 ret = -1;
842 goto done;
843 }
844
845 if (new_seqnum == old_seqnum) {
846 /* Recovery prevented all our changes: retry. */
847 goto again;
848 }
849 if (new_seqnum != (old_seqnum + 1)) {
850 DEBUG(0, (__location__ " ERROR: new_seqnum[%lu] != "
851 "old_seqnum[%lu] + (0 or 1) after failed "
852 "TRANS3_COMMIT - this should not happen!\n",
853 (unsigned long)new_seqnum,
854 (unsigned long)old_seqnum));
855 ret = -1;
856 goto done;
857 }
858 /*
859 * Recovery propagated our changes to all nodes, completing
860 * our commit for us - succeed.
861 */
862 }
863
864 ret = 0;
865
866 done:
867 h->ctx->transaction = NULL;
868 talloc_free(h);
869 return ret;
870 }
871
872
873 /*
874 cancel a transaction
875 */
876 static int db_ctdb_transaction_cancel(struct db_context *db)
877 {
878 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
879 struct db_ctdb_ctx);
880 struct db_ctdb_transaction_handle *h = ctx->transaction;
881
882 if (h == NULL) {
883 DEBUG(0,(__location__ " transaction cancel with no open transaction on db 0x%08x\n", ctx->db_id));
884 return -1;
885 }
886
887 if (h->nesting != 0) {
888 h->nesting--;
889 h->nested_cancel = true;
890 DEBUG(5, (__location__ " transaction cancel on db 0x%08x: nesting %d -> %d\n",
891 ctx->db_id, ctx->transaction->nesting + 1, ctx->transaction->nesting));
892 return 0;
893 }
894
895 DEBUG(5,(__location__ " Cancel transaction on db 0x%08x\n", ctx->db_id));
896
897 ctx->transaction = NULL;
898 talloc_free(h);
899 return 0;
900 }
901
902
903 static NTSTATUS db_ctdb_storev(struct db_record *rec,
904 const TDB_DATA *dbufs, int num_dbufs, int flag)
905 {
906 struct db_ctdb_rec *crec = talloc_get_type_abort(
907 rec->private_data, struct db_ctdb_rec);
908 NTSTATUS status;
909
910 status = db_ctdb_ltdb_store(crec->ctdb_ctx, rec->key, &(crec->header),
911 dbufs, num_dbufs);
912 return status;
913 }
914
915
916
917 static NTSTATUS db_ctdb_send_schedule_for_deletion(struct db_record *rec)
918 {
919 NTSTATUS status = NT_STATUS_OK;
920 int ret;
921 struct ctdb_control_schedule_for_deletion *dd;
922 TDB_DATA indata;
923 int32_t cstatus;
924 struct db_ctdb_rec *crec = talloc_get_type_abort(
925 rec->private_data, struct db_ctdb_rec);
926 struct db_ctdb_ctx *ctx = crec->ctdb_ctx;
927
928 indata.dsize = offsetof(struct ctdb_control_schedule_for_deletion, key) + rec->key.dsize;
929 indata.dptr = talloc_zero_array(crec, uint8_t, indata.dsize);
930 if (indata.dptr == NULL) {
931 DEBUG(0, (__location__ " talloc failed!\n"));
932 return NT_STATUS_NO_MEMORY;
933 }
934
935 dd = (struct ctdb_control_schedule_for_deletion *)(void *)indata.dptr;
936 dd->db_id = ctx->db_id;
937 dd->hdr = crec->header;
938 dd->keylen = rec->key.dsize;
939 memcpy(dd->key, rec->key.dptr, rec->key.dsize);
940
941 ret = ctdbd_control_local(messaging_ctdb_connection(),
942 CTDB_CONTROL_SCHEDULE_FOR_DELETION,
943 crec->ctdb_ctx->db_id,
944 CTDB_CTRL_FLAG_NOREPLY, /* flags */
945 indata,
946 NULL, /* outdata */
947 NULL, /* errmsg */
948 &cstatus);
949 talloc_free(indata.dptr);
950
951 if ((ret != 0) || cstatus != 0) {
952 DEBUG(1, (__location__ " Error sending local control "
953 "SCHEDULE_FOR_DELETION: %s, cstatus = %"PRIi32"\n",
954 strerror(ret), cstatus));
955 if (ret != 0) {
956 status = map_nt_error_from_unix(ret);
957 } else {
958 status = NT_STATUS_UNSUCCESSFUL;
959 }
960 }
961
962 return status;
963 }
964
965 static NTSTATUS db_ctdb_delete(struct db_record *rec)
966 {
967 NTSTATUS status;
968
969 /*
970 * We have to store the header with empty data. TODO: Fix the
971 * tdb-level cleanup
972 */
973
974 status = db_ctdb_storev(rec, &tdb_null, 1, 0);
975 if (!NT_STATUS_IS_OK(status)) {
976 return status;
977 }
978
979 status = db_ctdb_send_schedule_for_deletion(rec);
980 return status;
981 }
982
983 static int db_ctdb_record_destr(struct db_record* data)
984 {
985 struct db_ctdb_rec *crec = talloc_get_type_abort(
986 data->private_data, struct db_ctdb_rec);
987 int threshold;
988 int ret;
989 struct timeval before;
990 double timediff;
991
992 DEBUG(10, (DEBUGLEVEL > 10
993 ? "Unlocking db %u key %s\n"
994 : "Unlocking db %u key %.20s\n",
995 (int)crec->ctdb_ctx->db_id,
996 hex_encode_talloc(data, (unsigned char *)data->key.dptr,
997 data->key.dsize)));
998
999 before = timeval_current();
1000
1001 ret = tdb_chainunlock(crec->ctdb_ctx->wtdb->tdb, data->key);
1002
1003 timediff = timeval_elapsed(&before);
1004 timediff *= 1000; /* get us milliseconds */
1005
1006 if (timediff > crec->ctdb_ctx->warn_unlock_msecs) {
1007 char *key;
1008 key = hex_encode_talloc(talloc_tos(),
1009 (unsigned char *)data->key.dptr,
1010 data->key.dsize);
1011 DEBUG(0, ("tdb_chainunlock on db %s, key %s took %f milliseconds\n",
1012 tdb_name(crec->ctdb_ctx->wtdb->tdb), key,
1013 timediff));
1014 TALLOC_FREE(key);
1015 }
1016
1017 if (ret != 0) {
1018 DEBUG(0, ("tdb_chainunlock failed\n"));
1019 return -1;
1020 }
1021
1022 threshold = crec->ctdb_ctx->warn_locktime_msecs;
1023 if (threshold != 0) {
1024 timediff = timeval_elapsed(&crec->lock_time) * 1000;
1025 if (timediff > threshold) {
1026 const char *key;
1027
1028 key = hex_encode_talloc(data,
1029 (unsigned char *)data->key.dptr,
1030 data->key.dsize);
1031 DEBUG(0, ("Held tdb lock on db %s, key %s "
1032 "%f milliseconds\n",
1033 tdb_name(crec->ctdb_ctx->wtdb->tdb),
1034 key, timediff));
1035 }
1036 }
1037
1038 return 0;
1039 }
1040
1041 /**
1042 * Check whether we have a valid local copy of the given record,
1043 * either for reading or for writing.
1044 */
1045 static bool db_ctdb_can_use_local_hdr(const struct ctdb_ltdb_header *hdr,
1046 uint32_t my_vnn, bool read_only)
1047 {
1048 if (hdr->dmaster != my_vnn) {
1049 /* If we're not dmaster, it must be r/o copy. */
1050 return read_only && (hdr->flags & CTDB_REC_RO_HAVE_READONLY);
1051 }
1052
1053 /*
1054 * If we want write access, no one may have r/o copies.
1055 */
1056 return read_only || !(hdr->flags & CTDB_REC_RO_HAVE_DELEGATIONS);
1057 }
1058
1059 static bool db_ctdb_can_use_local_copy(TDB_DATA ctdb_data, uint32_t my_vnn,
1060 bool read_only)
1061 {
1062 if (ctdb_data.dptr == NULL) {
1063 return false;
1064 }
1065
1066 if (ctdb_data.dsize < sizeof(struct ctdb_ltdb_header)) {
1067 return false;
1068 }
1069
1070 return db_ctdb_can_use_local_hdr(
1071 (struct ctdb_ltdb_header *)ctdb_data.dptr, my_vnn, read_only);
1072 }
1073
1074 static struct db_record *fetch_locked_internal(struct db_ctdb_ctx *ctx,
1075 TALLOC_CTX *mem_ctx,
1076 TDB_DATA key,
1077 bool tryonly)
1078 {
1079 struct db_record *result;
1080 struct db_ctdb_rec *crec;
1081 TDB_DATA ctdb_data;
1082 int migrate_attempts;
1083 struct timeval migrate_start;
1084 struct timeval chainlock_start;
1085 struct timeval ctdb_start_time;
1086 double chainlock_time = 0;
1087 double ctdb_time = 0;
1088 int duration_msecs;
1089 int lockret;
1090 int ret;
1091
1092 if (!(result = talloc(mem_ctx, struct db_record))) {
1093 DEBUG(0, ("talloc failed\n"));
1094 return NULL;
1095 }
1096
1097 if (!(crec = talloc_zero(result, struct db_ctdb_rec))) {
1098 DEBUG(0, ("talloc failed\n"));
1099 TALLOC_FREE(result);
1100 return NULL;
1101 }
1102
1103 result->db = ctx->db;
1104 result->private_data = (void *)crec;
1105 crec->ctdb_ctx = ctx;
1106
1107 result->key.dsize = key.dsize;
1108 result->key.dptr = (uint8_t *)talloc_memdup(result, key.dptr,
1109 key.dsize);
1110 if (result->key.dptr == NULL) {
1111 DEBUG(0, ("talloc failed\n"));
1112 TALLOC_FREE(result);
1113 return NULL;
1114 }
1115
1116 migrate_attempts = 0;
1117 GetTimeOfDay(&migrate_start);
1118
1119 /*
1120 * Do a blocking lock on the record
1121 */
1122 again:
1123
1124 if (DEBUGLEVEL >= 10) {
1125 char *keystr = hex_encode_talloc(result, key.dptr, key.dsize);
1126 DEBUG(10, (DEBUGLEVEL > 10
1127 ? "Locking db %u key %s\n"
1128 : "Locking db %u key %.20s\n",
1129 (int)crec->ctdb_ctx->db_id, keystr));
1130 TALLOC_FREE(keystr);
1131 }
1132
1133 GetTimeOfDay(&chainlock_start);
1134 lockret = tryonly
1135 ? tdb_chainlock_nonblock(ctx->wtdb->tdb, key)
1136 : tdb_chainlock(ctx->wtdb->tdb, key);
1137 chainlock_time += timeval_elapsed(&chainlock_start);
1138
1139 if (lockret != 0) {
1140 DEBUG(3, ("tdb_chainlock failed\n"));
1141 TALLOC_FREE(result);
1142 return NULL;
1143 }
1144
1145 result->storev = db_ctdb_storev;
1146 result->delete_rec = db_ctdb_delete;
1147 talloc_set_destructor(result, db_ctdb_record_destr);
1148
1149 ctdb_data = tdb_fetch(ctx->wtdb->tdb, key);
1150
1151 /*
1152 * See if we have a valid record and we are the dmaster. If so, we can
1153 * take the shortcut and just return it.
1154 */
1155
1156 if (!db_ctdb_can_use_local_copy(ctdb_data, get_my_vnn(), false)) {
1157 SAFE_FREE(ctdb_data.dptr);
1158 tdb_chainunlock(ctx->wtdb->tdb, key);
1159 talloc_set_destructor(result, NULL);
1160
1161 if (tryonly && (migrate_attempts != 0)) {
1162 DEBUG(5, ("record migrated away again\n"));
1163 TALLOC_FREE(result);
1164 return NULL;
1165 }
1166
1167 migrate_attempts += 1;
1168
1169 DEBUG(10, ("ctdb_data.dptr = %p, dmaster = %"PRIu32" "
1170 "(%"PRIu32") %"PRIu32"\n",
1171 ctdb_data.dptr, ctdb_data.dptr ?
1172 ((struct ctdb_ltdb_header *)ctdb_data.dptr)->dmaster :
1173 UINT32_MAX,
1174 get_my_vnn(),
1175 ctdb_data.dptr ?
1176 ((struct ctdb_ltdb_header *)ctdb_data.dptr)->flags : 0));
1177
1178 GetTimeOfDay(&ctdb_start_time);
1179 ret = ctdbd_migrate(messaging_ctdb_connection(), ctx->db_id,
1180 key);
1181 ctdb_time += timeval_elapsed(&ctdb_start_time);
1182
1183 if (ret != 0) {
1184 DEBUG(5, ("ctdbd_migrate failed: %s\n",
1185 strerror(ret)));
1186 TALLOC_FREE(result);
1187 return NULL;
1188 }
1189 /* now its migrated, try again */
1190 goto again;
1191 }
1192
1193 {
1194 double duration;
1195 duration = timeval_elapsed(&migrate_start);
1196
1197 /*
1198 * Convert the duration to milliseconds to avoid a
1199 * floating-point division of
1200 * lp_parm_int("migrate_duration") by 1000.
1201 */
1202 duration_msecs = duration * 1000;
1203 }
1204
1205 if ((migrate_attempts > ctx->warn_migrate_attempts) ||
1206 (duration_msecs > ctx->warn_migrate_msecs)) {
1207 int chain = 0;
1208
1209 if (tdb_get_flags(ctx->wtdb->tdb) & TDB_INCOMPATIBLE_HASH) {
1210 chain = tdb_jenkins_hash(&key) %
1211 tdb_hash_size(ctx->wtdb->tdb);
1212 }
1213
1214 DEBUG(0, ("db_ctdb_fetch_locked for %s key %s, chain %d "
1215 "needed %d attempts, %d milliseconds, "
1216 "chainlock: %f ms, CTDB %f ms\n",
1217 tdb_name(ctx->wtdb->tdb),
1218 hex_encode_talloc(talloc_tos(),
1219 (unsigned char *)key.dptr,
1220 key.dsize),
1221 chain,
1222 migrate_attempts, duration_msecs,
1223 chainlock_time * 1000.0,
1224 ctdb_time * 1000.0));
1225 }
1226
1227 GetTimeOfDay(&crec->lock_time);
1228
1229 memcpy(&crec->header, ctdb_data.dptr, sizeof(crec->header));
1230
1231 result->value.dsize = ctdb_data.dsize - sizeof(crec->header);
1232 result->value.dptr = NULL;
1233
1234 if ((result->value.dsize != 0)
1235 && !(result->value.dptr = (uint8_t *)talloc_memdup(
1236 result, ctdb_data.dptr + sizeof(crec->header),
1237 result->value.dsize))) {
1238 DEBUG(0, ("talloc failed\n"));
1239 TALLOC_FREE(result);
1240 }
1241
1242 SAFE_FREE(ctdb_data.dptr);
1243
1244 return result;
1245 }
1246
1247 static struct db_record *db_ctdb_fetch_locked(struct db_context *db,
1248 TALLOC_CTX *mem_ctx,
1249 TDB_DATA key)
1250 {
1251 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
1252 struct db_ctdb_ctx);
1253
1254 if (ctx->transaction != NULL) {
1255 return db_ctdb_fetch_locked_transaction(ctx, mem_ctx, key);
1256 }
1257
1258 if (db->persistent) {
1259 return db_ctdb_fetch_locked_persistent(ctx, mem_ctx, key);
1260 }
1261
1262 return fetch_locked_internal(ctx, mem_ctx, key, false);
1263 }
1264
1265 static struct db_record *db_ctdb_try_fetch_locked(struct db_context *db,
1266 TALLOC_CTX *mem_ctx,
1267 TDB_DATA key)
1268 {
1269 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
1270 struct db_ctdb_ctx);
1271
1272 if (ctx->transaction != NULL) {
1273 return db_ctdb_fetch_locked_transaction(ctx, mem_ctx, key);
1274 }
1275
1276 if (db->persistent) {
1277 return db_ctdb_fetch_locked_persistent(ctx, mem_ctx, key);
1278 }
1279
1280 return fetch_locked_internal(ctx, mem_ctx, key, true);
1281 }
1282
1283 struct db_ctdb_parse_record_state {
1284 void (*parser)(TDB_DATA key, TDB_DATA data, void *private_data);
1285 void *private_data;
1286 uint32_t my_vnn;
1287 bool ask_for_readonly_copy;
1288 bool done;
1289 bool empty_record;
1290 };
1291
1292 static void db_ctdb_parse_record_parser(
1293 TDB_DATA key, struct ctdb_ltdb_header *header,
1294 TDB_DATA data, void *private_data)
1295 {
1296 struct db_ctdb_parse_record_state *state =
1297 (struct db_ctdb_parse_record_state *)private_data;
1298 state->parser(key, data, state->private_data);
1299 }
1300
1301 static void db_ctdb_parse_record_parser_nonpersistent(
1302 TDB_DATA key, struct ctdb_ltdb_header *header,
1303 TDB_DATA data, void *private_data)
1304 {
1305 struct db_ctdb_parse_record_state *state =
1306 (struct db_ctdb_parse_record_state *)private_data;
1307
1308 if (db_ctdb_can_use_local_hdr(header, state->my_vnn, true)) {
1309 /*
1310 * A record consisting only of the ctdb header can be
1311 * a validly created empty record or a tombstone
1312 * record of a deleted record (not vacuumed yet). Mark
1313 * it accordingly.
1314 */
1315 state->empty_record = (data.dsize == 0);
1316 if (!state->empty_record) {
1317 state->parser(key, data, state->private_data);
1318 }
1319 state->done = true;
1320 } else {
1321 /*
1322 * We found something in the db, so it seems that this record,
1323 * while not usable locally right now, is popular. Ask for a
1324 * R/O copy.
1325 */
1326 state->ask_for_readonly_copy = true;
1327 }
1328 }
1329
1330 static NTSTATUS db_ctdb_try_parse_local_record(struct db_ctdb_ctx *ctx,
1331 TDB_DATA key,
1332 struct db_ctdb_parse_record_state *state)
1333 {
1334 NTSTATUS status;
1335
1336 if (ctx->transaction != NULL) {
1337 struct db_ctdb_transaction_handle *h = ctx->transaction;
1338 bool found;
1339
1340 /*
1341 * Transactions only happen for persistent db's.
1342 */
1343
1344 found = parse_newest_in_marshall_buffer(
1345 h->m_write, key, db_ctdb_parse_record_parser, state);
1346
1347 if (found) {
1348 return NT_STATUS_OK;
1349 }
1350 }
1351
1352 if (ctx->db->persistent) {
1353 /*
1354 * Persistent db, but not found in the transaction buffer
1355 */
1356 return db_ctdb_ltdb_parse(
1357 ctx, key, db_ctdb_parse_record_parser, state);
1358 }
1359
1360 state->done = false;
1361 state->ask_for_readonly_copy = false;
1362
1363 status = db_ctdb_ltdb_parse(
1364 ctx, key, db_ctdb_parse_record_parser_nonpersistent, state);
1365 if (NT_STATUS_IS_OK(status) && state->done) {
1366 if (state->empty_record) {
1367 /*
1368 * We know authoritatively, that this is an empty
1369 * record. Since ctdb does not distinguish between empty
1370 * and deleted records, this can be a record stored as
1371 * empty or a not-yet-vacuumed tombstone record of a
1372 * deleted record. Now Samba right now can live without
1373 * empty records, so we can safely report this record
1374 * as non-existing.
1375 *
1376 * See bugs 10008 and 12005.
1377 */
1378 return NT_STATUS_NOT_FOUND;
1379 }
1380 return NT_STATUS_OK;
1381 }
1382
1383 return NT_STATUS_MORE_PROCESSING_REQUIRED;
1384 }
1385
1386 static NTSTATUS db_ctdb_parse_record(struct db_context *db, TDB_DATA key,
1387 void (*parser)(TDB_DATA key,
1388 TDB_DATA data,
1389 void *private_data),
1390 void *private_data)
1391 {
1392 struct db_ctdb_ctx *ctx = talloc_get_type_abort(
1393 db->private_data, struct db_ctdb_ctx);
1394 struct db_ctdb_parse_record_state state;
1395 NTSTATUS status;
1396 int ret;
1397
1398 state.parser = parser;
1399 state.private_data = private_data;
1400 state.my_vnn = get_my_vnn();
1401 state.empty_record = false;
1402
1403 status = db_ctdb_try_parse_local_record(ctx, key, &state);
1404 if (!NT_STATUS_EQUAL(status, NT_STATUS_MORE_PROCESSING_REQUIRED)) {
1405 return status;
1406 }
1407
1408 ret = ctdbd_parse(messaging_ctdb_connection(), ctx->db_id, key,
1409 state.ask_for_readonly_copy, parser, private_data);
1410 if (ret != 0) {
1411 if (ret == ENOENT) {
1412 /*
1413 * This maps to
1414 * NT_STATUS_OBJECT_NAME_NOT_FOUND. Our upper
1415 * layers expect NT_STATUS_NOT_FOUND for "no
1416 * record around". We need to convert dbwrap
1417 * to 0/errno away from NTSTATUS ... :-)
1418 */
1419 return NT_STATUS_NOT_FOUND;
1420 }
1421 return map_nt_error_from_unix(ret);
1422 }
1423 return NT_STATUS_OK;
1424 }
1425
1426 static void db_ctdb_parse_record_done(struct tevent_req *subreq);
1427
1428 static struct tevent_req *db_ctdb_parse_record_send(
1429 TALLOC_CTX *mem_ctx,
1430 struct tevent_context *ev,
1431 struct db_context *db,
1432 TDB_DATA key,
1433 void (*parser)(TDB_DATA key,
1434 TDB_DATA data,
1435 void *private_data),
1436 void *private_data,
1437 enum dbwrap_req_state *req_state)
1438 {
1439 struct db_ctdb_ctx *ctx = talloc_get_type_abort(
1440 db->private_data, struct db_ctdb_ctx);
1441 struct tevent_req *req = NULL;
1442 struct tevent_req *subreq = NULL;
1443 struct db_ctdb_parse_record_state *state = NULL;
1444 NTSTATUS status;
1445
1446 req = tevent_req_create(mem_ctx, &state,
1447 struct db_ctdb_parse_record_state);
1448 if (req == NULL) {
1449 *req_state = DBWRAP_REQ_ERROR;
1450 return NULL;
1451
1452 }
1453
1454 *state = (struct db_ctdb_parse_record_state) {
1455 .parser = parser,
1456 .private_data = private_data,
1457 .my_vnn = get_my_vnn(),
1458 .empty_record = false,
1459 };
1460
1461 status = db_ctdb_try_parse_local_record(ctx, key, state);
1462 if (!NT_STATUS_EQUAL(status, NT_STATUS_MORE_PROCESSING_REQUIRED)) {
1463 if (tevent_req_nterror(req, status)) {
1464 *req_state = DBWRAP_REQ_ERROR;
1465 return tevent_req_post(req, ev);
1466 }
1467 *req_state = DBWRAP_REQ_DONE;
1468 tevent_req_done(req);
1469 return tevent_req_post(req, ev);
1470 }
1471
1472 subreq = ctdbd_parse_send(state,
1473 ev,
1474 ctdb_async_ctx.async_conn,
1475 ctx->db_id,
1476 key,
1477 state->ask_for_readonly_copy,
1478 parser,
1479 private_data,
1480 req_state);
1481 if (tevent_req_nomem(subreq, req)) {
1482 *req_state = DBWRAP_REQ_ERROR;
1483 return tevent_req_post(req, ev);
1484 }
1485 tevent_req_set_callback(subreq, db_ctdb_parse_record_done, req);
1486
1487 return req;
1488 }
1489
1490 static void db_ctdb_parse_record_done(struct tevent_req *subreq)
1491 {
1492 struct tevent_req *req = tevent_req_callback_data(
1493 subreq, struct tevent_req);
1494 int ret;
1495
1496 ret = ctdbd_parse_recv(subreq);
1497 TALLOC_FREE(subreq);
1498 if (ret != 0) {
1499 if (ret == ENOENT) {
1500 /*
1501 * This maps to NT_STATUS_OBJECT_NAME_NOT_FOUND. Our
1502 * upper layers expect NT_STATUS_NOT_FOUND for "no
1503 * record around". We need to convert dbwrap to 0/errno
1504 * away from NTSTATUS ... :-)
1505 */
1506 tevent_req_nterror(req, NT_STATUS_NOT_FOUND);
1507 return;
1508 }
1509 tevent_req_nterror(req, map_nt_error_from_unix(ret));
1510 return;
1511 }
1512
1513 tevent_req_done(req);
1514 return;
1515 }
1516
1517 static NTSTATUS db_ctdb_parse_record_recv(struct tevent_req *req)
1518 {
1519 return tevent_req_simple_recv_ntstatus(req);
1520 }
1521
1522 struct traverse_state {
1523 struct db_context *db;
1524 int (*fn)(struct db_record *rec, void *private_data);
1525 void *private_data;
1526 int count;
1527 };
1528
1529 static void traverse_callback(TDB_DATA key, TDB_DATA data, void *private_data)
1530 {
1531 struct traverse_state *state = (struct traverse_state *)private_data;
1532 struct db_record *rec;
1533 TALLOC_CTX *tmp_ctx = talloc_new(state->db);
1534 /* we have to give them a locked record to prevent races */
1535 rec = db_ctdb_fetch_locked(state->db, tmp_ctx, key);
1536 if (rec && rec->value.dsize > 0) {
1537 state->fn(rec, state->private_data);
1538 }
1539 talloc_free(tmp_ctx);
1540 }
1541
1542 static int traverse_persistent_callback(TDB_CONTEXT *tdb, TDB_DATA kbuf, TDB_DATA dbuf,
1543 void *private_data)
1544 {
1545 struct traverse_state *state = (struct traverse_state *)private_data;
1546 struct db_record *rec;
1547 TALLOC_CTX *tmp_ctx = talloc_new(state->db);
1548 int ret = 0;
1549
1550 /*
1551 * Skip the __db_sequence_number__ key:
1552 * This is used for persistent transactions internally.
1553 */
1554 if (kbuf.dsize == strlen(CTDB_DB_SEQNUM_KEY) + 1 &&
1555 strcmp((const char*)kbuf.dptr, CTDB_DB_SEQNUM_KEY) == 0)
1556 {
1557 goto done;
1558 }
1559
1560 /* we have to give them a locked record to prevent races */
1561 rec = db_ctdb_fetch_locked(state->db, tmp_ctx, kbuf);
1562 if (rec && rec->value.dsize > 0) {
1563 ret = state->fn(rec, state->private_data);
1564 }
1565
1566 done:
1567 talloc_free(tmp_ctx);
1568 return ret;
1569 }
1570
1571 /* wrapper to use traverse_persistent_callback with dbwrap */
1572 static int traverse_persistent_callback_dbwrap(struct db_record *rec, void* data)
1573 {
1574 return traverse_persistent_callback(NULL, rec->key, rec->value, data);
1575 }
1576
1577 static int db_ctdbd_traverse(uint32_t db_id,
1578 void (*fn)(TDB_DATA key, TDB_DATA data,
1579 void *private_data),
1580 void *private_data)
1581 {
1582 struct ctdbd_connection *conn;
1583 int ret;
1584
1585 become_root();
1586 ret = ctdbd_init_connection(talloc_tos(), lp_ctdbd_socket(),
1587 lp_ctdb_timeout(), &conn);
1588 unbecome_root();
1589 if (ret != 0) {
1590 DBG_WARNING("ctdbd_init_connection failed: %s\n",
1591 strerror(ret));
1592 return ret;
1593 }
1594
1595 ret = ctdbd_traverse(conn, db_id, fn, private_data);
1596 TALLOC_FREE(conn);
1597
1598 if (ret != 0) {
1599 DBG_WARNING("ctdbd_traverse failed: %s\n",
1600 strerror(ret));
1601 return ret;
1602 }
1603
1604 return 0;
1605 }
1606
1607
1608 static int db_ctdb_traverse(struct db_context *db,
1609 int (*fn)(struct db_record *rec,
1610 void *private_data),
1611 void *private_data)
1612 {
1613 int ret;
1614 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
1615 struct db_ctdb_ctx);
1616 struct traverse_state state;
1617
1618 state.db = db;
1619 state.fn = fn;
1620 state.private_data = private_data;
1621 state.count = 0;
1622
1623 if (db->persistent) {
1624 struct tdb_context *ltdb = ctx->wtdb->tdb;
1625
1626 /* for persistent databases we don't need to do a ctdb traverse,
1627 we can do a faster local traverse */
1628 ret = tdb_traverse(ltdb, traverse_persistent_callback, &state);
1629 if (ret < 0) {
1630 return ret;
1631 }
1632 if (ctx->transaction && ctx->transaction->m_write) {
1633 /*
1634 * we now have to handle keys not yet
1635 * present at transaction start
1636 */
1637 struct db_context *newkeys = db_open_rbt(talloc_tos());
1638 struct ctdb_marshall_buffer *mbuf = ctx->transaction->m_write;
1639 struct ctdb_rec_data_old *rec=NULL;
1640 uint32_t i;
1641 int count = 0;
1642 NTSTATUS status;
1643
1644 if (newkeys == NULL) {
1645 return -1;
1646 }
1647
1648 for (i=0; i<mbuf->count; i++) {
1649 TDB_DATA key;
1650 rec = db_ctdb_marshall_loop_next_key(
1651 mbuf, rec, &key);
1652 SMB_ASSERT(rec != NULL);
1653
1654 if (!tdb_exists(ltdb, key)) {
1655 dbwrap_store(newkeys, key, tdb_null, 0);
1656 }
1657 }
1658 status = dbwrap_traverse(newkeys,
1659 traverse_persistent_callback_dbwrap,
1660 &state,
1661 &count);
1662 talloc_free(newkeys);
1663 if (!NT_STATUS_IS_OK(status)) {
1664 return -1;
1665 }
1666 ret += count;
1667 }
1668 return ret;
1669 }
1670
1671 ret = db_ctdbd_traverse(ctx->db_id, traverse_callback, &state);
1672 if (ret != 0) {
1673 return -1;
1674 }
1675 return state.count;
1676 }
1677
1678 static NTSTATUS db_ctdb_storev_deny(struct db_record *rec,
1679 const TDB_DATA *dbufs, int num_dbufs, int flag)
1680 {
1681 return NT_STATUS_MEDIA_WRITE_PROTECTED;
1682 }
1683
1684 static NTSTATUS db_ctdb_delete_deny(struct db_record *rec)
1685 {
1686 return NT_STATUS_MEDIA_WRITE_PROTECTED;
1687 }
1688
1689 static void traverse_read_callback(TDB_DATA key, TDB_DATA data, void *private_data)
1690 {
1691 struct traverse_state *state = (struct traverse_state *)private_data;
1692 struct db_record rec;
1693
1694 ZERO_STRUCT(rec);
1695 rec.db = state->db;
1696 rec.key = key;
1697 rec.value = data;
1698 rec.storev = db_ctdb_storev_deny;
1699 rec.delete_rec = db_ctdb_delete_deny;
1700 rec.private_data = NULL;
1701 state->fn(&rec, state->private_data);
1702 state->count++;
1703 }
1704
1705 static int traverse_persistent_callback_read(TDB_CONTEXT *tdb, TDB_DATA kbuf, TDB_DATA dbuf,
1706 void *private_data)
1707 {
1708 struct traverse_state *state = (struct traverse_state *)private_data;
1709 struct db_record rec;
1710
1711 /*
1712 * Skip the __db_sequence_number__ key:
1713 * This is used for persistent transactions internally.
1714 */
1715 if (kbuf.dsize == strlen(CTDB_DB_SEQNUM_KEY) + 1 &&
1716 strcmp((const char*)kbuf.dptr, CTDB_DB_SEQNUM_KEY) == 0)
1717 {
1718 return 0;
1719 }
1720
1721 ZERO_STRUCT(rec);
1722 rec.db = state->db;
1723 rec.key = kbuf;
1724 rec.value = dbuf;
1725 rec.storev = db_ctdb_storev_deny;
1726 rec.delete_rec = db_ctdb_delete_deny;
1727 rec.private_data = NULL;
1728
1729 if (rec.value.dsize <= sizeof(struct ctdb_ltdb_header)) {
1730 /* a deleted record */
1731 return 0;
1732 }
1733 rec.value.dsize -= sizeof(struct ctdb_ltdb_header);
1734 rec.value.dptr += sizeof(struct ctdb_ltdb_header);
1735
1736 state->count++;
1737 return state->fn(&rec, state->private_data);
1738 }
1739
1740 static int db_ctdb_traverse_read(struct db_context *db,
1741 int (*fn)(struct db_record *rec,
1742 void *private_data),
1743 void *private_data)
1744 {
1745 int ret;
1746 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
1747 struct db_ctdb_ctx);
1748 struct traverse_state state;
1749
1750 state.db = db;
1751 state.fn = fn;
1752 state.private_data = private_data;
1753 state.count = 0;
1754
1755 if (db->persistent) {
1756 /* for persistent databases we don't need to do a ctdb traverse,
1757 we can do a faster local traverse */
1758 return tdb_traverse_read(ctx->wtdb->tdb, traverse_persistent_callback_read, &state);
1759 }
1760
1761 ret = db_ctdbd_traverse(ctx->db_id, traverse_read_callback, &state);
1762 if (ret != 0) {
1763 return -1;
1764 }
1765 return state.count;
1766 }
1767
1768 static int db_ctdb_get_seqnum(struct db_context *db)
1769 {
1770 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
1771 struct db_ctdb_ctx);
1772 return tdb_get_seqnum(ctx->wtdb->tdb);
1773 }
1774
1775 static size_t db_ctdb_id(struct db_context *db, uint8_t *id, size_t idlen)
1776 {
1777 struct db_ctdb_ctx *ctx = talloc_get_type_abort(
1778 db->private_data, struct db_ctdb_ctx);
1779
1780 if (idlen >= sizeof(ctx->db_id)) {
1781 memcpy(id, &ctx->db_id, sizeof(ctx->db_id));
1782 }
1783
1784 return sizeof(ctx->db_id);
1785 }
1786
1787 struct db_context *db_open_ctdb(TALLOC_CTX *mem_ctx,
1788 struct messaging_context *msg_ctx,
1789 const char *name,
1790 int hash_size, int tdb_flags,
1791 int open_flags, mode_t mode,
1792 enum dbwrap_lock_order lock_order,
1793 uint64_t dbwrap_flags)
1794 {
1795 struct db_context *result;
1796 struct db_ctdb_ctx *db_ctdb;
1797 char *db_path;
1798 struct loadparm_context *lp_ctx;
1799 TDB_DATA data;
1800 bool persistent = (tdb_flags & TDB_CLEAR_IF_FIRST) == 0;
1801 int32_t cstatus;
1802 int ret;
1803
1804 if (!lp_clustering()) {
1805 DEBUG(10, ("Clustering disabled -- no ctdb\n"));
1806 return NULL;
1807 }
1808
1809 if (!(result = talloc_zero(mem_ctx, struct db_context))) {
1810 DEBUG(0, ("talloc failed\n"));
1811 TALLOC_FREE(result);
1812 return NULL;
1813 }
1814
1815 if (!(db_ctdb = talloc(result, struct db_ctdb_ctx))) {
1816 DEBUG(0, ("talloc failed\n"));
1817 TALLOC_FREE(result);
1818 return NULL;
1819 }
1820
1821 result->name = talloc_strdup(result, name);
1822 if (result->name == NULL) {
1823 DEBUG(0, ("talloc failed\n"));
1824 TALLOC_FREE(result);
1825 return NULL;
1826 }
1827
1828 db_ctdb->transaction = NULL;
1829 db_ctdb->db = result;
1830
1831 ret = ctdbd_db_attach(messaging_ctdb_connection(), name,
1832 &db_ctdb->db_id, persistent);
1833 if (ret != 0) {
1834 DEBUG(0, ("ctdbd_db_attach failed for %s: %s\n", name,
1835 strerror(ret)));
1836 TALLOC_FREE(result);
1837 return NULL;
1838 }
1839
1840 if (tdb_flags & TDB_SEQNUM) {
1841 data.dptr = (uint8_t *)&db_ctdb->db_id;
1842 data.dsize = sizeof(db_ctdb->db_id);
1843
1844 ret = ctdbd_control_local(messaging_ctdb_connection(),
1845 CTDB_CONTROL_ENABLE_SEQNUM,
1846 0, 0, data,
1847 NULL, NULL, &cstatus);
1848 if ((ret != 0) || cstatus != 0) {
1849 DBG_ERR("ctdb_control for enable seqnum "
1850 "failed: %s\n", strerror(ret));
1851 TALLOC_FREE(result);
1852 return NULL;
1853 }
1854 }
1855
1856 db_path = ctdbd_dbpath(messaging_ctdb_connection(), db_ctdb,
1857 db_ctdb->db_id);
1858
1859 result->persistent = persistent;
1860 result->lock_order = lock_order;
1861
1862 data.dptr = (uint8_t *)&db_ctdb->db_id;
1863 data.dsize = sizeof(db_ctdb->db_id);
1864
1865 ret = ctdbd_control_local(messaging_ctdb_connection(),
1866 CTDB_CONTROL_DB_OPEN_FLAGS,
1867 0, 0, data, NULL, &data, &cstatus);
1868 if (ret != 0) {
1869 DBG_ERR(" ctdb control for db_open_flags "
1870 "failed: %s\n", strerror(ret));
1871 TALLOC_FREE(result);
1872 return NULL;
1873 }
1874
1875 if (cstatus != 0 || data.dsize != sizeof(int)) {
1876 DBG_ERR("ctdb_control for db_open_flags failed\n");
1877 TALLOC_FREE(result);
1878 return NULL;
1879 }
1880
1881 tdb_flags = *(int *)data.dptr;
1882
1883 if (!result->persistent) {
1884 ret = ctdb_async_ctx_init(NULL, messaging_tevent_context(msg_ctx));
1885 if (ret != 0) {
1886 DBG_ERR("ctdb_async_ctx_init failed: %s\n", strerror(ret));
1887 TALLOC_FREE(result);
1888 return NULL;
1889 }
1890 }
1891
1892 if (!result->persistent &&
1893 (dbwrap_flags & DBWRAP_FLAG_OPTIMIZE_READONLY_ACCESS))
1894 {
1895 TDB_DATA indata;
1896
1897 indata = make_tdb_data((uint8_t *)&db_ctdb->db_id,
1898 sizeof(db_ctdb->db_id));
1899
1900 ret = ctdbd_control_local(
1901 messaging_ctdb_connection(),
1902 CTDB_CONTROL_SET_DB_READONLY, 0, 0,
1903 indata, NULL, NULL, &cstatus);
1904 if ((ret != 0) || (cstatus != 0)) {
1905 DEBUG(1, ("CTDB_CONTROL_SET_DB_READONLY failed: "
1906 "%s, %"PRIi32"\n", strerror(ret), cstatus));
1907 TALLOC_FREE(result);
1908 return NULL;
1909 }
1910 }
1911
1912 lp_ctx = loadparm_init_s3(db_path, loadparm_s3_helpers());
1913
1914 if (hash_size == 0) {
1915 hash_size = lpcfg_tdb_hash_size(lp_ctx, db_path);
1916 }
1917
1918 db_ctdb->wtdb = tdb_wrap_open(db_ctdb, db_path, hash_size,
1919 lpcfg_tdb_flags(lp_ctx, tdb_flags),
1920 O_RDWR, 0);
1921 talloc_unlink(db_path, lp_ctx);
1922 if (db_ctdb->wtdb == NULL) {
1923 DEBUG(0, ("Could not open tdb %s: %s\n", db_path, strerror(errno)));
1924 TALLOC_FREE(result);
1925 return NULL;
1926 }
1927 talloc_free(db_path);
1928
1929 /* honor permissions if user has specified O_CREAT */
1930 if (open_flags & O_CREAT) {
1931 int fd;
1932 fd = tdb_fd(db_ctdb->wtdb->tdb);
1933 ret = fchmod(fd, mode);
1934 if (ret == -1) {
1935 DBG_WARNING("fchmod failed: %s\n",
1936 strerror(errno));
1937 TALLOC_FREE(result);
1938 return NULL;
1939 }
1940 }
1941
1942 if (result->persistent) {
1943 db_ctdb->lock_ctx = g_lock_ctx_init(db_ctdb, msg_ctx);
1944 if (db_ctdb->lock_ctx == NULL) {
1945 DEBUG(0, ("g_lock_ctx_init failed\n"));
1946 TALLOC_FREE(result);
1947 return NULL;
1948 }
1949 }
1950
1951 db_ctdb->warn_unlock_msecs = lp_parm_int(-1, "ctdb",
1952 "unlock_warn_threshold", 5);
1953 db_ctdb->warn_migrate_attempts = lp_parm_int(-1, "ctdb",
1954 "migrate_attempts", 10);
1955 db_ctdb->warn_migrate_msecs = lp_parm_int(-1, "ctdb",
1956 "migrate_duration", 5000);
1957 db_ctdb->warn_locktime_msecs = lp_ctdb_locktime_warn_threshold();
1958
1959 result->private_data = (void *)db_ctdb;
1960 result->fetch_locked = db_ctdb_fetch_locked;
1961 result->try_fetch_locked = db_ctdb_try_fetch_locked;
1962 result->parse_record = db_ctdb_parse_record;
1963 result->parse_record_send = db_ctdb_parse_record_send;
1964 result->parse_record_recv = db_ctdb_parse_record_recv;
1965 result->traverse = db_ctdb_traverse;
1966 result->traverse_read = db_ctdb_traverse_read;
1967 result->get_seqnum = db_ctdb_get_seqnum;
1968 result->transaction_start = db_ctdb_transaction_start;
1969 result->transaction_commit = db_ctdb_transaction_commit;
1970 result->transaction_cancel = db_ctdb_transaction_cancel;
1971 result->id = db_ctdb_id;
1972
1973 DEBUG(3,("db_open_ctdb: opened database '%s' with dbid 0x%x\n",
1974 name, db_ctdb->db_id));
1975
1976 return result;
1977 }
Samba GIT mirror