s4-kcc: correctly populate the neighbor object when taking information from repsTo
[Samba/gebeck_regimport.git] / source3 / lib / dbwrap / dbwrap_ctdb.c
blob454a28399f9e18f9a497b352e43a99214034a4b6
1 /*
2 Unix SMB/CIFS implementation.
3 Database interface wrapper around ctdbd
4 Copyright (C) Volker Lendecke 2007-2009
5 Copyright (C) Michael Adam 2009
7 This program is free software; you can redistribute it and/or modify
8 it under the terms of the GNU General Public License as published by
9 the Free Software Foundation; either version 3 of the License, or
10 (at your option) any later version.
12 This program is distributed in the hope that it will be useful,
13 but WITHOUT ANY WARRANTY; without even the implied warranty of
14 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 GNU General Public License for more details.
17 You should have received a copy of the GNU General Public License
18 along with this program. If not, see <http://www.gnu.org/licenses/>.
21 #include "includes.h"
22 #include "system/filesys.h"
23 #include "lib/util/tdb_wrap.h"
24 #include "util_tdb.h"
25 #ifdef CLUSTER_SUPPORT
26 #include "ctdb.h"
27 #include "ctdb_private.h"
28 #include "ctdbd_conn.h"
29 #include "dbwrap/dbwrap.h"
30 #include "g_lock.h"
31 #include "messages.h"
33 struct db_ctdb_transaction_handle {
34 struct db_ctdb_ctx *ctx;
36 * we store the reads and writes done under a transaction:
37 * - one list stores both reads and writes (m_all),
38 * - the other just writes (m_write)
40 struct ctdb_marshall_buffer *m_all;
41 struct ctdb_marshall_buffer *m_write;
42 uint32_t nesting;
43 bool nested_cancel;
44 char *lock_name;
47 struct db_ctdb_ctx {
48 struct db_context *db;
49 struct tdb_wrap *wtdb;
50 uint32 db_id;
51 struct db_ctdb_transaction_handle *transaction;
52 struct g_lock_ctx *lock_ctx;
55 struct db_ctdb_rec {
56 struct db_ctdb_ctx *ctdb_ctx;
57 struct ctdb_ltdb_header header;
58 struct timeval lock_time;
61 static NTSTATUS tdb_error_to_ntstatus(struct tdb_context *tdb)
63 NTSTATUS status;
64 enum TDB_ERROR tret = tdb_error(tdb);
66 switch (tret) {
67 case TDB_ERR_EXISTS:
68 status = NT_STATUS_OBJECT_NAME_COLLISION;
69 break;
70 case TDB_ERR_NOEXIST:
71 status = NT_STATUS_OBJECT_NAME_NOT_FOUND;
72 break;
73 default:
74 status = NT_STATUS_INTERNAL_DB_CORRUPTION;
75 break;
78 return status;
82 /**
83 * fetch a record from the tdb, separating out the header
84 * information and returning the body of the record.
86 static NTSTATUS db_ctdb_ltdb_fetch(struct db_ctdb_ctx *db,
87 TDB_DATA key,
88 struct ctdb_ltdb_header *header,
89 TALLOC_CTX *mem_ctx,
90 TDB_DATA *data)
92 TDB_DATA rec;
93 NTSTATUS status;
95 rec = tdb_fetch_compat(db->wtdb->tdb, key);
96 if (rec.dsize < sizeof(struct ctdb_ltdb_header)) {
97 status = NT_STATUS_NOT_FOUND;
98 if (data) {
99 ZERO_STRUCTP(data);
101 if (header) {
102 header->dmaster = (uint32_t)-1;
103 header->rsn = 0;
105 goto done;
108 if (header) {
109 *header = *(struct ctdb_ltdb_header *)rec.dptr;
112 if (data) {
113 data->dsize = rec.dsize - sizeof(struct ctdb_ltdb_header);
114 if (data->dsize == 0) {
115 data->dptr = NULL;
116 } else {
117 data->dptr = (unsigned char *)talloc_memdup(mem_ctx,
118 rec.dptr
119 + sizeof(struct ctdb_ltdb_header),
120 data->dsize);
121 if (data->dptr == NULL) {
122 status = NT_STATUS_NO_MEMORY;
123 goto done;
128 status = NT_STATUS_OK;
130 done:
131 SAFE_FREE(rec.dptr);
132 return status;
136 * Store a record together with the ctdb record header
137 * in the local copy of the database.
139 static NTSTATUS db_ctdb_ltdb_store(struct db_ctdb_ctx *db,
140 TDB_DATA key,
141 struct ctdb_ltdb_header *header,
142 TDB_DATA data)
144 TALLOC_CTX *tmp_ctx = talloc_stackframe();
145 TDB_DATA rec;
146 int ret;
148 rec.dsize = data.dsize + sizeof(struct ctdb_ltdb_header);
149 rec.dptr = (uint8_t *)talloc_size(tmp_ctx, rec.dsize);
151 if (rec.dptr == NULL) {
152 talloc_free(tmp_ctx);
153 return NT_STATUS_NO_MEMORY;
156 memcpy(rec.dptr, header, sizeof(struct ctdb_ltdb_header));
157 memcpy(sizeof(struct ctdb_ltdb_header) + (uint8_t *)rec.dptr, data.dptr, data.dsize);
159 ret = tdb_store(db->wtdb->tdb, key, rec, TDB_REPLACE);
161 talloc_free(tmp_ctx);
163 return (ret == 0) ? NT_STATUS_OK
164 : tdb_error_to_ntstatus(db->wtdb->tdb);
169 form a ctdb_rec_data record from a key/data pair
171 note that header may be NULL. If not NULL then it is included in the data portion
172 of the record
174 static struct ctdb_rec_data *db_ctdb_marshall_record(TALLOC_CTX *mem_ctx, uint32_t reqid,
175 TDB_DATA key,
176 struct ctdb_ltdb_header *header,
177 TDB_DATA data)
179 size_t length;
180 struct ctdb_rec_data *d;
182 length = offsetof(struct ctdb_rec_data, data) + key.dsize +
183 data.dsize + (header?sizeof(*header):0);
184 d = (struct ctdb_rec_data *)talloc_size(mem_ctx, length);
185 if (d == NULL) {
186 return NULL;
188 d->length = length;
189 d->reqid = reqid;
190 d->keylen = key.dsize;
191 memcpy(&d->data[0], key.dptr, key.dsize);
192 if (header) {
193 d->datalen = data.dsize + sizeof(*header);
194 memcpy(&d->data[key.dsize], header, sizeof(*header));
195 memcpy(&d->data[key.dsize+sizeof(*header)], data.dptr, data.dsize);
196 } else {
197 d->datalen = data.dsize;
198 memcpy(&d->data[key.dsize], data.dptr, data.dsize);
200 return d;
204 /* helper function for marshalling multiple records */
205 static struct ctdb_marshall_buffer *db_ctdb_marshall_add(TALLOC_CTX *mem_ctx,
206 struct ctdb_marshall_buffer *m,
207 uint64_t db_id,
208 uint32_t reqid,
209 TDB_DATA key,
210 struct ctdb_ltdb_header *header,
211 TDB_DATA data)
213 struct ctdb_rec_data *r;
214 size_t m_size, r_size;
215 struct ctdb_marshall_buffer *m2 = NULL;
217 r = db_ctdb_marshall_record(talloc_tos(), reqid, key, header, data);
218 if (r == NULL) {
219 talloc_free(m);
220 return NULL;
223 if (m == NULL) {
224 m = (struct ctdb_marshall_buffer *)talloc_zero_size(
225 mem_ctx, offsetof(struct ctdb_marshall_buffer, data));
226 if (m == NULL) {
227 goto done;
229 m->db_id = db_id;
232 m_size = talloc_get_size(m);
233 r_size = talloc_get_size(r);
235 m2 = (struct ctdb_marshall_buffer *)talloc_realloc_size(
236 mem_ctx, m, m_size + r_size);
237 if (m2 == NULL) {
238 talloc_free(m);
239 goto done;
242 memcpy(m_size + (uint8_t *)m2, r, r_size);
244 m2->count++;
246 done:
247 talloc_free(r);
248 return m2;
251 /* we've finished marshalling, return a data blob with the marshalled records */
252 static TDB_DATA db_ctdb_marshall_finish(struct ctdb_marshall_buffer *m)
254 TDB_DATA data;
255 data.dptr = (uint8_t *)m;
256 data.dsize = talloc_get_size(m);
257 return data;
261 loop over a marshalling buffer
263 - pass r==NULL to start
264 - loop the number of times indicated by m->count
266 static struct ctdb_rec_data *db_ctdb_marshall_loop_next(struct ctdb_marshall_buffer *m, struct ctdb_rec_data *r,
267 uint32_t *reqid,
268 struct ctdb_ltdb_header *header,
269 TDB_DATA *key, TDB_DATA *data)
271 if (r == NULL) {
272 r = (struct ctdb_rec_data *)&m->data[0];
273 } else {
274 r = (struct ctdb_rec_data *)(r->length + (uint8_t *)r);
277 if (reqid != NULL) {
278 *reqid = r->reqid;
281 if (key != NULL) {
282 key->dptr = &r->data[0];
283 key->dsize = r->keylen;
285 if (data != NULL) {
286 data->dptr = &r->data[r->keylen];
287 data->dsize = r->datalen;
288 if (header != NULL) {
289 data->dptr += sizeof(*header);
290 data->dsize -= sizeof(*header);
294 if (header != NULL) {
295 if (r->datalen < sizeof(*header)) {
296 return NULL;
298 *header = *(struct ctdb_ltdb_header *)&r->data[r->keylen];
301 return r;
305 * CTDB transaction destructor
307 static int db_ctdb_transaction_destructor(struct db_ctdb_transaction_handle *h)
309 NTSTATUS status;
311 status = g_lock_unlock(h->ctx->lock_ctx, h->lock_name);
312 if (!NT_STATUS_IS_OK(status)) {
313 DEBUG(0, ("g_lock_unlock failed: %s\n", nt_errstr(status)));
314 return -1;
316 return 0;
320 * CTDB dbwrap API: transaction_start function
321 * starts a transaction on a persistent database
323 static int db_ctdb_transaction_start(struct db_context *db)
325 struct db_ctdb_transaction_handle *h;
326 NTSTATUS status;
327 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
328 struct db_ctdb_ctx);
330 if (!db->persistent) {
331 DEBUG(0,("transactions not supported on non-persistent database 0x%08x\n",
332 ctx->db_id));
333 return -1;
336 if (ctx->transaction) {
337 ctx->transaction->nesting++;
338 return 0;
341 h = talloc_zero(db, struct db_ctdb_transaction_handle);
342 if (h == NULL) {
343 DEBUG(0,(__location__ " oom for transaction handle\n"));
344 return -1;
347 h->ctx = ctx;
349 h->lock_name = talloc_asprintf(h, "transaction_db_0x%08x",
350 (unsigned int)ctx->db_id);
351 if (h->lock_name == NULL) {
352 DEBUG(0, ("talloc_asprintf failed\n"));
353 TALLOC_FREE(h);
354 return -1;
358 * Wait a day, i.e. forever...
360 status = g_lock_lock(ctx->lock_ctx, h->lock_name, G_LOCK_WRITE,
361 timeval_set(86400, 0));
362 if (!NT_STATUS_IS_OK(status)) {
363 DEBUG(0, ("g_lock_lock failed: %s\n", nt_errstr(status)));
364 TALLOC_FREE(h);
365 return -1;
368 talloc_set_destructor(h, db_ctdb_transaction_destructor);
370 ctx->transaction = h;
372 DEBUG(5,(__location__ " Started transaction on db 0x%08x\n", ctx->db_id));
374 return 0;
377 static bool pull_newest_from_marshall_buffer(struct ctdb_marshall_buffer *buf,
378 TDB_DATA key,
379 struct ctdb_ltdb_header *pheader,
380 TALLOC_CTX *mem_ctx,
381 TDB_DATA *pdata)
383 struct ctdb_rec_data *rec = NULL;
384 struct ctdb_ltdb_header h;
385 bool found = false;
386 TDB_DATA data;
387 int i;
389 if (buf == NULL) {
390 return false;
393 ZERO_STRUCT(h);
394 ZERO_STRUCT(data);
397 * Walk the list of records written during this
398 * transaction. If we want to read one we have already
399 * written, return the last written sample. Thus we do not do
400 * a "break;" for the first hit, this record might have been
401 * overwritten later.
404 for (i=0; i<buf->count; i++) {
405 TDB_DATA tkey, tdata;
406 uint32_t reqid;
407 struct ctdb_ltdb_header hdr;
409 ZERO_STRUCT(hdr);
411 rec = db_ctdb_marshall_loop_next(buf, rec, &reqid, &hdr, &tkey,
412 &tdata);
413 if (rec == NULL) {
414 return false;
417 if (tdb_data_equal(key, tkey)) {
418 found = true;
419 data = tdata;
420 h = hdr;
424 if (!found) {
425 return false;
428 if (pdata != NULL) {
429 data.dptr = (uint8_t *)talloc_memdup(mem_ctx, data.dptr,
430 data.dsize);
431 if ((data.dsize != 0) && (data.dptr == NULL)) {
432 return false;
434 *pdata = data;
437 if (pheader != NULL) {
438 *pheader = h;
441 return true;
445 fetch a record inside a transaction
447 static int db_ctdb_transaction_fetch(struct db_ctdb_ctx *db,
448 TALLOC_CTX *mem_ctx,
449 TDB_DATA key, TDB_DATA *data)
451 struct db_ctdb_transaction_handle *h = db->transaction;
452 NTSTATUS status;
453 bool found;
455 found = pull_newest_from_marshall_buffer(h->m_write, key, NULL,
456 mem_ctx, data);
457 if (found) {
458 return 0;
461 status = db_ctdb_ltdb_fetch(h->ctx, key, NULL, mem_ctx, data);
463 if (NT_STATUS_EQUAL(status, NT_STATUS_NOT_FOUND)) {
464 *data = tdb_null;
465 } else if (!NT_STATUS_IS_OK(status)) {
466 return -1;
469 h->m_all = db_ctdb_marshall_add(h, h->m_all, h->ctx->db_id, 1, key,
470 NULL, *data);
471 if (h->m_all == NULL) {
472 DEBUG(0,(__location__ " Failed to add to marshalling "
473 "record\n"));
474 data->dsize = 0;
475 talloc_free(data->dptr);
476 return -1;
479 return 0;
483 * Fetch a record from a persistent database
484 * without record locking and without an active transaction.
486 * This just fetches from the local database copy.
487 * Since the databases are kept in syc cluster-wide,
488 * there is no point in doing a ctdb call to fetch the
489 * record from the lmaster. It does even harm since migration
490 * of records bump their RSN and hence render the persistent
491 * database inconsistent.
493 static int db_ctdb_fetch_persistent(struct db_ctdb_ctx *db,
494 TALLOC_CTX *mem_ctx,
495 TDB_DATA key, TDB_DATA *data)
497 NTSTATUS status;
499 status = db_ctdb_ltdb_fetch(db, key, NULL, mem_ctx, data);
501 if (NT_STATUS_EQUAL(status, NT_STATUS_NOT_FOUND)) {
502 *data = tdb_null;
503 } else if (!NT_STATUS_IS_OK(status)) {
504 return -1;
507 return 0;
510 static NTSTATUS db_ctdb_store_transaction(struct db_record *rec, TDB_DATA data, int flag);
511 static NTSTATUS db_ctdb_delete_transaction(struct db_record *rec);
513 static struct db_record *db_ctdb_fetch_locked_transaction(struct db_ctdb_ctx *ctx,
514 TALLOC_CTX *mem_ctx,
515 TDB_DATA key)
517 struct db_record *result;
518 TDB_DATA ctdb_data;
520 if (!(result = talloc(mem_ctx, struct db_record))) {
521 DEBUG(0, ("talloc failed\n"));
522 return NULL;
525 result->private_data = ctx->transaction;
527 result->key.dsize = key.dsize;
528 result->key.dptr = (uint8 *)talloc_memdup(result, key.dptr, key.dsize);
529 if (result->key.dptr == NULL) {
530 DEBUG(0, ("talloc failed\n"));
531 TALLOC_FREE(result);
532 return NULL;
535 result->store = db_ctdb_store_transaction;
536 result->delete_rec = db_ctdb_delete_transaction;
538 if (pull_newest_from_marshall_buffer(ctx->transaction->m_write, key,
539 NULL, result, &result->value)) {
540 return result;
543 ctdb_data = tdb_fetch_compat(ctx->wtdb->tdb, key);
544 if (ctdb_data.dptr == NULL) {
545 /* create the record */
546 result->value = tdb_null;
547 return result;
550 result->value.dsize = ctdb_data.dsize - sizeof(struct ctdb_ltdb_header);
551 result->value.dptr = NULL;
553 if ((result->value.dsize != 0)
554 && !(result->value.dptr = (uint8 *)talloc_memdup(
555 result, ctdb_data.dptr + sizeof(struct ctdb_ltdb_header),
556 result->value.dsize))) {
557 DEBUG(0, ("talloc failed\n"));
558 TALLOC_FREE(result);
561 SAFE_FREE(ctdb_data.dptr);
563 return result;
566 static int db_ctdb_record_destructor(struct db_record **recp)
568 struct db_record *rec = talloc_get_type_abort(*recp, struct db_record);
569 struct db_ctdb_transaction_handle *h = talloc_get_type_abort(
570 rec->private_data, struct db_ctdb_transaction_handle);
571 int ret = h->ctx->db->transaction_commit(h->ctx->db);
572 if (ret != 0) {
573 DEBUG(0,(__location__ " transaction_commit failed\n"));
575 return 0;
579 auto-create a transaction for persistent databases
581 static struct db_record *db_ctdb_fetch_locked_persistent(struct db_ctdb_ctx *ctx,
582 TALLOC_CTX *mem_ctx,
583 TDB_DATA key)
585 int res;
586 struct db_record *rec, **recp;
588 res = db_ctdb_transaction_start(ctx->db);
589 if (res == -1) {
590 return NULL;
593 rec = db_ctdb_fetch_locked_transaction(ctx, mem_ctx, key);
594 if (rec == NULL) {
595 ctx->db->transaction_cancel(ctx->db);
596 return NULL;
599 /* destroy this transaction when we release the lock */
600 recp = talloc(rec, struct db_record *);
601 if (recp == NULL) {
602 ctx->db->transaction_cancel(ctx->db);
603 talloc_free(rec);
604 return NULL;
606 *recp = rec;
607 talloc_set_destructor(recp, db_ctdb_record_destructor);
608 return rec;
613 stores a record inside a transaction
615 static NTSTATUS db_ctdb_transaction_store(struct db_ctdb_transaction_handle *h,
616 TDB_DATA key, TDB_DATA data)
618 TALLOC_CTX *tmp_ctx = talloc_new(h);
619 TDB_DATA rec;
620 struct ctdb_ltdb_header header;
622 ZERO_STRUCT(header);
624 /* we need the header so we can update the RSN */
626 if (!pull_newest_from_marshall_buffer(h->m_write, key, &header,
627 NULL, NULL)) {
629 rec = tdb_fetch_compat(h->ctx->wtdb->tdb, key);
631 if (rec.dptr != NULL) {
632 memcpy(&header, rec.dptr,
633 sizeof(struct ctdb_ltdb_header));
634 rec.dsize -= sizeof(struct ctdb_ltdb_header);
637 * a special case, we are writing the same
638 * data that is there now
640 if (data.dsize == rec.dsize &&
641 memcmp(data.dptr,
642 rec.dptr + sizeof(struct ctdb_ltdb_header),
643 data.dsize) == 0) {
644 SAFE_FREE(rec.dptr);
645 talloc_free(tmp_ctx);
646 return NT_STATUS_OK;
649 SAFE_FREE(rec.dptr);
652 header.dmaster = get_my_vnn();
653 header.rsn++;
655 h->m_all = db_ctdb_marshall_add(h, h->m_all, h->ctx->db_id, 0, key,
656 NULL, data);
657 if (h->m_all == NULL) {
658 DEBUG(0,(__location__ " Failed to add to marshalling "
659 "record\n"));
660 talloc_free(tmp_ctx);
661 return NT_STATUS_NO_MEMORY;
664 h->m_write = db_ctdb_marshall_add(h, h->m_write, h->ctx->db_id, 0, key, &header, data);
665 if (h->m_write == NULL) {
666 DEBUG(0,(__location__ " Failed to add to marshalling record\n"));
667 talloc_free(tmp_ctx);
668 return NT_STATUS_NO_MEMORY;
671 talloc_free(tmp_ctx);
672 return NT_STATUS_OK;
677 a record store inside a transaction
679 static NTSTATUS db_ctdb_store_transaction(struct db_record *rec, TDB_DATA data, int flag)
681 struct db_ctdb_transaction_handle *h = talloc_get_type_abort(
682 rec->private_data, struct db_ctdb_transaction_handle);
683 NTSTATUS status;
685 status = db_ctdb_transaction_store(h, rec->key, data);
686 return status;
690 a record delete inside a transaction
692 static NTSTATUS db_ctdb_delete_transaction(struct db_record *rec)
694 struct db_ctdb_transaction_handle *h = talloc_get_type_abort(
695 rec->private_data, struct db_ctdb_transaction_handle);
696 NTSTATUS status;
698 status = db_ctdb_transaction_store(h, rec->key, tdb_null);
699 return status;
703 * Fetch the db sequence number of a persistent db directly from the db.
705 static NTSTATUS db_ctdb_fetch_db_seqnum_from_db(struct db_ctdb_ctx *db,
706 uint64_t *seqnum)
708 NTSTATUS status;
709 const char *keyname = CTDB_DB_SEQNUM_KEY;
710 TDB_DATA key;
711 TDB_DATA data;
712 struct ctdb_ltdb_header header;
713 TALLOC_CTX *mem_ctx = talloc_stackframe();
715 if (seqnum == NULL) {
716 return NT_STATUS_INVALID_PARAMETER;
719 key = string_term_tdb_data(keyname);
721 status = db_ctdb_ltdb_fetch(db, key, &header, mem_ctx, &data);
722 if (!NT_STATUS_IS_OK(status) &&
723 !NT_STATUS_EQUAL(status, NT_STATUS_NOT_FOUND))
725 goto done;
728 status = NT_STATUS_OK;
730 if (data.dsize != sizeof(uint64_t)) {
731 *seqnum = 0;
732 goto done;
735 *seqnum = *(uint64_t *)data.dptr;
737 done:
738 TALLOC_FREE(mem_ctx);
739 return status;
743 * Store the database sequence number inside a transaction.
745 static NTSTATUS db_ctdb_store_db_seqnum(struct db_ctdb_transaction_handle *h,
746 uint64_t seqnum)
748 NTSTATUS status;
749 const char *keyname = CTDB_DB_SEQNUM_KEY;
750 TDB_DATA key;
751 TDB_DATA data;
753 key = string_term_tdb_data(keyname);
755 data.dptr = (uint8_t *)&seqnum;
756 data.dsize = sizeof(uint64_t);
758 status = db_ctdb_transaction_store(h, key, data);
760 return status;
764 commit a transaction
766 static int db_ctdb_transaction_commit(struct db_context *db)
768 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
769 struct db_ctdb_ctx);
770 NTSTATUS rets;
771 int status;
772 struct db_ctdb_transaction_handle *h = ctx->transaction;
773 uint64_t old_seqnum, new_seqnum;
774 int ret;
776 if (h == NULL) {
777 DEBUG(0,(__location__ " transaction commit with no open transaction on db 0x%08x\n", ctx->db_id));
778 return -1;
781 if (h->nested_cancel) {
782 db->transaction_cancel(db);
783 DEBUG(5,(__location__ " Failed transaction commit after nested cancel\n"));
784 return -1;
787 if (h->nesting != 0) {
788 h->nesting--;
789 return 0;
792 if (h->m_write == NULL) {
794 * No changes were made, so don't change the seqnum,
795 * don't push to other node, just exit with success.
797 ret = 0;
798 goto done;
801 DEBUG(5,(__location__ " Commit transaction on db 0x%08x\n", ctx->db_id));
804 * As the last db action before committing, bump the database sequence
805 * number. Note that this undoes all changes to the seqnum records
806 * performed under the transaction. This record is not meant to be
807 * modified by user interaction. It is for internal use only...
809 rets = db_ctdb_fetch_db_seqnum_from_db(ctx, &old_seqnum);
810 if (!NT_STATUS_IS_OK(rets)) {
811 DEBUG(1, (__location__ " failed to fetch the db sequence number "
812 "in transaction commit on db 0x%08x\n", ctx->db_id));
813 ret = -1;
814 goto done;
817 new_seqnum = old_seqnum + 1;
819 rets = db_ctdb_store_db_seqnum(h, new_seqnum);
820 if (!NT_STATUS_IS_OK(rets)) {
821 DEBUG(1, (__location__ "failed to store the db sequence number "
822 " in transaction commit on db 0x%08x\n", ctx->db_id));
823 ret = -1;
824 goto done;
827 again:
828 /* tell ctdbd to commit to the other nodes */
829 rets = ctdbd_control_local(messaging_ctdbd_connection(),
830 CTDB_CONTROL_TRANS3_COMMIT,
831 h->ctx->db_id, 0,
832 db_ctdb_marshall_finish(h->m_write),
833 NULL, NULL, &status);
834 if (!NT_STATUS_IS_OK(rets) || status != 0) {
836 * The TRANS3_COMMIT control should only possibly fail when a
837 * recovery has been running concurrently. In any case, the db
838 * will be the same on all nodes, either the new copy or the
839 * old copy. This can be detected by comparing the old and new
840 * local sequence numbers.
842 rets = db_ctdb_fetch_db_seqnum_from_db(ctx, &new_seqnum);
843 if (!NT_STATUS_IS_OK(rets)) {
844 DEBUG(1, (__location__ " failed to refetch db sequence "
845 "number after failed TRANS3_COMMIT\n"));
846 ret = -1;
847 goto done;
850 if (new_seqnum == old_seqnum) {
851 /* Recovery prevented all our changes: retry. */
852 goto again;
853 } else if (new_seqnum != (old_seqnum + 1)) {
854 DEBUG(0, (__location__ " ERROR: new_seqnum[%lu] != "
855 "old_seqnum[%lu] + (0 or 1) after failed "
856 "TRANS3_COMMIT - this should not happen!\n",
857 (unsigned long)new_seqnum,
858 (unsigned long)old_seqnum));
859 ret = -1;
860 goto done;
863 * Recovery propagated our changes to all nodes, completing
864 * our commit for us - succeed.
868 ret = 0;
870 done:
871 h->ctx->transaction = NULL;
872 talloc_free(h);
873 return ret;
878 cancel a transaction
880 static int db_ctdb_transaction_cancel(struct db_context *db)
882 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
883 struct db_ctdb_ctx);
884 struct db_ctdb_transaction_handle *h = ctx->transaction;
886 if (h == NULL) {
887 DEBUG(0,(__location__ " transaction cancel with no open transaction on db 0x%08x\n", ctx->db_id));
888 return -1;
891 if (h->nesting != 0) {
892 h->nesting--;
893 h->nested_cancel = true;
894 return 0;
897 DEBUG(5,(__location__ " Cancel transaction on db 0x%08x\n", ctx->db_id));
899 ctx->transaction = NULL;
900 talloc_free(h);
901 return 0;
905 static NTSTATUS db_ctdb_store(struct db_record *rec, TDB_DATA data, int flag)
907 struct db_ctdb_rec *crec = talloc_get_type_abort(
908 rec->private_data, struct db_ctdb_rec);
910 return db_ctdb_ltdb_store(crec->ctdb_ctx, rec->key, &(crec->header), data);
915 #ifdef HAVE_CTDB_CONTROL_SCHEDULE_FOR_DELETION_DECL
916 static NTSTATUS db_ctdb_send_schedule_for_deletion(struct db_record *rec)
918 NTSTATUS status;
919 struct ctdb_control_schedule_for_deletion *dd;
920 TDB_DATA indata;
921 int cstatus;
922 struct db_ctdb_rec *crec = talloc_get_type_abort(
923 rec->private_data, struct db_ctdb_rec);
925 indata.dsize = offsetof(struct ctdb_control_schedule_for_deletion, key) + rec->key.dsize;
926 indata.dptr = talloc_zero_array(crec, uint8_t, indata.dsize);
927 if (indata.dptr == NULL) {
928 DEBUG(0, (__location__ " talloc failed!\n"));
929 return NT_STATUS_NO_MEMORY;
932 dd = (struct ctdb_control_schedule_for_deletion *)(void *)indata.dptr;
933 dd->db_id = crec->ctdb_ctx->db_id;
934 dd->hdr = crec->header;
935 dd->keylen = rec->key.dsize;
936 memcpy(dd->key, rec->key.dptr, rec->key.dsize);
938 status = ctdbd_control_local(messaging_ctdbd_connection(),
939 CTDB_CONTROL_SCHEDULE_FOR_DELETION,
940 crec->ctdb_ctx->db_id,
941 CTDB_CTRL_FLAG_NOREPLY, /* flags */
942 indata,
943 NULL, /* outdata */
944 NULL, /* errmsg */
945 &cstatus);
946 talloc_free(indata.dptr);
948 if (!NT_STATUS_IS_OK(status) || cstatus != 0) {
949 DEBUG(1, (__location__ " Error sending local control "
950 "SCHEDULE_FOR_DELETION: %s, cstatus = %d\n",
951 nt_errstr(status), cstatus));
952 if (NT_STATUS_IS_OK(status)) {
953 status = NT_STATUS_UNSUCCESSFUL;
957 return status;
959 #endif
961 static NTSTATUS db_ctdb_delete(struct db_record *rec)
963 TDB_DATA data;
964 NTSTATUS status;
967 * We have to store the header with empty data. TODO: Fix the
968 * tdb-level cleanup
971 ZERO_STRUCT(data);
973 status = db_ctdb_store(rec, data, 0);
974 if (!NT_STATUS_IS_OK(status)) {
975 return status;
978 #ifdef HAVE_CTDB_CONTROL_SCHEDULE_FOR_DELETION_DECL
979 status = db_ctdb_send_schedule_for_deletion(rec);
980 #endif
982 return status;
985 static int db_ctdb_record_destr(struct db_record* data)
987 struct db_ctdb_rec *crec = talloc_get_type_abort(
988 data->private_data, struct db_ctdb_rec);
989 int threshold;
991 DEBUG(10, (DEBUGLEVEL > 10
992 ? "Unlocking db %u key %s\n"
993 : "Unlocking db %u key %.20s\n",
994 (int)crec->ctdb_ctx->db_id,
995 hex_encode_talloc(data, (unsigned char *)data->key.dptr,
996 data->key.dsize)));
998 tdb_chainunlock(crec->ctdb_ctx->wtdb->tdb, data->key);
1000 threshold = lp_ctdb_locktime_warn_threshold();
1001 if (threshold != 0) {
1002 double timediff = timeval_elapsed(&crec->lock_time);
1003 if ((timediff * 1000) > threshold) {
1004 DEBUG(0, ("Held tdb lock %f seconds\n", timediff));
1008 return 0;
1011 static struct db_record *fetch_locked_internal(struct db_ctdb_ctx *ctx,
1012 TALLOC_CTX *mem_ctx,
1013 TDB_DATA key)
1015 struct db_record *result;
1016 struct db_ctdb_rec *crec;
1017 NTSTATUS status;
1018 TDB_DATA ctdb_data;
1019 int migrate_attempts = 0;
1021 if (!(result = talloc(mem_ctx, struct db_record))) {
1022 DEBUG(0, ("talloc failed\n"));
1023 return NULL;
1026 if (!(crec = talloc_zero(result, struct db_ctdb_rec))) {
1027 DEBUG(0, ("talloc failed\n"));
1028 TALLOC_FREE(result);
1029 return NULL;
1032 result->private_data = (void *)crec;
1033 crec->ctdb_ctx = ctx;
1035 result->key.dsize = key.dsize;
1036 result->key.dptr = (uint8 *)talloc_memdup(result, key.dptr, key.dsize);
1037 if (result->key.dptr == NULL) {
1038 DEBUG(0, ("talloc failed\n"));
1039 TALLOC_FREE(result);
1040 return NULL;
1044 * Do a blocking lock on the record
1046 again:
1048 if (DEBUGLEVEL >= 10) {
1049 char *keystr = hex_encode_talloc(result, key.dptr, key.dsize);
1050 DEBUG(10, (DEBUGLEVEL > 10
1051 ? "Locking db %u key %s\n"
1052 : "Locking db %u key %.20s\n",
1053 (int)crec->ctdb_ctx->db_id, keystr));
1054 TALLOC_FREE(keystr);
1057 if (tdb_chainlock(ctx->wtdb->tdb, key) != 0) {
1058 DEBUG(3, ("tdb_chainlock failed\n"));
1059 TALLOC_FREE(result);
1060 return NULL;
1063 result->store = db_ctdb_store;
1064 result->delete_rec = db_ctdb_delete;
1065 talloc_set_destructor(result, db_ctdb_record_destr);
1067 ctdb_data = tdb_fetch_compat(ctx->wtdb->tdb, key);
1070 * See if we have a valid record and we are the dmaster. If so, we can
1071 * take the shortcut and just return it.
1074 if ((ctdb_data.dptr == NULL) ||
1075 (ctdb_data.dsize < sizeof(struct ctdb_ltdb_header)) ||
1076 ((struct ctdb_ltdb_header *)ctdb_data.dptr)->dmaster != get_my_vnn()
1077 #if 0
1078 || (random() % 2 != 0)
1079 #endif
1081 SAFE_FREE(ctdb_data.dptr);
1082 tdb_chainunlock(ctx->wtdb->tdb, key);
1083 talloc_set_destructor(result, NULL);
1085 migrate_attempts += 1;
1087 DEBUG(10, ("ctdb_data.dptr = %p, dmaster = %u (%u)\n",
1088 ctdb_data.dptr, ctdb_data.dptr ?
1089 ((struct ctdb_ltdb_header *)ctdb_data.dptr)->dmaster : -1,
1090 get_my_vnn()));
1092 status = ctdbd_migrate(messaging_ctdbd_connection(), ctx->db_id,
1093 key);
1094 if (!NT_STATUS_IS_OK(status)) {
1095 DEBUG(5, ("ctdb_migrate failed: %s\n",
1096 nt_errstr(status)));
1097 TALLOC_FREE(result);
1098 return NULL;
1100 /* now its migrated, try again */
1101 goto again;
1104 if (migrate_attempts > 10) {
1105 DEBUG(0, ("db_ctdb_fetch_locked needed %d attempts\n",
1106 migrate_attempts));
1109 GetTimeOfDay(&crec->lock_time);
1111 memcpy(&crec->header, ctdb_data.dptr, sizeof(crec->header));
1113 result->value.dsize = ctdb_data.dsize - sizeof(crec->header);
1114 result->value.dptr = NULL;
1116 if ((result->value.dsize != 0)
1117 && !(result->value.dptr = (uint8 *)talloc_memdup(
1118 result, ctdb_data.dptr + sizeof(crec->header),
1119 result->value.dsize))) {
1120 DEBUG(0, ("talloc failed\n"));
1121 TALLOC_FREE(result);
1124 SAFE_FREE(ctdb_data.dptr);
1126 return result;
1129 static struct db_record *db_ctdb_fetch_locked(struct db_context *db,
1130 TALLOC_CTX *mem_ctx,
1131 TDB_DATA key)
1133 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
1134 struct db_ctdb_ctx);
1136 if (ctx->transaction != NULL) {
1137 return db_ctdb_fetch_locked_transaction(ctx, mem_ctx, key);
1140 if (db->persistent) {
1141 return db_ctdb_fetch_locked_persistent(ctx, mem_ctx, key);
1144 return fetch_locked_internal(ctx, mem_ctx, key);
1148 fetch (unlocked, no migration) operation on ctdb
1150 static int db_ctdb_fetch(struct db_context *db, TALLOC_CTX *mem_ctx,
1151 TDB_DATA key, TDB_DATA *data)
1153 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
1154 struct db_ctdb_ctx);
1155 NTSTATUS status;
1156 TDB_DATA ctdb_data;
1158 if (ctx->transaction) {
1159 return db_ctdb_transaction_fetch(ctx, mem_ctx, key, data);
1162 if (db->persistent) {
1163 return db_ctdb_fetch_persistent(ctx, mem_ctx, key, data);
1166 /* try a direct fetch */
1167 ctdb_data = tdb_fetch_compat(ctx->wtdb->tdb, key);
1170 * See if we have a valid record and we are the dmaster. If so, we can
1171 * take the shortcut and just return it.
1172 * we bypass the dmaster check for persistent databases
1174 if ((ctdb_data.dptr != NULL) &&
1175 (ctdb_data.dsize >= sizeof(struct ctdb_ltdb_header)) &&
1176 ((struct ctdb_ltdb_header *)ctdb_data.dptr)->dmaster == get_my_vnn())
1178 /* we are the dmaster - avoid the ctdb protocol op */
1180 data->dsize = ctdb_data.dsize - sizeof(struct ctdb_ltdb_header);
1181 if (data->dsize == 0) {
1182 SAFE_FREE(ctdb_data.dptr);
1183 data->dptr = NULL;
1184 return 0;
1187 data->dptr = (uint8 *)talloc_memdup(
1188 mem_ctx, ctdb_data.dptr+sizeof(struct ctdb_ltdb_header),
1189 data->dsize);
1191 SAFE_FREE(ctdb_data.dptr);
1193 if (data->dptr == NULL) {
1194 return -1;
1196 return 0;
1199 SAFE_FREE(ctdb_data.dptr);
1201 /* we weren't able to get it locally - ask ctdb to fetch it for us */
1202 status = ctdbd_fetch(messaging_ctdbd_connection(), ctx->db_id, key,
1203 mem_ctx, data);
1204 if (!NT_STATUS_IS_OK(status)) {
1205 DEBUG(5, ("ctdbd_fetch failed: %s\n", nt_errstr(status)));
1206 return -1;
1209 return 0;
1212 struct traverse_state {
1213 struct db_context *db;
1214 int (*fn)(struct db_record *rec, void *private_data);
1215 void *private_data;
1218 static void traverse_callback(TDB_DATA key, TDB_DATA data, void *private_data)
1220 struct traverse_state *state = (struct traverse_state *)private_data;
1221 struct db_record *rec;
1222 TALLOC_CTX *tmp_ctx = talloc_new(state->db);
1223 /* we have to give them a locked record to prevent races */
1224 rec = db_ctdb_fetch_locked(state->db, tmp_ctx, key);
1225 if (rec && rec->value.dsize > 0) {
1226 state->fn(rec, state->private_data);
1228 talloc_free(tmp_ctx);
1231 static int traverse_persistent_callback(TDB_CONTEXT *tdb, TDB_DATA kbuf, TDB_DATA dbuf,
1232 void *private_data)
1234 struct traverse_state *state = (struct traverse_state *)private_data;
1235 struct db_record *rec;
1236 TALLOC_CTX *tmp_ctx = talloc_new(state->db);
1237 int ret = 0;
1238 /* we have to give them a locked record to prevent races */
1239 rec = db_ctdb_fetch_locked(state->db, tmp_ctx, kbuf);
1240 if (rec && rec->value.dsize > 0) {
1241 ret = state->fn(rec, state->private_data);
1243 talloc_free(tmp_ctx);
1244 return ret;
1247 static int db_ctdb_traverse(struct db_context *db,
1248 int (*fn)(struct db_record *rec,
1249 void *private_data),
1250 void *private_data)
1252 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
1253 struct db_ctdb_ctx);
1254 struct traverse_state state;
1256 state.db = db;
1257 state.fn = fn;
1258 state.private_data = private_data;
1260 if (db->persistent) {
1261 /* for persistent databases we don't need to do a ctdb traverse,
1262 we can do a faster local traverse */
1263 return tdb_traverse(ctx->wtdb->tdb, traverse_persistent_callback, &state);
1267 ctdbd_traverse(ctx->db_id, traverse_callback, &state);
1268 return 0;
1271 static NTSTATUS db_ctdb_store_deny(struct db_record *rec, TDB_DATA data, int flag)
1273 return NT_STATUS_MEDIA_WRITE_PROTECTED;
1276 static NTSTATUS db_ctdb_delete_deny(struct db_record *rec)
1278 return NT_STATUS_MEDIA_WRITE_PROTECTED;
1281 static void traverse_read_callback(TDB_DATA key, TDB_DATA data, void *private_data)
1283 struct traverse_state *state = (struct traverse_state *)private_data;
1284 struct db_record rec;
1285 rec.key = key;
1286 rec.value = data;
1287 rec.store = db_ctdb_store_deny;
1288 rec.delete_rec = db_ctdb_delete_deny;
1289 rec.private_data = state->db;
1290 state->fn(&rec, state->private_data);
1293 static int traverse_persistent_callback_read(TDB_CONTEXT *tdb, TDB_DATA kbuf, TDB_DATA dbuf,
1294 void *private_data)
1296 struct traverse_state *state = (struct traverse_state *)private_data;
1297 struct db_record rec;
1298 rec.key = kbuf;
1299 rec.value = dbuf;
1300 rec.store = db_ctdb_store_deny;
1301 rec.delete_rec = db_ctdb_delete_deny;
1302 rec.private_data = state->db;
1304 if (rec.value.dsize <= sizeof(struct ctdb_ltdb_header)) {
1305 /* a deleted record */
1306 return 0;
1308 rec.value.dsize -= sizeof(struct ctdb_ltdb_header);
1309 rec.value.dptr += sizeof(struct ctdb_ltdb_header);
1311 return state->fn(&rec, state->private_data);
1314 static int db_ctdb_traverse_read(struct db_context *db,
1315 int (*fn)(struct db_record *rec,
1316 void *private_data),
1317 void *private_data)
1319 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
1320 struct db_ctdb_ctx);
1321 struct traverse_state state;
1323 state.db = db;
1324 state.fn = fn;
1325 state.private_data = private_data;
1327 if (db->persistent) {
1328 /* for persistent databases we don't need to do a ctdb traverse,
1329 we can do a faster local traverse */
1330 return tdb_traverse_read(ctx->wtdb->tdb, traverse_persistent_callback_read, &state);
1333 ctdbd_traverse(ctx->db_id, traverse_read_callback, &state);
1334 return 0;
1337 static int db_ctdb_get_seqnum(struct db_context *db)
1339 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
1340 struct db_ctdb_ctx);
1341 return tdb_get_seqnum(ctx->wtdb->tdb);
1344 static int db_ctdb_get_flags(struct db_context *db)
1346 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
1347 struct db_ctdb_ctx);
1348 return tdb_get_flags(ctx->wtdb->tdb);
1351 struct db_context *db_open_ctdb(TALLOC_CTX *mem_ctx,
1352 const char *name,
1353 int hash_size, int tdb_flags,
1354 int open_flags, mode_t mode)
1356 struct db_context *result;
1357 struct db_ctdb_ctx *db_ctdb;
1358 char *db_path;
1359 struct ctdbd_connection *conn;
1361 if (!lp_clustering()) {
1362 DEBUG(10, ("Clustering disabled -- no ctdb\n"));
1363 return NULL;
1366 if (!(result = talloc_zero(mem_ctx, struct db_context))) {
1367 DEBUG(0, ("talloc failed\n"));
1368 TALLOC_FREE(result);
1369 return NULL;
1372 if (!(db_ctdb = talloc(result, struct db_ctdb_ctx))) {
1373 DEBUG(0, ("talloc failed\n"));
1374 TALLOC_FREE(result);
1375 return NULL;
1378 db_ctdb->transaction = NULL;
1379 db_ctdb->db = result;
1381 conn = messaging_ctdbd_connection();
1382 if (conn == NULL) {
1383 DEBUG(1, ("Could not connect to ctdb\n"));
1384 TALLOC_FREE(result);
1385 return NULL;
1388 if (!NT_STATUS_IS_OK(ctdbd_db_attach(conn, name, &db_ctdb->db_id, tdb_flags))) {
1389 DEBUG(0, ("ctdbd_db_attach failed for %s\n", name));
1390 TALLOC_FREE(result);
1391 return NULL;
1394 db_path = ctdbd_dbpath(conn, db_ctdb, db_ctdb->db_id);
1396 result->persistent = ((tdb_flags & TDB_CLEAR_IF_FIRST) == 0);
1398 /* only pass through specific flags */
1399 tdb_flags &= TDB_SEQNUM;
1401 /* honor permissions if user has specified O_CREAT */
1402 if (open_flags & O_CREAT) {
1403 chmod(db_path, mode);
1406 db_ctdb->wtdb = tdb_wrap_open(db_ctdb, db_path, hash_size, tdb_flags, O_RDWR, 0);
1407 if (db_ctdb->wtdb == NULL) {
1408 DEBUG(0, ("Could not open tdb %s: %s\n", db_path, strerror(errno)));
1409 TALLOC_FREE(result);
1410 return NULL;
1412 talloc_free(db_path);
1414 if (result->persistent) {
1415 db_ctdb->lock_ctx = g_lock_ctx_init(db_ctdb,
1416 ctdb_conn_msg_ctx(conn));
1417 if (db_ctdb->lock_ctx == NULL) {
1418 DEBUG(0, ("g_lock_ctx_init failed\n"));
1419 TALLOC_FREE(result);
1420 return NULL;
1424 result->private_data = (void *)db_ctdb;
1425 result->fetch_locked = db_ctdb_fetch_locked;
1426 result->fetch = db_ctdb_fetch;
1427 result->traverse = db_ctdb_traverse;
1428 result->traverse_read = db_ctdb_traverse_read;
1429 result->get_seqnum = db_ctdb_get_seqnum;
1430 result->get_flags = db_ctdb_get_flags;
1431 result->transaction_start = db_ctdb_transaction_start;
1432 result->transaction_commit = db_ctdb_transaction_commit;
1433 result->transaction_cancel = db_ctdb_transaction_cancel;
1435 DEBUG(3,("db_open_ctdb: opened database '%s' with dbid 0x%x\n",
1436 name, db_ctdb->db_id));
1438 return result;
1440 #endif