examples/VFS: fix skel_transparent.c in reference to shadow_copy changes
[Samba/gebeck_regimport.git] / source3 / lib / dbwrap_ctdb.c
blob255a6730203ce397f6e103170378c3864c6fc282
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 "g_lock.h"
30 #include "messages.h"
32 struct db_ctdb_transaction_handle {
33 struct db_ctdb_ctx *ctx;
35 * we store the reads and writes done under a transaction:
36 * - one list stores both reads and writes (m_all),
37 * - the other just writes (m_write)
39 struct ctdb_marshall_buffer *m_all;
40 struct ctdb_marshall_buffer *m_write;
41 uint32_t nesting;
42 bool nested_cancel;
43 char *lock_name;
46 struct db_ctdb_ctx {
47 struct db_context *db;
48 struct tdb_wrap *wtdb;
49 uint32 db_id;
50 struct db_ctdb_transaction_handle *transaction;
51 struct g_lock_ctx *lock_ctx;
54 struct db_ctdb_rec {
55 struct db_ctdb_ctx *ctdb_ctx;
56 struct ctdb_ltdb_header header;
57 struct timeval lock_time;
60 static NTSTATUS tdb_error_to_ntstatus(struct tdb_context *tdb)
62 NTSTATUS status;
63 enum TDB_ERROR tret = tdb_error(tdb);
65 switch (tret) {
66 case TDB_ERR_EXISTS:
67 status = NT_STATUS_OBJECT_NAME_COLLISION;
68 break;
69 case TDB_ERR_NOEXIST:
70 status = NT_STATUS_OBJECT_NAME_NOT_FOUND;
71 break;
72 default:
73 status = NT_STATUS_INTERNAL_DB_CORRUPTION;
74 break;
77 return status;
81 /**
82 * fetch a record from the tdb, separating out the header
83 * information and returning the body of the record.
85 static NTSTATUS db_ctdb_ltdb_fetch(struct db_ctdb_ctx *db,
86 TDB_DATA key,
87 struct ctdb_ltdb_header *header,
88 TALLOC_CTX *mem_ctx,
89 TDB_DATA *data)
91 TDB_DATA rec;
92 NTSTATUS status;
94 rec = tdb_fetch_compat(db->wtdb->tdb, key);
95 if (rec.dsize < sizeof(struct ctdb_ltdb_header)) {
96 status = NT_STATUS_NOT_FOUND;
97 if (data) {
98 ZERO_STRUCTP(data);
100 if (header) {
101 header->dmaster = (uint32_t)-1;
102 header->rsn = 0;
104 goto done;
107 if (header) {
108 *header = *(struct ctdb_ltdb_header *)rec.dptr;
111 if (data) {
112 data->dsize = rec.dsize - sizeof(struct ctdb_ltdb_header);
113 if (data->dsize == 0) {
114 data->dptr = NULL;
115 } else {
116 data->dptr = (unsigned char *)talloc_memdup(mem_ctx,
117 rec.dptr
118 + sizeof(struct ctdb_ltdb_header),
119 data->dsize);
120 if (data->dptr == NULL) {
121 status = NT_STATUS_NO_MEMORY;
122 goto done;
127 status = NT_STATUS_OK;
129 done:
130 SAFE_FREE(rec.dptr);
131 return status;
135 * Store a record together with the ctdb record header
136 * in the local copy of the database.
138 static NTSTATUS db_ctdb_ltdb_store(struct db_ctdb_ctx *db,
139 TDB_DATA key,
140 struct ctdb_ltdb_header *header,
141 TDB_DATA data)
143 TALLOC_CTX *tmp_ctx = talloc_stackframe();
144 TDB_DATA rec;
145 int ret;
147 rec.dsize = data.dsize + sizeof(struct ctdb_ltdb_header);
148 rec.dptr = (uint8_t *)talloc_size(tmp_ctx, rec.dsize);
150 if (rec.dptr == NULL) {
151 talloc_free(tmp_ctx);
152 return NT_STATUS_NO_MEMORY;
155 memcpy(rec.dptr, header, sizeof(struct ctdb_ltdb_header));
156 memcpy(sizeof(struct ctdb_ltdb_header) + (uint8_t *)rec.dptr, data.dptr, data.dsize);
158 ret = tdb_store(db->wtdb->tdb, key, rec, TDB_REPLACE);
160 talloc_free(tmp_ctx);
162 return (ret == 0) ? NT_STATUS_OK
163 : tdb_error_to_ntstatus(db->wtdb->tdb);
168 form a ctdb_rec_data record from a key/data pair
170 note that header may be NULL. If not NULL then it is included in the data portion
171 of the record
173 static struct ctdb_rec_data *db_ctdb_marshall_record(TALLOC_CTX *mem_ctx, uint32_t reqid,
174 TDB_DATA key,
175 struct ctdb_ltdb_header *header,
176 TDB_DATA data)
178 size_t length;
179 struct ctdb_rec_data *d;
181 length = offsetof(struct ctdb_rec_data, data) + key.dsize +
182 data.dsize + (header?sizeof(*header):0);
183 d = (struct ctdb_rec_data *)talloc_size(mem_ctx, length);
184 if (d == NULL) {
185 return NULL;
187 d->length = length;
188 d->reqid = reqid;
189 d->keylen = key.dsize;
190 memcpy(&d->data[0], key.dptr, key.dsize);
191 if (header) {
192 d->datalen = data.dsize + sizeof(*header);
193 memcpy(&d->data[key.dsize], header, sizeof(*header));
194 memcpy(&d->data[key.dsize+sizeof(*header)], data.dptr, data.dsize);
195 } else {
196 d->datalen = data.dsize;
197 memcpy(&d->data[key.dsize], data.dptr, data.dsize);
199 return d;
203 /* helper function for marshalling multiple records */
204 static struct ctdb_marshall_buffer *db_ctdb_marshall_add(TALLOC_CTX *mem_ctx,
205 struct ctdb_marshall_buffer *m,
206 uint64_t db_id,
207 uint32_t reqid,
208 TDB_DATA key,
209 struct ctdb_ltdb_header *header,
210 TDB_DATA data)
212 struct ctdb_rec_data *r;
213 size_t m_size, r_size;
214 struct ctdb_marshall_buffer *m2 = NULL;
216 r = db_ctdb_marshall_record(talloc_tos(), reqid, key, header, data);
217 if (r == NULL) {
218 talloc_free(m);
219 return NULL;
222 if (m == NULL) {
223 m = (struct ctdb_marshall_buffer *)talloc_zero_size(
224 mem_ctx, offsetof(struct ctdb_marshall_buffer, data));
225 if (m == NULL) {
226 goto done;
228 m->db_id = db_id;
231 m_size = talloc_get_size(m);
232 r_size = talloc_get_size(r);
234 m2 = (struct ctdb_marshall_buffer *)talloc_realloc_size(
235 mem_ctx, m, m_size + r_size);
236 if (m2 == NULL) {
237 talloc_free(m);
238 goto done;
241 memcpy(m_size + (uint8_t *)m2, r, r_size);
243 m2->count++;
245 done:
246 talloc_free(r);
247 return m2;
250 /* we've finished marshalling, return a data blob with the marshalled records */
251 static TDB_DATA db_ctdb_marshall_finish(struct ctdb_marshall_buffer *m)
253 TDB_DATA data;
254 data.dptr = (uint8_t *)m;
255 data.dsize = talloc_get_size(m);
256 return data;
260 loop over a marshalling buffer
262 - pass r==NULL to start
263 - loop the number of times indicated by m->count
265 static struct ctdb_rec_data *db_ctdb_marshall_loop_next(struct ctdb_marshall_buffer *m, struct ctdb_rec_data *r,
266 uint32_t *reqid,
267 struct ctdb_ltdb_header *header,
268 TDB_DATA *key, TDB_DATA *data)
270 if (r == NULL) {
271 r = (struct ctdb_rec_data *)&m->data[0];
272 } else {
273 r = (struct ctdb_rec_data *)(r->length + (uint8_t *)r);
276 if (reqid != NULL) {
277 *reqid = r->reqid;
280 if (key != NULL) {
281 key->dptr = &r->data[0];
282 key->dsize = r->keylen;
284 if (data != NULL) {
285 data->dptr = &r->data[r->keylen];
286 data->dsize = r->datalen;
287 if (header != NULL) {
288 data->dptr += sizeof(*header);
289 data->dsize -= sizeof(*header);
293 if (header != NULL) {
294 if (r->datalen < sizeof(*header)) {
295 return NULL;
297 *header = *(struct ctdb_ltdb_header *)&r->data[r->keylen];
300 return r;
304 * CTDB transaction destructor
306 static int db_ctdb_transaction_destructor(struct db_ctdb_transaction_handle *h)
308 NTSTATUS status;
310 status = g_lock_unlock(h->ctx->lock_ctx, h->lock_name);
311 if (!NT_STATUS_IS_OK(status)) {
312 DEBUG(0, ("g_lock_unlock failed: %s\n", nt_errstr(status)));
313 return -1;
315 return 0;
319 * CTDB dbwrap API: transaction_start function
320 * starts a transaction on a persistent database
322 static int db_ctdb_transaction_start(struct db_context *db)
324 struct db_ctdb_transaction_handle *h;
325 NTSTATUS status;
326 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
327 struct db_ctdb_ctx);
329 if (!db->persistent) {
330 DEBUG(0,("transactions not supported on non-persistent database 0x%08x\n",
331 ctx->db_id));
332 return -1;
335 if (ctx->transaction) {
336 ctx->transaction->nesting++;
337 return 0;
340 h = talloc_zero(db, struct db_ctdb_transaction_handle);
341 if (h == NULL) {
342 DEBUG(0,(__location__ " oom for transaction handle\n"));
343 return -1;
346 h->ctx = ctx;
348 h->lock_name = talloc_asprintf(h, "transaction_db_0x%08x",
349 (unsigned int)ctx->db_id);
350 if (h->lock_name == NULL) {
351 DEBUG(0, ("talloc_asprintf failed\n"));
352 TALLOC_FREE(h);
353 return -1;
357 * Wait a day, i.e. forever...
359 status = g_lock_lock(ctx->lock_ctx, h->lock_name, G_LOCK_WRITE,
360 timeval_set(86400, 0));
361 if (!NT_STATUS_IS_OK(status)) {
362 DEBUG(0, ("g_lock_lock failed: %s\n", nt_errstr(status)));
363 TALLOC_FREE(h);
364 return -1;
367 talloc_set_destructor(h, db_ctdb_transaction_destructor);
369 ctx->transaction = h;
371 DEBUG(5,(__location__ " Started transaction on db 0x%08x\n", ctx->db_id));
373 return 0;
376 static bool pull_newest_from_marshall_buffer(struct ctdb_marshall_buffer *buf,
377 TDB_DATA key,
378 struct ctdb_ltdb_header *pheader,
379 TALLOC_CTX *mem_ctx,
380 TDB_DATA *pdata)
382 struct ctdb_rec_data *rec = NULL;
383 struct ctdb_ltdb_header h;
384 bool found = false;
385 TDB_DATA data;
386 int i;
388 if (buf == NULL) {
389 return false;
392 ZERO_STRUCT(h);
393 ZERO_STRUCT(data);
396 * Walk the list of records written during this
397 * transaction. If we want to read one we have already
398 * written, return the last written sample. Thus we do not do
399 * a "break;" for the first hit, this record might have been
400 * overwritten later.
403 for (i=0; i<buf->count; i++) {
404 TDB_DATA tkey, tdata;
405 uint32_t reqid;
406 struct ctdb_ltdb_header hdr;
408 ZERO_STRUCT(hdr);
410 rec = db_ctdb_marshall_loop_next(buf, rec, &reqid, &hdr, &tkey,
411 &tdata);
412 if (rec == NULL) {
413 return false;
416 if (tdb_data_equal(key, tkey)) {
417 found = true;
418 data = tdata;
419 h = hdr;
423 if (!found) {
424 return false;
427 if (pdata != NULL) {
428 data.dptr = (uint8_t *)talloc_memdup(mem_ctx, data.dptr,
429 data.dsize);
430 if ((data.dsize != 0) && (data.dptr == NULL)) {
431 return false;
433 *pdata = data;
436 if (pheader != NULL) {
437 *pheader = h;
440 return true;
444 fetch a record inside a transaction
446 static int db_ctdb_transaction_fetch(struct db_ctdb_ctx *db,
447 TALLOC_CTX *mem_ctx,
448 TDB_DATA key, TDB_DATA *data)
450 struct db_ctdb_transaction_handle *h = db->transaction;
451 NTSTATUS status;
452 bool found;
454 found = pull_newest_from_marshall_buffer(h->m_write, key, NULL,
455 mem_ctx, data);
456 if (found) {
457 return 0;
460 status = db_ctdb_ltdb_fetch(h->ctx, key, NULL, mem_ctx, data);
462 if (NT_STATUS_EQUAL(status, NT_STATUS_NOT_FOUND)) {
463 *data = tdb_null;
464 } else if (!NT_STATUS_IS_OK(status)) {
465 return -1;
468 h->m_all = db_ctdb_marshall_add(h, h->m_all, h->ctx->db_id, 1, key,
469 NULL, *data);
470 if (h->m_all == NULL) {
471 DEBUG(0,(__location__ " Failed to add to marshalling "
472 "record\n"));
473 data->dsize = 0;
474 talloc_free(data->dptr);
475 return -1;
478 return 0;
482 * Fetch a record from a persistent database
483 * without record locking and without an active transaction.
485 * This just fetches from the local database copy.
486 * Since the databases are kept in syc cluster-wide,
487 * there is no point in doing a ctdb call to fetch the
488 * record from the lmaster. It does even harm since migration
489 * of records bump their RSN and hence render the persistent
490 * database inconsistent.
492 static int db_ctdb_fetch_persistent(struct db_ctdb_ctx *db,
493 TALLOC_CTX *mem_ctx,
494 TDB_DATA key, TDB_DATA *data)
496 NTSTATUS status;
498 status = db_ctdb_ltdb_fetch(db, key, NULL, mem_ctx, data);
500 if (NT_STATUS_EQUAL(status, NT_STATUS_NOT_FOUND)) {
501 *data = tdb_null;
502 } else if (!NT_STATUS_IS_OK(status)) {
503 return -1;
506 return 0;
509 static NTSTATUS db_ctdb_store_transaction(struct db_record *rec, TDB_DATA data, int flag);
510 static NTSTATUS db_ctdb_delete_transaction(struct db_record *rec);
512 static struct db_record *db_ctdb_fetch_locked_transaction(struct db_ctdb_ctx *ctx,
513 TALLOC_CTX *mem_ctx,
514 TDB_DATA key)
516 struct db_record *result;
517 TDB_DATA ctdb_data;
519 if (!(result = talloc(mem_ctx, struct db_record))) {
520 DEBUG(0, ("talloc failed\n"));
521 return NULL;
524 result->private_data = ctx->transaction;
526 result->key.dsize = key.dsize;
527 result->key.dptr = (uint8 *)talloc_memdup(result, key.dptr, key.dsize);
528 if (result->key.dptr == NULL) {
529 DEBUG(0, ("talloc failed\n"));
530 TALLOC_FREE(result);
531 return NULL;
534 result->store = db_ctdb_store_transaction;
535 result->delete_rec = db_ctdb_delete_transaction;
537 if (pull_newest_from_marshall_buffer(ctx->transaction->m_write, key,
538 NULL, result, &result->value)) {
539 return result;
542 ctdb_data = tdb_fetch_compat(ctx->wtdb->tdb, key);
543 if (ctdb_data.dptr == NULL) {
544 /* create the record */
545 result->value = tdb_null;
546 return result;
549 result->value.dsize = ctdb_data.dsize - sizeof(struct ctdb_ltdb_header);
550 result->value.dptr = NULL;
552 if ((result->value.dsize != 0)
553 && !(result->value.dptr = (uint8 *)talloc_memdup(
554 result, ctdb_data.dptr + sizeof(struct ctdb_ltdb_header),
555 result->value.dsize))) {
556 DEBUG(0, ("talloc failed\n"));
557 TALLOC_FREE(result);
560 SAFE_FREE(ctdb_data.dptr);
562 return result;
565 static int db_ctdb_record_destructor(struct db_record **recp)
567 struct db_record *rec = talloc_get_type_abort(*recp, struct db_record);
568 struct db_ctdb_transaction_handle *h = talloc_get_type_abort(
569 rec->private_data, struct db_ctdb_transaction_handle);
570 int ret = h->ctx->db->transaction_commit(h->ctx->db);
571 if (ret != 0) {
572 DEBUG(0,(__location__ " transaction_commit failed\n"));
574 return 0;
578 auto-create a transaction for persistent databases
580 static struct db_record *db_ctdb_fetch_locked_persistent(struct db_ctdb_ctx *ctx,
581 TALLOC_CTX *mem_ctx,
582 TDB_DATA key)
584 int res;
585 struct db_record *rec, **recp;
587 res = db_ctdb_transaction_start(ctx->db);
588 if (res == -1) {
589 return NULL;
592 rec = db_ctdb_fetch_locked_transaction(ctx, mem_ctx, key);
593 if (rec == NULL) {
594 ctx->db->transaction_cancel(ctx->db);
595 return NULL;
598 /* destroy this transaction when we release the lock */
599 recp = talloc(rec, struct db_record *);
600 if (recp == NULL) {
601 ctx->db->transaction_cancel(ctx->db);
602 talloc_free(rec);
603 return NULL;
605 *recp = rec;
606 talloc_set_destructor(recp, db_ctdb_record_destructor);
607 return rec;
612 stores a record inside a transaction
614 static NTSTATUS db_ctdb_transaction_store(struct db_ctdb_transaction_handle *h,
615 TDB_DATA key, TDB_DATA data)
617 TALLOC_CTX *tmp_ctx = talloc_new(h);
618 TDB_DATA rec;
619 struct ctdb_ltdb_header header;
621 ZERO_STRUCT(header);
623 /* we need the header so we can update the RSN */
625 if (!pull_newest_from_marshall_buffer(h->m_write, key, &header,
626 NULL, NULL)) {
628 rec = tdb_fetch_compat(h->ctx->wtdb->tdb, key);
630 if (rec.dptr != NULL) {
631 memcpy(&header, rec.dptr,
632 sizeof(struct ctdb_ltdb_header));
633 rec.dsize -= sizeof(struct ctdb_ltdb_header);
636 * a special case, we are writing the same
637 * data that is there now
639 if (data.dsize == rec.dsize &&
640 memcmp(data.dptr,
641 rec.dptr + sizeof(struct ctdb_ltdb_header),
642 data.dsize) == 0) {
643 SAFE_FREE(rec.dptr);
644 talloc_free(tmp_ctx);
645 return NT_STATUS_OK;
648 SAFE_FREE(rec.dptr);
651 header.dmaster = get_my_vnn();
652 header.rsn++;
654 h->m_all = db_ctdb_marshall_add(h, h->m_all, h->ctx->db_id, 0, key,
655 NULL, data);
656 if (h->m_all == NULL) {
657 DEBUG(0,(__location__ " Failed to add to marshalling "
658 "record\n"));
659 talloc_free(tmp_ctx);
660 return NT_STATUS_NO_MEMORY;
663 h->m_write = db_ctdb_marshall_add(h, h->m_write, h->ctx->db_id, 0, key, &header, data);
664 if (h->m_write == NULL) {
665 DEBUG(0,(__location__ " Failed to add to marshalling record\n"));
666 talloc_free(tmp_ctx);
667 return NT_STATUS_NO_MEMORY;
670 talloc_free(tmp_ctx);
671 return NT_STATUS_OK;
676 a record store inside a transaction
678 static NTSTATUS db_ctdb_store_transaction(struct db_record *rec, TDB_DATA data, int flag)
680 struct db_ctdb_transaction_handle *h = talloc_get_type_abort(
681 rec->private_data, struct db_ctdb_transaction_handle);
682 NTSTATUS status;
684 status = db_ctdb_transaction_store(h, rec->key, data);
685 return status;
689 a record delete inside a transaction
691 static NTSTATUS db_ctdb_delete_transaction(struct db_record *rec)
693 struct db_ctdb_transaction_handle *h = talloc_get_type_abort(
694 rec->private_data, struct db_ctdb_transaction_handle);
695 NTSTATUS status;
697 status = db_ctdb_transaction_store(h, rec->key, tdb_null);
698 return status;
702 * Fetch the db sequence number of a persistent db directly from the db.
704 static NTSTATUS db_ctdb_fetch_db_seqnum_from_db(struct db_ctdb_ctx *db,
705 uint64_t *seqnum)
707 NTSTATUS status;
708 const char *keyname = CTDB_DB_SEQNUM_KEY;
709 TDB_DATA key;
710 TDB_DATA data;
711 struct ctdb_ltdb_header header;
712 TALLOC_CTX *mem_ctx = talloc_stackframe();
714 if (seqnum == NULL) {
715 return NT_STATUS_INVALID_PARAMETER;
718 key = string_term_tdb_data(keyname);
720 status = db_ctdb_ltdb_fetch(db, key, &header, mem_ctx, &data);
721 if (!NT_STATUS_IS_OK(status) &&
722 !NT_STATUS_EQUAL(status, NT_STATUS_NOT_FOUND))
724 goto done;
727 status = NT_STATUS_OK;
729 if (data.dsize != sizeof(uint64_t)) {
730 *seqnum = 0;
731 goto done;
734 *seqnum = *(uint64_t *)data.dptr;
736 done:
737 TALLOC_FREE(mem_ctx);
738 return status;
742 * Store the database sequence number inside a transaction.
744 static NTSTATUS db_ctdb_store_db_seqnum(struct db_ctdb_transaction_handle *h,
745 uint64_t seqnum)
747 NTSTATUS status;
748 const char *keyname = CTDB_DB_SEQNUM_KEY;
749 TDB_DATA key;
750 TDB_DATA data;
752 key = string_term_tdb_data(keyname);
754 data.dptr = (uint8_t *)&seqnum;
755 data.dsize = sizeof(uint64_t);
757 status = db_ctdb_transaction_store(h, key, data);
759 return status;
763 commit a transaction
765 static int db_ctdb_transaction_commit(struct db_context *db)
767 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
768 struct db_ctdb_ctx);
769 NTSTATUS rets;
770 int status;
771 struct db_ctdb_transaction_handle *h = ctx->transaction;
772 uint64_t old_seqnum, new_seqnum;
773 int ret;
775 if (h == NULL) {
776 DEBUG(0,(__location__ " transaction commit with no open transaction on db 0x%08x\n", ctx->db_id));
777 return -1;
780 if (h->nested_cancel) {
781 db->transaction_cancel(db);
782 DEBUG(5,(__location__ " Failed transaction commit after nested cancel\n"));
783 return -1;
786 if (h->nesting != 0) {
787 h->nesting--;
788 return 0;
791 if (h->m_write == NULL) {
793 * No changes were made, so don't change the seqnum,
794 * don't push to other node, just exit with success.
796 ret = 0;
797 goto done;
800 DEBUG(5,(__location__ " Commit transaction on db 0x%08x\n", ctx->db_id));
803 * As the last db action before committing, bump the database sequence
804 * number. Note that this undoes all changes to the seqnum records
805 * performed under the transaction. This record is not meant to be
806 * modified by user interaction. It is for internal use only...
808 rets = db_ctdb_fetch_db_seqnum_from_db(ctx, &old_seqnum);
809 if (!NT_STATUS_IS_OK(rets)) {
810 DEBUG(1, (__location__ " failed to fetch the db sequence number "
811 "in transaction commit on db 0x%08x\n", ctx->db_id));
812 ret = -1;
813 goto done;
816 new_seqnum = old_seqnum + 1;
818 rets = db_ctdb_store_db_seqnum(h, new_seqnum);
819 if (!NT_STATUS_IS_OK(rets)) {
820 DEBUG(1, (__location__ "failed to store the db sequence number "
821 " in transaction commit on db 0x%08x\n", ctx->db_id));
822 ret = -1;
823 goto done;
826 again:
827 /* tell ctdbd to commit to the other nodes */
828 rets = ctdbd_control_local(messaging_ctdbd_connection(),
829 CTDB_CONTROL_TRANS3_COMMIT,
830 h->ctx->db_id, 0,
831 db_ctdb_marshall_finish(h->m_write),
832 NULL, NULL, &status);
833 if (!NT_STATUS_IS_OK(rets) || status != 0) {
835 * The TRANS3_COMMIT control should only possibly fail when a
836 * recovery has been running concurrently. In any case, the db
837 * will be the same on all nodes, either the new copy or the
838 * old copy. This can be detected by comparing the old and new
839 * local sequence numbers.
841 rets = db_ctdb_fetch_db_seqnum_from_db(ctx, &new_seqnum);
842 if (!NT_STATUS_IS_OK(rets)) {
843 DEBUG(1, (__location__ " failed to refetch db sequence "
844 "number after failed TRANS3_COMMIT\n"));
845 ret = -1;
846 goto done;
849 if (new_seqnum == old_seqnum) {
850 /* Recovery prevented all our changes: retry. */
851 goto again;
852 } else if (new_seqnum != (old_seqnum + 1)) {
853 DEBUG(0, (__location__ " ERROR: new_seqnum[%lu] != "
854 "old_seqnum[%lu] + (0 or 1) after failed "
855 "TRANS3_COMMIT - this should not happen!\n",
856 (unsigned long)new_seqnum,
857 (unsigned long)old_seqnum));
858 ret = -1;
859 goto done;
862 * Recovery propagated our changes to all nodes, completing
863 * our commit for us - succeed.
867 ret = 0;
869 done:
870 h->ctx->transaction = NULL;
871 talloc_free(h);
872 return ret;
877 cancel a transaction
879 static int db_ctdb_transaction_cancel(struct db_context *db)
881 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
882 struct db_ctdb_ctx);
883 struct db_ctdb_transaction_handle *h = ctx->transaction;
885 if (h == NULL) {
886 DEBUG(0,(__location__ " transaction cancel with no open transaction on db 0x%08x\n", ctx->db_id));
887 return -1;
890 if (h->nesting != 0) {
891 h->nesting--;
892 h->nested_cancel = true;
893 return 0;
896 DEBUG(5,(__location__ " Cancel transaction on db 0x%08x\n", ctx->db_id));
898 ctx->transaction = NULL;
899 talloc_free(h);
900 return 0;
904 static NTSTATUS db_ctdb_store(struct db_record *rec, TDB_DATA data, int flag)
906 struct db_ctdb_rec *crec = talloc_get_type_abort(
907 rec->private_data, struct db_ctdb_rec);
909 return db_ctdb_ltdb_store(crec->ctdb_ctx, rec->key, &(crec->header), data);
914 #ifdef CTDB_CONTROL_SCHEDULE_FOR_DELETION
915 static NTSTATUS db_ctdb_send_schedule_for_deletion(struct db_record *rec)
917 NTSTATUS status;
918 struct ctdb_control_schedule_for_deletion *dd;
919 TDB_DATA indata;
920 int cstatus;
921 struct db_ctdb_rec *crec = talloc_get_type_abort(
922 rec->private_data, struct db_ctdb_rec);
924 indata.dsize = offsetof(struct ctdb_control_schedule_for_deletion, key) + rec->key.dsize;
925 indata.dptr = talloc_zero_array(crec, uint8_t, indata.dsize);
926 if (indata.dptr == NULL) {
927 DEBUG(0, (__location__ " talloc failed!\n"));
928 return NT_STATUS_NO_MEMORY;
931 dd = (struct ctdb_control_schedule_for_deletion *)(void *)indata.dptr;
932 dd->db_id = crec->ctdb_ctx->db_id;
933 dd->hdr = crec->header;
934 dd->keylen = rec->key.dsize;
935 memcpy(dd->key, rec->key.dptr, rec->key.dsize);
937 status = ctdbd_control_local(messaging_ctdbd_connection(),
938 CTDB_CONTROL_SCHEDULE_FOR_DELETION,
939 crec->ctdb_ctx->db_id,
940 CTDB_CTRL_FLAG_NOREPLY, /* flags */
941 indata,
942 NULL, /* outdata */
943 NULL, /* errmsg */
944 &cstatus);
945 talloc_free(indata.dptr);
947 if (!NT_STATUS_IS_OK(status) || cstatus != 0) {
948 DEBUG(1, (__location__ " Error sending local control "
949 "SCHEDULE_FOR_DELETION: %s, cstatus = %d\n",
950 nt_errstr(status), cstatus));
951 if (NT_STATUS_IS_OK(status)) {
952 status = NT_STATUS_UNSUCCESSFUL;
956 return status;
958 #endif
960 static NTSTATUS db_ctdb_delete(struct db_record *rec)
962 TDB_DATA data;
963 NTSTATUS status;
966 * We have to store the header with empty data. TODO: Fix the
967 * tdb-level cleanup
970 ZERO_STRUCT(data);
972 status = db_ctdb_store(rec, data, 0);
973 if (!NT_STATUS_IS_OK(status)) {
974 return status;
977 #ifdef CTDB_CONTROL_SCHEDULE_FOR_DELETION
978 status = db_ctdb_send_schedule_for_deletion(rec);
979 #endif
981 return status;
984 static int db_ctdb_record_destr(struct db_record* data)
986 struct db_ctdb_rec *crec = talloc_get_type_abort(
987 data->private_data, struct db_ctdb_rec);
988 int threshold;
990 DEBUG(10, (DEBUGLEVEL > 10
991 ? "Unlocking db %u key %s\n"
992 : "Unlocking db %u key %.20s\n",
993 (int)crec->ctdb_ctx->db_id,
994 hex_encode_talloc(data, (unsigned char *)data->key.dptr,
995 data->key.dsize)));
997 tdb_chainunlock(crec->ctdb_ctx->wtdb->tdb, data->key);
999 threshold = lp_ctdb_locktime_warn_threshold();
1000 if (threshold != 0) {
1001 double timediff = timeval_elapsed(&crec->lock_time);
1002 if ((timediff * 1000) > threshold) {
1003 DEBUG(0, ("Held tdb lock %f seconds\n", timediff));
1007 return 0;
1010 static struct db_record *fetch_locked_internal(struct db_ctdb_ctx *ctx,
1011 TALLOC_CTX *mem_ctx,
1012 TDB_DATA key)
1014 struct db_record *result;
1015 struct db_ctdb_rec *crec;
1016 NTSTATUS status;
1017 TDB_DATA ctdb_data;
1018 int migrate_attempts = 0;
1020 if (!(result = talloc(mem_ctx, struct db_record))) {
1021 DEBUG(0, ("talloc failed\n"));
1022 return NULL;
1025 if (!(crec = talloc_zero(result, struct db_ctdb_rec))) {
1026 DEBUG(0, ("talloc failed\n"));
1027 TALLOC_FREE(result);
1028 return NULL;
1031 result->private_data = (void *)crec;
1032 crec->ctdb_ctx = ctx;
1034 result->key.dsize = key.dsize;
1035 result->key.dptr = (uint8 *)talloc_memdup(result, key.dptr, key.dsize);
1036 if (result->key.dptr == NULL) {
1037 DEBUG(0, ("talloc failed\n"));
1038 TALLOC_FREE(result);
1039 return NULL;
1043 * Do a blocking lock on the record
1045 again:
1047 if (DEBUGLEVEL >= 10) {
1048 char *keystr = hex_encode_talloc(result, key.dptr, key.dsize);
1049 DEBUG(10, (DEBUGLEVEL > 10
1050 ? "Locking db %u key %s\n"
1051 : "Locking db %u key %.20s\n",
1052 (int)crec->ctdb_ctx->db_id, keystr));
1053 TALLOC_FREE(keystr);
1056 if (tdb_chainlock(ctx->wtdb->tdb, key) != 0) {
1057 DEBUG(3, ("tdb_chainlock failed\n"));
1058 TALLOC_FREE(result);
1059 return NULL;
1062 result->store = db_ctdb_store;
1063 result->delete_rec = db_ctdb_delete;
1064 talloc_set_destructor(result, db_ctdb_record_destr);
1066 ctdb_data = tdb_fetch_compat(ctx->wtdb->tdb, key);
1069 * See if we have a valid record and we are the dmaster. If so, we can
1070 * take the shortcut and just return it.
1073 if ((ctdb_data.dptr == NULL) ||
1074 (ctdb_data.dsize < sizeof(struct ctdb_ltdb_header)) ||
1075 ((struct ctdb_ltdb_header *)ctdb_data.dptr)->dmaster != get_my_vnn()
1076 #if 0
1077 || (random() % 2 != 0)
1078 #endif
1080 SAFE_FREE(ctdb_data.dptr);
1081 tdb_chainunlock(ctx->wtdb->tdb, key);
1082 talloc_set_destructor(result, NULL);
1084 migrate_attempts += 1;
1086 DEBUG(10, ("ctdb_data.dptr = %p, dmaster = %u (%u)\n",
1087 ctdb_data.dptr, ctdb_data.dptr ?
1088 ((struct ctdb_ltdb_header *)ctdb_data.dptr)->dmaster : -1,
1089 get_my_vnn()));
1091 status = ctdbd_migrate(messaging_ctdbd_connection(), ctx->db_id,
1092 key);
1093 if (!NT_STATUS_IS_OK(status)) {
1094 DEBUG(5, ("ctdb_migrate failed: %s\n",
1095 nt_errstr(status)));
1096 TALLOC_FREE(result);
1097 return NULL;
1099 /* now its migrated, try again */
1100 goto again;
1103 if (migrate_attempts > 10) {
1104 DEBUG(0, ("db_ctdb_fetch_locked needed %d attempts\n",
1105 migrate_attempts));
1108 GetTimeOfDay(&crec->lock_time);
1110 memcpy(&crec->header, ctdb_data.dptr, sizeof(crec->header));
1112 result->value.dsize = ctdb_data.dsize - sizeof(crec->header);
1113 result->value.dptr = NULL;
1115 if ((result->value.dsize != 0)
1116 && !(result->value.dptr = (uint8 *)talloc_memdup(
1117 result, ctdb_data.dptr + sizeof(crec->header),
1118 result->value.dsize))) {
1119 DEBUG(0, ("talloc failed\n"));
1120 TALLOC_FREE(result);
1123 SAFE_FREE(ctdb_data.dptr);
1125 return result;
1128 static struct db_record *db_ctdb_fetch_locked(struct db_context *db,
1129 TALLOC_CTX *mem_ctx,
1130 TDB_DATA key)
1132 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
1133 struct db_ctdb_ctx);
1135 if (ctx->transaction != NULL) {
1136 return db_ctdb_fetch_locked_transaction(ctx, mem_ctx, key);
1139 if (db->persistent) {
1140 return db_ctdb_fetch_locked_persistent(ctx, mem_ctx, key);
1143 return fetch_locked_internal(ctx, mem_ctx, key);
1147 fetch (unlocked, no migration) operation on ctdb
1149 static int db_ctdb_fetch(struct db_context *db, TALLOC_CTX *mem_ctx,
1150 TDB_DATA key, TDB_DATA *data)
1152 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
1153 struct db_ctdb_ctx);
1154 NTSTATUS status;
1155 TDB_DATA ctdb_data;
1157 if (ctx->transaction) {
1158 return db_ctdb_transaction_fetch(ctx, mem_ctx, key, data);
1161 if (db->persistent) {
1162 return db_ctdb_fetch_persistent(ctx, mem_ctx, key, data);
1165 /* try a direct fetch */
1166 ctdb_data = tdb_fetch_compat(ctx->wtdb->tdb, key);
1169 * See if we have a valid record and we are the dmaster. If so, we can
1170 * take the shortcut and just return it.
1171 * we bypass the dmaster check for persistent databases
1173 if ((ctdb_data.dptr != NULL) &&
1174 (ctdb_data.dsize >= sizeof(struct ctdb_ltdb_header)) &&
1175 ((struct ctdb_ltdb_header *)ctdb_data.dptr)->dmaster == get_my_vnn())
1177 /* we are the dmaster - avoid the ctdb protocol op */
1179 data->dsize = ctdb_data.dsize - sizeof(struct ctdb_ltdb_header);
1180 if (data->dsize == 0) {
1181 SAFE_FREE(ctdb_data.dptr);
1182 data->dptr = NULL;
1183 return 0;
1186 data->dptr = (uint8 *)talloc_memdup(
1187 mem_ctx, ctdb_data.dptr+sizeof(struct ctdb_ltdb_header),
1188 data->dsize);
1190 SAFE_FREE(ctdb_data.dptr);
1192 if (data->dptr == NULL) {
1193 return -1;
1195 return 0;
1198 SAFE_FREE(ctdb_data.dptr);
1200 /* we weren't able to get it locally - ask ctdb to fetch it for us */
1201 status = ctdbd_fetch(messaging_ctdbd_connection(), ctx->db_id, key,
1202 mem_ctx, data);
1203 if (!NT_STATUS_IS_OK(status)) {
1204 DEBUG(5, ("ctdbd_fetch failed: %s\n", nt_errstr(status)));
1205 return -1;
1208 return 0;
1211 struct traverse_state {
1212 struct db_context *db;
1213 int (*fn)(struct db_record *rec, void *private_data);
1214 void *private_data;
1217 static void traverse_callback(TDB_DATA key, TDB_DATA data, void *private_data)
1219 struct traverse_state *state = (struct traverse_state *)private_data;
1220 struct db_record *rec;
1221 TALLOC_CTX *tmp_ctx = talloc_new(state->db);
1222 /* we have to give them a locked record to prevent races */
1223 rec = db_ctdb_fetch_locked(state->db, tmp_ctx, key);
1224 if (rec && rec->value.dsize > 0) {
1225 state->fn(rec, state->private_data);
1227 talloc_free(tmp_ctx);
1230 static int traverse_persistent_callback(TDB_CONTEXT *tdb, TDB_DATA kbuf, TDB_DATA dbuf,
1231 void *private_data)
1233 struct traverse_state *state = (struct traverse_state *)private_data;
1234 struct db_record *rec;
1235 TALLOC_CTX *tmp_ctx = talloc_new(state->db);
1236 int ret = 0;
1237 /* we have to give them a locked record to prevent races */
1238 rec = db_ctdb_fetch_locked(state->db, tmp_ctx, kbuf);
1239 if (rec && rec->value.dsize > 0) {
1240 ret = state->fn(rec, state->private_data);
1242 talloc_free(tmp_ctx);
1243 return ret;
1246 static int db_ctdb_traverse(struct db_context *db,
1247 int (*fn)(struct db_record *rec,
1248 void *private_data),
1249 void *private_data)
1251 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
1252 struct db_ctdb_ctx);
1253 struct traverse_state state;
1255 state.db = db;
1256 state.fn = fn;
1257 state.private_data = private_data;
1259 if (db->persistent) {
1260 /* for persistent databases we don't need to do a ctdb traverse,
1261 we can do a faster local traverse */
1262 return tdb_traverse(ctx->wtdb->tdb, traverse_persistent_callback, &state);
1266 ctdbd_traverse(ctx->db_id, traverse_callback, &state);
1267 return 0;
1270 static NTSTATUS db_ctdb_store_deny(struct db_record *rec, TDB_DATA data, int flag)
1272 return NT_STATUS_MEDIA_WRITE_PROTECTED;
1275 static NTSTATUS db_ctdb_delete_deny(struct db_record *rec)
1277 return NT_STATUS_MEDIA_WRITE_PROTECTED;
1280 static void traverse_read_callback(TDB_DATA key, TDB_DATA data, void *private_data)
1282 struct traverse_state *state = (struct traverse_state *)private_data;
1283 struct db_record rec;
1284 rec.key = key;
1285 rec.value = data;
1286 rec.store = db_ctdb_store_deny;
1287 rec.delete_rec = db_ctdb_delete_deny;
1288 rec.private_data = state->db;
1289 state->fn(&rec, state->private_data);
1292 static int traverse_persistent_callback_read(TDB_CONTEXT *tdb, TDB_DATA kbuf, TDB_DATA dbuf,
1293 void *private_data)
1295 struct traverse_state *state = (struct traverse_state *)private_data;
1296 struct db_record rec;
1297 rec.key = kbuf;
1298 rec.value = dbuf;
1299 rec.store = db_ctdb_store_deny;
1300 rec.delete_rec = db_ctdb_delete_deny;
1301 rec.private_data = state->db;
1303 if (rec.value.dsize <= sizeof(struct ctdb_ltdb_header)) {
1304 /* a deleted record */
1305 return 0;
1307 rec.value.dsize -= sizeof(struct ctdb_ltdb_header);
1308 rec.value.dptr += sizeof(struct ctdb_ltdb_header);
1310 return state->fn(&rec, state->private_data);
1313 static int db_ctdb_traverse_read(struct db_context *db,
1314 int (*fn)(struct db_record *rec,
1315 void *private_data),
1316 void *private_data)
1318 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
1319 struct db_ctdb_ctx);
1320 struct traverse_state state;
1322 state.db = db;
1323 state.fn = fn;
1324 state.private_data = private_data;
1326 if (db->persistent) {
1327 /* for persistent databases we don't need to do a ctdb traverse,
1328 we can do a faster local traverse */
1329 return tdb_traverse_read(ctx->wtdb->tdb, traverse_persistent_callback_read, &state);
1332 ctdbd_traverse(ctx->db_id, traverse_read_callback, &state);
1333 return 0;
1336 static int db_ctdb_get_seqnum(struct db_context *db)
1338 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
1339 struct db_ctdb_ctx);
1340 return tdb_get_seqnum(ctx->wtdb->tdb);
1343 static int db_ctdb_get_flags(struct db_context *db)
1345 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
1346 struct db_ctdb_ctx);
1347 return tdb_get_flags(ctx->wtdb->tdb);
1350 struct db_context *db_open_ctdb(TALLOC_CTX *mem_ctx,
1351 const char *name,
1352 int hash_size, int tdb_flags,
1353 int open_flags, mode_t mode)
1355 struct db_context *result;
1356 struct db_ctdb_ctx *db_ctdb;
1357 char *db_path;
1358 struct ctdbd_connection *conn;
1360 if (!lp_clustering()) {
1361 DEBUG(10, ("Clustering disabled -- no ctdb\n"));
1362 return NULL;
1365 if (!(result = talloc_zero(mem_ctx, struct db_context))) {
1366 DEBUG(0, ("talloc failed\n"));
1367 TALLOC_FREE(result);
1368 return NULL;
1371 if (!(db_ctdb = talloc(result, struct db_ctdb_ctx))) {
1372 DEBUG(0, ("talloc failed\n"));
1373 TALLOC_FREE(result);
1374 return NULL;
1377 db_ctdb->transaction = NULL;
1378 db_ctdb->db = result;
1380 conn = messaging_ctdbd_connection();
1381 if (conn == NULL) {
1382 DEBUG(1, ("Could not connect to ctdb\n"));
1383 TALLOC_FREE(result);
1384 return NULL;
1387 if (!NT_STATUS_IS_OK(ctdbd_db_attach(conn, name, &db_ctdb->db_id, tdb_flags))) {
1388 DEBUG(0, ("ctdbd_db_attach failed for %s\n", name));
1389 TALLOC_FREE(result);
1390 return NULL;
1393 db_path = ctdbd_dbpath(conn, db_ctdb, db_ctdb->db_id);
1395 result->persistent = ((tdb_flags & TDB_CLEAR_IF_FIRST) == 0);
1397 /* only pass through specific flags */
1398 tdb_flags &= TDB_SEQNUM;
1400 /* honor permissions if user has specified O_CREAT */
1401 if (open_flags & O_CREAT) {
1402 chmod(db_path, mode);
1405 db_ctdb->wtdb = tdb_wrap_open(db_ctdb, db_path, hash_size, tdb_flags, O_RDWR, 0);
1406 if (db_ctdb->wtdb == NULL) {
1407 DEBUG(0, ("Could not open tdb %s: %s\n", db_path, strerror(errno)));
1408 TALLOC_FREE(result);
1409 return NULL;
1411 talloc_free(db_path);
1413 if (result->persistent) {
1414 db_ctdb->lock_ctx = g_lock_ctx_init(db_ctdb,
1415 ctdb_conn_msg_ctx(conn));
1416 if (db_ctdb->lock_ctx == NULL) {
1417 DEBUG(0, ("g_lock_ctx_init failed\n"));
1418 TALLOC_FREE(result);
1419 return NULL;
1423 result->private_data = (void *)db_ctdb;
1424 result->fetch_locked = db_ctdb_fetch_locked;
1425 result->fetch = db_ctdb_fetch;
1426 result->traverse = db_ctdb_traverse;
1427 result->traverse_read = db_ctdb_traverse_read;
1428 result->get_seqnum = db_ctdb_get_seqnum;
1429 result->get_flags = db_ctdb_get_flags;
1430 result->transaction_start = db_ctdb_transaction_start;
1431 result->transaction_commit = db_ctdb_transaction_commit;
1432 result->transaction_cancel = db_ctdb_transaction_cancel;
1434 DEBUG(3,("db_open_ctdb: opened database '%s' with dbid 0x%x\n",
1435 name, db_ctdb->db_id));
1437 return result;
1439 #endif