summaryrefslogtreecommitdiff
path: root/source3/lib/tdb/common/transaction.c
blob: 7acda640c8d32338de6e8651971d5f848740aae6 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
 /* 
   Unix SMB/CIFS implementation.

   trivial database library

   Copyright (C) Andrew Tridgell              2005

     ** NOTE! The following LGPL license applies to the tdb
     ** library. This does NOT imply that all of Samba is released
     ** under the LGPL
   
   This library is free software; you can redistribute it and/or
   modify it under the terms of the GNU Lesser General Public
   License as published by the Free Software Foundation; either
   version 3 of the License, or (at your option) any later version.

   This library is distributed in the hope that it will be useful,
   but WITHOUT ANY WARRANTY; without even the implied warranty of
   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
   Lesser General Public License for more details.

   You should have received a copy of the GNU Lesser General Public
   License along with this library; if not, see <http://www.gnu.org/licenses/>.
*/

#include "tdb_private.h"

/*
  transaction design:

  - only allow a single transaction at a time per database. This makes
    using the transaction API simpler, as otherwise the caller would
    have to cope with temporary failures in transactions that conflict
    with other current transactions

  - keep the transaction recovery information in the same file as the
    database, using a special 'transaction recovery' record pointed at
    by the header. This removes the need for extra journal files as
    used by some other databases

  - dynamically allocated the transaction recover record, re-using it
    for subsequent transactions. If a larger record is needed then
    tdb_free() the old record to place it on the normal tdb freelist
    before allocating the new record

  - during transactions, keep a linked list of writes all that have
    been performed by intercepting all tdb_write() calls. The hooked
    transaction versions of tdb_read() and tdb_write() check this
    linked list and try to use the elements of the list in preference
    to the real database.

  - don't allow any locks to be held when a transaction starts,
    otherwise we can end up with deadlock (plus lack of lock nesting
    in posix locks would mean the lock is lost)

  - if the caller gains a lock during the transaction but doesn't
    release it then fail the commit

  - allow for nested calls to tdb_transaction_start(), re-using the
    existing transaction record. If the inner transaction is cancelled
    then a subsequent commit will fail
 
  - keep a mirrored copy of the tdb hash chain heads to allow for the
    fast hash heads scan on traverse, updating the mirrored copy in
    the transaction version of tdb_write

  - allow callers to mix transaction and non-transaction use of tdb,
    although once a transaction is started then an exclusive lock is
    gained until the transaction is committed or cancelled

  - the commit stategy involves first saving away all modified data
    into a linearised buffer in the transaction recovery area, then
    marking the transaction recovery area with a magic value to
    indicate a valid recovery record. In total 4 fsync/msync calls are
    needed per commit to prevent race conditions. It might be possible
    to reduce this to 3 or even 2 with some more work.

  - check for a valid recovery record on open of the tdb, while the
    global lock is held. Automatically recover from the transaction
    recovery area if needed, then continue with the open as
    usual. This allows for smooth crash recovery with no administrator
    intervention.

  - if TDB_NOSYNC is passed to flags in tdb_open then transactions are
    still available, but no transaction recovery area is used and no
    fsync/msync calls are made.

*/


/*
  hold the context of any current transaction
*/
struct tdb_transaction {
	/* we keep a mirrored copy of the tdb hash heads here so
	   tdb_next_hash_chain() can operate efficiently */
	uint32_t *hash_heads;

	/* the original io methods - used to do IOs to the real db */
	const struct tdb_methods *io_methods;

	/* the list of transaction blocks. When a block is first
	   written to, it gets created in this list */
	uint8_t **blocks;
	uint32_t num_blocks;
	uint32_t block_size;      /* bytes in each block */
	uint32_t last_block_size; /* number of valid bytes in the last block */

	/* non-zero when an internal transaction error has
	   occurred. All write operations will then fail until the
	   transaction is ended */
	int transaction_error;

	/* when inside a transaction we need to keep track of any
	   nested tdb_transaction_start() calls, as these are allowed,
	   but don't create a new transaction */
	int nesting;

	/* old file size before transaction */
	tdb_len_t old_map_size;
};


/*
  read while in a transaction. We need to check first if the data is in our list
  of transaction elements, then if not do a real read
*/
static int transaction_read(struct tdb_context *tdb, tdb_off_t off, void *buf, 
			    tdb_len_t len, int cv)
{
	uint32_t blk;

	/* break it down into block sized ops */
	while (len + (off % tdb->transaction->block_size) > tdb->transaction->block_size) {
		tdb_len_t len2 = tdb->transaction->block_size - (off % tdb->transaction->block_size);
		if (transaction_read(tdb, off, buf, len2, cv) != 0) {
			return -1;
		}
		len -= len2;
		off += len2;
		buf = (void *)(len2 + (char *)buf);
	}

	if (len == 0) {
		return 0;
	}

	blk = off / tdb->transaction->block_size;

	/* see if we have it in the block list */
	if (tdb->transaction->num_blocks <= blk ||
	    tdb->transaction->blocks[blk] == NULL) {
		/* nope, do a real read */
		if (tdb->transaction->io_methods->tdb_read(tdb, off, buf, len, cv) != 0) {
			goto fail;
		}
		return 0;
	}

	/* it is in the block list. Now check for the last block */
	if (blk == tdb->transaction->num_blocks-1) {
		if (len > tdb->transaction->last_block_size) {
			goto fail;
		}
	}
	
	/* now copy it out of this block */
	memcpy(buf, tdb->transaction->blocks[blk] + (off % tdb->transaction->block_size), len);
	if (cv) {
		tdb_convert(buf, len);
	}
	return 0;

fail:
	TDB_LOG((tdb, TDB_DEBUG_FATAL, "transaction_read: failed at off=%d len=%d\n", off, len));
	tdb->ecode = TDB_ERR_IO;
	tdb->transaction->transaction_error = 1;
	return -1;
}


/*
  write while in a transaction
*/
static int transaction_write(struct tdb_context *tdb, tdb_off_t off, 
			     const void *buf, tdb_len_t len)
{
	uint32_t blk;

	/* if the write is to a hash head, then update the transaction
	   hash heads */
	if (len == sizeof(tdb_off_t) && off >= FREELIST_TOP &&
	    off < FREELIST_TOP+TDB_HASHTABLE_SIZE(tdb)) {
		uint32_t chain = (off-FREELIST_TOP) / sizeof(tdb_off_t);
		memcpy(&tdb->transaction->hash_heads[chain], buf, len);
	}

	/* break it up into block sized chunks */
	while (len + (off % tdb->transaction->block_size) > tdb->transaction->block_size) {
		tdb_len_t len2 = tdb->transaction->block_size - (off % tdb->transaction->block_size);
		if (transaction_write(tdb, off, buf, len2) != 0) {
			return -1;
		}
		len -= len2;
		off += len2;
		if (buf != NULL) {
			buf = (const void *)(len2 + (const char *)buf);
		}
	}

	if (len == 0) {
		return 0;
	}

	blk = off / tdb->transaction->block_size;
	off = off % tdb->transaction->block_size;

	if (tdb->transaction->num_blocks <= blk) {
		uint8_t **new_blocks;
		/* expand the blocks array */
		if (tdb->transaction->blocks == NULL) {
			new_blocks = (uint8_t **)malloc(
				(blk+1)*sizeof(uint8_t *));
		} else {
			new_blocks = (uint8_t **)realloc(
				tdb->transaction->blocks,
				(blk+1)*sizeof(uint8_t *));
		}
		if (new_blocks == NULL) {
			tdb->ecode = TDB_ERR_OOM;
			goto fail;
		}
		memset(&new_blocks[tdb->transaction->num_blocks], 0, 
		       (1+(blk - tdb->transaction->num_blocks))*sizeof(uint8_t *));
		tdb->transaction->blocks = new_blocks;
		tdb->transaction->num_blocks = blk+1;
		tdb->transaction->last_block_size = 0;
	}

	/* allocate and fill a block? */
	if (tdb->transaction->blocks[blk] == NULL) {
		tdb->transaction->blocks[blk] = (uint8_t *)calloc(tdb->transaction->block_size, 1);
		if (tdb->transaction->blocks[blk] == NULL) {
			tdb->ecode = TDB_ERR_OOM;
			tdb->transaction->transaction_error = 1;
			return -1;			
		}
		if (tdb->transaction->old_map_size > blk * tdb->transaction->block_size) {
			tdb_len_t len2 = tdb->transaction->block_size;
			if (len2 + (blk * tdb->transaction->block_size) > tdb->transaction->old_map_size) {
				len2 = tdb->transaction->old_map_size - (blk * tdb->transaction->block_size);
			}
			if (tdb->transaction->io_methods->tdb_read(tdb, blk * tdb->transaction->block_size, 
								   tdb->transaction->blocks[blk], 
								   len2, 0) != 0) {
				SAFE_FREE(tdb->transaction->blocks[blk]);				
				tdb->ecode = TDB_ERR_IO;
				goto fail;
			}
			if (blk == tdb->transaction->num_blocks-1) {
				tdb->transaction->last_block_size = len2;
			}			
		}
	}
	
	/* overwrite part of an existing block */
	if (buf == NULL) {
		memset(tdb->transaction->blocks[blk] + off, 0, len);
	} else {
		memcpy(tdb->transaction->blocks[blk] + off, buf, len);
	}
	if (blk == tdb->transaction->num_blocks-1) {
		if (len + off > tdb->transaction->last_block_size) {
			tdb->transaction->last_block_size = len + off;
		}
	}

	return 0;

fail:
	TDB_LOG((tdb, TDB_DEBUG_FATAL, "transaction_write: failed at off=%d len=%d\n", 
		 (blk*tdb->transaction->block_size) + off, len));
	tdb->transaction->transaction_error = 1;
	return -1;
}


/*
  write while in a transaction - this varient never expands the transaction blocks, it only
  updates existing blocks. This means it cannot change the recovery size
*/
static int transaction_write_existing(struct tdb_context *tdb, tdb_off_t off, 
				      const void *buf, tdb_len_t len)
{
	uint32_t blk;

	/* break it up into block sized chunks */
	while (len + (off % tdb->transaction->block_size) > tdb->transaction->block_size) {
		tdb_len_t len2 = tdb->transaction->block_size - (off % tdb->transaction->block_size);
		if (transaction_write_existing(tdb, off, buf, len2) != 0) {
			return -1;
		}
		len -= len2;
		off += len2;
		if (buf != NULL) {
			buf = (const void *)(len2 + (const char *)buf);
		}
	}

	if (len == 0) {
		return 0;
	}

	blk = off / tdb->transaction->block_size;
	off = off % tdb->transaction->block_size;

	if (tdb->transaction->num_blocks <= blk ||
	    tdb->transaction->blocks[blk] == NULL) {
		return 0;
	}

	if (blk == tdb->transaction->num_blocks-1 &&
	    off + len > tdb->transaction->last_block_size) {
		if (off >= tdb->transaction->last_block_size) {
			return 0;
		}
		len = tdb->transaction->last_block_size - off;
	}

	/* overwrite part of an existing block */
	memcpy(tdb->transaction->blocks[blk] + off, buf, len);

	return 0;
}


/*
  accelerated hash chain head search, using the cached hash heads
*/
static void transaction_next_hash_chain(struct tdb_context *tdb, uint32_t *chain)
{
	uint32_t h = *chain;
	for (;h < tdb->header.hash_size;h++) {
		/* the +1 takes account of the freelist */
		if (0 != tdb->transaction->hash_heads[h+1]) {
			break;
		}
	}
	(*chain) = h;
}

/*
  out of bounds check during a transaction
*/
static int transaction_oob(struct tdb_context *tdb, tdb_off_t len, int probe)
{
	if (len <= tdb->map_size) {
		return 0;
	}
	return TDB_ERRCODE(TDB_ERR_IO, -1);
}

/*
  transaction version of tdb_expand().
*/
static int transaction_expand_file(struct tdb_context *tdb, tdb_off_t size, 
				   tdb_off_t addition)
{
	/* add a write to the transaction elements, so subsequent
	   reads see the zero data */
	if (transaction_write(tdb, size, NULL, addition) != 0) {
		return -1;
	}

	return 0;
}

/*
  brlock during a transaction - ignore them
*/
static int transaction_brlock(struct tdb_context *tdb, tdb_off_t offset, 
			      int rw_type, int lck_type, int probe, size_t len)
{
	return 0;
}

static const struct tdb_methods transaction_methods = {
	transaction_read,
	transaction_write,
	transaction_next_hash_chain,
	transaction_oob,
	transaction_expand_file,
	transaction_brlock
};


/*
  start a tdb transaction. No token is returned, as only a single
  transaction is allowed to be pending per tdb_context
*/
int tdb_transaction_start(struct tdb_context *tdb)
{
	/* some sanity checks */
	if (tdb->read_only || (tdb->flags & TDB_INTERNAL) || tdb->traverse_read) {
		TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_transaction_start: cannot start a transaction on a read-only or internal db\n"));
		tdb->ecode = TDB_ERR_EINVAL;
		return -1;
	}

	/* cope with nested tdb_transaction_start() calls */
	if (tdb->transaction != NULL) {
		tdb->transaction->nesting++;
		TDB_LOG((tdb, TDB_DEBUG_TRACE, "tdb_transaction_start: nesting %d\n", 
			 tdb->transaction->nesting));
		return 0;
	}

	if (tdb->num_locks != 0 || tdb->global_lock.count) {
		/* the caller must not have any locks when starting a
		   transaction as otherwise we'll be screwed by lack
		   of nested locks in posix */
		TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_transaction_start: cannot start a transaction with locks held\n"));
		tdb->ecode = TDB_ERR_LOCK;
		return -1;
	}

	if (tdb->travlocks.next != NULL) {
		/* you cannot use transactions inside a traverse (although you can use
		   traverse inside a transaction) as otherwise you can end up with
		   deadlock */
		TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_transaction_start: cannot start a transaction within a traverse\n"));
		tdb->ecode = TDB_ERR_LOCK;
		return -1;
	}

	tdb->transaction = (struct tdb_transaction *)
		calloc(sizeof(struct tdb_transaction), 1);
	if (tdb->transaction == NULL) {
		tdb->ecode = TDB_ERR_OOM;
		return -1;
	}

	/* a page at a time seems like a reasonable compromise between compactness and efficiency */
	tdb->transaction->block_size = tdb->page_size;

	/* get the transaction write lock. This is a blocking lock. As
	   discussed with Volker, there are a number of ways we could
	   make this async, which we will probably do in the future */
	if (tdb_transaction_lock(tdb, F_WRLCK) == -1) {
		SAFE_FREE(tdb->transaction->blocks);
		SAFE_FREE(tdb->transaction);
		return -1;
	}
	
	/* get a read lock from the freelist to the end of file. This
	   is upgraded to a write lock during the commit */
	if (tdb_brlock(tdb, FREELIST_TOP, F_RDLCK, F_SETLKW, 0, 0) == -1) {
		TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_transaction_start: failed to get hash locks\n"));
		tdb->ecode = TDB_ERR_LOCK;
		goto fail;
	}

	/* setup a copy of the hash table heads so the hash scan in
	   traverse can be fast */
	tdb->transaction->hash_heads = (uint32_t *)
		calloc(tdb->header.hash_size+1, sizeof(uint32_t));
	if (tdb->transaction->hash_heads == NULL) {
		tdb->ecode = TDB_ERR_OOM;
		goto fail;
	}
	if (tdb->methods->tdb_read(tdb, FREELIST_TOP, tdb->transaction->hash_heads,
				   TDB_HASHTABLE_SIZE(tdb), 0) != 0) {
		TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_start: failed to read hash heads\n"));
		tdb->ecode = TDB_ERR_IO;
		goto fail;
	}

	/* make sure we know about any file expansions already done by
	   anyone else */
	tdb->methods->tdb_oob(tdb, tdb->map_size + 1, 1);
	tdb->transaction->old_map_size = tdb->map_size;

	/* finally hook the io methods, replacing them with
	   transaction specific methods */
	tdb->transaction->io_methods = tdb->methods;
	tdb->methods = &transaction_methods;

	return 0;
	
fail:
	tdb_brlock(tdb, FREELIST_TOP, F_UNLCK, F_SETLKW, 0, 0);
	tdb_transaction_unlock(tdb);
	SAFE_FREE(tdb->transaction->blocks);
	SAFE_FREE(tdb->transaction->hash_heads);
	SAFE_FREE(tdb->transaction);
	return -1;
}


/*
  cancel the current transaction
*/
int tdb_transaction_cancel(struct tdb_context *tdb)
{	
	int i;

	if (tdb->transaction == NULL) {
		TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_transaction_cancel: no transaction\n"));
		return -1;
	}

	if (tdb->transaction->nesting != 0) {
		tdb->transaction->transaction_error = 1;
		tdb->transaction->nesting--;
		return 0;
	}		

	tdb->map_size = tdb->transaction->old_map_size;

	/* free all the transaction blocks */
	for (i=0;i<tdb->transaction->num_blocks;i++) {
		if (tdb->transaction->blocks[i] != NULL) {
			free(tdb->transaction->blocks[i]);
		}
	}
	SAFE_FREE(tdb->transaction->blocks);

	/* remove any global lock created during the transaction */
	if (tdb->global_lock.count != 0) {
		tdb_brlock(tdb, FREELIST_TOP, F_UNLCK, F_SETLKW, 0, 4*tdb->header.hash_size);
		tdb->global_lock.count = 0;
	}

	/* remove any locks created during the transaction */
	if (tdb->num_locks != 0) {
		for (i=0;i<tdb->num_lockrecs;i++) {
			tdb_brlock(tdb,FREELIST_TOP+4*tdb->lockrecs[i].list,
				   F_UNLCK,F_SETLKW, 0, 1);
		}
		tdb->num_locks = 0;
		tdb->num_lockrecs = 0;
		SAFE_FREE(tdb->lockrecs);
	}

	/* restore the normal io methods */
	tdb->methods = tdb->transaction->io_methods;

	tdb_brlock(tdb, FREELIST_TOP, F_UNLCK, F_SETLKW, 0, 0);
	tdb_transaction_unlock(tdb);
	SAFE_FREE(tdb->transaction->hash_heads);
	SAFE_FREE(tdb->transaction);
	
	return 0;
}

/*
  sync to disk
*/
static int transaction_sync(struct tdb_context *tdb, tdb_off_t offset, tdb_len_t length)
{	
	if (fsync(tdb->fd) != 0) {
		tdb->ecode = TDB_ERR_IO;
		TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction: fsync failed\n"));
		return -1;
	}
#ifdef HAVE_MMAP
	if (tdb->map_ptr) {
		tdb_off_t moffset = offset & ~(tdb->page_size-1);
		if (msync(moffset + (char *)tdb->map_ptr, 
			  length + (offset - moffset), MS_SYNC) != 0) {
			tdb->ecode = TDB_ERR_IO;
			TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction: msync failed - %s\n",
				 strerror(errno)));
			return -1;
		}
	}
#endif
	return 0;
}


/*
  work out how much space the linearised recovery data will consume
*/
static tdb_len_t tdb_recovery_size(struct tdb_context *tdb)
{
	tdb_len_t recovery_size = 0;
	int i;

	recovery_size = sizeof(uint32_t);
	for (i=0;i<tdb->transaction->num_blocks;i++) {
		if (i * tdb->transaction->block_size >= tdb->transaction->old_map_size) {
			break;
		}
		if (tdb->transaction->blocks[i] == NULL) {
			continue;
		}
		recovery_size += 2*sizeof(tdb_off_t);
		if (i == tdb->transaction->num_blocks-1) {
			recovery_size += tdb->transaction->last_block_size;
		} else {
			recovery_size += tdb->transaction->block_size;
		}
	}	

	return recovery_size;
}

/*
  allocate the recovery area, or use an existing recovery area if it is
  large enough
*/
static int tdb_recovery_allocate(struct tdb_context *tdb, 
				 tdb_len_t *recovery_size,
				 tdb_off_t *recovery_offset,
				 tdb_len_t *recovery_max_size)
{
	struct list_struct rec;
	const struct tdb_methods *methods = tdb->transaction->io_methods;
	tdb_off_t recovery_head;

	if (tdb_ofs_read(tdb, TDB_RECOVERY_HEAD, &recovery_head) == -1) {
		TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_recovery_allocate: failed to read recovery head\n"));
		return -1;
	}

	rec.rec_len = 0;

	if (recovery_head != 0 && 
	    methods->tdb_read(tdb, recovery_head, &rec, sizeof(rec), DOCONV()) == -1) {
		TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_recovery_allocate: failed to read recovery record\n"));
		return -1;
	}

	*recovery_size = tdb_recovery_size(tdb);

	if (recovery_head != 0 && *recovery_size <= rec.rec_len) {
		/* it fits in the existing area */
		*recovery_max_size = rec.rec_len;
		*recovery_offset = recovery_head;
		return 0;
	}

	/* we need to free up the old recovery area, then allocate a
	   new one at the end of the file. Note that we cannot use
	   tdb_allocate() to allocate the new one as that might return
	   us an area that is being currently used (as of the start of
	   the transaction) */
	if (recovery_head != 0) {
		if (tdb_free(tdb, recovery_head, &rec) == -1) {
			TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_recovery_allocate: failed to free previous recovery area\n"));
			return -1;
		}
	}

	/* the tdb_free() call might have increased the recovery size */
	*recovery_size = tdb_recovery_size(tdb);

	/* round up to a multiple of page size */
	*recovery_max_size = TDB_ALIGN(sizeof(rec) + *recovery_size, tdb->page_size) - sizeof(rec);
	*recovery_offset = tdb->map_size;
	recovery_head = *recovery_offset;

	if (methods->tdb_expand_file(tdb, tdb->transaction->old_map_size, 
				     (tdb->map_size - tdb->transaction->old_map_size) +
				     sizeof(rec) + *recovery_max_size) == -1) {
		TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_recovery_allocate: failed to create recovery area\n"));
		return -1;
	}

	/* remap the file (if using mmap) */
	methods->tdb_oob(tdb, tdb->map_size + 1, 1);

	/* we have to reset the old map size so that we don't try to expand the file
	   again in the transaction commit, which would destroy the recovery area */
	tdb->transaction->old_map_size = tdb->map_size;

	/* write the recovery header offset and sync - we can sync without a race here
	   as the magic ptr in the recovery record has not been set */
	CONVERT(recovery_head);
	if (methods->tdb_write(tdb, TDB_RECOVERY_HEAD, 
			       &recovery_head, sizeof(tdb_off_t)) == -1) {
		TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_recovery_allocate: failed to write recovery head\n"));
		return -1;
	}
	if (transaction_write_existing(tdb, TDB_RECOVERY_HEAD, &recovery_head, sizeof(tdb_off_t)) == -1) {
		TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_recovery_allocate: failed to write recovery head\n"));
		return -1;
	}

	return 0;
}


/*
  setup the recovery data that will be used on a crash during commit
*/
static int transaction_setup_recovery(struct tdb_context *tdb, 
				      tdb_off_t *magic_offset)
{
	tdb_len_t recovery_size;
	unsigned char *data, *p;
	const struct tdb_methods *methods = tdb->transaction->io_methods;
	struct list_struct *rec;
	tdb_off_t recovery_offset, recovery_max_size;
	tdb_off_t old_map_size = tdb->transaction->old_map_size;
	uint32_t magic, tailer;
	int i;

	/*
	  check that the recovery area has enough space
	*/
	if (tdb_recovery_allocate(tdb, &recovery_size, 
				  &recovery_offset, &recovery_max_size) == -1) {
		return -1;
	}

	data = (unsigned char *)malloc(recovery_size + sizeof(*rec));
	if (data == NULL) {
		tdb->ecode = TDB_ERR_OOM;
		return -1;
	}

	rec = (struct list_struct *)data;
	memset(rec, 0, sizeof(*rec));

	rec->magic    = 0;
	rec->data_len = recovery_size;
	rec->rec_len  = recovery_max_size;
	rec->key_len  = old_map_size;
	CONVERT(rec);

	/* build the recovery data into a single blob to allow us to do a single
	   large write, which should be more efficient */
	p = data + sizeof(*rec);
	for (i=0;i<tdb->transaction->num_blocks;i++) {
		tdb_off_t offset;
		tdb_len_t length;

		if (tdb->transaction->blocks[i] == NULL) {
			continue;
		}

		offset = i * tdb->transaction->block_size;
		length = tdb->transaction->block_size;
		if (i == tdb->transaction->num_blocks-1) {
			length = tdb->transaction->last_block_size;
		}
		
		if (offset >= old_map_size) {
			continue;
		}
		if (offset + length > tdb->transaction->old_map_size) {
			TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_setup_recovery: transaction data over new region boundary\n"));
			free(data);
			tdb->ecode = TDB_ERR_CORRUPT;
			return -1;
		}
		memcpy(p, &offset, 4);
		memcpy(p+4, &length, 4);
		if (DOCONV()) {
			tdb_convert(p, 8);
		}
		/* the recovery area contains the old data, not the
		   new data, so we have to call the original tdb_read
		   method to get it */
		if (methods->tdb_read(tdb, offset, p + 8, length, 0) != 0) {
			free(data);
			tdb->ecode = TDB_ERR_IO;
			return -1;
		}
		p += 8 + length;
	}

	/* and the tailer */
	tailer = sizeof(*rec) + recovery_max_size;
	memcpy(p, &tailer, 4);
	CONVERT(p);

	/* write the recovery data to the recovery area */
	if (methods->tdb_write(tdb, recovery_offset, data, sizeof(*rec) + recovery_size) == -1) {
		TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_setup_recovery: failed to write recovery data\n"));
		free(data);
		tdb->ecode = TDB_ERR_IO;
		return -1;
	}
	if (transaction_write_existing(tdb, recovery_offset, data, sizeof(*rec) + recovery_size) == -1) {
		TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_setup_recovery: failed to write secondary recovery data\n"));
		free(data);
		tdb->ecode = TDB_ERR_IO;
		return -1;
	}

	/* as we don't have ordered writes, we have to sync the recovery
	   data before we update the magic to indicate that the recovery
	   data is present */
	if (transaction_sync(tdb, recovery_offset, sizeof(*rec) + recovery_size) == -1) {
		free(data);
		return -1;
	}

	free(data);

	magic = TDB_RECOVERY_MAGIC;
	CONVERT(magic);

	*magic_offset = recovery_offset + offsetof(struct list_struct, magic);

	if (methods->tdb_write(tdb, *magic_offset, &magic, sizeof(magic)) == -1) {
		TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_setup_recovery: failed to write recovery magic\n"));
		tdb->ecode = TDB_ERR_IO;
		return -1;
	}
	if (transaction_write_existing(tdb, *magic_offset, &magic, sizeof(magic)) == -1) {
		TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_setup_recovery: failed to write secondary recovery magic\n"));
		tdb->ecode = TDB_ERR_IO;
		return -1;
	}

	/* ensure the recovery magic marker is on disk */
	if (transaction_sync(tdb, *magic_offset, sizeof(magic)) == -1) {
		return -1;
	}

	return 0;
}

/*
  commit the current transaction
*/
int tdb_transaction_commit(struct tdb_context *tdb)
{	
	const struct tdb_methods *methods;
	tdb_off_t magic_offset = 0;
	uint32_t zero = 0;
	int i;

	if (tdb->transaction == NULL) {
		TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_transaction_commit: no transaction\n"));
		return -1;
	}

	if (tdb->transaction->transaction_error) {
		tdb->ecode = TDB_ERR_IO;
		tdb_transaction_cancel(tdb);
		TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_transaction_commit: transaction error pending\n"));
		return -1;
	}


	if (tdb->transaction->nesting != 0) {
		tdb->transaction->nesting--;
		return 0;
	}		

	/* check for a null transaction */
	if (tdb->transaction->blocks == NULL) {
		tdb_transaction_cancel(tdb);
		return 0;
	}

	methods = tdb->transaction->io_methods;
	
	/* if there are any locks pending then the caller has not
	   nested their locks properly, so fail the transaction */
	if (tdb->num_locks || tdb->global_lock.count) {
		tdb->ecode = TDB_ERR_LOCK;
		TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_transaction_commit: locks pending on commit\n"));
		tdb_transaction_cancel(tdb);
		return -1;
	}

	/* upgrade the main transaction lock region to a write lock */
	if (tdb_brlock_upgrade(tdb, FREELIST_TOP, 0) == -1) {
		TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_transaction_start: failed to upgrade hash locks\n"));
		tdb->ecode = TDB_ERR_LOCK;
		tdb_transaction_cancel(tdb);
		return -1;
	}

	/* get the global lock - this prevents new users attaching to the database
	   during the commit */
	if (tdb_brlock(tdb, GLOBAL_LOCK, F_WRLCK, F_SETLKW, 0, 1) == -1) {
		TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_transaction_commit: failed to get global lock\n"));
		tdb->ecode = TDB_ERR_LOCK;
		tdb_transaction_cancel(tdb);
		return -1;
	}

	if (!(tdb->flags & TDB_NOSYNC)) {
		/* write the recovery data to the end of the file */
		if (transaction_setup_recovery(tdb, &magic_offset) == -1) {
			TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_commit: failed to setup recovery data\n"));
			tdb_brlock(tdb, GLOBAL_LOCK, F_UNLCK, F_SETLKW, 0, 1);
			tdb_transaction_cancel(tdb);
			return -1;
		}
	}

	/* expand the file to the new size if needed */
	if (tdb->map_size != tdb->transaction->old_map_size) {
		if (methods->tdb_expand_file(tdb, tdb->transaction->old_map_size, 
					     tdb->map_size - 
					     tdb->transaction->old_map_size) == -1) {
			tdb->ecode = TDB_ERR_IO;
			TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_commit: expansion failed\n"));
			tdb_brlock(tdb, GLOBAL_LOCK, F_UNLCK, F_SETLKW, 0, 1);
			tdb_transaction_cancel(tdb);
			return -1;
		}
		tdb->map_size = tdb->transaction->old_map_size;
		methods->tdb_oob(tdb, tdb->map_size + 1, 1);
	}

	/* perform all the writes */
	for (i=0;i<tdb->transaction->num_blocks;i++) {
		tdb_off_t offset;
		tdb_len_t length;

		if (tdb->transaction->blocks[i] == NULL) {
			continue;
		}

		offset = i * tdb->transaction->block_size;
		length = tdb->transaction->block_size;
		if (i == tdb->transaction->num_blocks-1) {
			length = tdb->transaction->last_block_size;
		}

		if (methods->tdb_write(tdb, offset, tdb->transaction->blocks[i], length) == -1) {
			TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_commit: write failed during commit\n"));
			
			/* we've overwritten part of the data and
			   possibly expanded the file, so we need to
			   run the crash recovery code */
			tdb->methods = methods;
			tdb_transaction_recover(tdb); 

			tdb_transaction_cancel(tdb);
			tdb_brlock(tdb, GLOBAL_LOCK, F_UNLCK, F_SETLKW, 0, 1);

			TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_commit: write failed\n"));
			return -1;
		}
		SAFE_FREE(tdb->transaction->blocks[i]);
	} 

	SAFE_FREE(tdb->transaction->blocks);
	tdb->transaction->num_blocks = 0;

	if (!(tdb->flags & TDB_NOSYNC)) {
		/* ensure the new data is on disk */
		if (transaction_sync(tdb, 0, tdb->map_size) == -1) {
			return -1;
		}

		/* remove the recovery marker */
		if (methods->tdb_write(tdb, magic_offset, &zero, 4) == -1) {
			TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_commit: failed to remove recovery magic\n"));
			return -1;
		}

		/* ensure the recovery marker has been removed on disk */
		if (transaction_sync(tdb, magic_offset, 4) == -1) {
			return -1;
		}
	}

	tdb_brlock(tdb, GLOBAL_LOCK, F_UNLCK, F_SETLKW, 0, 1);

	/*
	  TODO: maybe write to some dummy hdr field, or write to magic
	  offset without mmap, before the last sync, instead of the
	  utime() call
	*/

	/* on some systems (like Linux 2.6.x) changes via mmap/msync
	   don't change the mtime of the file, this means the file may
	   not be backed up (as tdb rounding to block sizes means that
	   file size changes are quite rare too). The following forces
	   mtime changes when a transaction completes */
#ifdef HAVE_UTIME
	utime(tdb->name, NULL);
#endif

	/* use a transaction cancel to free memory and remove the
	   transaction locks */
	tdb_transaction_cancel(tdb);

	return 0;
}


/*
  recover from an aborted transaction. Must be called with exclusive
  database write access already established (including the global
  lock to prevent new processes attaching)
*/
int tdb_transaction_recover(struct tdb_context *tdb)
{
	tdb_off_t recovery_head, recovery_eof;
	unsigned char *data, *p;
	uint32_t zero = 0;
	struct list_struct rec;

	/* find the recovery area */
	if (tdb_ofs_read(tdb, TDB_RECOVERY_HEAD, &recovery_head) == -1) {
		TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_recover: failed to read recovery head\n"));
		tdb->ecode = TDB_ERR_IO;
		return -1;
	}

	if (recovery_head == 0) {
		/* we have never allocated a recovery record */
		return 0;
	}

	/* read the recovery record */
	if (tdb->methods->tdb_read(tdb, recovery_head, &rec, 
				   sizeof(rec), DOCONV()) == -1) {
		TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_recover: failed to read recovery record\n"));		
		tdb->ecode = TDB_ERR_IO;
		return -1;
	}

	if (rec.magic != TDB_RECOVERY_MAGIC) {
		/* there is no valid recovery data */
		return 0;
	}

	if (tdb->read_only) {
		TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_recover: attempt to recover read only database\n"));
		tdb->ecode = TDB_ERR_CORRUPT;
		return -1;
	}

	recovery_eof = rec.key_len;

	data = (unsigned char *)malloc(rec.data_len);
	if (data == NULL) {
		TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_recover: failed to allocate recovery data\n"));		
		tdb->ecode = TDB_ERR_OOM;
		return -1;
	}

	/* read the full recovery data */
	if (tdb->methods->tdb_read(tdb, recovery_head + sizeof(rec), data,
				   rec.data_len, 0) == -1) {
		TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_recover: failed to read recovery data\n"));		
		tdb->ecode = TDB_ERR_IO;
		return -1;
	}

	/* recover the file data */
	p = data;
	while (p+8 < data + rec.data_len) {
		uint32_t ofs, len;
		if (DOCONV()) {
			tdb_convert(p, 8);
		}
		memcpy(&ofs, p, 4);
		memcpy(&len, p+4, 4);

		if (tdb->methods->tdb_write(tdb, ofs, p+8, len) == -1) {
			free(data);
			TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_recover: failed to recover %d bytes at offset %d\n", len, ofs));
			tdb->ecode = TDB_ERR_IO;
			return -1;
		}
		p += 8 + len;
	}

	free(data);

	if (transaction_sync(tdb, 0, tdb->map_size) == -1) {
		TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_recover: failed to sync recovery\n"));
		tdb->ecode = TDB_ERR_IO;
		return -1;
	}

	/* if the recovery area is after the recovered eof then remove it */
	if (recovery_eof <= recovery_head) {
		if (tdb_ofs_write(tdb, TDB_RECOVERY_HEAD, &zero) == -1) {
			TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_recover: failed to remove recovery head\n"));
			tdb->ecode = TDB_ERR_IO;
			return -1;			
		}
	}

	/* remove the recovery magic */
	if (tdb_ofs_write(tdb, recovery_head + offsetof(struct list_struct, magic), 
			  &zero) == -1) {
		TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_recover: failed to remove recovery magic\n"));
		tdb->ecode = TDB_ERR_IO;
		return -1;			
	}
	
	/* reduce the file size to the old size */
	tdb_munmap(tdb);
	if (ftruncate(tdb->fd, recovery_eof) != 0) {
		TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_recover: failed to reduce to recovery size\n"));
		tdb->ecode = TDB_ERR_IO;
		return -1;			
	}
	tdb->map_size = recovery_eof;
	tdb_mmap(tdb);

	if (transaction_sync(tdb, 0, recovery_eof) == -1) {
		TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_recover: failed to sync2 recovery\n"));
		tdb->ecode = TDB_ERR_IO;
		return -1;
	}

	TDB_LOG((tdb, TDB_DEBUG_TRACE, "tdb_transaction_recover: recovered %d byte database\n", 
		 recovery_eof));

	/* all done */
	return 0;
}