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
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
1388
1389
1390
1391
1392
1393
1394
1395
1396
1397
1398
1399
1400
1401
1402
1403
1404
1405
1406
1407
1408
1409
1410
1411
1412
1413
1414
1415
1416
1417
1418
1419
1420
1421
1422
1423
1424
1425
1426
1427
|
/*****************************************************************************
Copyright (c) 1997, 2017, Oracle and/or its affiliates. All Rights Reserved.
Copyright (c) 2017, 2020, MariaDB Corporation.
This program is free software; you can redistribute it and/or modify it under
the terms of the GNU General Public License as published by the Free Software
Foundation; version 2 of the License.
This program 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 General Public License for more details.
You should have received a copy of the GNU General Public License along with
this program; if not, write to the Free Software Foundation, Inc.,
51 Franklin Street, Fifth Floor, Boston, MA 02110-1335 USA
*****************************************************************************/
/**************************************************//**
@file row/row0umod.cc
Undo modify of a row
Created 2/27/1997 Heikki Tuuri
*******************************************************/
#include "row0umod.h"
#include "dict0dict.h"
#include "dict0stats.h"
#include "dict0boot.h"
#include "trx0undo.h"
#include "trx0roll.h"
#include "trx0purge.h"
#include "btr0btr.h"
#include "mach0data.h"
#include "ibuf0ibuf.h"
#include "row0undo.h"
#include "row0vers.h"
#include "row0log.h"
#include "trx0trx.h"
#include "trx0rec.h"
#include "row0row.h"
#include "row0upd.h"
#include "que0que.h"
#include "log0log.h"
/* Considerations on undoing a modify operation.
(1) Undoing a delete marking: all index records should be found. Some of
them may have delete mark already FALSE, if the delete mark operation was
stopped underway, or if the undo operation ended prematurely because of a
system crash.
(2) Undoing an update of a delete unmarked record: the newer version of
an updated secondary index entry should be removed if no prior version
of the clustered index record requires its existence. Otherwise, it should
be delete marked.
(3) Undoing an update of a delete marked record. In this kind of update a
delete marked clustered index record was delete unmarked and possibly also
some of its fields were changed. Now, it is possible that the delete marked
version has become obsolete at the time the undo is started. */
/*************************************************************************
IMPORTANT NOTE: Any operation that generates redo MUST check that there
is enough space in the redo log before for that operation. This is
done by calling log_free_check(). The reason for checking the
availability of the redo log space before the start of the operation is
that we MUST not hold any synchonization objects when performing the
check.
If you make a change in this module make sure that no codepath is
introduced where a call to log_free_check() is bypassed. */
/***********************************************************//**
Undoes a modify in a clustered index record.
@return DB_SUCCESS, DB_FAIL, or error code: we may run out of file space */
static MY_ATTRIBUTE((nonnull, warn_unused_result))
dberr_t
row_undo_mod_clust_low(
/*===================*/
undo_node_t* node, /*!< in: row undo node */
rec_offs** offsets,/*!< out: rec_get_offsets() on the record */
mem_heap_t** offsets_heap,
/*!< in/out: memory heap that can be emptied */
mem_heap_t* heap, /*!< in/out: memory heap */
const dtuple_t**rebuilt_old_pk,
/*!< out: row_log_table_get_pk()
before the update, or NULL if
the table is not being rebuilt online or
the PRIMARY KEY definition does not change */
byte* sys, /*!< out: DB_TRX_ID, DB_ROLL_PTR
for row_log_table_delete() */
que_thr_t* thr, /*!< in: query thread */
mtr_t* mtr, /*!< in: mtr; must be committed before
latching any further pages */
ulint mode) /*!< in: BTR_MODIFY_LEAF or BTR_MODIFY_TREE */
{
btr_pcur_t* pcur;
btr_cur_t* btr_cur;
dberr_t err;
#ifdef UNIV_DEBUG
ibool success;
#endif /* UNIV_DEBUG */
pcur = &node->pcur;
btr_cur = btr_pcur_get_btr_cur(pcur);
#ifdef UNIV_DEBUG
success =
#endif /* UNIV_DEBUG */
btr_pcur_restore_position(mode, pcur, mtr);
ut_ad(success);
ut_ad(rec_get_trx_id(btr_cur_get_rec(btr_cur),
btr_cur_get_index(btr_cur))
== thr_get_trx(thr)->id);
ut_ad(node->ref != &trx_undo_metadata
|| node->update->info_bits == REC_INFO_METADATA_ADD
|| node->update->info_bits == REC_INFO_METADATA_ALTER);
if (mode != BTR_MODIFY_LEAF
&& dict_index_is_online_ddl(btr_cur_get_index(btr_cur))) {
*rebuilt_old_pk = row_log_table_get_pk(
btr_cur_get_rec(btr_cur),
btr_cur_get_index(btr_cur), NULL, sys, &heap);
} else {
*rebuilt_old_pk = NULL;
}
if (mode != BTR_MODIFY_TREE) {
ut_ad((mode & ulint(~BTR_ALREADY_S_LATCHED))
== BTR_MODIFY_LEAF);
err = btr_cur_optimistic_update(
BTR_NO_LOCKING_FLAG | BTR_NO_UNDO_LOG_FLAG
| BTR_KEEP_SYS_FLAG,
btr_cur, offsets, offsets_heap,
node->update, node->cmpl_info,
thr, thr_get_trx(thr)->id, mtr);
ut_ad(err != DB_SUCCESS || node->ref != &trx_undo_metadata);
} else {
big_rec_t* dummy_big_rec;
err = btr_cur_pessimistic_update(
BTR_NO_LOCKING_FLAG
| BTR_NO_UNDO_LOG_FLAG
| BTR_KEEP_SYS_FLAG,
btr_cur, offsets, offsets_heap, heap,
&dummy_big_rec, node->update,
node->cmpl_info, thr, thr_get_trx(thr)->id, mtr);
ut_a(!dummy_big_rec);
if (err == DB_SUCCESS
&& node->ref == &trx_undo_metadata
&& btr_cur_get_index(btr_cur)->table->instant
&& node->update->info_bits == REC_INFO_METADATA_ADD) {
btr_reset_instant(*btr_cur_get_index(btr_cur), false,
mtr);
}
}
if (err == DB_SUCCESS
&& btr_cur_get_index(btr_cur)->table->id == DICT_COLUMNS_ID) {
/* This is rolling back an UPDATE or DELETE on SYS_COLUMNS.
If it was part of an instant ALTER TABLE operation, we
must evict the table definition, so that it can be
reloaded after the dictionary operation has been
completed. At this point, any corresponding operation
to the metadata record will have been rolled back. */
const dfield_t& table_id = *dtuple_get_nth_field(node->row, 0);
ut_ad(dfield_get_len(&table_id) == 8);
node->trx->evict_table(mach_read_from_8(static_cast<byte*>(
table_id.data)));
}
return(err);
}
/** Get the byte offset of the DB_TRX_ID column
@param[in] rec clustered index record
@param[in] index clustered index
@return the byte offset of DB_TRX_ID, from the start of rec */
static ulint row_trx_id_offset(const rec_t* rec, const dict_index_t* index)
{
ut_ad(index->n_uniq <= MAX_REF_PARTS);
ulint trx_id_offset = index->trx_id_offset;
if (!trx_id_offset) {
/* Reserve enough offsets for the PRIMARY KEY and 2 columns
so that we can access DB_TRX_ID, DB_ROLL_PTR. */
rec_offs offsets_[REC_OFFS_HEADER_SIZE + MAX_REF_PARTS + 2];
rec_offs_init(offsets_);
mem_heap_t* heap = NULL;
const ulint trx_id_pos = index->n_uniq ? index->n_uniq : 1;
rec_offs* offsets = rec_get_offsets(rec, index, offsets_, true,
trx_id_pos + 1, &heap);
ut_ad(!heap);
ulint len;
trx_id_offset = rec_get_nth_field_offs(
offsets, trx_id_pos, &len);
ut_ad(len == DATA_TRX_ID_LEN);
}
return trx_id_offset;
}
/** Determine if rollback must execute a purge-like operation.
@param[in,out] node row undo
@param[in,out] mtr mini-transaction
@return whether the record should be purged */
static bool row_undo_mod_must_purge(undo_node_t* node, mtr_t* mtr)
{
ut_ad(node->rec_type == TRX_UNDO_UPD_DEL_REC);
ut_ad(!node->table->is_temporary());
btr_cur_t* btr_cur = btr_pcur_get_btr_cur(&node->pcur);
ut_ad(btr_cur->index->is_primary());
DEBUG_SYNC_C("rollback_purge_clust");
if (!purge_sys.changes_visible(node->new_trx_id, node->table->name)) {
return false;
}
const rec_t* rec = btr_cur_get_rec(btr_cur);
return trx_read_trx_id(rec + row_trx_id_offset(rec, btr_cur->index))
== node->new_trx_id;
}
/***********************************************************//**
Undoes a modify in a clustered index record. Sets also the node state for the
next round of undo.
@return DB_SUCCESS or error code: we may run out of file space */
static MY_ATTRIBUTE((nonnull, warn_unused_result))
dberr_t
row_undo_mod_clust(
/*===============*/
undo_node_t* node, /*!< in: row undo node */
que_thr_t* thr) /*!< in: query thread */
{
btr_pcur_t* pcur;
mtr_t mtr;
bool have_latch = false;
dberr_t err;
dict_index_t* index;
bool online;
ut_ad(thr_get_trx(thr) == node->trx);
ut_ad(node->trx->dict_operation_lock_mode);
ut_ad(node->trx->in_rollback);
log_free_check();
pcur = &node->pcur;
index = btr_cur_get_index(btr_pcur_get_btr_cur(pcur));
ut_ad(index->is_primary());
mtr.start();
if (index->table->is_temporary()) {
mtr.set_log_mode(MTR_LOG_NO_REDO);
} else {
index->set_modified(mtr);
ut_ad(lock_table_has_locks(index->table));
}
online = dict_index_is_online_ddl(index);
if (online) {
ut_ad(node->trx->dict_operation_lock_mode != RW_X_LATCH);
mtr_s_lock_index(index, &mtr);
}
mem_heap_t* heap = mem_heap_create(1024);
mem_heap_t* offsets_heap = NULL;
rec_offs* offsets = NULL;
const dtuple_t* rebuilt_old_pk;
byte sys[DATA_TRX_ID_LEN + DATA_ROLL_PTR_LEN];
/* Try optimistic processing of the record, keeping changes within
the index page */
err = row_undo_mod_clust_low(node, &offsets, &offsets_heap,
heap, &rebuilt_old_pk, sys,
thr, &mtr, online
? BTR_MODIFY_LEAF | BTR_ALREADY_S_LATCHED
: BTR_MODIFY_LEAF);
if (err != DB_SUCCESS) {
btr_pcur_commit_specify_mtr(pcur, &mtr);
/* We may have to modify tree structure: do a pessimistic
descent down the index tree */
mtr.start();
if (index->table->is_temporary()) {
mtr.set_log_mode(MTR_LOG_NO_REDO);
} else {
index->set_modified(mtr);
}
err = row_undo_mod_clust_low(
node, &offsets, &offsets_heap,
heap, &rebuilt_old_pk, sys,
thr, &mtr, BTR_MODIFY_TREE);
ut_ad(err == DB_SUCCESS || err == DB_OUT_OF_FILE_SPACE);
}
/* Online rebuild cannot be initiated while we are holding
dict_sys.latch and index->lock. (It can be aborted.) */
ut_ad(online || !dict_index_is_online_ddl(index));
if (err == DB_SUCCESS && online) {
ut_ad(rw_lock_own_flagged(
&index->lock,
RW_LOCK_FLAG_S | RW_LOCK_FLAG_X
| RW_LOCK_FLAG_SX));
switch (node->rec_type) {
case TRX_UNDO_DEL_MARK_REC:
row_log_table_insert(
btr_pcur_get_rec(pcur), index, offsets);
break;
case TRX_UNDO_UPD_EXIST_REC:
row_log_table_update(
btr_pcur_get_rec(pcur), index, offsets,
rebuilt_old_pk);
break;
case TRX_UNDO_UPD_DEL_REC:
row_log_table_delete(
btr_pcur_get_rec(pcur), index, offsets, sys);
break;
default:
ut_ad(0);
break;
}
}
/**
* when scrubbing, and records gets cleared,
* the transaction id is not present afterwards.
* this is safe as: since the record is on free-list
* it can be reallocated at any time after this mtr-commits
* which is just below
*/
ut_ad(srv_immediate_scrub_data_uncompressed
|| row_get_rec_trx_id(btr_pcur_get_rec(pcur), index, offsets)
== node->new_trx_id);
btr_pcur_commit_specify_mtr(pcur, &mtr);
DEBUG_SYNC_C("rollback_undo_pk");
if (err != DB_SUCCESS) {
goto func_exit;
}
/* FIXME: Perform the below operations in the above
mini-transaction when possible. */
if (node->rec_type == TRX_UNDO_UPD_DEL_REC) {
/* In delete-marked records, DB_TRX_ID must
always refer to an existing update_undo log record. */
ut_ad(node->new_trx_id);
mtr.start();
if (!btr_pcur_restore_position(BTR_MODIFY_LEAF, pcur, &mtr)) {
goto mtr_commit_exit;
}
ut_ad(rec_get_deleted_flag(btr_pcur_get_rec(pcur),
dict_table_is_comp(node->table)));
if (index->table->is_temporary()) {
mtr.set_log_mode(MTR_LOG_NO_REDO);
if (btr_cur_optimistic_delete(&pcur->btr_cur, 0,
&mtr)) {
goto mtr_commit_exit;
}
btr_pcur_commit_specify_mtr(pcur, &mtr);
} else {
index->set_modified(mtr);
have_latch = true;
purge_sys.latch.rd_lock(SRW_LOCK_CALL);
if (!row_undo_mod_must_purge(node, &mtr)) {
goto mtr_commit_exit;
}
if (btr_cur_optimistic_delete(&pcur->btr_cur, 0,
&mtr)) {
goto mtr_commit_exit;
}
purge_sys.latch.rd_unlock();
btr_pcur_commit_specify_mtr(pcur, &mtr);
have_latch = false;
}
mtr.start();
if (!btr_pcur_restore_position(
BTR_MODIFY_TREE | BTR_LATCH_FOR_DELETE,
pcur, &mtr)) {
goto mtr_commit_exit;
}
ut_ad(rec_get_deleted_flag(btr_pcur_get_rec(pcur),
dict_table_is_comp(node->table)));
if (index->table->is_temporary()) {
mtr.set_log_mode(MTR_LOG_NO_REDO);
} else {
have_latch = true;
purge_sys.latch.rd_lock(SRW_LOCK_CALL);
if (!row_undo_mod_must_purge(node, &mtr)) {
goto mtr_commit_exit;
}
index->set_modified(mtr);
}
/* This operation is analogous to purge, we can free
also inherited externally stored fields. We can also
assume that the record was complete (including BLOBs),
because it had been delete-marked after it had been
completely inserted. Therefore, we are passing
rollback=false, just like purge does. */
btr_cur_pessimistic_delete(&err, FALSE, &pcur->btr_cur, 0,
false, &mtr);
ut_ad(err == DB_SUCCESS || err == DB_OUT_OF_FILE_SPACE);
} else if (!index->table->is_temporary() && node->new_trx_id) {
/* We rolled back a record so that it still exists.
We must reset the DB_TRX_ID if the history is no
longer accessible by any active read view. */
mtr.start();
if (!btr_pcur_restore_position(BTR_MODIFY_LEAF, pcur, &mtr)) {
goto mtr_commit_exit;
}
rec_t* rec = btr_pcur_get_rec(pcur);
have_latch = true;
purge_sys.latch.rd_lock(SRW_LOCK_CALL);
if (!purge_sys.changes_visible(node->new_trx_id,
node->table->name)) {
goto mtr_commit_exit;
}
ulint trx_id_offset = index->trx_id_offset;
ulint trx_id_pos = index->n_uniq ? index->n_uniq : 1;
/* Reserve enough offsets for the PRIMARY KEY and
2 columns so that we can access DB_TRX_ID, DB_ROLL_PTR. */
rec_offs offsets_[REC_OFFS_HEADER_SIZE + MAX_REF_PARTS + 2];
if (trx_id_offset) {
#ifdef UNIV_DEBUG
ut_ad(rec_offs_validate(NULL, index, offsets));
if (buf_block_get_page_zip(
btr_pcur_get_block(&node->pcur))) {
/* Below, page_zip_write_trx_id_and_roll_ptr()
needs offsets to access DB_TRX_ID,DB_ROLL_PTR.
We already computed offsets for possibly
another record in the clustered index.
Because the PRIMARY KEY is fixed-length,
the offsets for the PRIMARY KEY and
DB_TRX_ID,DB_ROLL_PTR are still valid.
Silence the rec_offs_validate() assertion. */
rec_offs_make_valid(rec, index, true, offsets);
}
#endif
} else if (rec_is_metadata(rec, *index)) {
ut_ad(!buf_block_get_page_zip(btr_pcur_get_block(
pcur)));
for (unsigned i = index->first_user_field(); i--; ) {
trx_id_offset += index->fields[i].fixed_len;
}
} else {
ut_ad(index->n_uniq <= MAX_REF_PARTS);
rec_offs_init(offsets_);
offsets = rec_get_offsets(
rec, index, offsets_, true, trx_id_pos + 2,
&heap);
ulint len;
trx_id_offset = rec_get_nth_field_offs(
offsets, trx_id_pos, &len);
ut_ad(len == DATA_TRX_ID_LEN);
}
if (trx_read_trx_id(rec + trx_id_offset) == node->new_trx_id) {
ut_ad(!rec_get_deleted_flag(
rec, dict_table_is_comp(node->table))
|| rec_is_alter_metadata(rec, *index));
index->set_modified(mtr);
buf_block_t* block = btr_pcur_get_block(pcur);
if (UNIV_LIKELY_NULL(block->page.zip.data)) {
page_zip_write_trx_id_and_roll_ptr(
block, rec, offsets, trx_id_pos,
0, 1ULL << ROLL_PTR_INSERT_FLAG_POS,
&mtr);
} else {
size_t offs = page_offset(rec + trx_id_offset);
mtr.memset(block, offs, DATA_TRX_ID_LEN, 0);
offs += DATA_TRX_ID_LEN;
mtr.write<1,mtr_t::MAYBE_NOP>(*block,
block->frame
+ offs, 0x80U);
mtr.memset(block, offs + 1,
DATA_ROLL_PTR_LEN - 1, 0);
}
}
} else {
goto func_exit;
}
mtr_commit_exit:
if (have_latch) {
purge_sys.latch.rd_unlock();
}
btr_pcur_commit_specify_mtr(pcur, &mtr);
func_exit:
if (offsets_heap) {
mem_heap_free(offsets_heap);
}
mem_heap_free(heap);
return(err);
}
/***********************************************************//**
Delete marks or removes a secondary index entry if found.
@return DB_SUCCESS, DB_FAIL, or DB_OUT_OF_FILE_SPACE */
static MY_ATTRIBUTE((nonnull, warn_unused_result))
dberr_t
row_undo_mod_del_mark_or_remove_sec_low(
/*====================================*/
undo_node_t* node, /*!< in: row undo node */
que_thr_t* thr, /*!< in: query thread */
dict_index_t* index, /*!< in: index */
dtuple_t* entry, /*!< in: index entry */
ulint mode) /*!< in: latch mode BTR_MODIFY_LEAF or
BTR_MODIFY_TREE */
{
btr_pcur_t pcur;
btr_cur_t* btr_cur;
ibool success;
dberr_t err = DB_SUCCESS;
mtr_t mtr;
mtr_t mtr_vers;
row_search_result search_result;
const bool modify_leaf = mode == BTR_MODIFY_LEAF;
row_mtr_start(&mtr, index, !modify_leaf);
if (!index->is_committed()) {
/* The index->online_status may change if the index is
or was being created online, but not committed yet. It
is protected by index->lock. */
if (modify_leaf) {
mode = BTR_MODIFY_LEAF | BTR_ALREADY_S_LATCHED;
mtr_s_lock_index(index, &mtr);
} else {
ut_ad(mode == (BTR_MODIFY_TREE | BTR_LATCH_FOR_DELETE));
mtr_sx_lock_index(index, &mtr);
}
if (row_log_online_op_try(index, entry, 0)) {
goto func_exit_no_pcur;
}
} else {
/* For secondary indexes,
index->online_status==ONLINE_INDEX_COMPLETE if
index->is_committed(). */
ut_ad(!dict_index_is_online_ddl(index));
}
btr_cur = btr_pcur_get_btr_cur(&pcur);
if (dict_index_is_spatial(index)) {
if (modify_leaf) {
btr_cur->thr = thr;
mode |= BTR_RTREE_DELETE_MARK;
}
mode |= BTR_RTREE_UNDO_INS;
}
search_result = row_search_index_entry(index, entry, mode,
&pcur, &mtr);
switch (UNIV_EXPECT(search_result, ROW_FOUND)) {
case ROW_NOT_FOUND:
/* In crash recovery, the secondary index record may
be missing if the UPDATE did not have time to insert
the secondary index records before the crash. When we
are undoing that UPDATE in crash recovery, the record
may be missing.
In normal processing, if an update ends in a deadlock
before it has inserted all updated secondary index
records, then the undo will not find those records. */
goto func_exit;
case ROW_FOUND:
break;
case ROW_BUFFERED:
case ROW_NOT_DELETED_REF:
/* These are invalid outcomes, because the mode passed
to row_search_index_entry() did not include any of the
flags BTR_INSERT, BTR_DELETE, or BTR_DELETE_MARK. */
ut_error;
}
/* We should remove the index record if no prior version of the row,
which cannot be purged yet, requires its existence. If some requires,
we should delete mark the record. */
mtr_vers.start();
success = btr_pcur_restore_position(BTR_SEARCH_LEAF, &(node->pcur),
&mtr_vers);
ut_a(success);
/* For temporary table, we can skip to check older version of
clustered index entry, because there is no MVCC or purge. */
if (node->table->is_temporary()
|| row_vers_old_has_index_entry(
false, btr_pcur_get_rec(&node->pcur),
&mtr_vers, index, entry, 0, 0)) {
btr_rec_set_deleted<true>(btr_cur_get_block(btr_cur),
btr_cur_get_rec(btr_cur), &mtr);
} else {
/* Remove the index record */
if (dict_index_is_spatial(index)) {
rec_t* rec = btr_pcur_get_rec(&pcur);
if (rec_get_deleted_flag(rec,
dict_table_is_comp(index->table))) {
ib::error() << "Record found in index "
<< index->name << " is deleted marked"
" on rollback update.";
ut_ad(0);
}
}
if (modify_leaf) {
err = btr_cur_optimistic_delete(btr_cur, 0, &mtr)
? DB_SUCCESS : DB_FAIL;
} else {
/* Passing rollback=false,
because we are deleting a secondary index record:
the distinction only matters when deleting a
record that contains externally stored columns. */
ut_ad(!index->is_primary());
btr_cur_pessimistic_delete(&err, FALSE, btr_cur, 0,
false, &mtr);
/* The delete operation may fail if we have little
file space left: TODO: easiest to crash the database
and restart with more file space */
}
}
btr_pcur_commit_specify_mtr(&(node->pcur), &mtr_vers);
func_exit:
btr_pcur_close(&pcur);
func_exit_no_pcur:
mtr_commit(&mtr);
return(err);
}
/***********************************************************//**
Delete marks or removes a secondary index entry if found.
NOTE that if we updated the fields of a delete-marked secondary index record
so that alphabetically they stayed the same, e.g., 'abc' -> 'aBc', we cannot
return to the original values because we do not know them. But this should
not cause problems because in row0sel.cc, in queries we always retrieve the
clustered index record or an earlier version of it, if the secondary index
record through which we do the search is delete-marked.
@return DB_SUCCESS or DB_OUT_OF_FILE_SPACE */
static MY_ATTRIBUTE((nonnull, warn_unused_result))
dberr_t
row_undo_mod_del_mark_or_remove_sec(
/*================================*/
undo_node_t* node, /*!< in: row undo node */
que_thr_t* thr, /*!< in: query thread */
dict_index_t* index, /*!< in: index */
dtuple_t* entry) /*!< in: index entry */
{
dberr_t err;
err = row_undo_mod_del_mark_or_remove_sec_low(node, thr, index,
entry, BTR_MODIFY_LEAF);
if (err == DB_SUCCESS) {
return(err);
}
err = row_undo_mod_del_mark_or_remove_sec_low(node, thr, index,
entry, BTR_MODIFY_TREE | BTR_LATCH_FOR_DELETE);
return(err);
}
/***********************************************************//**
Delete unmarks a secondary index entry which must be found. It might not be
delete-marked at the moment, but it does not harm to unmark it anyway. We also
need to update the fields of the secondary index record if we updated its
fields but alphabetically they stayed the same, e.g., 'abc' -> 'aBc'.
@retval DB_SUCCESS on success
@retval DB_FAIL if BTR_MODIFY_TREE should be tried
@retval DB_OUT_OF_FILE_SPACE when running out of tablespace
@retval DB_DUPLICATE_KEY if the value was missing
and an insert would lead to a duplicate exists */
static MY_ATTRIBUTE((nonnull, warn_unused_result))
dberr_t
row_undo_mod_del_unmark_sec_and_undo_update(
/*========================================*/
ulint mode, /*!< in: search mode: BTR_MODIFY_LEAF or
BTR_MODIFY_TREE */
que_thr_t* thr, /*!< in: query thread */
dict_index_t* index, /*!< in: index */
dtuple_t* entry) /*!< in: index entry */
{
btr_pcur_t pcur;
btr_cur_t* btr_cur = btr_pcur_get_btr_cur(&pcur);
upd_t* update;
dberr_t err = DB_SUCCESS;
big_rec_t* dummy_big_rec;
mtr_t mtr;
trx_t* trx = thr_get_trx(thr);
const ulint flags
= BTR_KEEP_SYS_FLAG | BTR_NO_LOCKING_FLAG;
row_search_result search_result;
ulint orig_mode = mode;
ut_ad(trx->id != 0);
if (dict_index_is_spatial(index)) {
/* FIXME: Currently we do a 2-pass search for the undo
due to avoid undel-mark a wrong rec in rolling back in
partial update. Later, we could log some info in
secondary index updates to avoid this. */
ut_ad(mode & BTR_MODIFY_LEAF);
mode |= BTR_RTREE_DELETE_MARK;
}
try_again:
row_mtr_start(&mtr, index, !(mode & BTR_MODIFY_LEAF));
if (!index->is_committed()) {
/* The index->online_status may change if the index is
or was being created online, but not committed yet. It
is protected by index->lock. */
if (mode == BTR_MODIFY_LEAF) {
mode = BTR_MODIFY_LEAF | BTR_ALREADY_S_LATCHED;
mtr_s_lock_index(index, &mtr);
} else {
ut_ad(mode == BTR_MODIFY_TREE);
mtr_sx_lock_index(index, &mtr);
}
if (row_log_online_op_try(index, entry, trx->id)) {
goto func_exit_no_pcur;
}
} else {
/* For secondary indexes,
index->online_status==ONLINE_INDEX_COMPLETE if
index->is_committed(). */
ut_ad(!dict_index_is_online_ddl(index));
}
btr_cur->thr = thr;
search_result = row_search_index_entry(index, entry, mode,
&pcur, &mtr);
switch (search_result) {
mem_heap_t* heap;
mem_heap_t* offsets_heap;
rec_offs* offsets;
case ROW_BUFFERED:
case ROW_NOT_DELETED_REF:
/* These are invalid outcomes, because the mode passed
to row_search_index_entry() did not include any of the
flags BTR_INSERT, BTR_DELETE, or BTR_DELETE_MARK. */
ut_error;
case ROW_NOT_FOUND:
/* For spatial index, if first search didn't find an
undel-marked rec, try to find a del-marked rec. */
if (dict_index_is_spatial(index) && btr_cur->rtr_info->fd_del) {
if (mode != orig_mode) {
mode = orig_mode;
btr_pcur_close(&pcur);
mtr_commit(&mtr);
goto try_again;
}
}
if (index->is_committed()) {
/* During online secondary index creation, it
is possible that MySQL is waiting for a
meta-data lock upgrade before invoking
ha_innobase::commit_inplace_alter_table()
while this ROLLBACK is executing. InnoDB has
finished building the index, but it does not
yet exist in MySQL. In this case, we suppress
the printout to the error log. */
ib::warn() << "Record in index " << index->name
<< " of table " << index->table->name
<< " was not found on rollback, trying to"
" insert: " << *entry
<< " at: " << rec_index_print(
btr_cur_get_rec(btr_cur), index);
}
if (btr_cur->up_match >= dict_index_get_n_unique(index)
|| btr_cur->low_match >= dict_index_get_n_unique(index)) {
if (index->is_committed()) {
ib::warn() << "Record in index " << index->name
<< " was not found on rollback, and"
" a duplicate exists";
}
err = DB_DUPLICATE_KEY;
break;
}
/* Insert the missing record that we were trying to
delete-unmark. */
big_rec_t* big_rec;
rec_t* insert_rec;
offsets = NULL;
offsets_heap = NULL;
err = btr_cur_optimistic_insert(
flags, btr_cur, &offsets, &offsets_heap,
entry, &insert_rec, &big_rec,
0, thr, &mtr);
ut_ad(!big_rec);
if (err == DB_FAIL && mode == BTR_MODIFY_TREE) {
err = btr_cur_pessimistic_insert(
flags, btr_cur,
&offsets, &offsets_heap,
entry, &insert_rec, &big_rec,
0, thr, &mtr);
/* There are no off-page columns in
secondary indexes. */
ut_ad(!big_rec);
}
if (err == DB_SUCCESS) {
page_update_max_trx_id(
btr_cur_get_block(btr_cur),
btr_cur_get_page_zip(btr_cur),
trx->id, &mtr);
}
if (offsets_heap) {
mem_heap_free(offsets_heap);
}
break;
case ROW_FOUND:
btr_rec_set_deleted<false>(btr_cur_get_block(btr_cur),
btr_cur_get_rec(btr_cur), &mtr);
heap = mem_heap_create(
sizeof(upd_t)
+ dtuple_get_n_fields(entry) * sizeof(upd_field_t));
offsets_heap = NULL;
offsets = rec_get_offsets(
btr_cur_get_rec(btr_cur),
index, NULL, true, ULINT_UNDEFINED, &offsets_heap);
update = row_upd_build_sec_rec_difference_binary(
btr_cur_get_rec(btr_cur), index, offsets, entry, heap);
if (upd_get_n_fields(update) == 0) {
/* Do nothing */
} else if (mode != BTR_MODIFY_TREE) {
/* Try an optimistic updating of the record, keeping
changes within the page */
/* TODO: pass offsets, not &offsets */
err = btr_cur_optimistic_update(
flags, btr_cur, &offsets, &offsets_heap,
update, 0, thr, thr_get_trx(thr)->id, &mtr);
switch (err) {
case DB_OVERFLOW:
case DB_UNDERFLOW:
case DB_ZIP_OVERFLOW:
err = DB_FAIL;
default:
break;
}
} else {
err = btr_cur_pessimistic_update(
flags, btr_cur, &offsets, &offsets_heap,
heap, &dummy_big_rec,
update, 0, thr, thr_get_trx(thr)->id, &mtr);
ut_a(!dummy_big_rec);
}
mem_heap_free(heap);
mem_heap_free(offsets_heap);
}
btr_pcur_close(&pcur);
func_exit_no_pcur:
mtr_commit(&mtr);
return(err);
}
/***********************************************************//**
Flags a secondary index corrupted. */
static MY_ATTRIBUTE((nonnull))
void
row_undo_mod_sec_flag_corrupted(
/*============================*/
trx_t* trx, /*!< in/out: transaction */
dict_index_t* index) /*!< in: secondary index */
{
ut_ad(!dict_index_is_clust(index));
switch (trx->dict_operation_lock_mode) {
case RW_S_LATCH:
/* Because row_undo() is holding an S-latch
on the data dictionary during normal rollback,
we can only mark the index corrupted in the
data dictionary cache. TODO: fix this somehow.*/
mutex_enter(&dict_sys.mutex);
dict_set_corrupted_index_cache_only(index);
mutex_exit(&dict_sys.mutex);
break;
default:
ut_ad(0);
/* fall through */
case RW_X_LATCH:
/* This should be the rollback of a data dictionary
transaction. */
dict_set_corrupted(index, trx, "rollback");
}
}
/***********************************************************//**
Undoes a modify in secondary indexes when undo record type is UPD_DEL.
@return DB_SUCCESS or DB_OUT_OF_FILE_SPACE */
static MY_ATTRIBUTE((nonnull, warn_unused_result))
dberr_t
row_undo_mod_upd_del_sec(
/*=====================*/
undo_node_t* node, /*!< in: row undo node */
que_thr_t* thr) /*!< in: query thread */
{
mem_heap_t* heap;
dberr_t err = DB_SUCCESS;
ut_ad(node->rec_type == TRX_UNDO_UPD_DEL_REC);
ut_ad(!node->undo_row);
heap = mem_heap_create(1024);
while (node->index != NULL) {
dict_index_t* index = node->index;
dtuple_t* entry;
if (index->type & DICT_FTS) {
dict_table_next_uncorrupted_index(node->index);
continue;
}
/* During online index creation,
HA_ALTER_INPLACE_COPY_NO_LOCK or HA_ALTER_INPLACE_NOCOPY_NO_LOCk
should guarantee that any active transaction has not modified
indexed columns such that col->ord_part was 0 at the
time when the undo log record was written. When we get
to roll back an undo log entry TRX_UNDO_DEL_MARK_REC,
it should always cover all affected indexes. */
entry = row_build_index_entry(
node->row, node->ext, index, heap);
if (UNIV_UNLIKELY(!entry)) {
/* The database must have crashed after
inserting a clustered index record but before
writing all the externally stored columns of
that record. Because secondary index entries
are inserted after the clustered index record,
we may assume that the secondary index record
does not exist. However, this situation may
only occur during the rollback of incomplete
transactions. */
ut_a(thr_get_trx(thr) == trx_roll_crash_recv_trx);
} else {
err = row_undo_mod_del_mark_or_remove_sec(
node, thr, index, entry);
if (UNIV_UNLIKELY(err != DB_SUCCESS)) {
break;
}
}
mem_heap_empty(heap);
dict_table_next_uncorrupted_index(node->index);
}
mem_heap_free(heap);
return(err);
}
/***********************************************************//**
Undoes a modify in secondary indexes when undo record type is DEL_MARK.
@return DB_SUCCESS or DB_OUT_OF_FILE_SPACE */
static MY_ATTRIBUTE((nonnull, warn_unused_result))
dberr_t
row_undo_mod_del_mark_sec(
/*======================*/
undo_node_t* node, /*!< in: row undo node */
que_thr_t* thr) /*!< in: query thread */
{
mem_heap_t* heap;
dberr_t err = DB_SUCCESS;
ut_ad(!node->undo_row);
heap = mem_heap_create(1024);
while (node->index != NULL) {
dict_index_t* index = node->index;
dtuple_t* entry;
if (index->type == DICT_FTS) {
dict_table_next_uncorrupted_index(node->index);
continue;
}
/* During online index creation,
HA_ALTER_INPLACE_COPY_NO_LOCK or HA_ALTER_INPLACE_NOCOPY_NO_LOCK
should guarantee that any active transaction has not modified
indexed columns such that col->ord_part was 0 at the
time when the undo log record was written. When we get
to roll back an undo log entry TRX_UNDO_DEL_MARK_REC,
it should always cover all affected indexes. */
entry = row_build_index_entry(
node->row, node->ext, index, heap);
ut_a(entry);
err = row_undo_mod_del_unmark_sec_and_undo_update(
BTR_MODIFY_LEAF, thr, index, entry);
if (err == DB_FAIL) {
err = row_undo_mod_del_unmark_sec_and_undo_update(
BTR_MODIFY_TREE, thr, index, entry);
}
if (err == DB_DUPLICATE_KEY) {
row_undo_mod_sec_flag_corrupted(
thr_get_trx(thr), index);
err = DB_SUCCESS;
/* Do not return any error to the caller. The
duplicate will be reported by ALTER TABLE or
CREATE UNIQUE INDEX. Unfortunately we cannot
report the duplicate key value to the DDL
thread, because the altered_table object is
private to its call stack. */
} else if (err != DB_SUCCESS) {
break;
}
mem_heap_empty(heap);
dict_table_next_uncorrupted_index(node->index);
}
mem_heap_free(heap);
return(err);
}
/***********************************************************//**
Undoes a modify in secondary indexes when undo record type is UPD_EXIST.
@return DB_SUCCESS or DB_OUT_OF_FILE_SPACE */
static MY_ATTRIBUTE((nonnull, warn_unused_result))
dberr_t
row_undo_mod_upd_exist_sec(
/*=======================*/
undo_node_t* node, /*!< in: row undo node */
que_thr_t* thr) /*!< in: query thread */
{
mem_heap_t* heap;
dberr_t err = DB_SUCCESS;
if (node->index == NULL
|| ((node->cmpl_info & UPD_NODE_NO_ORD_CHANGE))) {
/* No change in secondary indexes */
return(err);
}
heap = mem_heap_create(1024);
while (node->index != NULL) {
dict_index_t* index = node->index;
dtuple_t* entry;
if (dict_index_is_spatial(index)) {
if (!row_upd_changes_ord_field_binary_func(
index, node->update,
#ifdef UNIV_DEBUG
thr,
#endif /* UNIV_DEBUG */
node->row,
node->ext, ROW_BUILD_FOR_UNDO)) {
dict_table_next_uncorrupted_index(node->index);
continue;
}
} else {
if (index->type == DICT_FTS
|| !row_upd_changes_ord_field_binary(index,
node->update,
thr, node->row,
node->ext)) {
dict_table_next_uncorrupted_index(node->index);
continue;
}
}
/* Build the newest version of the index entry */
entry = row_build_index_entry(node->row, node->ext,
index, heap);
if (UNIV_UNLIKELY(!entry)) {
/* The server must have crashed in
row_upd_clust_rec_by_insert() before
the updated externally stored columns (BLOBs)
of the new clustered index entry were written. */
/* The table must be in DYNAMIC or COMPRESSED
format. REDUNDANT and COMPACT formats
store a local 768-byte prefix of each
externally stored column. */
ut_a(dict_table_has_atomic_blobs(index->table));
/* This is only legitimate when
rolling back an incomplete transaction
after crash recovery. */
ut_a(thr_get_trx(thr)->is_recovered);
/* The server must have crashed before
completing the insert of the new
clustered index entry and before
inserting to the secondary indexes.
Because node->row was not yet written
to this index, we can ignore it. But
we must restore node->undo_row. */
} else {
/* NOTE that if we updated the fields of a
delete-marked secondary index record so that
alphabetically they stayed the same, e.g.,
'abc' -> 'aBc', we cannot return to the
original values because we do not know them.
But this should not cause problems because
in row0sel.cc, in queries we always retrieve
the clustered index record or an earlier
version of it, if the secondary index record
through which we do the search is
delete-marked. */
err = row_undo_mod_del_mark_or_remove_sec(
node, thr, index, entry);
if (err != DB_SUCCESS) {
break;
}
}
mem_heap_empty(heap);
/* We may have to update the delete mark in the
secondary index record of the previous version of
the row. We also need to update the fields of
the secondary index record if we updated its fields
but alphabetically they stayed the same, e.g.,
'abc' -> 'aBc'. */
if (dict_index_is_spatial(index)) {
entry = row_build_index_entry_low(node->undo_row,
node->undo_ext,
index, heap,
ROW_BUILD_FOR_UNDO);
} else {
entry = row_build_index_entry(node->undo_row,
node->undo_ext,
index, heap);
}
ut_a(entry);
err = row_undo_mod_del_unmark_sec_and_undo_update(
BTR_MODIFY_LEAF, thr, index, entry);
if (err == DB_FAIL) {
err = row_undo_mod_del_unmark_sec_and_undo_update(
BTR_MODIFY_TREE, thr, index, entry);
}
if (err == DB_DUPLICATE_KEY) {
row_undo_mod_sec_flag_corrupted(
thr_get_trx(thr), index);
err = DB_SUCCESS;
} else if (err != DB_SUCCESS) {
break;
}
mem_heap_empty(heap);
dict_table_next_uncorrupted_index(node->index);
}
mem_heap_free(heap);
return(err);
}
/** Parse an update undo record.
@param[in,out] node row rollback state
@param[in] dict_locked whether the data dictionary cache is locked */
static bool row_undo_mod_parse_undo_rec(undo_node_t* node, bool dict_locked)
{
dict_index_t* clust_index;
byte* ptr;
undo_no_t undo_no;
table_id_t table_id;
trx_id_t trx_id;
roll_ptr_t roll_ptr;
byte info_bits;
ulint type;
ulint cmpl_info;
bool dummy_extern;
ut_ad(node->state == UNDO_UPDATE_PERSISTENT
|| node->state == UNDO_UPDATE_TEMPORARY);
ut_ad(node->trx->in_rollback);
ut_ad(!trx_undo_roll_ptr_is_insert(node->roll_ptr));
ptr = trx_undo_rec_get_pars(node->undo_rec, &type, &cmpl_info,
&dummy_extern, &undo_no, &table_id);
node->rec_type = type;
if (node->state == UNDO_UPDATE_PERSISTENT) {
node->table = dict_table_open_on_id(table_id, dict_locked,
DICT_TABLE_OP_NORMAL);
} else if (!dict_locked) {
mutex_enter(&dict_sys.mutex);
node->table = dict_sys.get_temporary_table(table_id);
mutex_exit(&dict_sys.mutex);
} else {
node->table = dict_sys.get_temporary_table(table_id);
}
if (!node->table) {
return false;
}
ut_ad(!node->table->skip_alter_undo);
if (UNIV_UNLIKELY(!node->table->is_accessible())) {
close_table:
/* Normally, tables should not disappear or become
unaccessible during ROLLBACK, because they should be
protected by InnoDB table locks. Corruption could be
a valid exception.
FIXME: When running out of temporary tablespace, it
would probably be better to just drop all temporary
tables (and temporary undo log records) of the current
connection, instead of doing this rollback. */
dict_table_close(node->table, dict_locked, FALSE);
node->table = NULL;
return false;
}
clust_index = dict_table_get_first_index(node->table);
ptr = trx_undo_update_rec_get_sys_cols(ptr, &trx_id, &roll_ptr,
&info_bits);
ptr = trx_undo_rec_get_row_ref(ptr, clust_index, &(node->ref),
node->heap);
ptr = trx_undo_update_rec_get_update(ptr, clust_index, type, trx_id,
roll_ptr, info_bits,
node->heap, &(node->update));
node->new_trx_id = trx_id;
node->cmpl_info = cmpl_info;
ut_ad(!node->ref->info_bits);
if (node->update->info_bits & REC_INFO_MIN_REC_FLAG) {
if ((node->update->info_bits & ~REC_INFO_DELETED_FLAG)
!= REC_INFO_MIN_REC_FLAG) {
ut_ad("wrong info_bits in undo log record" == 0);
goto close_table;
}
/* This must be an undo log record for a subsequent
instant ALTER TABLE, extending the metadata record. */
ut_ad(clust_index->is_instant());
ut_ad(clust_index->table->instant
|| !(node->update->info_bits & REC_INFO_DELETED_FLAG));
node->ref = &trx_undo_metadata;
node->update->info_bits = (node->update->info_bits
& REC_INFO_DELETED_FLAG)
? REC_INFO_METADATA_ALTER
: REC_INFO_METADATA_ADD;
}
if (!row_undo_search_clust_to_pcur(node)) {
/* As long as this rolling-back transaction exists,
the PRIMARY KEY value pointed to by the undo log
record should exist.
However, if InnoDB is killed during a rollback, or
shut down during the rollback of recovered
transactions, then after restart we may try to roll
back some of the same undo log records again, because
trx_roll_try_truncate() is not being invoked after
every undo log record.
It is also possible that the record
was not modified yet (the DB_ROLL_PTR does not match
node->roll_ptr) and thus there is nothing to roll back.
btr_cur_upd_lock_and_undo() only writes the undo log
record after successfully acquiring an exclusive lock
on the the clustered index record. That lock will not
be released before the transaction is committed or
fully rolled back. (Exception: if the server was
killed, restarted, and shut down again before the
rollback of the recovered transaction was completed,
it is possible that the transaction was partially
rolled back and locks released.) */
goto close_table;
}
/* Extract indexed virtual columns from undo log */
if (node->ref != &trx_undo_metadata && node->table->n_v_cols) {
row_upd_replace_vcol(node->row, node->table,
node->update, false, node->undo_row,
(node->cmpl_info & UPD_NODE_NO_ORD_CHANGE)
? NULL : ptr);
}
return true;
}
/***********************************************************//**
Undoes a modify operation on a row of a table.
@return DB_SUCCESS or error code */
dberr_t
row_undo_mod(
/*=========*/
undo_node_t* node, /*!< in: row undo node */
que_thr_t* thr) /*!< in: query thread */
{
dberr_t err;
ut_ad(thr_get_trx(thr) == node->trx);
const bool dict_locked = node->trx->dict_operation_lock_mode
== RW_X_LATCH;
if (!row_undo_mod_parse_undo_rec(node, dict_locked)) {
return DB_SUCCESS;
}
node->index = dict_table_get_first_index(node->table);
ut_ad(dict_index_is_clust(node->index));
if (node->ref->info_bits) {
ut_ad(node->ref->is_metadata());
goto rollback_clust;
}
/* Skip the clustered index (the first index) */
node->index = dict_table_get_next_index(node->index);
/* Skip all corrupted secondary index */
dict_table_skip_corrupt_index(node->index);
switch (node->rec_type) {
case TRX_UNDO_UPD_EXIST_REC:
err = row_undo_mod_upd_exist_sec(node, thr);
break;
case TRX_UNDO_DEL_MARK_REC:
err = row_undo_mod_del_mark_sec(node, thr);
break;
case TRX_UNDO_UPD_DEL_REC:
err = row_undo_mod_upd_del_sec(node, thr);
break;
default:
ut_error;
err = DB_ERROR;
}
if (err == DB_SUCCESS) {
rollback_clust:
err = row_undo_mod_clust(node, thr);
bool update_statistics
= !(node->cmpl_info & UPD_NODE_NO_ORD_CHANGE);
if (err == DB_SUCCESS && node->table->stat_initialized) {
switch (node->rec_type) {
case TRX_UNDO_UPD_EXIST_REC:
break;
case TRX_UNDO_DEL_MARK_REC:
dict_table_n_rows_inc(node->table);
update_statistics = update_statistics
|| !srv_stats_include_delete_marked;
break;
case TRX_UNDO_UPD_DEL_REC:
dict_table_n_rows_dec(node->table);
update_statistics = update_statistics
|| !srv_stats_include_delete_marked;
break;
}
/* Do not attempt to update statistics when
executing ROLLBACK in the InnoDB SQL
interpreter, because in that case we would
already be holding dict_sys.mutex, which
would be acquired when updating statistics. */
if (update_statistics && !dict_locked) {
dict_stats_update_if_needed(node->table,
*node->trx);
} else {
node->table->stat_modified_counter++;
}
}
}
dict_table_close(node->table, dict_locked, FALSE);
node->table = NULL;
return(err);
}
|