Repository navigation
Expand file tree
/
Copy pathbtree.c
More file actions
8277 lines (7694 loc) · 276 KB
/
Copy pathbtree.c
File metadata and controls
8277 lines (7694 loc) · 276 KB
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
/*
** 2004 April 6
**
** The author disclaims copyright to this source code. In place of
** a legal notice, here is a blessing:
**
** May you do good and not evil.
** May you find forgiveness for yourself and forgive others.
** May you share freely, never taking more than you give.
**
*************************************************************************
** This file implements a external (disk-based) database using BTrees.
** See the header comment on "btreeInt.h" for additional information.
** Including a description of file format and an overview of operation.
*/
#include "btreeInt.h"
/*
** The header string that appears at the beginning of every
** SQLite database.
*/
static const char zMagicHeader[] = SQLITE_FILE_HEADER;
/*
** Set this global variable to 1 to enable tracing using the TRACE
** macro.
*/
#if 0
int sqlite3BtreeTrace=1; /* True to enable tracing */
# define TRACE(X) if(sqlite3BtreeTrace){printf X;fflush(stdout);}
#else
# define TRACE(X)
#endif
/*
** Extract a 2-byte big-endian integer from an array of unsigned bytes.
** But if the value is zero, make it 65536.
**
** This routine is used to extract the "offset to cell content area" value
** from the header of a btree page. If the page size is 65536 and the page
** is empty, the offset should be 65536, but the 2-byte value stores zero.
** This routine makes the necessary adjustment to 65536.
*/
#define get2byteNotZero(X) (((((int)get2byte(X))-1)&0xffff)+1)
#ifndef SQLITE_OMIT_SHARED_CACHE
/*
** A list of BtShared objects that are eligible for participation
** in shared cache. This variable has file scope during normal builds,
** but the test harness needs to access it so we make it global for
** test builds.
**
** Access to this variable is protected by SQLITE_MUTEX_STATIC_MASTER.
*/
#ifdef SQLITE_TEST
BtShared *SQLITE_WSD sqlite3SharedCacheList = 0;
#else
static BtShared *SQLITE_WSD sqlite3SharedCacheList = 0;
#endif
#endif /* SQLITE_OMIT_SHARED_CACHE */
#ifndef SQLITE_OMIT_SHARED_CACHE
/*
** Enable or disable the shared pager and schema features.
**
** This routine has no effect on existing database connections.
** The shared cache setting effects only future calls to
** sqlite3_open(), sqlite3_open16(), or sqlite3_open_v2().
*/
int sqlite3_enable_shared_cache(int enable){
sqlite3GlobalConfig.sharedCacheEnabled = enable;
return SQLITE_OK;
}
#endif
#ifdef SQLITE_OMIT_SHARED_CACHE
/*
** The functions querySharedCacheTableLock(), setSharedCacheTableLock(),
** and clearAllSharedCacheTableLocks()
** manipulate entries in the BtShared.pLock linked list used to store
** shared-cache table level locks. If the library is compiled with the
** shared-cache feature disabled, then there is only ever one user
** of each BtShared structure and so this locking is not necessary.
** So define the lock related functions as no-ops.
*/
#define querySharedCacheTableLock(a,b,c) SQLITE_OK
#define setSharedCacheTableLock(a,b,c) SQLITE_OK
#define clearAllSharedCacheTableLocks(a)
#define downgradeAllSharedCacheTableLocks(a)
#define hasSharedCacheTableLock(a,b,c,d) 1
#define hasReadConflicts(a, b) 0
#endif
#ifndef SQLITE_OMIT_SHARED_CACHE
#ifdef SQLITE_DEBUG
/*
**** This function is only used as part of an assert() statement. ***
**
** Check to see if pBtree holds the required locks to read or write to the
** table with root page iRoot. Return 1 if it does and 0 if not.
**
** For example, when writing to a table with root-page iRoot via
** Btree connection pBtree:
**
** assert( hasSharedCacheTableLock(pBtree, iRoot, 0, WRITE_LOCK) );
**
** When writing to an index that resides in a sharable database, the
** caller should have first obtained a lock specifying the root page of
** the corresponding table. This makes things a bit more complicated,
** as this module treats each table as a separate structure. To determine
** the table corresponding to the index being written, this
** function has to search through the database schema.
**
** Instead of a lock on the table/index rooted at page iRoot, the caller may
** hold a write-lock on the schema table (root page 1). This is also
** acceptable.
*/
static int hasSharedCacheTableLock(
Btree *pBtree, /* Handle that must hold lock */
Pgno iRoot, /* Root page of b-tree */
int isIndex, /* True if iRoot is the root of an index b-tree */
int eLockType /* Required lock type (READ_LOCK or WRITE_LOCK) */
){
Schema *pSchema = (Schema *)pBtree->pBt->pSchema;
Pgno iTab = 0;
BtLock *pLock;
/* If this database is not shareable, or if the client is reading
** and has the read-uncommitted flag set, then no lock is required.
** Return true immediately.
*/
if( (pBtree->sharable==0)
|| (eLockType==READ_LOCK && (pBtree->db->flags & SQLITE_ReadUncommitted))
){
return 1;
}
/* If the client is reading or writing an index and the schema is
** not loaded, then it is too difficult to actually check to see if
** the correct locks are held. So do not bother - just return true.
** This case does not come up very often anyhow.
*/
if( isIndex && (!pSchema || (pSchema->flags&DB_SchemaLoaded)==0) ){
return 1;
}
/* Figure out the root-page that the lock should be held on. For table
** b-trees, this is just the root page of the b-tree being read or
** written. For index b-trees, it is the root page of the associated
** table. */
if( isIndex ){
HashElem *p;
for(p=sqliteHashFirst(&pSchema->idxHash); p; p=sqliteHashNext(p)){
Index *pIdx = (Index *)sqliteHashData(p);
if( pIdx->tnum==(int)iRoot ){
iTab = pIdx->pTable->tnum;
}
}
}else{
iTab = iRoot;
}
/* Search for the required lock. Either a write-lock on root-page iTab, a
** write-lock on the schema table, or (if the client is reading) a
** read-lock on iTab will suffice. Return 1 if any of these are found. */
for(pLock=pBtree->pBt->pLock; pLock; pLock=pLock->pNext){
if( pLock->pBtree==pBtree
&& (pLock->iTable==iTab || (pLock->eLock==WRITE_LOCK && pLock->iTable==1))
&& pLock->eLock>=eLockType
){
return 1;
}
}
/* Failed to find the required lock. */
return 0;
}
#endif /* SQLITE_DEBUG */
#ifdef SQLITE_DEBUG
/*
**** This function may be used as part of assert() statements only. ****
**
** Return true if it would be illegal for pBtree to write into the
** table or index rooted at iRoot because other shared connections are
** simultaneously reading that same table or index.
**
** It is illegal for pBtree to write if some other Btree object that
** shares the same BtShared object is currently reading or writing
** the iRoot table. Except, if the other Btree object has the
** read-uncommitted flag set, then it is OK for the other object to
** have a read cursor.
**
** For example, before writing to any part of the table or index
** rooted at page iRoot, one should call:
**
** assert( !hasReadConflicts(pBtree, iRoot) );
*/
static int hasReadConflicts(Btree *pBtree, Pgno iRoot){
BtCursor *p;
for(p=pBtree->pBt->pCursor; p; p=p->pNext){
if( p->pgnoRoot==iRoot
&& p->pBtree!=pBtree
&& 0==(p->pBtree->db->flags & SQLITE_ReadUncommitted)
){
return 1;
}
}
return 0;
}
#endif /* #ifdef SQLITE_DEBUG */
/*
** Query to see if Btree handle p may obtain a lock of type eLock
** (READ_LOCK or WRITE_LOCK) on the table with root-page iTab. Return
** SQLITE_OK if the lock may be obtained (by calling
** setSharedCacheTableLock()), or SQLITE_LOCKED if not.
*/
static int querySharedCacheTableLock(Btree *p, Pgno iTab, u8 eLock){
BtShared *pBt = p->pBt;
BtLock *pIter;
assert( sqlite3BtreeHoldsMutex(p) );
assert( eLock==READ_LOCK || eLock==WRITE_LOCK );
assert( p->db!=0 );
assert( !(p->db->flags&SQLITE_ReadUncommitted)||eLock==WRITE_LOCK||iTab==1 );
/* If requesting a write-lock, then the Btree must have an open write
** transaction on this file. And, obviously, for this to be so there
** must be an open write transaction on the file itself.
*/
assert( eLock==READ_LOCK || (p==pBt->pWriter && p->inTrans==TRANS_WRITE) );
assert( eLock==READ_LOCK || pBt->inTransaction==TRANS_WRITE );
/* This routine is a no-op if the shared-cache is not enabled */
if( !p->sharable ){
return SQLITE_OK;
}
/* If some other connection is holding an exclusive lock, the
** requested lock may not be obtained.
*/
if( pBt->pWriter!=p && (pBt->btsFlags & BTS_EXCLUSIVE)!=0 ){
sqlite3ConnectionBlocked(p->db, pBt->pWriter->db);
return SQLITE_LOCKED_SHAREDCACHE;
}
for(pIter=pBt->pLock; pIter; pIter=pIter->pNext){
/* The condition (pIter->eLock!=eLock) in the following if(...)
** statement is a simplification of:
**
** (eLock==WRITE_LOCK || pIter->eLock==WRITE_LOCK)
**
** since we know that if eLock==WRITE_LOCK, then no other connection
** may hold a WRITE_LOCK on any table in this file (since there can
** only be a single writer).
*/
assert( pIter->eLock==READ_LOCK || pIter->eLock==WRITE_LOCK );
assert( eLock==READ_LOCK || pIter->pBtree==p || pIter->eLock==READ_LOCK);
if( pIter->pBtree!=p && pIter->iTable==iTab && pIter->eLock!=eLock ){
sqlite3ConnectionBlocked(p->db, pIter->pBtree->db);
if( eLock==WRITE_LOCK ){
assert( p==pBt->pWriter );
pBt->btsFlags |= BTS_PENDING;
}
return SQLITE_LOCKED_SHAREDCACHE;
}
}
return SQLITE_OK;
}
#endif /* !SQLITE_OMIT_SHARED_CACHE */
#ifndef SQLITE_OMIT_SHARED_CACHE
/*
** Add a lock on the table with root-page iTable to the shared-btree used
** by Btree handle p. Parameter eLock must be either READ_LOCK or
** WRITE_LOCK.
**
** This function assumes the following:
**
** (a) The specified Btree object p is connected to a sharable
** database (one with the BtShared.sharable flag set), and
**
** (b) No other Btree objects hold a lock that conflicts
** with the requested lock (i.e. querySharedCacheTableLock() has
** already been called and returned SQLITE_OK).
**
** SQLITE_OK is returned if the lock is added successfully. SQLITE_NOMEM
** is returned if a malloc attempt fails.
*/
static int setSharedCacheTableLock(Btree *p, Pgno iTable, u8 eLock){
BtShared *pBt = p->pBt;
BtLock *pLock = 0;
BtLock *pIter;
assert( sqlite3BtreeHoldsMutex(p) );
assert( eLock==READ_LOCK || eLock==WRITE_LOCK );
assert( p->db!=0 );
/* A connection with the read-uncommitted flag set will never try to
** obtain a read-lock using this function. The only read-lock obtained
** by a connection in read-uncommitted mode is on the sqlite_master
** table, and that lock is obtained in BtreeBeginTrans(). */
assert( 0==(p->db->flags&SQLITE_ReadUncommitted) || eLock==WRITE_LOCK );
/* This function should only be called on a sharable b-tree after it
** has been determined that no other b-tree holds a conflicting lock. */
assert( p->sharable );
assert( SQLITE_OK==querySharedCacheTableLock(p, iTable, eLock) );
/* First search the list for an existing lock on this table. */
for(pIter=pBt->pLock; pIter; pIter=pIter->pNext){
if( pIter->iTable==iTable && pIter->pBtree==p ){
pLock = pIter;
break;
}
}
/* If the above search did not find a BtLock struct associating Btree p
** with table iTable, allocate one and link it into the list.
*/
if( !pLock ){
pLock = (BtLock *)sqlite3MallocZero(sizeof(BtLock));
if( !pLock ){
return SQLITE_NOMEM;
}
pLock->iTable = iTable;
pLock->pBtree = p;
pLock->pNext = pBt->pLock;
pBt->pLock = pLock;
}
/* Set the BtLock.eLock variable to the maximum of the current lock
** and the requested lock. This means if a write-lock was already held
** and a read-lock requested, we don't incorrectly downgrade the lock.
*/
assert( WRITE_LOCK>READ_LOCK );
if( eLock>pLock->eLock ){
pLock->eLock = eLock;
}
return SQLITE_OK;
}
#endif /* !SQLITE_OMIT_SHARED_CACHE */
#ifndef SQLITE_OMIT_SHARED_CACHE
/*
** Release all the table locks (locks obtained via calls to
** the setSharedCacheTableLock() procedure) held by Btree object p.
**
** This function assumes that Btree p has an open read or write
** transaction. If it does not, then the BTS_PENDING flag
** may be incorrectly cleared.
*/
static void clearAllSharedCacheTableLocks(Btree *p){
BtShared *pBt = p->pBt;
BtLock **ppIter = &pBt->pLock;
assert( sqlite3BtreeHoldsMutex(p) );
assert( p->sharable || 0==*ppIter );
assert( p->inTrans>0 );
while( *ppIter ){
BtLock *pLock = *ppIter;
assert( (pBt->btsFlags & BTS_EXCLUSIVE)==0 || pBt->pWriter==pLock->pBtree );
assert( pLock->pBtree->inTrans>=pLock->eLock );
if( pLock->pBtree==p ){
*ppIter = pLock->pNext;
assert( pLock->iTable!=1 || pLock==&p->lock );
if( pLock->iTable!=1 ){
sqlite3_free(pLock);
}
}else{
ppIter = &pLock->pNext;
}
}
assert( (pBt->btsFlags & BTS_PENDING)==0 || pBt->pWriter );
if( pBt->pWriter==p ){
pBt->pWriter = 0;
pBt->btsFlags &= ~(BTS_EXCLUSIVE|BTS_PENDING);
}else if( pBt->nTransaction==2 ){
/* This function is called when Btree p is concluding its
** transaction. If there currently exists a writer, and p is not
** that writer, then the number of locks held by connections other
** than the writer must be about to drop to zero. In this case
** set the BTS_PENDING flag to 0.
**
** If there is not currently a writer, then BTS_PENDING must
** be zero already. So this next line is harmless in that case.
*/
pBt->btsFlags &= ~BTS_PENDING;
}
}
/*
** This function changes all write-locks held by Btree p into read-locks.
*/
static void downgradeAllSharedCacheTableLocks(Btree *p){
BtShared *pBt = p->pBt;
if( pBt->pWriter==p ){
BtLock *pLock;
pBt->pWriter = 0;
pBt->btsFlags &= ~(BTS_EXCLUSIVE|BTS_PENDING);
for(pLock=pBt->pLock; pLock; pLock=pLock->pNext){
assert( pLock->eLock==READ_LOCK || pLock->pBtree==p );
pLock->eLock = READ_LOCK;
}
}
}
#endif /* SQLITE_OMIT_SHARED_CACHE */
static void releasePage(MemPage *pPage); /* Forward reference */
/*
***** This routine is used inside of assert() only ****
**
** Verify that the cursor holds the mutex on its BtShared
*/
#ifdef SQLITE_DEBUG
static int cursorHoldsMutex(BtCursor *p){
return sqlite3_mutex_held(p->pBt->mutex);
}
#endif
#ifndef SQLITE_OMIT_INCRBLOB
/*
** Invalidate the overflow page-list cache for cursor pCur, if any.
*/
static void invalidateOverflowCache(BtCursor *pCur){
assert( cursorHoldsMutex(pCur) );
sqlite3_free(pCur->aOverflow);
pCur->aOverflow = 0;
}
/*
** Invalidate the overflow page-list cache for all cursors opened
** on the shared btree structure pBt.
*/
static void invalidateAllOverflowCache(BtShared *pBt){
BtCursor *p;
assert( sqlite3_mutex_held(pBt->mutex) );
for(p=pBt->pCursor; p; p=p->pNext){
invalidateOverflowCache(p);
}
}
/*
** This function is called before modifying the contents of a table
** to invalidate any incrblob cursors that are open on the
** row or one of the rows being modified.
**
** If argument isClearTable is true, then the entire contents of the
** table is about to be deleted. In this case invalidate all incrblob
** cursors open on any row within the table with root-page pgnoRoot.
**
** Otherwise, if argument isClearTable is false, then the row with
** rowid iRow is being replaced or deleted. In this case invalidate
** only those incrblob cursors open on that specific row.
*/
static void invalidateIncrblobCursors(
Btree *pBtree, /* The database file to check */
i64 iRow, /* The rowid that might be changing */
int isClearTable /* True if all rows are being deleted */
){
BtCursor *p;
BtShared *pBt = pBtree->pBt;
assert( sqlite3BtreeHoldsMutex(pBtree) );
for(p=pBt->pCursor; p; p=p->pNext){
if( p->isIncrblobHandle && (isClearTable || p->info.nKey==iRow) ){
p->eState = CURSOR_INVALID;
}
}
}
#else
/* Stub functions when INCRBLOB is omitted */
#define invalidateOverflowCache(x)
#define invalidateAllOverflowCache(x)
#define invalidateIncrblobCursors(x,y,z)
#endif /* SQLITE_OMIT_INCRBLOB */
/*
** Set bit pgno of the BtShared.pHasContent bitvec. This is called
** when a page that previously contained data becomes a free-list leaf
** page.
**
** The BtShared.pHasContent bitvec exists to work around an obscure
** bug caused by the interaction of two useful IO optimizations surrounding
** free-list leaf pages:
**
** 1) When all data is deleted from a page and the page becomes
** a free-list leaf page, the page is not written to the database
** (as free-list leaf pages contain no meaningful data). Sometimes
** such a page is not even journalled (as it will not be modified,
** why bother journalling it?).
**
** 2) When a free-list leaf page is reused, its content is not read
** from the database or written to the journal file (why should it
** be, if it is not at all meaningful?).
**
** By themselves, these optimizations work fine and provide a handy
** performance boost to bulk delete or insert operations. However, if
** a page is moved to the free-list and then reused within the same
** transaction, a problem comes up. If the page is not journalled when
** it is moved to the free-list and it is also not journalled when it
** is extracted from the free-list and reused, then the original data
** may be lost. In the event of a rollback, it may not be possible
** to restore the database to its original configuration.
**
** The solution is the BtShared.pHasContent bitvec. Whenever a page is
** moved to become a free-list leaf page, the corresponding bit is
** set in the bitvec. Whenever a leaf page is extracted from the free-list,
** optimization 2 above is omitted if the corresponding bit is already
** set in BtShared.pHasContent. The contents of the bitvec are cleared
** at the end of every transaction.
*/
static int btreeSetHasContent(BtShared *pBt, Pgno pgno){
int rc = SQLITE_OK;
if( !pBt->pHasContent ){
assert( pgno<=pBt->nPage );
pBt->pHasContent = sqlite3BitvecCreate(pBt->nPage);
if( !pBt->pHasContent ){
rc = SQLITE_NOMEM;
}
}
if( rc==SQLITE_OK && pgno<=sqlite3BitvecSize(pBt->pHasContent) ){
rc = sqlite3BitvecSet(pBt->pHasContent, pgno);
}
return rc;
}
/*
** Query the BtShared.pHasContent vector.
**
** This function is called when a free-list leaf page is removed from the
** free-list for reuse. It returns false if it is safe to retrieve the
** page from the pager layer with the 'no-content' flag set. True otherwise.
*/
static int btreeGetHasContent(BtShared *pBt, Pgno pgno){
Bitvec *p = pBt->pHasContent;
return (p && (pgno>sqlite3BitvecSize(p) || sqlite3BitvecTest(p, pgno)));
}
/*
** Clear (destroy) the BtShared.pHasContent bitvec. This should be
** invoked at the conclusion of each write-transaction.
*/
static void btreeClearHasContent(BtShared *pBt){
sqlite3BitvecDestroy(pBt->pHasContent);
pBt->pHasContent = 0;
}
/*
** Save the current cursor position in the variables BtCursor.nKey
** and BtCursor.pKey. The cursor's state is set to CURSOR_REQUIRESEEK.
**
** The caller must ensure that the cursor is valid (has eState==CURSOR_VALID)
** prior to calling this routine.
*/
static int saveCursorPosition(BtCursor *pCur){
int rc;
assert( CURSOR_VALID==pCur->eState );
assert( 0==pCur->pKey );
assert( cursorHoldsMutex(pCur) );
rc = sqlite3BtreeKeySize(pCur, &pCur->nKey);
assert( rc==SQLITE_OK ); /* KeySize() cannot fail */
/* If this is an intKey table, then the above call to BtreeKeySize()
** stores the integer key in pCur->nKey. In this case this value is
** all that is required. Otherwise, if pCur is not open on an intKey
** table, then malloc space for and store the pCur->nKey bytes of key
** data.
*/
if( 0==pCur->apPage[0]->intKey ){
void *pKey = sqlite3Malloc( (int)pCur->nKey );
if( pKey ){
rc = sqlite3BtreeKey(pCur, 0, (int)pCur->nKey, pKey);
if( rc==SQLITE_OK ){
pCur->pKey = pKey;
}else{
sqlite3_free(pKey);
}
}else{
rc = SQLITE_NOMEM;
}
}
assert( !pCur->apPage[0]->intKey || !pCur->pKey );
if( rc==SQLITE_OK ){
int i;
for(i=0; i<=pCur->iPage; i++){
releasePage(pCur->apPage[i]);
pCur->apPage[i] = 0;
}
pCur->iPage = -1;
pCur->eState = CURSOR_REQUIRESEEK;
}
invalidateOverflowCache(pCur);
return rc;
}
/*
** Save the positions of all cursors (except pExcept) that are open on
** the table with root-page iRoot. Usually, this is called just before cursor
** pExcept is used to modify the table (BtreeDelete() or BtreeInsert()).
*/
static int saveAllCursors(BtShared *pBt, Pgno iRoot, BtCursor *pExcept){
BtCursor *p;
assert( sqlite3_mutex_held(pBt->mutex) );
assert( pExcept==0 || pExcept->pBt==pBt );
for(p=pBt->pCursor; p; p=p->pNext){
if( p!=pExcept && (0==iRoot || p->pgnoRoot==iRoot) &&
p->eState==CURSOR_VALID ){
int rc = saveCursorPosition(p);
if( SQLITE_OK!=rc ){
return rc;
}
}
}
return SQLITE_OK;
}
/*
** Clear the current cursor position.
*/
void sqlite3BtreeClearCursor(BtCursor *pCur){
assert( cursorHoldsMutex(pCur) );
sqlite3_free(pCur->pKey);
pCur->pKey = 0;
pCur->eState = CURSOR_INVALID;
}
/*
** In this version of BtreeMoveto, pKey is a packed index record
** such as is generated by the OP_MakeRecord opcode. Unpack the
** record and then call BtreeMovetoUnpacked() to do the work.
*/
static int btreeMoveto(
BtCursor *pCur, /* Cursor open on the btree to be searched */
const void *pKey, /* Packed key if the btree is an index */
i64 nKey, /* Integer key for tables. Size of pKey for indices */
int bias, /* Bias search to the high end */
int *pRes /* Write search results here */
){
int rc; /* Status code */
UnpackedRecord *pIdxKey; /* Unpacked index key */
char aSpace[150]; /* Temp space for pIdxKey - to avoid a malloc */
char *pFree = 0;
if( pKey ){
assert( nKey==(i64)(int)nKey );
pIdxKey = sqlite3VdbeAllocUnpackedRecord(
pCur->pKeyInfo, aSpace, sizeof(aSpace), &pFree
);
if( pIdxKey==0 ) return SQLITE_NOMEM;
sqlite3VdbeRecordUnpack(pCur->pKeyInfo, (int)nKey, pKey, pIdxKey);
}else{
pIdxKey = 0;
}
rc = sqlite3BtreeMovetoUnpacked(pCur, pIdxKey, nKey, bias, pRes);
if( pFree ){
sqlite3DbFree(pCur->pKeyInfo->db, pFree);
}
return rc;
}
/*
** Restore the cursor to the position it was in (or as close to as possible)
** when saveCursorPosition() was called. Note that this call deletes the
** saved position info stored by saveCursorPosition(), so there can be
** at most one effective restoreCursorPosition() call after each
** saveCursorPosition().
*/
static int btreeRestoreCursorPosition(BtCursor *pCur){
int rc;
assert( cursorHoldsMutex(pCur) );
assert( pCur->eState>=CURSOR_REQUIRESEEK );
if( pCur->eState==CURSOR_FAULT ){
return pCur->skipNext;
}
pCur->eState = CURSOR_INVALID;
rc = btreeMoveto(pCur, pCur->pKey, pCur->nKey, 0, &pCur->skipNext);
if( rc==SQLITE_OK ){
sqlite3_free(pCur->pKey);
pCur->pKey = 0;
assert( pCur->eState==CURSOR_VALID || pCur->eState==CURSOR_INVALID );
}
return rc;
}
#define restoreCursorPosition(p) \
(p->eState>=CURSOR_REQUIRESEEK ? \
btreeRestoreCursorPosition(p) : \
SQLITE_OK)
/*
** Determine whether or not a cursor has moved from the position it
** was last placed at. Cursors can move when the row they are pointing
** at is deleted out from under them.
**
** This routine returns an error code if something goes wrong. The
** integer *pHasMoved is set to one if the cursor has moved and 0 if not.
*/
int sqlite3BtreeCursorHasMoved(BtCursor *pCur, int *pHasMoved){
int rc;
rc = restoreCursorPosition(pCur);
if( rc ){
*pHasMoved = 1;
return rc;
}
if( pCur->eState!=CURSOR_VALID || pCur->skipNext!=0 ){
*pHasMoved = 1;
}else{
*pHasMoved = 0;
}
return SQLITE_OK;
}
#ifndef SQLITE_OMIT_AUTOVACUUM
/*
** Given a page number of a regular database page, return the page
** number for the pointer-map page that contains the entry for the
** input page number.
**
** Return 0 (not a valid page) for pgno==1 since there is
** no pointer map associated with page 1. The integrity_check logic
** requires that ptrmapPageno(*,1)!=1.
*/
static Pgno ptrmapPageno(BtShared *pBt, Pgno pgno){
int nPagesPerMapPage;
Pgno iPtrMap, ret;
assert( sqlite3_mutex_held(pBt->mutex) );
if( pgno<2 ) return 0;
nPagesPerMapPage = (pBt->usableSize/5)+1;
iPtrMap = (pgno-2)/nPagesPerMapPage;
ret = (iPtrMap*nPagesPerMapPage) + 2;
if( ret==PENDING_BYTE_PAGE(pBt) ){
ret++;
}
return ret;
}
/*
** Write an entry into the pointer map.
**
** This routine updates the pointer map entry for page number 'key'
** so that it maps to type 'eType' and parent page number 'pgno'.
**
** If *pRC is initially non-zero (non-SQLITE_OK) then this routine is
** a no-op. If an error occurs, the appropriate error code is written
** into *pRC.
*/
static void ptrmapPut(BtShared *pBt, Pgno key, u8 eType, Pgno parent, int *pRC){
DbPage *pDbPage; /* The pointer map page */
u8 *pPtrmap; /* The pointer map data */
Pgno iPtrmap; /* The pointer map page number */
int offset; /* Offset in pointer map page */
int rc; /* Return code from subfunctions */
if( *pRC ) return;
assert( sqlite3_mutex_held(pBt->mutex) );
/* The master-journal page number must never be used as a pointer map page */
assert( 0==PTRMAP_ISPAGE(pBt, PENDING_BYTE_PAGE(pBt)) );
assert( pBt->autoVacuum );
if( key==0 ){
*pRC = SQLITE_CORRUPT_BKPT;
return;
}
iPtrmap = PTRMAP_PAGENO(pBt, key);
rc = sqlite3PagerGet(pBt->pPager, iPtrmap, &pDbPage);
if( rc!=SQLITE_OK ){
*pRC = rc;
return;
}
offset = PTRMAP_PTROFFSET(iPtrmap, key);
if( offset<0 ){
*pRC = SQLITE_CORRUPT_BKPT;
goto ptrmap_exit;
}
assert( offset <= (int)pBt->usableSize-5 );
pPtrmap = (u8 *)sqlite3PagerGetData(pDbPage);
if( eType!=pPtrmap[offset] || get4byte(&pPtrmap[offset+1])!=parent ){
TRACE(("PTRMAP_UPDATE: %d->(%d,%d)\n", key, eType, parent));
*pRC= rc = sqlite3PagerWrite(pDbPage);
if( rc==SQLITE_OK ){
pPtrmap[offset] = eType;
put4byte(&pPtrmap[offset+1], parent);
}
}
ptrmap_exit:
sqlite3PagerUnref(pDbPage);
}
/*
** Read an entry from the pointer map.
**
** This routine retrieves the pointer map entry for page 'key', writing
** the type and parent page number to *pEType and *pPgno respectively.
** An error code is returned if something goes wrong, otherwise SQLITE_OK.
*/
static int ptrmapGet(BtShared *pBt, Pgno key, u8 *pEType, Pgno *pPgno){
DbPage *pDbPage; /* The pointer map page */
int iPtrmap; /* Pointer map page index */
u8 *pPtrmap; /* Pointer map page data */
int offset; /* Offset of entry in pointer map */
int rc;
assert( sqlite3_mutex_held(pBt->mutex) );
iPtrmap = PTRMAP_PAGENO(pBt, key);
rc = sqlite3PagerGet(pBt->pPager, iPtrmap, &pDbPage);
if( rc!=0 ){
return rc;
}
pPtrmap = (u8 *)sqlite3PagerGetData(pDbPage);
offset = PTRMAP_PTROFFSET(iPtrmap, key);
if( offset<0 ){
sqlite3PagerUnref(pDbPage);
return SQLITE_CORRUPT_BKPT;
}
assert( offset <= (int)pBt->usableSize-5 );
assert( pEType!=0 );
*pEType = pPtrmap[offset];
if( pPgno ) *pPgno = get4byte(&pPtrmap[offset+1]);
sqlite3PagerUnref(pDbPage);
if( *pEType<1 || *pEType>5 ) return SQLITE_CORRUPT_BKPT;
return SQLITE_OK;
}
#else /* if defined SQLITE_OMIT_AUTOVACUUM */
#define ptrmapPut(w,x,y,z,rc)
#define ptrmapGet(w,x,y,z) SQLITE_OK
#define ptrmapPutOvflPtr(x, y, rc)
#endif
/*
** Given a btree page and a cell index (0 means the first cell on
** the page, 1 means the second cell, and so forth) return a pointer
** to the cell content.
**
** This routine works only for pages that do not contain overflow cells.
*/
#define findCell(P,I) \
((P)->aData + ((P)->maskPage & get2byte(&(P)->aCellIdx[2*(I)])))
#define findCellv2(D,M,O,I) (D+(M&get2byte(D+(O+2*(I)))))
/*
** This a more complex version of findCell() that works for
** pages that do contain overflow cells.
*/
static u8 *findOverflowCell(MemPage *pPage, int iCell){
int i;
assert( sqlite3_mutex_held(pPage->pBt->mutex) );
for(i=pPage->nOverflow-1; i>=0; i--){
int k;
struct _OvflCell *pOvfl;
pOvfl = &pPage->aOvfl[i];
k = pOvfl->idx;
if( k<=iCell ){
if( k==iCell ){
return pOvfl->pCell;
}
iCell--;
}
}
return findCell(pPage, iCell);
}
/*
** Parse a cell content block and fill in the CellInfo structure. There
** are two versions of this function. btreeParseCell() takes a
** cell index as the second argument and btreeParseCellPtr()
** takes a pointer to the body of the cell as its second argument.
**
** Within this file, the parseCell() macro can be called instead of
** btreeParseCellPtr(). Using some compilers, this will be faster.
*/
static void btreeParseCellPtr(
MemPage *pPage, /* Page containing the cell */
u8 *pCell, /* Pointer to the cell text. */
CellInfo *pInfo /* Fill in this structure */
){
u16 n; /* Number bytes in cell content header */
u32 nPayload; /* Number of bytes of cell payload */
assert( sqlite3_mutex_held(pPage->pBt->mutex) );
pInfo->pCell = pCell;
assert( pPage->leaf==0 || pPage->leaf==1 );
n = pPage->childPtrSize;
assert( n==4-4*pPage->leaf );
if( pPage->intKey ){
if( pPage->hasData ){
n += getVarint32(&pCell[n], nPayload);
}else{
nPayload = 0;
}
n += getVarint(&pCell[n], (u64*)&pInfo->nKey);
pInfo->nData = nPayload;
}else{
pInfo->nData = 0;
n += getVarint32(&pCell[n], nPayload);
pInfo->nKey = nPayload;
}
pInfo->nPayload = nPayload;
pInfo->nHeader = n;
testcase( nPayload==pPage->maxLocal );
testcase( nPayload==pPage->maxLocal+1 );
if( likely(nPayload<=pPage->maxLocal) ){
/* This is the (easy) common case where the entire payload fits
** on the local page. No overflow is required.
*/
if( (pInfo->nSize = (u16)(n+nPayload))<4 ) pInfo->nSize = 4;
pInfo->nLocal = (u16)nPayload;
pInfo->iOverflow = 0;
}else{
/* If the payload will not fit completely on the local page, we have
** to decide how much to store locally and how much to spill onto
** overflow pages. The strategy is to minimize the amount of unused
** space on overflow pages while keeping the amount of local storage
** in between minLocal and maxLocal.
**
** Warning: changing the way overflow payload is distributed in any
** way will result in an incompatible file format.
*/
int minLocal; /* Minimum amount of payload held locally */
int maxLocal; /* Maximum amount of payload held locally */
int surplus; /* Overflow payload available for local storage */
minLocal = pPage->minLocal;
maxLocal = pPage->maxLocal;
surplus = minLocal + (nPayload - minLocal)%(pPage->pBt->usableSize - 4);
testcase( surplus==maxLocal );
testcase( surplus==maxLocal+1 );
if( surplus <= maxLocal ){
pInfo->nLocal = (u16)surplus;
}else{
pInfo->nLocal = (u16)minLocal;
}
pInfo->iOverflow = (u16)(pInfo->nLocal + n);
pInfo->nSize = pInfo->iOverflow + 4;
}
}
#define parseCell(pPage, iCell, pInfo) \
btreeParseCellPtr((pPage), findCell((pPage), (iCell)), (pInfo))
static void btreeParseCell(
MemPage *pPage, /* Page containing the cell */
int iCell, /* The cell index. First cell is 0 */
CellInfo *pInfo /* Fill in this structure */
){
parseCell(pPage, iCell, pInfo);
}
/*
** Compute the total number of bytes that a Cell needs in the cell
** data area of the btree-page. The return number includes the cell
** data header and the local payload, but not any overflow page or
** the space used by the cell pointer.
*/
static u16 cellSizePtr(MemPage *pPage, u8 *pCell){
u8 *pIter = &pCell[pPage->childPtrSize];
u32 nSize;
#ifdef SQLITE_DEBUG
/* The value returned by this function should always be the same as
** the (CellInfo.nSize) value found by doing a full parse of the
** cell. If SQLITE_DEBUG is defined, an assert() at the bottom of
** this function verifies that this invariant is not violated. */
CellInfo debuginfo;
btreeParseCellPtr(pPage, pCell, &debuginfo);
#endif
if( pPage->intKey ){
u8 *pEnd;
if( pPage->hasData ){
pIter += getVarint32(pIter, nSize);
}else{
nSize = 0;
}
/* pIter now points at the 64-bit integer key value, a variable length
** integer. The following block moves pIter to point at the first byte
** past the end of the key value. */