Repository navigation
Expand file tree
/
Copy pathpcache1.c
More file actions
1287 lines (1207 loc) · 43.7 KB
/
Copy pathpcache1.c
File metadata and controls
1287 lines (1207 loc) · 43.7 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
/*
** 2008 November 05
**
** 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 the default page cache implementation (the
** sqlite3_pcache interface). It also contains part of the implementation
** of the SQLITE_CONFIG_PAGECACHE and sqlite3_release_memory() features.
** If the default page cache implementation is overridden, then neither of
** these two features are available.
**
** A Page cache line looks like this:
**
** -------------------------------------------------------------
** | database page content | PgHdr1 | MemPage | PgHdr |
** -------------------------------------------------------------
**
** The database page content is up front (so that buffer overreads tend to
** flow harmlessly into the PgHdr1, MemPage, and PgHdr extensions). MemPage
** is the extension added by the btree.c module containing information such
** as the database page number and how that database page is used. PgHdr
** is added by the pcache.c layer and contains information used to keep track
** of which pages are "dirty". PgHdr1 is an extension added by this
** module (pcache1.c). The PgHdr1 header is a subclass of sqlite3_pcache_page.
** PgHdr1 contains information needed to look up a page by its page number.
** The superclass sqlite3_pcache_page.pBuf points to the start of the
** database page content and sqlite3_pcache_page.pExtra points to PgHdr.
**
** The size of the extension (MemPage+PgHdr+PgHdr1) can be determined at
** runtime using sqlite3_config(SQLITE_CONFIG_PCACHE_HDRSZ, &size). The
** sizes of the extensions sum to 272 bytes on x64 for 3.8.10, but this
** size can vary according to architecture, compile-time options, and
** SQLite library version number.
**
** Historical note: It used to be that if the SQLITE_PCACHE_SEPARATE_HEADER
** was defined, then the page content would be held in a separate memory
** allocation from the PgHdr1. This was intended to avoid clownshoe memory
** allocations. However, the btree layer needs a small (16-byte) overrun
** area after the page content buffer. The header serves as that overrun
** area. Therefore SQLITE_PCACHE_SEPARATE_HEADER was discontinued to avoid
** any possibility of a memory error.
**
** This module tracks pointers to PgHdr1 objects. Only pcache.c communicates
** with this module. Information is passed back and forth as PgHdr1 pointers.
**
** The pcache.c and pager.c modules deal pointers to PgHdr objects.
** The btree.c module deals with pointers to MemPage objects.
**
** SOURCE OF PAGE CACHE MEMORY:
**
** Memory for a page might come from any of three sources:
**
** (1) The general-purpose memory allocator - sqlite3Malloc()
** (2) Global page-cache memory provided using sqlite3_config() with
** SQLITE_CONFIG_PAGECACHE.
** (3) PCache-local bulk allocation.
**
** The third case is a chunk of heap memory (defaulting to 100 pages worth)
** that is allocated when the page cache is created. The size of the local
** bulk allocation can be adjusted using
**
** sqlite3_config(SQLITE_CONFIG_PAGECACHE, (void*)0, 0, N).
**
** If N is positive, then N pages worth of memory are allocated using a single
** sqlite3Malloc() call and that memory is used for the first N pages allocated.
** Or if N is negative, then -1024*N bytes of memory are allocated and used
** for as many pages as can be accommodated.
**
** Only one of (2) or (3) can be used. Once the memory available to (2) or
** (3) is exhausted, subsequent allocations fail over to the general-purpose
** memory allocator (1).
**
** Earlier versions of SQLite used only methods (1) and (2). But experiments
** show that method (3) with N==100 provides about a 5% performance boost for
** common workloads.
*/
#include "sqliteInt.h"
typedef struct PCache1 PCache1;
typedef struct PgHdr1 PgHdr1;
typedef struct PgFreeslot PgFreeslot;
typedef struct PGroup PGroup;
/*
** Each cache entry is represented by an instance of the following
** structure. A buffer of PgHdr1.pCache->szPage bytes is allocated
** directly before this structure and is used to cache the page content.
**
** When reading a corrupt database file, it is possible that SQLite might
** read a few bytes (no more than 16 bytes) past the end of the page buffer.
** It will only read past the end of the page buffer, never write. This
** object is positioned immediately after the page buffer to serve as an
** overrun area, so that overreads are harmless.
**
** Variables isBulkLocal and isAnchor were once type "u8". That works,
** but causes a 2-byte gap in the structure for most architectures (since
** pointers must be either 4 or 8-byte aligned). As this structure is located
** in memory directly after the associated page data, if the database is
** corrupt, code at the b-tree layer may overread the page buffer and
** read part of this structure before the corruption is detected. This
** can cause a valgrind error if the uninitialized gap is accessed. Using u16
** ensures there is no such gap, and therefore no bytes of uninitialized
** memory in the structure.
**
** The pLruNext and pLruPrev pointers form a double-linked circular list
** of all pages that are unpinned. The PGroup.lru element (which should be
** the only element on the list with PgHdr1.isAnchor set to 1) forms the
** beginning and the end of the list.
*/
struct PgHdr1 {
sqlite3_pcache_page page; /* Base class. Must be first. pBuf & pExtra */
unsigned int iKey; /* Key value (page number) */
u16 isBulkLocal; /* This page from bulk local storage */
u16 isAnchor; /* This is the PGroup.lru element */
PgHdr1 *pNext; /* Next in hash table chain */
PCache1 *pCache; /* Cache that currently owns this page */
PgHdr1 *pLruNext; /* Next in circular LRU list of unpinned pages */
PgHdr1 *pLruPrev; /* Previous in LRU list of unpinned pages */
/* NB: pLruPrev is only valid if pLruNext!=0 */
};
/*
** A page is pinned if it is not on the LRU list. To be "pinned" means
** that the page is in active use and must not be deallocated.
*/
#define PAGE_IS_PINNED(p) ((p)->pLruNext==0)
#define PAGE_IS_UNPINNED(p) ((p)->pLruNext!=0)
/* Each page cache (or PCache) belongs to a PGroup. A PGroup is a set
** of one or more PCaches that are able to recycle each other's unpinned
** pages when they are under memory pressure. A PGroup is an instance of
** the following object.
**
** This page cache implementation works in one of two modes:
**
** (1) Every PCache is the sole member of its own PGroup. There is
** one PGroup per PCache.
**
** (2) There is a single global PGroup that all PCaches are a member
** of.
**
** Mode 1 uses more memory (since PCache instances are not able to rob
** unused pages from other PCaches) but it also operates without a mutex,
** and is therefore often faster. Mode 2 requires a mutex in order to be
** threadsafe, but recycles pages more efficiently.
**
** For mode (1), PGroup.mutex is NULL. For mode (2) there is only a single
** PGroup which is the pcache1.grp global variable and its mutex is
** SQLITE_MUTEX_STATIC_LRU.
*/
struct PGroup {
sqlite3_mutex *mutex; /* MUTEX_STATIC_LRU or NULL */
unsigned int nMaxPage; /* Sum of nMax for purgeable caches */
unsigned int nMinPage; /* Sum of nMin for purgeable caches */
unsigned int mxPinned; /* nMaxpage + 10 - nMinPage */
unsigned int nPurgeable; /* Number of purgeable pages allocated */
PgHdr1 lru; /* The beginning and end of the LRU list */
};
/* Each page cache is an instance of the following object. Every
** open database file (including each in-memory database and each
** temporary or transient database) has a single page cache which
** is an instance of this object.
**
** Pointers to structures of this type are cast and returned as
** opaque sqlite3_pcache* handles.
*/
struct PCache1 {
/* Cache configuration parameters. Page size (szPage) and the purgeable
** flag (bPurgeable) and the pnPurgeable pointer are all set when the
** cache is created and are never changed thereafter. nMax may be
** modified at any time by a call to the pcache1Cachesize() method.
** The PGroup mutex must be held when accessing nMax.
*/
PGroup *pGroup; /* PGroup this cache belongs to */
unsigned int *pnPurgeable; /* Pointer to pGroup->nPurgeable */
int szPage; /* Size of database content section */
int szExtra; /* sizeof(MemPage)+sizeof(PgHdr) */
int szAlloc; /* Total size of one pcache line */
int bPurgeable; /* True if cache is purgeable */
unsigned int nMin; /* Minimum number of pages reserved */
unsigned int nMax; /* Configured "cache_size" value */
unsigned int n90pct; /* nMax*9/10 */
unsigned int iMaxKey; /* Largest key seen since xTruncate() */
unsigned int nPurgeableDummy; /* pnPurgeable points here when not used*/
/* Hash table of all pages. The following variables may only be accessed
** when the accessor is holding the PGroup mutex.
*/
unsigned int nRecyclable; /* Number of pages in the LRU list */
unsigned int nPage; /* Total number of pages in apHash */
unsigned int nHash; /* Number of slots in apHash[] */
PgHdr1 **apHash; /* Hash table for fast lookup by key */
PgHdr1 *pFree; /* List of unused pcache-local pages */
void *pBulk; /* Bulk memory used by pcache-local */
};
/*
** Free slots in the allocator used to divide up the global page cache
** buffer provided using the SQLITE_CONFIG_PAGECACHE mechanism.
*/
struct PgFreeslot {
PgFreeslot *pNext; /* Next free slot */
};
/*
** Global data used by this cache.
*/
static SQLITE_WSD struct PCacheGlobal {
PGroup grp; /* The global PGroup for mode (2) */
/* Variables related to SQLITE_CONFIG_PAGECACHE settings. The
** szSlot, nSlot, pStart, pEnd, nReserve, and isInit values are all
** fixed at sqlite3_initialize() time and do not require mutex protection.
** The nFreeSlot and pFree values do require mutex protection.
*/
int isInit; /* True if initialized */
int separateCache; /* Use a new PGroup for each PCache */
int nInitPage; /* Initial bulk allocation size */
int szSlot; /* Size of each free slot */
int nSlot; /* The number of pcache slots */
int nReserve; /* Try to keep nFreeSlot above this */
void *pStart, *pEnd; /* Bounds of global page cache memory */
/* Above requires no mutex. Use mutex below for variable that follow. */
sqlite3_mutex *mutex; /* Mutex for accessing the following: */
PgFreeslot *pFree; /* Free page blocks */
int nFreeSlot; /* Number of unused pcache slots */
int bUnderPressure; /* True if low on PAGECACHE memory */
} pcache1_g;
/*
** All code in this file should access the global structure above via the
** alias "pcache1". This ensures that the WSD emulation is used when
** compiling for systems that do not support real WSD.
*/
#define pcache1 (GLOBAL(struct PCacheGlobal, pcache1_g))
/*
** Macros to enter and leave the PCache LRU mutex.
*/
#if !defined(SQLITE_ENABLE_MEMORY_MANAGEMENT) || SQLITE_THREADSAFE==0
# define pcache1EnterMutex(X) assert((X)->mutex==0)
# define pcache1LeaveMutex(X) assert((X)->mutex==0)
# define PCACHE1_MIGHT_USE_GROUP_MUTEX 0
#else
# define pcache1EnterMutex(X) sqlite3_mutex_enter((X)->mutex)
# define pcache1LeaveMutex(X) sqlite3_mutex_leave((X)->mutex)
# define PCACHE1_MIGHT_USE_GROUP_MUTEX 1
#endif
/******************************************************************************/
/******** Page Allocation/SQLITE_CONFIG_PCACHE Related Functions **************/
/*
** This function is called during initialization if a static buffer is
** supplied to use for the page-cache by passing the SQLITE_CONFIG_PAGECACHE
** verb to sqlite3_config(). Parameter pBuf points to an allocation large
** enough to contain 'n' buffers of 'sz' bytes each.
**
** This routine is called from sqlite3_initialize() and so it is guaranteed
** to be serialized already. There is no need for further mutexing.
*/
void sqlite3PCacheBufferSetup(void *pBuf, int sz, int n){
if( pcache1.isInit ){
PgFreeslot *p;
if( pBuf==0 ) sz = n = 0;
if( n==0 ) sz = 0;
sz = ROUNDDOWN8(sz);
pcache1.szSlot = sz;
pcache1.nSlot = pcache1.nFreeSlot = n;
pcache1.nReserve = n>90 ? 10 : (n/10 + 1);
pcache1.pStart = pBuf;
pcache1.pFree = 0;
AtomicStore(&pcache1.bUnderPressure,0);
while( n-- ){
p = (PgFreeslot*)pBuf;
p->pNext = pcache1.pFree;
pcache1.pFree = p;
pBuf = (void*)&((char*)pBuf)[sz];
}
pcache1.pEnd = pBuf;
}
}
/*
** Try to initialize the pCache->pFree and pCache->pBulk fields. Return
** true if pCache->pFree ends up containing one or more free pages.
*/
static int pcache1InitBulk(PCache1 *pCache){
i64 szBulk;
char *zBulk;
if( pcache1.nInitPage==0 ) return 0;
/* Do not bother with a bulk allocation if the cache size very small */
if( pCache->nMax<3 ) return 0;
sqlite3BeginBenignMalloc();
if( pcache1.nInitPage>0 ){
szBulk = pCache->szAlloc * (i64)pcache1.nInitPage;
}else{
szBulk = -1024 * (i64)pcache1.nInitPage;
}
if( szBulk > pCache->szAlloc*(i64)pCache->nMax ){
szBulk = pCache->szAlloc*(i64)pCache->nMax;
}
if( szBulk>=pCache->szAlloc ){
zBulk = pCache->pBulk = sqlite3Malloc( szBulk );
sqlite3EndBenignMalloc();
if( zBulk ){
int nBulk = sqlite3MallocSize(zBulk)/pCache->szAlloc;
do{
PgHdr1 *pX = (PgHdr1*)&zBulk[pCache->szPage];
pX->page.pBuf = zBulk;
pX->page.pExtra = (u8*)pX + ROUND8(sizeof(*pX));
assert( EIGHT_BYTE_ALIGNMENT( pX->page.pExtra ) );
pX->isBulkLocal = 1;
pX->isAnchor = 0;
pX->pNext = pCache->pFree;
pX->pLruPrev = 0; /* Initializing this saves a valgrind error */
pCache->pFree = pX;
zBulk += pCache->szAlloc;
}while( --nBulk );
}
}
return pCache->pFree!=0;
}
/*
** Malloc function used within this file to allocate space from the buffer
** configured using sqlite3_config(SQLITE_CONFIG_PAGECACHE) option. If no
** such buffer exists or there is no space left in it, this function falls
** back to sqlite3Malloc().
**
** Multiple threads can run this routine at the same time. Global variables
** in pcache1 need to be protected via mutex.
*/
static void *pcache1Alloc(int nByte){
void *p = 0;
assert( sqlite3_mutex_notheld(pcache1.grp.mutex) );
if( nByte<=pcache1.szSlot ){
sqlite3_mutex_enter(pcache1.mutex);
p = (PgHdr1 *)pcache1.pFree;
if( p ){
pcache1.pFree = pcache1.pFree->pNext;
pcache1.nFreeSlot--;
AtomicStore(&pcache1.bUnderPressure,pcache1.nFreeSlot<pcache1.nReserve);
assert( pcache1.nFreeSlot>=0 );
sqlite3StatusHighwater(SQLITE_STATUS_PAGECACHE_SIZE, nByte);
sqlite3StatusUp(SQLITE_STATUS_PAGECACHE_USED, 1);
}
sqlite3_mutex_leave(pcache1.mutex);
}
if( p==0 ){
/* Memory is not available in the SQLITE_CONFIG_PAGECACHE pool. Get
** it from sqlite3Malloc instead.
*/
p = sqlite3Malloc(nByte);
#ifndef SQLITE_DISABLE_PAGECACHE_OVERFLOW_STATS
if( p ){
int sz = sqlite3MallocSize(p);
sqlite3_mutex_enter(pcache1.mutex);
sqlite3StatusHighwater(SQLITE_STATUS_PAGECACHE_SIZE, nByte);
sqlite3StatusUp(SQLITE_STATUS_PAGECACHE_OVERFLOW, sz);
sqlite3_mutex_leave(pcache1.mutex);
}
#endif
sqlite3MemdebugSetType(p, MEMTYPE_PCACHE);
}
return p;
}
/*
** Free an allocated buffer obtained from pcache1Alloc().
*/
static void pcache1Free(void *p){
if( p==0 ) return;
if( SQLITE_WITHIN(p, pcache1.pStart, pcache1.pEnd) ){
PgFreeslot *pSlot;
sqlite3_mutex_enter(pcache1.mutex);
sqlite3StatusDown(SQLITE_STATUS_PAGECACHE_USED, 1);
pSlot = (PgFreeslot*)p;
pSlot->pNext = pcache1.pFree;
pcache1.pFree = pSlot;
pcache1.nFreeSlot++;
AtomicStore(&pcache1.bUnderPressure,pcache1.nFreeSlot<pcache1.nReserve);
assert( pcache1.nFreeSlot<=pcache1.nSlot );
sqlite3_mutex_leave(pcache1.mutex);
}else{
assert( sqlite3MemdebugHasType(p, MEMTYPE_PCACHE) );
sqlite3MemdebugSetType(p, MEMTYPE_HEAP);
#ifndef SQLITE_DISABLE_PAGECACHE_OVERFLOW_STATS
{
int nFreed = 0;
nFreed = sqlite3MallocSize(p);
sqlite3_mutex_enter(pcache1.mutex);
sqlite3StatusDown(SQLITE_STATUS_PAGECACHE_OVERFLOW, nFreed);
sqlite3_mutex_leave(pcache1.mutex);
}
#endif
sqlite3_free(p);
}
}
#ifdef SQLITE_ENABLE_MEMORY_MANAGEMENT
/*
** Return the size of a pcache allocation
*/
static int pcache1MemSize(void *p){
if( p>=pcache1.pStart && p<pcache1.pEnd ){
return pcache1.szSlot;
}else{
int iSize;
assert( sqlite3MemdebugHasType(p, MEMTYPE_PCACHE) );
sqlite3MemdebugSetType(p, MEMTYPE_HEAP);
iSize = sqlite3MallocSize(p);
sqlite3MemdebugSetType(p, MEMTYPE_PCACHE);
return iSize;
}
}
#endif /* SQLITE_ENABLE_MEMORY_MANAGEMENT */
/*
** Allocate a new page object initially associated with cache pCache.
*/
static PgHdr1 *pcache1AllocPage(PCache1 *pCache, int benignMalloc){
PgHdr1 *p = 0;
void *pPg;
assert( sqlite3_mutex_held(pCache->pGroup->mutex) );
if( pCache->pFree || (pCache->nPage==0 && pcache1InitBulk(pCache)) ){
assert( pCache->pFree!=0 );
p = pCache->pFree;
pCache->pFree = p->pNext;
p->pNext = 0;
}else{
#ifdef SQLITE_ENABLE_MEMORY_MANAGEMENT
/* The group mutex must be released before pcache1Alloc() is called. This
** is because it might call sqlite3_release_memory(), which assumes that
** this mutex is not held. */
assert( pcache1.separateCache==0 );
assert( pCache->pGroup==&pcache1.grp );
pcache1LeaveMutex(pCache->pGroup);
#endif
if( benignMalloc ){ sqlite3BeginBenignMalloc(); }
pPg = pcache1Alloc(pCache->szAlloc);
if( benignMalloc ){ sqlite3EndBenignMalloc(); }
#ifdef SQLITE_ENABLE_MEMORY_MANAGEMENT
pcache1EnterMutex(pCache->pGroup);
#endif
if( pPg==0 ) return 0;
p = (PgHdr1 *)&((u8 *)pPg)[pCache->szPage];
p->page.pBuf = pPg;
p->page.pExtra = (u8*)p + ROUND8(sizeof(*p));
assert( EIGHT_BYTE_ALIGNMENT( p->page.pExtra ) );
p->isBulkLocal = 0;
p->isAnchor = 0;
p->pLruPrev = 0; /* Initializing this saves a valgrind error */
}
(*pCache->pnPurgeable)++;
return p;
}
/*
** Free a page object allocated by pcache1AllocPage().
*/
static void pcache1FreePage(PgHdr1 *p){
PCache1 *pCache;
assert( p!=0 );
pCache = p->pCache;
assert( sqlite3_mutex_held(p->pCache->pGroup->mutex) );
if( p->isBulkLocal ){
p->pNext = pCache->pFree;
pCache->pFree = p;
}else{
pcache1Free(p->page.pBuf);
}
(*pCache->pnPurgeable)--;
}
/*
** Malloc function used by SQLite to obtain space from the buffer configured
** using sqlite3_config(SQLITE_CONFIG_PAGECACHE) option. If no such buffer
** exists, this function falls back to sqlite3Malloc().
*/
void *sqlite3PageMalloc(int sz){
assert( sz<=65536+8 ); /* These allocations are never very large */
return pcache1Alloc(sz);
}
/*
** Free an allocated buffer obtained from sqlite3PageMalloc().
*/
void sqlite3PageFree(void *p){
pcache1Free(p);
}
/*
** Return true if it desirable to avoid allocating a new page cache
** entry.
**
** If memory was allocated specifically to the page cache using
** SQLITE_CONFIG_PAGECACHE but that memory has all been used, then
** it is desirable to avoid allocating a new page cache entry because
** presumably SQLITE_CONFIG_PAGECACHE was suppose to be sufficient
** for all page cache needs and we should not need to spill the
** allocation onto the heap.
**
** Or, the heap is used for all page cache memory but the heap is
** under memory pressure, then again it is desirable to avoid
** allocating a new page cache entry in order to avoid stressing
** the heap even further.
*/
static int pcache1UnderMemoryPressure(PCache1 *pCache){
if( pcache1.nSlot && (pCache->szPage+pCache->szExtra)<=pcache1.szSlot ){
return AtomicLoad(&pcache1.bUnderPressure);
}else{
return sqlite3HeapNearlyFull();
}
}
/******************************************************************************/
/******** General Implementation Functions ************************************/
/*
** This function is used to resize the hash table used by the cache passed
** as the first argument.
**
** The PCache mutex must be held when this function is called.
*/
static void pcache1ResizeHash(PCache1 *p){
PgHdr1 **apNew;
u64 nNew;
u32 i;
assert( sqlite3_mutex_held(p->pGroup->mutex) );
nNew = 2*(u64)p->nHash;
if( nNew<256 ){
nNew = 256;
}
assert( (nNew & (nNew-1))==0 ); /* nNew is always a power of two */
pcache1LeaveMutex(p->pGroup);
if( p->nHash ){ sqlite3BeginBenignMalloc(); }
apNew = (PgHdr1 **)sqlite3MallocZero(sizeof(PgHdr1 *)*nNew);
if( p->nHash ){ sqlite3EndBenignMalloc(); }
pcache1EnterMutex(p->pGroup);
if( apNew ){
for(i=0; i<p->nHash; i++){
PgHdr1 *pPage;
PgHdr1 *pNext = p->apHash[i];
while( (pPage = pNext)!=0 ){
unsigned int h = (unsigned int)(pPage->iKey & (nNew-1));
pNext = pPage->pNext;
pPage->pNext = apNew[h];
apNew[h] = pPage;
}
}
sqlite3_free(p->apHash);
p->apHash = apNew;
p->nHash = nNew;
}
}
/*
** This function is used internally to remove the page pPage from the
** PGroup LRU list, if is part of it. If pPage is not part of the PGroup
** LRU list, then this function is a no-op.
**
** The PGroup mutex must be held when this function is called.
*/
static PgHdr1 *pcache1PinPage(PgHdr1 *pPage){
assert( pPage!=0 );
assert( PAGE_IS_UNPINNED(pPage) );
assert( pPage->pLruNext );
assert( pPage->pLruPrev );
assert( sqlite3_mutex_held(pPage->pCache->pGroup->mutex) );
pPage->pLruPrev->pLruNext = pPage->pLruNext;
pPage->pLruNext->pLruPrev = pPage->pLruPrev;
pPage->pLruNext = 0;
/* pPage->pLruPrev = 0;
** No need to clear pLruPrev as it is never accessed if pLruNext is 0 */
assert( pPage->isAnchor==0 );
assert( pPage->pCache->pGroup->lru.isAnchor==1 );
pPage->pCache->nRecyclable--;
return pPage;
}
/*
** Remove the page supplied as an argument from the hash table
** (PCache1.apHash structure) that it is currently stored in.
** Also free the page if freePage is true.
**
** The PGroup mutex must be held when this function is called.
*/
static void pcache1RemoveFromHash(PgHdr1 *pPage, int freeFlag){
unsigned int h;
PCache1 *pCache = pPage->pCache;
PgHdr1 **pp;
assert( sqlite3_mutex_held(pCache->pGroup->mutex) );
assert( pCache->nHash>0 && (pCache->nHash & (pCache->nHash-1))==0 );
h = pPage->iKey & (pCache->nHash-1);
for(pp=&pCache->apHash[h]; (*pp)!=pPage; pp=&(*pp)->pNext);
*pp = (*pp)->pNext;
pCache->nPage--;
if( freeFlag ) pcache1FreePage(pPage);
}
/*
** If there are currently more than nMaxPage pages allocated, try
** to recycle pages to reduce the number allocated to nMaxPage.
*/
static void pcache1EnforceMaxPage(PCache1 *pCache){
PGroup *pGroup = pCache->pGroup;
PgHdr1 *p;
assert( sqlite3_mutex_held(pGroup->mutex) );
while( pGroup->nPurgeable>pGroup->nMaxPage
&& (p=pGroup->lru.pLruPrev)->isAnchor==0
){
assert( p->pCache->pGroup==pGroup );
assert( PAGE_IS_UNPINNED(p) );
pcache1PinPage(p);
pcache1RemoveFromHash(p, 1);
}
if( pCache->nPage==0 && pCache->pBulk ){
sqlite3_free(pCache->pBulk);
pCache->pBulk = pCache->pFree = 0;
}
}
/*
** Discard all pages from cache pCache with a page number (key value)
** greater than or equal to iLimit. Any pinned pages that meet this
** criteria are unpinned before they are discarded.
**
** The PCache mutex must be held when this function is called.
*/
static void pcache1TruncateUnsafe(
PCache1 *pCache, /* The cache to truncate */
unsigned int iLimit /* Drop pages with this pgno or larger */
){
TESTONLY( int nPage = 0; ) /* To assert pCache->nPage is correct */
unsigned int h, iStop;
assert( sqlite3_mutex_held(pCache->pGroup->mutex) );
assert( pCache->iMaxKey >= iLimit );
assert( pCache->nHash>0 && (pCache->nHash & (pCache->nHash-1))==0 );
if( pCache->iMaxKey - iLimit < pCache->nHash ){
/* If we are just shaving the last few pages off the end of the
** cache, then there is no point in scanning the entire hash table.
** Only scan those hash slots that might contain pages that need to
** be removed. */
h = iLimit & (pCache->nHash-1);
iStop = pCache->iMaxKey & (pCache->nHash-1);
TESTONLY( nPage = -10; ) /* Disable the pCache->nPage validity check */
}else{
/* This is the general case where many pages are being removed.
** It is necessary to scan the entire hash table */
h = pCache->nHash/2;
iStop = h - 1;
}
for(;;){
PgHdr1 **pp;
PgHdr1 *pPage;
assert( h<pCache->nHash );
pp = &pCache->apHash[h];
while( (pPage = *pp)!=0 ){
if( pPage->iKey>=iLimit ){
pCache->nPage--;
*pp = pPage->pNext;
if( PAGE_IS_UNPINNED(pPage) ) pcache1PinPage(pPage);
pcache1FreePage(pPage);
}else{
pp = &pPage->pNext;
TESTONLY( if( nPage>=0 ) nPage++; )
}
}
if( h==iStop ) break;
h = (h+1) & (pCache->nHash-1);
}
assert( nPage<0 || pCache->nPage==(unsigned)nPage );
}
/******************************************************************************/
/******** sqlite3_pcache Methods **********************************************/
/*
** Implementation of the sqlite3_pcache.xInit method.
*/
static int pcache1Init(void *NotUsed){
UNUSED_PARAMETER(NotUsed);
assert( pcache1.isInit==0 );
memset(&pcache1, 0, sizeof(pcache1));
/*
** The pcache1.separateCache variable is true if each PCache has its own
** private PGroup (mode-1). pcache1.separateCache is false if the single
** PGroup in pcache1.grp is used for all page caches (mode-2).
**
** * Always use a unified cache (mode-2) if ENABLE_MEMORY_MANAGEMENT
**
** * Use a unified cache in single-threaded applications that have
** configured a start-time buffer for use as page-cache memory using
** sqlite3_config(SQLITE_CONFIG_PAGECACHE, pBuf, sz, N) with non-NULL
** pBuf argument.
**
** * Otherwise use separate caches (mode-1)
*/
#if defined(SQLITE_ENABLE_MEMORY_MANAGEMENT)
pcache1.separateCache = 0;
#elif SQLITE_THREADSAFE
pcache1.separateCache = sqlite3GlobalConfig.pPage==0
|| sqlite3GlobalConfig.bCoreMutex>0;
#else
pcache1.separateCache = sqlite3GlobalConfig.pPage==0;
#endif
#if SQLITE_THREADSAFE
if( sqlite3GlobalConfig.bCoreMutex ){
pcache1.grp.mutex = sqlite3MutexAlloc(SQLITE_MUTEX_STATIC_LRU);
pcache1.mutex = sqlite3MutexAlloc(SQLITE_MUTEX_STATIC_PMEM);
}
#endif
if( pcache1.separateCache
&& sqlite3GlobalConfig.nPage!=0
&& sqlite3GlobalConfig.pPage==0
){
pcache1.nInitPage = sqlite3GlobalConfig.nPage;
}else{
pcache1.nInitPage = 0;
}
pcache1.grp.mxPinned = 10;
pcache1.isInit = 1;
return SQLITE_OK;
}
/*
** Implementation of the sqlite3_pcache.xShutdown method.
** Note that the static mutex allocated in xInit does
** not need to be freed.
*/
static void pcache1Shutdown(void *NotUsed){
UNUSED_PARAMETER(NotUsed);
assert( pcache1.isInit!=0 );
memset(&pcache1, 0, sizeof(pcache1));
}
/* forward declaration */
static void pcache1Destroy(sqlite3_pcache *p);
/*
** Implementation of the sqlite3_pcache.xCreate method.
**
** Allocate a new cache.
*/
static sqlite3_pcache *pcache1Create(int szPage, int szExtra, int bPurgeable){
PCache1 *pCache; /* The newly created page cache */
PGroup *pGroup; /* The group the new page cache will belong to */
i64 sz; /* Bytes of memory required to allocate the new cache */
assert( (szPage & (szPage-1))==0 && szPage>=512 && szPage<=65536 );
assert( szExtra < 300 );
sz = sizeof(PCache1) + sizeof(PGroup)*pcache1.separateCache;
pCache = (PCache1 *)sqlite3MallocZero(sz);
if( pCache ){
if( pcache1.separateCache ){
pGroup = (PGroup*)&pCache[1];
pGroup->mxPinned = 10;
}else{
pGroup = &pcache1.grp;
}
pcache1EnterMutex(pGroup);
if( pGroup->lru.isAnchor==0 ){
pGroup->lru.isAnchor = 1;
pGroup->lru.pLruPrev = pGroup->lru.pLruNext = &pGroup->lru;
}
pCache->pGroup = pGroup;
pCache->szPage = szPage;
pCache->szExtra = szExtra;
pCache->szAlloc = szPage + szExtra + ROUND8(sizeof(PgHdr1));
pCache->bPurgeable = (bPurgeable ? 1 : 0);
pcache1ResizeHash(pCache);
if( bPurgeable ){
pCache->nMin = 10;
pGroup->nMinPage += pCache->nMin;
pGroup->mxPinned = pGroup->nMaxPage + 10 - pGroup->nMinPage;
pCache->pnPurgeable = &pGroup->nPurgeable;
}else{
pCache->pnPurgeable = &pCache->nPurgeableDummy;
}
pcache1LeaveMutex(pGroup);
if( pCache->nHash==0 ){
pcache1Destroy((sqlite3_pcache*)pCache);
pCache = 0;
}
}
return (sqlite3_pcache *)pCache;
}
/*
** Implementation of the sqlite3_pcache.xCachesize method.
**
** Configure the cache_size limit for a cache.
*/
static void pcache1Cachesize(sqlite3_pcache *p, int nMax){
PCache1 *pCache = (PCache1 *)p;
u32 n;
assert( nMax>=0 );
if( pCache->bPurgeable ){
PGroup *pGroup = pCache->pGroup;
pcache1EnterMutex(pGroup);
n = (u32)nMax;
if( n > 0x7fff0000 - pGroup->nMaxPage + pCache->nMax ){
n = 0x7fff0000 - pGroup->nMaxPage + pCache->nMax;
}
pGroup->nMaxPage += (n - pCache->nMax);
pGroup->mxPinned = pGroup->nMaxPage + 10 - pGroup->nMinPage;
pCache->nMax = n;
pCache->n90pct = pCache->nMax*9/10;
pcache1EnforceMaxPage(pCache);
pcache1LeaveMutex(pGroup);
}
}
/*
** Implementation of the sqlite3_pcache.xShrink method.
**
** Free up as much memory as possible.
*/
static void pcache1Shrink(sqlite3_pcache *p){
PCache1 *pCache = (PCache1*)p;
if( pCache->bPurgeable ){
PGroup *pGroup = pCache->pGroup;
unsigned int savedMaxPage;
pcache1EnterMutex(pGroup);
savedMaxPage = pGroup->nMaxPage;
pGroup->nMaxPage = 0;
pcache1EnforceMaxPage(pCache);
pGroup->nMaxPage = savedMaxPage;
pcache1LeaveMutex(pGroup);
}
}
/*
** Implementation of the sqlite3_pcache.xPagecount method.
*/
static int pcache1Pagecount(sqlite3_pcache *p){
int n;
PCache1 *pCache = (PCache1*)p;
pcache1EnterMutex(pCache->pGroup);
n = pCache->nPage;
pcache1LeaveMutex(pCache->pGroup);
return n;
}
/*
** Implement steps 3, 4, and 5 of the pcache1Fetch() algorithm described
** in the header of the pcache1Fetch() procedure.
**
** This steps are broken out into a separate procedure because they are
** usually not needed, and by avoiding the stack initialization required
** for these steps, the main pcache1Fetch() procedure can run faster.
*/
static SQLITE_NOINLINE PgHdr1 *pcache1FetchStage2(
PCache1 *pCache,
unsigned int iKey,
int createFlag
){
unsigned int nPinned;
PGroup *pGroup = pCache->pGroup;
PgHdr1 *pPage = 0;
/* Step 3: Abort if createFlag is 1 but the cache is nearly full */
assert( pCache->nPage >= pCache->nRecyclable );
nPinned = pCache->nPage - pCache->nRecyclable;
assert( pGroup->mxPinned == pGroup->nMaxPage + 10 - pGroup->nMinPage );
assert( pCache->n90pct == pCache->nMax*9/10 );
if( createFlag==1 && (
nPinned>=pGroup->mxPinned
|| nPinned>=pCache->n90pct
|| (pcache1UnderMemoryPressure(pCache) && pCache->nRecyclable<nPinned)
)){
return 0;
}
if( pCache->nPage>=pCache->nHash ) pcache1ResizeHash(pCache);
assert( pCache->nHash>0 && pCache->apHash );
assert( (pCache->nHash & (pCache->nHash-1))==0 );
/* Step 4. Try to recycle a page. */
if( pCache->bPurgeable
&& !pGroup->lru.pLruPrev->isAnchor
&& ((pCache->nPage+1>=pCache->nMax) || pcache1UnderMemoryPressure(pCache))
){
PCache1 *pOther;
pPage = pGroup->lru.pLruPrev;
assert( PAGE_IS_UNPINNED(pPage) );
pcache1RemoveFromHash(pPage, 0);
pcache1PinPage(pPage);
pOther = pPage->pCache;
if( pOther->szAlloc != pCache->szAlloc ){
pcache1FreePage(pPage);
pPage = 0;
}else{
pGroup->nPurgeable -= (pOther->bPurgeable - pCache->bPurgeable);
}
}
/* Step 5. If a usable page buffer has still not been found,
** attempt to allocate a new one.
*/
if( !pPage ){
pPage = pcache1AllocPage(pCache, createFlag==1);
}
if( pPage ){
unsigned int h = iKey & (pCache->nHash-1);
pCache->nPage++;
pPage->iKey = iKey;
pPage->pNext = pCache->apHash[h];
pPage->pCache = pCache;
pPage->pLruNext = 0;
/* pPage->pLruPrev = 0;
** No need to clear pLruPrev since it is not accessed when pLruNext==0 */
*(void **)pPage->page.pExtra = 0;
pCache->apHash[h] = pPage;
if( iKey>pCache->iMaxKey ){
pCache->iMaxKey = iKey;
}
}
return pPage;
}
/*
** Implementation of the sqlite3_pcache.xFetch method.
**
** Fetch a page by key value.
**
** Whether or not a new page may be allocated by this function depends on
** the value of the createFlag argument. 0 means do not allocate a new
** page. 1 means allocate a new page if space is easily available. 2
** means to try really hard to allocate a new page.
**
** For a non-purgeable cache (a cache used as the storage for an in-memory
** database) there is really no difference between createFlag 1 and 2. So
** the calling function (pcache.c) will never have a createFlag of 1 on
** a non-purgeable cache.
**
** There are three different approaches to obtaining space for a page,
** depending on the value of parameter createFlag (which may be 0, 1 or 2).
**
** 1. Regardless of the value of createFlag, the cache is searched for a
** copy of the requested page. If one is found, it is returned.
**
** 2. If createFlag==0 and the page is not already in the cache, NULL is
** returned.
**
** 3. If createFlag is 1, and the page is not already in the cache, then
** return NULL (do not allocate a new page) if any of the following
** conditions are true:
**
** (a) the number of pages pinned by the cache is greater than
** PCache1.nMax, or
**
** (b) the number of pages pinned by the cache is greater than
** the sum of nMax for all purgeable caches, less the sum of
** nMin for all other purgeable caches, or
**
** 4. If none of the first three conditions apply and the cache is marked
** as purgeable, and if one of the following is true:
**
** (a) The number of pages allocated for the cache is already
** PCache1.nMax, or
**
** (b) The number of pages allocated for all purgeable caches is
** already equal to or greater than the sum of nMax for all
** purgeable caches,
**
** (c) The system is under memory pressure and wants to avoid
** unnecessary pages cache entry allocations
**
** then attempt to recycle a page from the LRU list. If it is the right
** size, return the recycled buffer. Otherwise, free the buffer and
** proceed to step 5.
**
** 5. Otherwise, allocate and return a new page buffer.
**
** There are two versions of this routine. pcache1FetchWithMutex() is