1/*
2** 2008 August 05
3**
4** The author disclaims copyright to this source code.  In place of
5** a legal notice, here is a blessing:
6**
7**    May you do good and not evil.
8**    May you find forgiveness for yourself and forgive others.
9**    May you share freely, never taking more than you give.
10**
11*************************************************************************
12** This file implements that page cache.
13*/
14#include "sqliteInt.h"
15
16/*
17** A complete page cache is an instance of this structure.
18*/
19struct PCache {
20  PgHdr *pDirty, *pDirtyTail;         /* List of dirty pages in LRU order */
21  PgHdr *pSynced;                     /* Last synced page in dirty page list */
22  int nRef;                           /* Number of referenced pages */
23  int nMax;                           /* Configured cache size */
24  int szPage;                         /* Size of every page in this cache */
25  int szExtra;                        /* Size of extra space for each page */
26  int bPurgeable;                     /* True if pages are on backing store */
27  int (*xStress)(void*,PgHdr*);       /* Call to try make a page clean */
28  void *pStress;                      /* Argument to xStress */
29  sqlite3_pcache *pCache;             /* Pluggable cache module */
30  PgHdr *pPage1;                      /* Reference to page 1 */
31};
32
33/*
34** Some of the assert() macros in this code are too expensive to run
35** even during normal debugging.  Use them only rarely on long-running
36** tests.  Enable the expensive asserts using the
37** -DSQLITE_ENABLE_EXPENSIVE_ASSERT=1 compile-time option.
38*/
39#ifdef SQLITE_ENABLE_EXPENSIVE_ASSERT
40# define expensive_assert(X)  assert(X)
41#else
42# define expensive_assert(X)
43#endif
44
45/********************************** Linked List Management ********************/
46
47#if !defined(NDEBUG) && defined(SQLITE_ENABLE_EXPENSIVE_ASSERT)
48/*
49** Check that the pCache->pSynced variable is set correctly. If it
50** is not, either fail an assert or return zero. Otherwise, return
51** non-zero. This is only used in debugging builds, as follows:
52**
53**   expensive_assert( pcacheCheckSynced(pCache) );
54*/
55static int pcacheCheckSynced(PCache *pCache){
56  PgHdr *p;
57  for(p=pCache->pDirtyTail; p!=pCache->pSynced; p=p->pDirtyPrev){
58    assert( p->nRef || (p->flags&PGHDR_NEED_SYNC) );
59  }
60  return (p==0 || p->nRef || (p->flags&PGHDR_NEED_SYNC)==0);
61}
62#endif /* !NDEBUG && SQLITE_ENABLE_EXPENSIVE_ASSERT */
63
64/*
65** Remove page pPage from the list of dirty pages.
66*/
67static void pcacheRemoveFromDirtyList(PgHdr *pPage){
68  PCache *p = pPage->pCache;
69
70  assert( pPage->pDirtyNext || pPage==p->pDirtyTail );
71  assert( pPage->pDirtyPrev || pPage==p->pDirty );
72
73  /* Update the PCache1.pSynced variable if necessary. */
74  if( p->pSynced==pPage ){
75    PgHdr *pSynced = pPage->pDirtyPrev;
76    while( pSynced && (pSynced->flags&PGHDR_NEED_SYNC) ){
77      pSynced = pSynced->pDirtyPrev;
78    }
79    p->pSynced = pSynced;
80  }
81
82  if( pPage->pDirtyNext ){
83    pPage->pDirtyNext->pDirtyPrev = pPage->pDirtyPrev;
84  }else{
85    assert( pPage==p->pDirtyTail );
86    p->pDirtyTail = pPage->pDirtyPrev;
87  }
88  if( pPage->pDirtyPrev ){
89    pPage->pDirtyPrev->pDirtyNext = pPage->pDirtyNext;
90  }else{
91    assert( pPage==p->pDirty );
92    p->pDirty = pPage->pDirtyNext;
93  }
94  pPage->pDirtyNext = 0;
95  pPage->pDirtyPrev = 0;
96
97  expensive_assert( pcacheCheckSynced(p) );
98}
99
100/*
101** Add page pPage to the head of the dirty list (PCache1.pDirty is set to
102** pPage).
103*/
104static void pcacheAddToDirtyList(PgHdr *pPage){
105  PCache *p = pPage->pCache;
106
107  assert( pPage->pDirtyNext==0 && pPage->pDirtyPrev==0 && p->pDirty!=pPage );
108
109  pPage->pDirtyNext = p->pDirty;
110  if( pPage->pDirtyNext ){
111    assert( pPage->pDirtyNext->pDirtyPrev==0 );
112    pPage->pDirtyNext->pDirtyPrev = pPage;
113  }
114  p->pDirty = pPage;
115  if( !p->pDirtyTail ){
116    p->pDirtyTail = pPage;
117  }
118  if( !p->pSynced && 0==(pPage->flags&PGHDR_NEED_SYNC) ){
119    p->pSynced = pPage;
120  }
121  expensive_assert( pcacheCheckSynced(p) );
122}
123
124/*
125** Wrapper around the pluggable caches xUnpin method. If the cache is
126** being used for an in-memory database, this function is a no-op.
127*/
128static void pcacheUnpin(PgHdr *p){
129  PCache *pCache = p->pCache;
130  if( pCache->bPurgeable ){
131    if( p->pgno==1 ){
132      pCache->pPage1 = 0;
133    }
134    sqlite3GlobalConfig.pcache.xUnpin(pCache->pCache, p, 0);
135  }
136}
137
138/*************************************************** General Interfaces ******
139**
140** Initialize and shutdown the page cache subsystem. Neither of these
141** functions are threadsafe.
142*/
143int sqlite3PcacheInitialize(void){
144  if( sqlite3GlobalConfig.pcache.xInit==0 ){
145    /* IMPLEMENTATION-OF: R-26801-64137 If the xInit() method is NULL, then the
146    ** built-in default page cache is used instead of the application defined
147    ** page cache. */
148    sqlite3PCacheSetDefault();
149  }
150  return sqlite3GlobalConfig.pcache.xInit(sqlite3GlobalConfig.pcache.pArg);
151}
152void sqlite3PcacheShutdown(void){
153  if( sqlite3GlobalConfig.pcache.xShutdown ){
154    /* IMPLEMENTATION-OF: R-26000-56589 The xShutdown() method may be NULL. */
155    sqlite3GlobalConfig.pcache.xShutdown(sqlite3GlobalConfig.pcache.pArg);
156  }
157}
158
159/*
160** Return the size in bytes of a PCache object.
161*/
162int sqlite3PcacheSize(void){ return sizeof(PCache); }
163
164/*
165** Create a new PCache object. Storage space to hold the object
166** has already been allocated and is passed in as the p pointer.
167** The caller discovers how much space needs to be allocated by
168** calling sqlite3PcacheSize().
169*/
170void sqlite3PcacheOpen(
171  int szPage,                  /* Size of every page */
172  int szExtra,                 /* Extra space associated with each page */
173  int bPurgeable,              /* True if pages are on backing store */
174  int (*xStress)(void*,PgHdr*),/* Call to try to make pages clean */
175  void *pStress,               /* Argument to xStress */
176  PCache *p                    /* Preallocated space for the PCache */
177){
178  memset(p, 0, sizeof(PCache));
179  p->szPage = szPage;
180  p->szExtra = szExtra;
181  p->bPurgeable = bPurgeable;
182  p->xStress = xStress;
183  p->pStress = pStress;
184  p->nMax = 100;
185}
186
187/*
188** Change the page size for PCache object. The caller must ensure that there
189** are no outstanding page references when this function is called.
190*/
191void sqlite3PcacheSetPageSize(PCache *pCache, int szPage){
192  assert( pCache->nRef==0 && pCache->pDirty==0 );
193  if( pCache->pCache ){
194    sqlite3GlobalConfig.pcache.xDestroy(pCache->pCache);
195    pCache->pCache = 0;
196    pCache->pPage1 = 0;
197  }
198  pCache->szPage = szPage;
199}
200
201/*
202** Try to obtain a page from the cache.
203*/
204int sqlite3PcacheFetch(
205  PCache *pCache,       /* Obtain the page from this cache */
206  Pgno pgno,            /* Page number to obtain */
207  int createFlag,       /* If true, create page if it does not exist already */
208  PgHdr **ppPage        /* Write the page here */
209){
210  PgHdr *pPage = 0;
211  int eCreate;
212
213  assert( pCache!=0 );
214  assert( createFlag==1 || createFlag==0 );
215  assert( pgno>0 );
216
217  /* If the pluggable cache (sqlite3_pcache*) has not been allocated,
218  ** allocate it now.
219  */
220  if( !pCache->pCache && createFlag ){
221    sqlite3_pcache *p;
222    int nByte;
223    nByte = pCache->szPage + pCache->szExtra + sizeof(PgHdr);
224    p = sqlite3GlobalConfig.pcache.xCreate(nByte, pCache->bPurgeable);
225    if( !p ){
226      return SQLITE_NOMEM;
227    }
228    sqlite3GlobalConfig.pcache.xCachesize(p, pCache->nMax);
229    pCache->pCache = p;
230  }
231
232  eCreate = createFlag * (1 + (!pCache->bPurgeable || !pCache->pDirty));
233  if( pCache->pCache ){
234    pPage = sqlite3GlobalConfig.pcache.xFetch(pCache->pCache, pgno, eCreate);
235  }
236
237  if( !pPage && eCreate==1 ){
238    PgHdr *pPg;
239
240    /* Find a dirty page to write-out and recycle. First try to find a
241    ** page that does not require a journal-sync (one with PGHDR_NEED_SYNC
242    ** cleared), but if that is not possible settle for any other
243    ** unreferenced dirty page.
244    */
245    expensive_assert( pcacheCheckSynced(pCache) );
246    for(pPg=pCache->pSynced;
247        pPg && (pPg->nRef || (pPg->flags&PGHDR_NEED_SYNC));
248        pPg=pPg->pDirtyPrev
249    );
250    pCache->pSynced = pPg;
251    if( !pPg ){
252      for(pPg=pCache->pDirtyTail; pPg && pPg->nRef; pPg=pPg->pDirtyPrev);
253    }
254    if( pPg ){
255      int rc;
256      rc = pCache->xStress(pCache->pStress, pPg);
257      if( rc!=SQLITE_OK && rc!=SQLITE_BUSY ){
258        return rc;
259      }
260    }
261
262    pPage = sqlite3GlobalConfig.pcache.xFetch(pCache->pCache, pgno, 2);
263  }
264
265  if( pPage ){
266    if( !pPage->pData ){
267      memset(pPage, 0, sizeof(PgHdr));
268      pPage->pData = (void *)&pPage[1];
269      pPage->pExtra = (void*)&((char *)pPage->pData)[pCache->szPage];
270      memset(pPage->pExtra, 0, pCache->szExtra);
271      pPage->pCache = pCache;
272      pPage->pgno = pgno;
273    }
274    assert( pPage->pCache==pCache );
275    assert( pPage->pgno==pgno );
276    assert( pPage->pData==(void *)&pPage[1] );
277    assert( pPage->pExtra==(void *)&((char *)&pPage[1])[pCache->szPage] );
278
279    if( 0==pPage->nRef ){
280      pCache->nRef++;
281    }
282    pPage->nRef++;
283    if( pgno==1 ){
284      pCache->pPage1 = pPage;
285    }
286  }
287  *ppPage = pPage;
288  return (pPage==0 && eCreate) ? SQLITE_NOMEM : SQLITE_OK;
289}
290
291/*
292** Decrement the reference count on a page. If the page is clean and the
293** reference count drops to 0, then it is made elible for recycling.
294*/
295void sqlite3PcacheRelease(PgHdr *p){
296  assert( p->nRef>0 );
297  p->nRef--;
298  if( p->nRef==0 ){
299    PCache *pCache = p->pCache;
300    pCache->nRef--;
301    if( (p->flags&PGHDR_DIRTY)==0 ){
302      pcacheUnpin(p);
303    }else{
304      /* Move the page to the head of the dirty list. */
305      pcacheRemoveFromDirtyList(p);
306      pcacheAddToDirtyList(p);
307    }
308  }
309}
310
311/*
312** Increase the reference count of a supplied page by 1.
313*/
314void sqlite3PcacheRef(PgHdr *p){
315  assert(p->nRef>0);
316  p->nRef++;
317}
318
319/*
320** Drop a page from the cache. There must be exactly one reference to the
321** page. This function deletes that reference, so after it returns the
322** page pointed to by p is invalid.
323*/
324void sqlite3PcacheDrop(PgHdr *p){
325  PCache *pCache;
326  assert( p->nRef==1 );
327  if( p->flags&PGHDR_DIRTY ){
328    pcacheRemoveFromDirtyList(p);
329  }
330  pCache = p->pCache;
331  pCache->nRef--;
332  if( p->pgno==1 ){
333    pCache->pPage1 = 0;
334  }
335  sqlite3GlobalConfig.pcache.xUnpin(pCache->pCache, p, 1);
336}
337
338/*
339** Make sure the page is marked as dirty. If it isn't dirty already,
340** make it so.
341*/
342void sqlite3PcacheMakeDirty(PgHdr *p){
343  p->flags &= ~PGHDR_DONT_WRITE;
344  assert( p->nRef>0 );
345  if( 0==(p->flags & PGHDR_DIRTY) ){
346    p->flags |= PGHDR_DIRTY;
347    pcacheAddToDirtyList( p);
348  }
349}
350
351/*
352** Make sure the page is marked as clean. If it isn't clean already,
353** make it so.
354*/
355void sqlite3PcacheMakeClean(PgHdr *p){
356  if( (p->flags & PGHDR_DIRTY) ){
357    pcacheRemoveFromDirtyList(p);
358    p->flags &= ~(PGHDR_DIRTY|PGHDR_NEED_SYNC);
359    if( p->nRef==0 ){
360      pcacheUnpin(p);
361    }
362  }
363}
364
365/*
366** Make every page in the cache clean.
367*/
368void sqlite3PcacheCleanAll(PCache *pCache){
369  PgHdr *p;
370  while( (p = pCache->pDirty)!=0 ){
371    sqlite3PcacheMakeClean(p);
372  }
373}
374
375/*
376** Clear the PGHDR_NEED_SYNC flag from all dirty pages.
377*/
378void sqlite3PcacheClearSyncFlags(PCache *pCache){
379  PgHdr *p;
380  for(p=pCache->pDirty; p; p=p->pDirtyNext){
381    p->flags &= ~PGHDR_NEED_SYNC;
382  }
383  pCache->pSynced = pCache->pDirtyTail;
384}
385
386/*
387** Change the page number of page p to newPgno.
388*/
389void sqlite3PcacheMove(PgHdr *p, Pgno newPgno){
390  PCache *pCache = p->pCache;
391  assert( p->nRef>0 );
392  assert( newPgno>0 );
393  sqlite3GlobalConfig.pcache.xRekey(pCache->pCache, p, p->pgno, newPgno);
394  p->pgno = newPgno;
395  if( (p->flags&PGHDR_DIRTY) && (p->flags&PGHDR_NEED_SYNC) ){
396    pcacheRemoveFromDirtyList(p);
397    pcacheAddToDirtyList(p);
398  }
399}
400
401/*
402** Drop every cache entry whose page number is greater than "pgno". The
403** caller must ensure that there are no outstanding references to any pages
404** other than page 1 with a page number greater than pgno.
405**
406** If there is a reference to page 1 and the pgno parameter passed to this
407** function is 0, then the data area associated with page 1 is zeroed, but
408** the page object is not dropped.
409*/
410void sqlite3PcacheTruncate(PCache *pCache, Pgno pgno){
411  if( pCache->pCache ){
412    PgHdr *p;
413    PgHdr *pNext;
414    for(p=pCache->pDirty; p; p=pNext){
415      pNext = p->pDirtyNext;
416      /* This routine never gets call with a positive pgno except right
417      ** after sqlite3PcacheCleanAll().  So if there are dirty pages,
418      ** it must be that pgno==0.
419      */
420      assert( p->pgno>0 );
421      if( ALWAYS(p->pgno>pgno) ){
422        assert( p->flags&PGHDR_DIRTY );
423        sqlite3PcacheMakeClean(p);
424      }
425    }
426    if( pgno==0 && pCache->pPage1 ){
427      memset(pCache->pPage1->pData, 0, pCache->szPage);
428      pgno = 1;
429    }
430    sqlite3GlobalConfig.pcache.xTruncate(pCache->pCache, pgno+1);
431  }
432}
433
434/*
435** Close a cache.
436*/
437void sqlite3PcacheClose(PCache *pCache){
438  if( pCache->pCache ){
439    sqlite3GlobalConfig.pcache.xDestroy(pCache->pCache);
440  }
441}
442
443/*
444** Discard the contents of the cache.
445*/
446void sqlite3PcacheClear(PCache *pCache){
447  sqlite3PcacheTruncate(pCache, 0);
448}
449
450/*
451** Merge two lists of pages connected by pDirty and in pgno order.
452** Do not both fixing the pDirtyPrev pointers.
453*/
454static PgHdr *pcacheMergeDirtyList(PgHdr *pA, PgHdr *pB){
455  PgHdr result, *pTail;
456  pTail = &result;
457  while( pA && pB ){
458    if( pA->pgno<pB->pgno ){
459      pTail->pDirty = pA;
460      pTail = pA;
461      pA = pA->pDirty;
462    }else{
463      pTail->pDirty = pB;
464      pTail = pB;
465      pB = pB->pDirty;
466    }
467  }
468  if( pA ){
469    pTail->pDirty = pA;
470  }else if( pB ){
471    pTail->pDirty = pB;
472  }else{
473    pTail->pDirty = 0;
474  }
475  return result.pDirty;
476}
477
478/*
479** Sort the list of pages in accending order by pgno.  Pages are
480** connected by pDirty pointers.  The pDirtyPrev pointers are
481** corrupted by this sort.
482**
483** Since there cannot be more than 2^31 distinct pages in a database,
484** there cannot be more than 31 buckets required by the merge sorter.
485** One extra bucket is added to catch overflow in case something
486** ever changes to make the previous sentence incorrect.
487*/
488#define N_SORT_BUCKET  32
489static PgHdr *pcacheSortDirtyList(PgHdr *pIn){
490  PgHdr *a[N_SORT_BUCKET], *p;
491  int i;
492  memset(a, 0, sizeof(a));
493  while( pIn ){
494    p = pIn;
495    pIn = p->pDirty;
496    p->pDirty = 0;
497    for(i=0; ALWAYS(i<N_SORT_BUCKET-1); i++){
498      if( a[i]==0 ){
499        a[i] = p;
500        break;
501      }else{
502        p = pcacheMergeDirtyList(a[i], p);
503        a[i] = 0;
504      }
505    }
506    if( NEVER(i==N_SORT_BUCKET-1) ){
507      /* To get here, there need to be 2^(N_SORT_BUCKET) elements in
508      ** the input list.  But that is impossible.
509      */
510      a[i] = pcacheMergeDirtyList(a[i], p);
511    }
512  }
513  p = a[0];
514  for(i=1; i<N_SORT_BUCKET; i++){
515    p = pcacheMergeDirtyList(p, a[i]);
516  }
517  return p;
518}
519
520/*
521** Return a list of all dirty pages in the cache, sorted by page number.
522*/
523PgHdr *sqlite3PcacheDirtyList(PCache *pCache){
524  PgHdr *p;
525  for(p=pCache->pDirty; p; p=p->pDirtyNext){
526    p->pDirty = p->pDirtyNext;
527  }
528  return pcacheSortDirtyList(pCache->pDirty);
529}
530
531/*
532** Return the total number of referenced pages held by the cache.
533*/
534int sqlite3PcacheRefCount(PCache *pCache){
535  return pCache->nRef;
536}
537
538/*
539** Return the number of references to the page supplied as an argument.
540*/
541int sqlite3PcachePageRefcount(PgHdr *p){
542  return p->nRef;
543}
544
545/*
546** Return the total number of pages in the cache.
547*/
548int sqlite3PcachePagecount(PCache *pCache){
549  int nPage = 0;
550  if( pCache->pCache ){
551    nPage = sqlite3GlobalConfig.pcache.xPagecount(pCache->pCache);
552  }
553  return nPage;
554}
555
556#ifdef SQLITE_TEST
557/*
558** Get the suggested cache-size value.
559*/
560int sqlite3PcacheGetCachesize(PCache *pCache){
561  return pCache->nMax;
562}
563#endif
564
565/*
566** Set the suggested cache-size value.
567*/
568void sqlite3PcacheSetCachesize(PCache *pCache, int mxPage){
569  pCache->nMax = mxPage;
570  if( pCache->pCache ){
571    sqlite3GlobalConfig.pcache.xCachesize(pCache->pCache, mxPage);
572  }
573}
574
575#if defined(SQLITE_CHECK_PAGES) || defined(SQLITE_DEBUG)
576/*
577** For all dirty pages currently in the cache, invoke the specified
578** callback. This is only used if the SQLITE_CHECK_PAGES macro is
579** defined.
580*/
581void sqlite3PcacheIterateDirty(PCache *pCache, void (*xIter)(PgHdr *)){
582  PgHdr *pDirty;
583  for(pDirty=pCache->pDirty; pDirty; pDirty=pDirty->pDirtyNext){
584    xIter(pDirty);
585  }
586}
587#endif
588