monotone

monotone Mtn Source Tree

Root/sqlite/pager.c

1/*
2** 2001 September 15
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 is the implementation of the page cache subsystem or "pager".
13**
14** The pager is used to access a database disk file. It implements
15** atomic commit and rollback through the use of a journal file that
16** is separate from the database file. The pager also implements file
17** locking to prevent two processes from writing the same database
18** file simultaneously, or one process from reading the database while
19** another is writing.
20**
21** @(#) $Id: pager.c,v 1.348 2007/06/18 17:25:18 drh Exp $
22*/
23#ifndef SQLITE_OMIT_DISKIO
24#include "sqliteInt.h"
25#include "os.h"
26#include "pager.h"
27#include <assert.h>
28#include <string.h>
29
30/*
31** Macros for troubleshooting. Normally turned off
32*/
33#if 0
34#define sqlite3DebugPrintf printf
35#define PAGERTRACE1(X) sqlite3DebugPrintf(X)
36#define PAGERTRACE2(X,Y) sqlite3DebugPrintf(X,Y)
37#define PAGERTRACE3(X,Y,Z) sqlite3DebugPrintf(X,Y,Z)
38#define PAGERTRACE4(X,Y,Z,W) sqlite3DebugPrintf(X,Y,Z,W)
39#define PAGERTRACE5(X,Y,Z,W,V) sqlite3DebugPrintf(X,Y,Z,W,V)
40#else
41#define PAGERTRACE1(X)
42#define PAGERTRACE2(X,Y)
43#define PAGERTRACE3(X,Y,Z)
44#define PAGERTRACE4(X,Y,Z,W)
45#define PAGERTRACE5(X,Y,Z,W,V)
46#endif
47
48/*
49** The following two macros are used within the PAGERTRACEX() macros above
50** to print out file-descriptors.
51**
52** PAGERID() takes a pointer to a Pager struct as it's argument. The
53** associated file-descriptor is returned. FILEHANDLEID() takes an OsFile
54** struct as it's argument.
55*/
56#define PAGERID(p) ((int)(p->fd))
57#define FILEHANDLEID(fd) ((int)fd)
58
59/*
60** The page cache as a whole is always in one of the following
61** states:
62**
63** PAGER_UNLOCK The page cache is not currently reading or
64** writing the database file. There is no
65** data held in memory. This is the initial
66** state.
67**
68** PAGER_SHARED The page cache is reading the database.
69** Writing is not permitted. There can be
70** multiple readers accessing the same database
71** file at the same time.
72**
73** PAGER_RESERVED This process has reserved the database for writing
74** but has not yet made any changes. Only one process
75** at a time can reserve the database. The original
76** database file has not been modified so other
77** processes may still be reading the on-disk
78** database file.
79**
80** PAGER_EXCLUSIVE The page cache is writing the database.
81** Access is exclusive. No other processes or
82** threads can be reading or writing while one
83** process is writing.
84**
85** PAGER_SYNCED The pager moves to this state from PAGER_EXCLUSIVE
86** after all dirty pages have been written to the
87** database file and the file has been synced to
88** disk. All that remains to do is to remove or
89** truncate the journal file and the transaction
90** will be committed.
91**
92** The page cache comes up in PAGER_UNLOCK. The first time a
93** sqlite3PagerGet() occurs, the state transitions to PAGER_SHARED.
94** After all pages have been released using sqlite_page_unref(),
95** the state transitions back to PAGER_UNLOCK. The first time
96** that sqlite3PagerWrite() is called, the state transitions to
97** PAGER_RESERVED. (Note that sqlite3PagerWrite() can only be
98** called on an outstanding page which means that the pager must
99** be in PAGER_SHARED before it transitions to PAGER_RESERVED.)
100** PAGER_RESERVED means that there is an open rollback journal.
101** The transition to PAGER_EXCLUSIVE occurs before any changes
102** are made to the database file, though writes to the rollback
103** journal occurs with just PAGER_RESERVED. After an sqlite3PagerRollback()
104** or sqlite3PagerCommitPhaseTwo(), the state can go back to PAGER_SHARED,
105** or it can stay at PAGER_EXCLUSIVE if we are in exclusive access mode.
106*/
107#define PAGER_UNLOCK 0
108#define PAGER_SHARED 1 /* same as SHARED_LOCK */
109#define PAGER_RESERVED 2 /* same as RESERVED_LOCK */
110#define PAGER_EXCLUSIVE 4 /* same as EXCLUSIVE_LOCK */
111#define PAGER_SYNCED 5
112
113/*
114** If the SQLITE_BUSY_RESERVED_LOCK macro is set to true at compile-time,
115** then failed attempts to get a reserved lock will invoke the busy callback.
116** This is off by default. To see why, consider the following scenario:
117**
118** Suppose thread A already has a shared lock and wants a reserved lock.
119** Thread B already has a reserved lock and wants an exclusive lock. If
120** both threads are using their busy callbacks, it might be a long time
121** be for one of the threads give up and allows the other to proceed.
122** But if the thread trying to get the reserved lock gives up quickly
123** (if it never invokes its busy callback) then the contention will be
124** resolved quickly.
125*/
126#ifndef SQLITE_BUSY_RESERVED_LOCK
127# define SQLITE_BUSY_RESERVED_LOCK 0
128#endif
129
130/*
131** This macro rounds values up so that if the value is an address it
132** is guaranteed to be an address that is aligned to an 8-byte boundary.
133*/
134#define FORCE_ALIGNMENT(X) (((X)+7)&~7)
135
136/*
137** Each in-memory image of a page begins with the following header.
138** This header is only visible to this pager module. The client
139** code that calls pager sees only the data that follows the header.
140**
141** Client code should call sqlite3PagerWrite() on a page prior to making
142** any modifications to that page. The first time sqlite3PagerWrite()
143** is called, the original page contents are written into the rollback
144** journal and PgHdr.inJournal and PgHdr.needSync are set. Later, once
145** the journal page has made it onto the disk surface, PgHdr.needSync
146** is cleared. The modified page cannot be written back into the original
147** database file until the journal pages has been synced to disk and the
148** PgHdr.needSync has been cleared.
149**
150** The PgHdr.dirty flag is set when sqlite3PagerWrite() is called and
151** is cleared again when the page content is written back to the original
152** database file.
153**
154** Details of important structure elements:
155**
156** needSync
157**
158** If this is true, this means that it is not safe to write the page
159** content to the database because the original content needed
160** for rollback has not by synced to the main rollback journal.
161** The original content may have been written to the rollback journal
162** but it has not yet been synced. So we cannot write to the database
163** file because power failure might cause the page in the journal file
164** to never reach the disk. It is as if the write to the journal file
165** does not occur until the journal file is synced.
166**
167** This flag is false if the page content exactly matches what
168** currently exists in the database file. The needSync flag is also
169** false if the original content has been written to the main rollback
170** journal and synced. If the page represents a new page that has
171** been added onto the end of the database during the current
172** transaction, the needSync flag is true until the original database
173** size in the journal header has been synced to disk.
174**
175** inJournal
176**
177** This is true if the original page has been written into the main
178** rollback journal. This is always false for new pages added to
179** the end of the database file during the current transaction.
180** And this flag says nothing about whether or not the journal
181** has been synced to disk. For pages that are in the original
182** database file, the following expression should always be true:
183**
184** inJournal = (pPager->aInJournal[(pgno-1)/8] & (1<<((pgno-1)%8))!=0
185**
186** The pPager->aInJournal[] array is only valid for the original
187** pages of the database, not new pages that are added to the end
188** of the database, so obviously the above expression cannot be
189** valid for new pages. For new pages inJournal is always 0.
190**
191** dirty
192**
193** When true, this means that the content of the page has been
194** modified and needs to be written back to the database file.
195** If false, it means that either the content of the page is
196** unchanged or else the content is unimportant and we do not
197** care whether or not it is preserved.
198**
199** alwaysRollback
200**
201** This means that the sqlite3PagerDontRollback() API should be
202** ignored for this page. The DontRollback() API attempts to say
203** that the content of the page on disk is unimportant (it is an
204** unused page on the freelist) so that it is unnecessary to
205** rollback changes to this page because the content of the page
206** can change without changing the meaning of the database. This
207** flag overrides any DontRollback() attempt. This flag is set
208** when a page that originally contained valid data is added to
209** the freelist. Later in the same transaction, this page might
210** be pulled from the freelist and reused for something different
211** and at that point the DontRollback() API will be called because
212** pages taken from the freelist do not need to be protected by
213** the rollback journal. But this flag says that the page was
214** not originally part of the freelist so that it still needs to
215** be rolled back in spite of any subsequent DontRollback() calls.
216**
217** needRead
218**
219** This flag means (when true) that the content of the page has
220** not yet been loaded from disk. The in-memory content is just
221** garbage. (Actually, we zero the content, but you should not
222** make any assumptions about the content nevertheless.) If the
223** content is needed in the future, it should be read from the
224** original database file.
225*/
226typedef struct PgHdr PgHdr;
227struct PgHdr {
228 Pager *pPager; /* The pager to which this page belongs */
229 Pgno pgno; /* The page number for this page */
230 PgHdr *pNextHash, *pPrevHash; /* Hash collision chain for PgHdr.pgno */
231 PgHdr *pNextFree, *pPrevFree; /* Freelist of pages where nRef==0 */
232 PgHdr *pNextAll; /* A list of all pages */
233 u8 inJournal; /* TRUE if has been written to journal */
234 u8 dirty; /* TRUE if we need to write back changes */
235 u8 needSync; /* Sync journal before writing this page */
236 u8 alwaysRollback; /* Disable DontRollback() for this page */
237 u8 needRead; /* Read content if PagerWrite() is called */
238 short int nRef; /* Number of users of this page */
239 PgHdr *pDirty, *pPrevDirty; /* Dirty pages */
240 u32 notUsed; /* Buffer space */
241#ifdef SQLITE_CHECK_PAGES
242 u32 pageHash;
243#endif
244 /* pPager->pageSize bytes of page data follow this header */
245 /* Pager.nExtra bytes of local data follow the page data */
246};
247
248/*
249** For an in-memory only database, some extra information is recorded about
250** each page so that changes can be rolled back. (Journal files are not
251** used for in-memory databases.) The following information is added to
252** the end of every EXTRA block for in-memory databases.
253**
254** This information could have been added directly to the PgHdr structure.
255** But then it would take up an extra 8 bytes of storage on every PgHdr
256** even for disk-based databases. Splitting it out saves 8 bytes. This
257** is only a savings of 0.8% but those percentages add up.
258*/
259typedef struct PgHistory PgHistory;
260struct PgHistory {
261 u8 *pOrig; /* Original page text. Restore to this on a full rollback */
262 u8 *pStmt; /* Text as it was at the beginning of the current statement */
263 PgHdr *pNextStmt, *pPrevStmt; /* List of pages in the statement journal */
264 u8 inStmt; /* TRUE if in the statement subjournal */
265};
266
267/*
268** A macro used for invoking the codec if there is one
269*/
270#ifdef SQLITE_HAS_CODEC
271# define CODEC1(P,D,N,X) if( P->xCodec!=0 ){ P->xCodec(P->pCodecArg,D,N,X); }
272# define CODEC2(P,D,N,X) ((char*)(P->xCodec!=0?P->xCodec(P->pCodecArg,D,N,X):D))
273#else
274# define CODEC1(P,D,N,X) /* NO-OP */
275# define CODEC2(P,D,N,X) ((char*)D)
276#endif
277
278/*
279** Convert a pointer to a PgHdr into a pointer to its data
280** and back again.
281*/
282#define PGHDR_TO_DATA(P) ((void*)(&(P)[1]))
283#define DATA_TO_PGHDR(D) (&((PgHdr*)(D))[-1])
284#define PGHDR_TO_EXTRA(G,P) ((void*)&((char*)(&(G)[1]))[(P)->pageSize])
285#define PGHDR_TO_HIST(P,PGR) \
286 ((PgHistory*)&((char*)(&(P)[1]))[(PGR)->pageSize+(PGR)->nExtra])
287
288/*
289** A open page cache is an instance of the following structure.
290**
291** Pager.errCode may be set to SQLITE_IOERR, SQLITE_CORRUPT, or
292** or SQLITE_FULL. Once one of the first three errors occurs, it persists
293** and is returned as the result of every major pager API call. The
294** SQLITE_FULL return code is slightly different. It persists only until the
295** next successful rollback is performed on the pager cache. Also,
296** SQLITE_FULL does not affect the sqlite3PagerGet() and sqlite3PagerLookup()
297** APIs, they may still be used successfully.
298*/
299struct Pager {
300 u8 journalOpen; /* True if journal file descriptors is valid */
301 u8 journalStarted; /* True if header of journal is synced */
302 u8 useJournal; /* Use a rollback journal on this file */
303 u8 noReadlock; /* Do not bother to obtain readlocks */
304 u8 stmtOpen; /* True if the statement subjournal is open */
305 u8 stmtInUse; /* True we are in a statement subtransaction */
306 u8 stmtAutoopen; /* Open stmt journal when main journal is opened*/
307 u8 noSync; /* Do not sync the journal if true */
308 u8 fullSync; /* Do extra syncs of the journal for robustness */
309 u8 full_fsync; /* Use F_FULLFSYNC when available */
310 u8 state; /* PAGER_UNLOCK, _SHARED, _RESERVED, etc. */
311 u8 tempFile; /* zFilename is a temporary file */
312 u8 readOnly; /* True for a read-only database */
313 u8 needSync; /* True if an fsync() is needed on the journal */
314 u8 dirtyCache; /* True if cached pages have changed */
315 u8 alwaysRollback; /* Disable DontRollback() for all pages */
316 u8 memDb; /* True to inhibit all file I/O */
317 u8 setMaster; /* True if a m-j name has been written to jrnl */
318 u8 doNotSync; /* Boolean. While true, do not spill the cache */
319 u8 exclusiveMode; /* Boolean. True if locking_mode==EXCLUSIVE */
320 u8 changeCountDone; /* Set after incrementing the change-counter */
321 int errCode; /* One of several kinds of errors */
322 int dbSize; /* Number of pages in the file */
323 int origDbSize; /* dbSize before the current change */
324 int stmtSize; /* Size of database (in pages) at stmt_begin() */
325 int nRec; /* Number of pages written to the journal */
326 u32 cksumInit; /* Quasi-random value added to every checksum */
327 int stmtNRec; /* Number of records in stmt subjournal */
328 int nExtra; /* Add this many bytes to each in-memory page */
329 int pageSize; /* Number of bytes in a page */
330 int nPage; /* Total number of in-memory pages */
331 int nRef; /* Number of in-memory pages with PgHdr.nRef>0 */
332 int mxPage; /* Maximum number of pages to hold in cache */
333 Pgno mxPgno; /* Maximum allowed size of the database */
334 u8 *aInJournal; /* One bit for each page in the database file */
335 u8 *aInStmt; /* One bit for each page in the database */
336 char *zFilename; /* Name of the database file */
337 char *zJournal; /* Name of the journal file */
338 char *zDirectory; /* Directory hold database and journal files */
339 OsFile *fd, *jfd; /* File descriptors for database and journal */
340 OsFile *stfd; /* File descriptor for the statement subjournal*/
341 BusyHandler *pBusyHandler; /* Pointer to sqlite.busyHandler */
342 PgHdr *pFirst, *pLast; /* List of free pages */
343 PgHdr *pFirstSynced; /* First free page with PgHdr.needSync==0 */
344 PgHdr *pAll; /* List of all pages */
345 PgHdr *pStmt; /* List of pages in the statement subjournal */
346 PgHdr *pDirty; /* List of all dirty pages */
347 i64 journalOff; /* Current byte offset in the journal file */
348 i64 journalHdr; /* Byte offset to previous journal header */
349 i64 stmtHdrOff; /* First journal header written this statement */
350 i64 stmtCksum; /* cksumInit when statement was started */
351 i64 stmtJSize; /* Size of journal at stmt_begin() */
352 int sectorSize; /* Assumed sector size during rollback */
353#ifdef SQLITE_TEST
354 int nHit, nMiss; /* Cache hits and missing */
355 int nRead, nWrite; /* Database pages read/written */
356#endif
357 void (*xDestructor)(DbPage*,int); /* Call this routine when freeing pages */
358 void (*xReiniter)(DbPage*,int); /* Call this routine when reloading pages */
359#ifdef SQLITE_HAS_CODEC
360 void *(*xCodec)(void*,void*,Pgno,int); /* Routine for en/decoding data */
361 void *pCodecArg; /* First argument to xCodec() */
362#endif
363 int nHash; /* Size of the pager hash table */
364 PgHdr **aHash; /* Hash table to map page number to PgHdr */
365#ifdef SQLITE_ENABLE_MEMORY_MANAGEMENT
366 Pager *pNext; /* Linked list of pagers in this thread */
367#endif
368 char *pTmpSpace; /* Pager.pageSize bytes of space for tmp use */
369 char dbFileVers[16]; /* Changes whenever database file changes */
370};
371
372/*
373** The following global variables hold counters used for
374** testing purposes only. These variables do not exist in
375** a non-testing build. These variables are not thread-safe.
376*/
377#ifdef SQLITE_TEST
378int sqlite3_pager_readdb_count = 0; /* Number of full pages read from DB */
379int sqlite3_pager_writedb_count = 0; /* Number of full pages written to DB */
380int sqlite3_pager_writej_count = 0; /* Number of pages written to journal */
381int sqlite3_pager_pgfree_count = 0; /* Number of cache pages freed */
382# define PAGER_INCR(v) v++
383#else
384# define PAGER_INCR(v)
385#endif
386
387
388
389/*
390** Journal files begin with the following magic string. The data
391** was obtained from /dev/random. It is used only as a sanity check.
392**
393** Since version 2.8.0, the journal format contains additional sanity
394** checking information. If the power fails while the journal is begin
395** written, semi-random garbage data might appear in the journal
396** file after power is restored. If an attempt is then made
397** to roll the journal back, the database could be corrupted. The additional
398** sanity checking data is an attempt to discover the garbage in the
399** journal and ignore it.
400**
401** The sanity checking information for the new journal format consists
402** of a 32-bit checksum on each page of data. The checksum covers both
403** the page number and the pPager->pageSize bytes of data for the page.
404** This cksum is initialized to a 32-bit random value that appears in the
405** journal file right after the header. The random initializer is important,
406** because garbage data that appears at the end of a journal is likely
407** data that was once in other files that have now been deleted. If the
408** garbage data came from an obsolete journal file, the checksums might
409** be correct. But by initializing the checksum to random value which
410** is different for every journal, we minimize that risk.
411*/
412static const unsigned char aJournalMagic[] = {
413 0xd9, 0xd5, 0x05, 0xf9, 0x20, 0xa1, 0x63, 0xd7,
414};
415
416/*
417** The size of the header and of each page in the journal is determined
418** by the following macros.
419*/
420#define JOURNAL_PG_SZ(pPager) ((pPager->pageSize) + 8)
421
422/*
423** The journal header size for this pager. In the future, this could be
424** set to some value read from the disk controller. The important
425** characteristic is that it is the same size as a disk sector.
426*/
427#define JOURNAL_HDR_SZ(pPager) (pPager->sectorSize)
428
429/*
430** The macro MEMDB is true if we are dealing with an in-memory database.
431** We do this as a macro so that if the SQLITE_OMIT_MEMORYDB macro is set,
432** the value of MEMDB will be a constant and the compiler will optimize
433** out code that would never execute.
434*/
435#ifdef SQLITE_OMIT_MEMORYDB
436# define MEMDB 0
437#else
438# define MEMDB pPager->memDb
439#endif
440
441/*
442** Page number PAGER_MJ_PGNO is never used in an SQLite database (it is
443** reserved for working around a windows/posix incompatibility). It is
444** used in the journal to signify that the remainder of the journal file
445** is devoted to storing a master journal name - there are no more pages to
446** roll back. See comments for function writeMasterJournal() for details.
447*/
448/* #define PAGER_MJ_PGNO(x) (PENDING_BYTE/((x)->pageSize)) */
449#define PAGER_MJ_PGNO(x) ((PENDING_BYTE/((x)->pageSize))+1)
450
451/*
452** The maximum legal page number is (2^31 - 1).
453*/
454#define PAGER_MAX_PGNO 2147483647
455
456/*
457** Enable reference count tracking (for debugging) here:
458*/
459#ifdef SQLITE_DEBUG
460 int pager3_refinfo_enable = 0;
461 static void pager_refinfo(PgHdr *p){
462 static int cnt = 0;
463 if( !pager3_refinfo_enable ) return;
464 sqlite3DebugPrintf(
465 "REFCNT: %4d addr=%p nRef=%-3d total=%d\n",
466 p->pgno, PGHDR_TO_DATA(p), p->nRef, p->pPager->nRef
467 );
468 cnt++; /* Something to set a breakpoint on */
469 }
470# define REFINFO(X) pager_refinfo(X)
471#else
472# define REFINFO(X)
473#endif
474
475/*
476** Return true if page *pPg has already been written to the statement
477** journal (or statement snapshot has been created, if *pPg is part
478** of an in-memory database).
479*/
480static int pageInStatement(PgHdr *pPg){
481 Pager *pPager = pPg->pPager;
482 if( MEMDB ){
483 return PGHDR_TO_HIST(pPg, pPager)->inStmt;
484 }else{
485 Pgno pgno = pPg->pgno;
486 u8 *a = pPager->aInStmt;
487 return (a && (int)pgno<=pPager->stmtSize && (a[pgno/8] & (1<<(pgno&7))));
488 }
489}
490
491/*
492** Change the size of the pager hash table to N. N must be a power
493** of two.
494*/
495static void pager_resize_hash_table(Pager *pPager, int N){
496 PgHdr **aHash, *pPg;
497 assert( N>0 && (N&(N-1))==0 );
498 aHash = sqliteMalloc( sizeof(aHash[0])*N );
499 if( aHash==0 ){
500 /* Failure to rehash is not an error. It is only a performance hit. */
501 return;
502 }
503 sqliteFree(pPager->aHash);
504 pPager->nHash = N;
505 pPager->aHash = aHash;
506 for(pPg=pPager->pAll; pPg; pPg=pPg->pNextAll){
507 int h;
508 if( pPg->pgno==0 ){
509 assert( pPg->pNextHash==0 && pPg->pPrevHash==0 );
510 continue;
511 }
512 h = pPg->pgno & (N-1);
513 pPg->pNextHash = aHash[h];
514 if( aHash[h] ){
515 aHash[h]->pPrevHash = pPg;
516 }
517 aHash[h] = pPg;
518 pPg->pPrevHash = 0;
519 }
520}
521
522/*
523** Read a 32-bit integer from the given file descriptor. Store the integer
524** that is read in *pRes. Return SQLITE_OK if everything worked, or an
525** error code is something goes wrong.
526**
527** All values are stored on disk as big-endian.
528*/
529static int read32bits(OsFile *fd, u32 *pRes){
530 unsigned char ac[4];
531 int rc = sqlite3OsRead(fd, ac, sizeof(ac));
532 if( rc==SQLITE_OK ){
533 *pRes = sqlite3Get4byte(ac);
534 }
535 return rc;
536}
537
538/*
539** Write a 32-bit integer into a string buffer in big-endian byte order.
540*/
541#define put32bits(A,B) sqlite3Put4byte((u8*)A,B)
542
543/*
544** Write a 32-bit integer into the given file descriptor. Return SQLITE_OK
545** on success or an error code is something goes wrong.
546*/
547static int write32bits(OsFile *fd, u32 val){
548 char ac[4];
549 put32bits(ac, val);
550 return sqlite3OsWrite(fd, ac, 4);
551}
552
553/*
554** Read a 32-bit integer at offset 'offset' from the page identified by
555** page header 'p'.
556*/
557static u32 retrieve32bits(PgHdr *p, int offset){
558 unsigned char *ac;
559 ac = &((unsigned char*)PGHDR_TO_DATA(p))[offset];
560 return sqlite3Get4byte(ac);
561}
562
563
564/*
565** This function should be called when an error occurs within the pager
566** code. The first argument is a pointer to the pager structure, the
567** second the error-code about to be returned by a pager API function.
568** The value returned is a copy of the second argument to this function.
569**
570** If the second argument is SQLITE_IOERR, SQLITE_CORRUPT, or SQLITE_FULL
571** the error becomes persistent. All subsequent API calls on this Pager
572** will immediately return the same error code.
573*/
574static int pager_error(Pager *pPager, int rc){
575 int rc2 = rc & 0xff;
576 assert( pPager->errCode==SQLITE_FULL || pPager->errCode==SQLITE_OK );
577 if(
578 rc2==SQLITE_FULL ||
579 rc2==SQLITE_IOERR ||
580 rc2==SQLITE_CORRUPT
581 ){
582 pPager->errCode = rc;
583 }
584 return rc;
585}
586
587/*
588** If SQLITE_CHECK_PAGES is defined then we do some sanity checking
589** on the cache using a hash function. This is used for testing
590** and debugging only.
591*/
592#ifdef SQLITE_CHECK_PAGES
593/*
594** Return a 32-bit hash of the page data for pPage.
595*/
596static u32 pager_datahash(int nByte, unsigned char *pData){
597 u32 hash = 0;
598 int i;
599 for(i=0; i<nByte; i++){
600 hash = (hash*1039) + pData[i];
601 }
602 return hash;
603}
604static u32 pager_pagehash(PgHdr *pPage){
605 return pager_datahash(pPage->pPager->pageSize,
606 (unsigned char *)PGHDR_TO_DATA(pPage));
607}
608
609/*
610** The CHECK_PAGE macro takes a PgHdr* as an argument. If SQLITE_CHECK_PAGES
611** is defined, and NDEBUG is not defined, an assert() statement checks
612** that the page is either dirty or still matches the calculated page-hash.
613*/
614#define CHECK_PAGE(x) checkPage(x)
615static void checkPage(PgHdr *pPg){
616 Pager *pPager = pPg->pPager;
617 assert( !pPg->pageHash || pPager->errCode || MEMDB || pPg->dirty ||
618 pPg->pageHash==pager_pagehash(pPg) );
619}
620
621#else
622#define pager_datahash(X,Y) 0
623#define pager_pagehash(X) 0
624#define CHECK_PAGE(x)
625#endif
626
627/*
628** When this is called the journal file for pager pPager must be open.
629** The master journal file name is read from the end of the file and
630** written into memory obtained from sqliteMalloc(). *pzMaster is
631** set to point at the memory and SQLITE_OK returned. The caller must
632** sqliteFree() *pzMaster.
633**
634** If no master journal file name is present *pzMaster is set to 0 and
635** SQLITE_OK returned.
636*/
637static int readMasterJournal(OsFile *pJrnl, char **pzMaster){
638 int rc;
639 u32 len;
640 i64 szJ;
641 u32 cksum;
642 int i;
643 unsigned char aMagic[8]; /* A buffer to hold the magic header */
644
645 *pzMaster = 0;
646
647 rc = sqlite3OsFileSize(pJrnl, &szJ);
648 if( rc!=SQLITE_OK || szJ<16 ) return rc;
649
650 rc = sqlite3OsSeek(pJrnl, szJ-16);
651 if( rc!=SQLITE_OK ) return rc;
652
653 rc = read32bits(pJrnl, &len);
654 if( rc!=SQLITE_OK ) return rc;
655
656 rc = read32bits(pJrnl, &cksum);
657 if( rc!=SQLITE_OK ) return rc;
658
659 rc = sqlite3OsRead(pJrnl, aMagic, 8);
660 if( rc!=SQLITE_OK || memcmp(aMagic, aJournalMagic, 8) ) return rc;
661
662 rc = sqlite3OsSeek(pJrnl, szJ-16-len);
663 if( rc!=SQLITE_OK ) return rc;
664
665 *pzMaster = (char *)sqliteMalloc(len+1);
666 if( !*pzMaster ){
667 return SQLITE_NOMEM;
668 }
669 rc = sqlite3OsRead(pJrnl, *pzMaster, len);
670 if( rc!=SQLITE_OK ){
671 sqliteFree(*pzMaster);
672 *pzMaster = 0;
673 return rc;
674 }
675
676 /* See if the checksum matches the master journal name */
677 for(i=0; i<len; i++){
678 cksum -= (*pzMaster)[i];
679 }
680 if( cksum ){
681 /* If the checksum doesn't add up, then one or more of the disk sectors
682 ** containing the master journal filename is corrupted. This means
683 ** definitely roll back, so just return SQLITE_OK and report a (nul)
684 ** master-journal filename.
685 */
686 sqliteFree(*pzMaster);
687 *pzMaster = 0;
688 }else{
689 (*pzMaster)[len] = '\0';
690 }
691
692 return SQLITE_OK;
693}
694
695/*
696** Seek the journal file descriptor to the next sector boundary where a
697** journal header may be read or written. Pager.journalOff is updated with
698** the new seek offset.
699**
700** i.e for a sector size of 512:
701**
702** Input Offset Output Offset
703** ---------------------------------------
704** 0 0
705** 512 512
706** 100 512
707** 2000 2048
708**
709*/
710static int seekJournalHdr(Pager *pPager){
711 i64 offset = 0;
712 i64 c = pPager->journalOff;
713 if( c ){
714 offset = ((c-1)/JOURNAL_HDR_SZ(pPager) + 1) * JOURNAL_HDR_SZ(pPager);
715 }
716 assert( offset%JOURNAL_HDR_SZ(pPager)==0 );
717 assert( offset>=c );
718 assert( (offset-c)<JOURNAL_HDR_SZ(pPager) );
719 pPager->journalOff = offset;
720 return sqlite3OsSeek(pPager->jfd, pPager->journalOff);
721}
722
723/*
724** The journal file must be open when this routine is called. A journal
725** header (JOURNAL_HDR_SZ bytes) is written into the journal file at the
726** current location.
727**
728** The format for the journal header is as follows:
729** - 8 bytes: Magic identifying journal format.
730** - 4 bytes: Number of records in journal, or -1 no-sync mode is on.
731** - 4 bytes: Random number used for page hash.
732** - 4 bytes: Initial database page count.
733** - 4 bytes: Sector size used by the process that wrote this journal.
734**
735** Followed by (JOURNAL_HDR_SZ - 24) bytes of unused space.
736*/
737static int writeJournalHdr(Pager *pPager){
738 char zHeader[sizeof(aJournalMagic)+16];
739 int rc;
740
741 if( pPager->stmtHdrOff==0 ){
742 pPager->stmtHdrOff = pPager->journalOff;
743 }
744
745 rc = seekJournalHdr(pPager);
746 if( rc ) return rc;
747
748 pPager->journalHdr = pPager->journalOff;
749 pPager->journalOff += JOURNAL_HDR_SZ(pPager);
750
751 /* FIX ME:
752 **
753 ** Possibly for a pager not in no-sync mode, the journal magic should not
754 ** be written until nRec is filled in as part of next syncJournal().
755 **
756 ** Actually maybe the whole journal header should be delayed until that
757 ** point. Think about this.
758 */
759 memcpy(zHeader, aJournalMagic, sizeof(aJournalMagic));
760 /* The nRec Field. 0xFFFFFFFF for no-sync journals. */
761 put32bits(&zHeader[sizeof(aJournalMagic)], pPager->noSync ? 0xffffffff : 0);
762 /* The random check-hash initialiser */
763 sqlite3Randomness(sizeof(pPager->cksumInit), &pPager->cksumInit);
764 put32bits(&zHeader[sizeof(aJournalMagic)+4], pPager->cksumInit);
765 /* The initial database size */
766 put32bits(&zHeader[sizeof(aJournalMagic)+8], pPager->dbSize);
767 /* The assumed sector size for this process */
768 put32bits(&zHeader[sizeof(aJournalMagic)+12], pPager->sectorSize);
769 IOTRACE(("JHDR %p %lld %d\n", pPager, pPager->journalHdr, sizeof(zHeader)))
770 rc = sqlite3OsWrite(pPager->jfd, zHeader, sizeof(zHeader));
771
772 /* The journal header has been written successfully. Seek the journal
773 ** file descriptor to the end of the journal header sector.
774 */
775 if( rc==SQLITE_OK ){
776 IOTRACE(("JTAIL %p %lld\n", pPager, pPager->journalOff-1))
777 rc = sqlite3OsSeek(pPager->jfd, pPager->journalOff-1);
778 if( rc==SQLITE_OK ){
779 rc = sqlite3OsWrite(pPager->jfd, "\000", 1);
780 }
781 }
782 return rc;
783}
784
785/*
786** The journal file must be open when this is called. A journal header file
787** (JOURNAL_HDR_SZ bytes) is read from the current location in the journal
788** file. See comments above function writeJournalHdr() for a description of
789** the journal header format.
790**
791** If the header is read successfully, *nRec is set to the number of
792** page records following this header and *dbSize is set to the size of the
793** database before the transaction began, in pages. Also, pPager->cksumInit
794** is set to the value read from the journal header. SQLITE_OK is returned
795** in this case.
796**
797** If the journal header file appears to be corrupted, SQLITE_DONE is
798** returned and *nRec and *dbSize are not set. If JOURNAL_HDR_SZ bytes
799** cannot be read from the journal file an error code is returned.
800*/
801static int readJournalHdr(
802 Pager *pPager,
803 i64 journalSize,
804 u32 *pNRec,
805 u32 *pDbSize
806){
807 int rc;
808 unsigned char aMagic[8]; /* A buffer to hold the magic header */
809
810 rc = seekJournalHdr(pPager);
811 if( rc ) return rc;
812
813 if( pPager->journalOff+JOURNAL_HDR_SZ(pPager) > journalSize ){
814 return SQLITE_DONE;
815 }
816
817 rc = sqlite3OsRead(pPager->jfd, aMagic, sizeof(aMagic));
818 if( rc ) return rc;
819
820 if( memcmp(aMagic, aJournalMagic, sizeof(aMagic))!=0 ){
821 return SQLITE_DONE;
822 }
823
824 rc = read32bits(pPager->jfd, pNRec);
825 if( rc ) return rc;
826
827 rc = read32bits(pPager->jfd, &pPager->cksumInit);
828 if( rc ) return rc;
829
830 rc = read32bits(pPager->jfd, pDbSize);
831 if( rc ) return rc;
832
833 /* Update the assumed sector-size to match the value used by
834 ** the process that created this journal. If this journal was
835 ** created by a process other than this one, then this routine
836 ** is being called from within pager_playback(). The local value
837 ** of Pager.sectorSize is restored at the end of that routine.
838 */
839 rc = read32bits(pPager->jfd, (u32 *)&pPager->sectorSize);
840 if( rc ) return rc;
841
842 pPager->journalOff += JOURNAL_HDR_SZ(pPager);
843 rc = sqlite3OsSeek(pPager->jfd, pPager->journalOff);
844 return rc;
845}
846
847
848/*
849** Write the supplied master journal name into the journal file for pager
850** pPager at the current location. The master journal name must be the last
851** thing written to a journal file. If the pager is in full-sync mode, the
852** journal file descriptor is advanced to the next sector boundary before
853** anything is written. The format is:
854**
855** + 4 bytes: PAGER_MJ_PGNO.
856** + N bytes: length of master journal name.
857** + 4 bytes: N
858** + 4 bytes: Master journal name checksum.
859** + 8 bytes: aJournalMagic[].
860**
861** The master journal page checksum is the sum of the bytes in the master
862** journal name.
863**
864** If zMaster is a NULL pointer (occurs for a single database transaction),
865** this call is a no-op.
866*/
867static int writeMasterJournal(Pager *pPager, const char *zMaster){
868 int rc;
869 int len;
870 int i;
871 u32 cksum = 0;
872 char zBuf[sizeof(aJournalMagic)+2*4];
873
874 if( !zMaster || pPager->setMaster) return SQLITE_OK;
875 pPager->setMaster = 1;
876
877 len = strlen(zMaster);
878 for(i=0; i<len; i++){
879 cksum += zMaster[i];
880 }
881
882 /* If in full-sync mode, advance to the next disk sector before writing
883 ** the master journal name. This is in case the previous page written to
884 ** the journal has already been synced.
885 */
886 if( pPager->fullSync ){
887 rc = seekJournalHdr(pPager);
888 if( rc!=SQLITE_OK ) return rc;
889 }
890 pPager->journalOff += (len+20);
891
892 rc = write32bits(pPager->jfd, PAGER_MJ_PGNO(pPager));
893 if( rc!=SQLITE_OK ) return rc;
894
895 rc = sqlite3OsWrite(pPager->jfd, zMaster, len);
896 if( rc!=SQLITE_OK ) return rc;
897
898 put32bits(zBuf, len);
899 put32bits(&zBuf[4], cksum);
900 memcpy(&zBuf[8], aJournalMagic, sizeof(aJournalMagic));
901 rc = sqlite3OsWrite(pPager->jfd, zBuf, 8+sizeof(aJournalMagic));
902 pPager->needSync = !pPager->noSync;
903 return rc;
904}
905
906/*
907** Add or remove a page from the list of all pages that are in the
908** statement journal.
909**
910** The Pager keeps a separate list of pages that are currently in
911** the statement journal. This helps the sqlite3PagerStmtCommit()
912** routine run MUCH faster for the common case where there are many
913** pages in memory but only a few are in the statement journal.
914*/
915static void page_add_to_stmt_list(PgHdr *pPg){
916 Pager *pPager = pPg->pPager;
917 PgHistory *pHist = PGHDR_TO_HIST(pPg, pPager);
918 assert( MEMDB );
919 if( !pHist->inStmt ){
920 assert( pHist->pPrevStmt==0 && pHist->pNextStmt==0 );
921 if( pPager->pStmt ){
922 PGHDR_TO_HIST(pPager->pStmt, pPager)->pPrevStmt = pPg;
923 }
924 pHist->pNextStmt = pPager->pStmt;
925 pPager->pStmt = pPg;
926 pHist->inStmt = 1;
927 }
928}
929
930/*
931** Find a page in the hash table given its page number. Return
932** a pointer to the page or NULL if not found.
933*/
934static PgHdr *pager_lookup(Pager *pPager, Pgno pgno){
935 PgHdr *p;
936 if( pPager->aHash==0 ) return 0;
937 p = pPager->aHash[pgno & (pPager->nHash-1)];
938 while( p && p->pgno!=pgno ){
939 p = p->pNextHash;
940 }
941 return p;
942}
943
944/*
945** Unlock the database file.
946*/
947static void pager_unlock(Pager *pPager){
948 if( !pPager->exclusiveMode ){
949 if( !MEMDB ){
950 sqlite3OsUnlock(pPager->fd, NO_LOCK);
951 pPager->dbSize = -1;
952 IOTRACE(("UNLOCK %p\n", pPager))
953 }
954 pPager->state = PAGER_UNLOCK;
955 pPager->changeCountDone = 0;
956 }
957}
958
959/*
960** Execute a rollback if a transaction is active and unlock the
961** database file. This is a no-op if the pager has already entered
962** the error-state.
963*/
964static void pagerUnlockAndRollback(Pager *p){
965 if( p->errCode ) return;
966 assert( p->state>=PAGER_RESERVED || p->journalOpen==0 );
967 if( p->state>=PAGER_RESERVED ){
968 sqlite3PagerRollback(p);
969 }
970 pager_unlock(p);
971 assert( p->errCode || !p->journalOpen || (p->exclusiveMode&&!p->journalOff) );
972 assert( p->errCode || !p->stmtOpen || p->exclusiveMode );
973}
974
975
976/*
977** Clear the in-memory cache. This routine
978** sets the state of the pager back to what it was when it was first
979** opened. Any outstanding pages are invalidated and subsequent attempts
980** to access those pages will likely result in a coredump.
981*/
982static void pager_reset(Pager *pPager){
983 PgHdr *pPg, *pNext;
984 if( pPager->errCode ) return;
985 for(pPg=pPager->pAll; pPg; pPg=pNext){
986 IOTRACE(("PGFREE %p %d\n", pPager, pPg->pgno));
987 PAGER_INCR(sqlite3_pager_pgfree_count);
988 pNext = pPg->pNextAll;
989 sqliteFree(pPg);
990 }
991 pPager->pStmt = 0;
992 pPager->pFirst = 0;
993 pPager->pFirstSynced = 0;
994 pPager->pLast = 0;
995 pPager->pAll = 0;
996 pPager->nHash = 0;
997 sqliteFree(pPager->aHash);
998 pPager->nPage = 0;
999 pPager->aHash = 0;
1000 pPager->nRef = 0;
1001}
1002
1003/*
1004** This routine ends a transaction. A transaction is ended by either
1005** a COMMIT or a ROLLBACK.
1006**
1007** When this routine is called, the pager has the journal file open and
1008** a RESERVED or EXCLUSIVE lock on the database. This routine will release
1009** the database lock and acquires a SHARED lock in its place if that is
1010** the appropriate thing to do. Release locks usually is appropriate,
1011** unless we are in exclusive access mode or unless this is a
1012** COMMIT AND BEGIN or ROLLBACK AND BEGIN operation.
1013**
1014** The journal file is either deleted or truncated.
1015**
1016** TODO: Consider keeping the journal file open for temporary databases.
1017** This might give a performance improvement on windows where opening
1018** a file is an expensive operation.
1019*/
1020static int pager_end_transaction(Pager *pPager){
1021 PgHdr *pPg;
1022 int rc = SQLITE_OK;
1023 int rc2 = SQLITE_OK;
1024 assert( !MEMDB );
1025 if( pPager->state<PAGER_RESERVED ){
1026 return SQLITE_OK;
1027 }
1028 sqlite3PagerStmtCommit(pPager);
1029 if( pPager->stmtOpen && !pPager->exclusiveMode ){
1030 sqlite3OsClose(&pPager->stfd);
1031 pPager->stmtOpen = 0;
1032 }
1033 if( pPager->journalOpen ){
1034 if( pPager->exclusiveMode
1035 && (rc = sqlite3OsTruncate(pPager->jfd, 0))==SQLITE_OK ){;
1036 sqlite3OsSeek(pPager->jfd, 0);
1037 pPager->journalOff = 0;
1038 pPager->journalStarted = 0;
1039 }else{
1040 sqlite3OsClose(&pPager->jfd);
1041 pPager->journalOpen = 0;
1042 if( rc==SQLITE_OK ){
1043 rc = sqlite3OsDelete(pPager->zJournal);
1044 }
1045 }
1046 sqliteFree( pPager->aInJournal );
1047 pPager->aInJournal = 0;
1048 for(pPg=pPager->pAll; pPg; pPg=pPg->pNextAll){
1049 pPg->inJournal = 0;
1050 pPg->dirty = 0;
1051 pPg->needSync = 0;
1052 pPg->alwaysRollback = 0;
1053#ifdef SQLITE_CHECK_PAGES
1054 pPg->pageHash = pager_pagehash(pPg);
1055#endif
1056 }
1057 pPager->pDirty = 0;
1058 pPager->dirtyCache = 0;
1059 pPager->nRec = 0;
1060 }else{
1061 assert( pPager->aInJournal==0 );
1062 assert( pPager->dirtyCache==0 || pPager->useJournal==0 );
1063 }
1064
1065 if( !pPager->exclusiveMode ){
1066 rc2 = sqlite3OsUnlock(pPager->fd, SHARED_LOCK);
1067 pPager->state = PAGER_SHARED;
1068 }else if( pPager->state==PAGER_SYNCED ){
1069 pPager->state = PAGER_EXCLUSIVE;
1070 }
1071 pPager->origDbSize = 0;
1072 pPager->setMaster = 0;
1073 pPager->needSync = 0;
1074 pPager->pFirstSynced = pPager->pFirst;
1075 pPager->dbSize = -1;
1076
1077 return (rc==SQLITE_OK?rc2:rc);
1078}
1079
1080/*
1081** Compute and return a checksum for the page of data.
1082**
1083** This is not a real checksum. It is really just the sum of the
1084** random initial value and the page number. We experimented with
1085** a checksum of the entire data, but that was found to be too slow.
1086**
1087** Note that the page number is stored at the beginning of data and
1088** the checksum is stored at the end. This is important. If journal
1089** corruption occurs due to a power failure, the most likely scenario
1090** is that one end or the other of the record will be changed. It is
1091** much less likely that the two ends of the journal record will be
1092** correct and the middle be corrupt. Thus, this "checksum" scheme,
1093** though fast and simple, catches the mostly likely kind of corruption.
1094**
1095** FIX ME: Consider adding every 200th (or so) byte of the data to the
1096** checksum. That way if a single page spans 3 or more disk sectors and
1097** only the middle sector is corrupt, we will still have a reasonable
1098** chance of failing the checksum and thus detecting the problem.
1099*/
1100static u32 pager_cksum(Pager *pPager, const u8 *aData){
1101 u32 cksum = pPager->cksumInit;
1102 int i = pPager->pageSize-200;
1103 while( i>0 ){
1104 cksum += aData[i];
1105 i -= 200;
1106 }
1107 return cksum;
1108}
1109
1110/* Forward declaration */
1111static void makeClean(PgHdr*);
1112
1113/*
1114** Read a single page from the journal file opened on file descriptor
1115** jfd. Playback this one page.
1116**
1117** If useCksum==0 it means this journal does not use checksums. Checksums
1118** are not used in statement journals because statement journals do not
1119** need to survive power failures.
1120*/
1121static int pager_playback_one_page(Pager *pPager, OsFile *jfd, int useCksum){
1122 int rc;
1123 PgHdr *pPg; /* An existing page in the cache */
1124 Pgno pgno; /* The page number of a page in journal */
1125 u32 cksum; /* Checksum used for sanity checking */
1126 u8 *aData = (u8 *)pPager->pTmpSpace; /* Temp storage for a page */
1127
1128 /* useCksum should be true for the main journal and false for
1129 ** statement journals. Verify that this is always the case
1130 */
1131 assert( jfd == (useCksum ? pPager->jfd : pPager->stfd) );
1132 assert( aData );
1133
1134 rc = read32bits(jfd, &pgno);
1135 if( rc!=SQLITE_OK ) return rc;
1136 rc = sqlite3OsRead(jfd, aData, pPager->pageSize);
1137 if( rc!=SQLITE_OK ) return rc;
1138 pPager->journalOff += pPager->pageSize + 4;
1139
1140 /* Sanity checking on the page. This is more important that I originally
1141 ** thought. If a power failure occurs while the journal is being written,
1142 ** it could cause invalid data to be written into the journal. We need to
1143 ** detect this invalid data (with high probability) and ignore it.
1144 */
1145 if( pgno==0 || pgno==PAGER_MJ_PGNO(pPager) ){
1146 return SQLITE_DONE;
1147 }
1148 if( pgno>(unsigned)pPager->dbSize ){
1149 return SQLITE_OK;
1150 }
1151 if( useCksum ){
1152 rc = read32bits(jfd, &cksum);
1153 if( rc ) return rc;
1154 pPager->journalOff += 4;
1155 if( pager_cksum(pPager, aData)!=cksum ){
1156 return SQLITE_DONE;
1157 }
1158 }
1159
1160 assert( pPager->state==PAGER_RESERVED || pPager->state>=PAGER_EXCLUSIVE );
1161
1162 /* If the pager is in RESERVED state, then there must be a copy of this
1163 ** page in the pager cache. In this case just update the pager cache,
1164 ** not the database file. The page is left marked dirty in this case.
1165 **
1166 ** An exception to the above rule: If the database is in no-sync mode
1167 ** and a page is moved during an incremental vacuum then the page may
1168 ** not be in the pager cache. Later: if a malloc() or IO error occurs
1169 ** during a Movepage() call, then the page may not be in the cache
1170 ** either. So the condition described in the above paragraph is not
1171 ** assert()able.
1172 **
1173 ** If in EXCLUSIVE state, then we update the pager cache if it exists
1174 ** and the main file. The page is then marked not dirty.
1175 **
1176 ** Ticket #1171: The statement journal might contain page content that is
1177 ** different from the page content at the start of the transaction.
1178 ** This occurs when a page is changed prior to the start of a statement
1179 ** then changed again within the statement. When rolling back such a
1180 ** statement we must not write to the original database unless we know
1181 ** for certain that original page contents are synced into the main rollback
1182 ** journal. Otherwise, a power loss might leave modified data in the
1183 ** database file without an entry in the rollback journal that can
1184 ** restore the database to its original form. Two conditions must be
1185 ** met before writing to the database files. (1) the database must be
1186 ** locked. (2) we know that the original page content is fully synced
1187 ** in the main journal either because the page is not in cache or else
1188 ** the page is marked as needSync==0.
1189 */
1190 pPg = pager_lookup(pPager, pgno);
1191 PAGERTRACE4("PLAYBACK %d page %d hash(%08x)\n",
1192 PAGERID(pPager), pgno, pager_datahash(pPager->pageSize, aData));
1193 if( pPager->state>=PAGER_EXCLUSIVE && (pPg==0 || pPg->needSync==0) ){
1194 rc = sqlite3OsSeek(pPager->fd, (pgno-1)*(i64)pPager->pageSize);
1195 if( rc==SQLITE_OK ){
1196 rc = sqlite3OsWrite(pPager->fd, aData, pPager->pageSize);
1197 }
1198 if( pPg ){
1199 makeClean(pPg);
1200 }
1201 }
1202 if( pPg ){
1203 /* No page should ever be explicitly rolled back that is in use, except
1204 ** for page 1 which is held in use in order to keep the lock on the
1205 ** database active. However such a page may be rolled back as a result
1206 ** of an internal error resulting in an automatic call to
1207 ** sqlite3PagerRollback().
1208 */
1209 void *pData;
1210 /* assert( pPg->nRef==0 || pPg->pgno==1 ); */
1211 pData = PGHDR_TO_DATA(pPg);
1212 memcpy(pData, aData, pPager->pageSize);
1213 if( pPager->xReiniter ){
1214 pPager->xReiniter(pPg, pPager->pageSize);
1215 }
1216#ifdef SQLITE_CHECK_PAGES
1217 pPg->pageHash = pager_pagehash(pPg);
1218#endif
1219 /* If this was page 1, then restore the value of Pager.dbFileVers.
1220 ** Do this before any decoding. */
1221 if( pgno==1 ){
1222 memcpy(&pPager->dbFileVers, &((u8*)pData)[24],sizeof(pPager->dbFileVers));
1223 }
1224
1225 /* Decode the page just read from disk */
1226 CODEC1(pPager, pData, pPg->pgno, 3);
1227 }
1228 return rc;
1229}
1230
1231/*
1232** Parameter zMaster is the name of a master journal file. A single journal
1233** file that referred to the master journal file has just been rolled back.
1234** This routine checks if it is possible to delete the master journal file,
1235** and does so if it is.
1236**
1237** The master journal file contains the names of all child journals.
1238** To tell if a master journal can be deleted, check to each of the
1239** children. If all children are either missing or do not refer to
1240** a different master journal, then this master journal can be deleted.
1241*/
1242static int pager_delmaster(const char *zMaster){
1243 int rc;
1244 int master_open = 0;
1245 OsFile *master = 0;
1246 char *zMasterJournal = 0; /* Contents of master journal file */
1247 i64 nMasterJournal; /* Size of master journal file */
1248
1249 /* Open the master journal file exclusively in case some other process
1250 ** is running this routine also. Not that it makes too much difference.
1251 */
1252 rc = sqlite3OsOpenReadOnly(zMaster, &master);
1253 assert( rc!=SQLITE_OK || master );
1254 if( rc!=SQLITE_OK ) goto delmaster_out;
1255 master_open = 1;
1256 rc = sqlite3OsFileSize(master, &nMasterJournal);
1257 if( rc!=SQLITE_OK ) goto delmaster_out;
1258
1259 if( nMasterJournal>0 ){
1260 char *zJournal;
1261 char *zMasterPtr = 0;
1262
1263 /* Load the entire master journal file into space obtained from
1264 ** sqliteMalloc() and pointed to by zMasterJournal.
1265 */
1266 zMasterJournal = (char *)sqliteMalloc(nMasterJournal);
1267 if( !zMasterJournal ){
1268 rc = SQLITE_NOMEM;
1269 goto delmaster_out;
1270 }
1271 rc = sqlite3OsRead(master, zMasterJournal, nMasterJournal);
1272 if( rc!=SQLITE_OK ) goto delmaster_out;
1273
1274 zJournal = zMasterJournal;
1275 while( (zJournal-zMasterJournal)<nMasterJournal ){
1276 if( sqlite3OsFileExists(zJournal) ){
1277 /* One of the journals pointed to by the master journal exists.
1278 ** Open it and check if it points at the master journal. If
1279 ** so, return without deleting the master journal file.
1280 */
1281 OsFile *journal = 0;
1282 int c;
1283
1284 rc = sqlite3OsOpenReadOnly(zJournal, &journal);
1285 assert( rc!=SQLITE_OK || journal );
1286 if( rc!=SQLITE_OK ){
1287 goto delmaster_out;
1288 }
1289
1290 rc = readMasterJournal(journal, &zMasterPtr);
1291 sqlite3OsClose(&journal);
1292 if( rc!=SQLITE_OK ){
1293 goto delmaster_out;
1294 }
1295
1296 c = zMasterPtr!=0 && strcmp(zMasterPtr, zMaster)==0;
1297 sqliteFree(zMasterPtr);
1298 if( c ){
1299 /* We have a match. Do not delete the master journal file. */
1300 goto delmaster_out;
1301 }
1302 }
1303 zJournal += (strlen(zJournal)+1);
1304 }
1305 }
1306
1307 rc = sqlite3OsDelete(zMaster);
1308
1309delmaster_out:
1310 if( zMasterJournal ){
1311 sqliteFree(zMasterJournal);
1312 }
1313 if( master_open ){
1314 sqlite3OsClose(&master);
1315 }
1316 return rc;
1317}
1318
1319
1320static void pager_truncate_cache(Pager *pPager);
1321
1322/*
1323** Truncate the main file of the given pager to the number of pages
1324** indicated. Also truncate the cached representation of the file.
1325*/
1326static int pager_truncate(Pager *pPager, int nPage){
1327 int rc = SQLITE_OK;
1328 if( pPager->state>=PAGER_EXCLUSIVE ){
1329 rc = sqlite3OsTruncate(pPager->fd, pPager->pageSize*(i64)nPage);
1330 }
1331 if( rc==SQLITE_OK ){
1332 pPager->dbSize = nPage;
1333 pager_truncate_cache(pPager);
1334 }
1335 return rc;
1336}
1337
1338/*
1339** Set the sectorSize for the given pager.
1340**
1341** The sector size is the larger of the sector size reported
1342** by sqlite3OsSectorSize() and the pageSize.
1343*/
1344static void setSectorSize(Pager *pPager){
1345 pPager->sectorSize = sqlite3OsSectorSize(pPager->fd);
1346 if( pPager->sectorSize<pPager->pageSize ){
1347 pPager->sectorSize = pPager->pageSize;
1348 }
1349}
1350
1351/*
1352** Playback the journal and thus restore the database file to
1353** the state it was in before we started making changes.
1354**
1355** The journal file format is as follows:
1356**
1357** (1) 8 byte prefix. A copy of aJournalMagic[].
1358** (2) 4 byte big-endian integer which is the number of valid page records
1359** in the journal. If this value is 0xffffffff, then compute the
1360** number of page records from the journal size.
1361** (3) 4 byte big-endian integer which is the initial value for the
1362** sanity checksum.
1363** (4) 4 byte integer which is the number of pages to truncate the
1364** database to during a rollback.
1365** (5) 4 byte integer which is the number of bytes in the master journal
1366** name. The value may be zero (indicate that there is no master
1367** journal.)
1368** (6) N bytes of the master journal name. The name will be nul-terminated
1369** and might be shorter than the value read from (5). If the first byte
1370** of the name is \000 then there is no master journal. The master
1371** journal name is stored in UTF-8.
1372** (7) Zero or more pages instances, each as follows:
1373** + 4 byte page number.
1374** + pPager->pageSize bytes of data.
1375** + 4 byte checksum
1376**
1377** When we speak of the journal header, we mean the first 6 items above.
1378** Each entry in the journal is an instance of the 7th item.
1379**
1380** Call the value from the second bullet "nRec". nRec is the number of
1381** valid page entries in the journal. In most cases, you can compute the
1382** value of nRec from the size of the journal file. But if a power
1383** failure occurred while the journal was being written, it could be the
1384** case that the size of the journal file had already been increased but
1385** the extra entries had not yet made it safely to disk. In such a case,
1386** the value of nRec computed from the file size would be too large. For
1387** that reason, we always use the nRec value in the header.
1388**
1389** If the nRec value is 0xffffffff it means that nRec should be computed
1390** from the file size. This value is used when the user selects the
1391** no-sync option for the journal. A power failure could lead to corruption
1392** in this case. But for things like temporary table (which will be
1393** deleted when the power is restored) we don't care.
1394**
1395** If the file opened as the journal file is not a well-formed
1396** journal file then all pages up to the first corrupted page are rolled
1397** back (or no pages if the journal header is corrupted). The journal file
1398** is then deleted and SQLITE_OK returned, just as if no corruption had
1399** been encountered.
1400**
1401** If an I/O or malloc() error occurs, the journal-file is not deleted
1402** and an error code is returned.
1403*/
1404static int pager_playback(Pager *pPager, int isHot){
1405 i64 szJ; /* Size of the journal file in bytes */
1406 u32 nRec; /* Number of Records in the journal */
1407 int i; /* Loop counter */
1408 Pgno mxPg = 0; /* Size of the original file in pages */
1409 int rc; /* Result code of a subroutine */
1410 char *zMaster = 0; /* Name of master journal file if any */
1411
1412 /* Figure out how many records are in the journal. Abort early if
1413 ** the journal is empty.
1414 */
1415 assert( pPager->journalOpen );
1416 rc = sqlite3OsFileSize(pPager->jfd, &szJ);
1417 if( rc!=SQLITE_OK || szJ==0 ){
1418 goto end_playback;
1419 }
1420
1421 /* Read the master journal name from the journal, if it is present.
1422 ** If a master journal file name is specified, but the file is not
1423 ** present on disk, then the journal is not hot and does not need to be
1424 ** played back.
1425 */
1426 rc = readMasterJournal(pPager->jfd, &zMaster);
1427 assert( rc!=SQLITE_DONE );
1428 if( rc!=SQLITE_OK || (zMaster && !sqlite3OsFileExists(zMaster)) ){
1429 sqliteFree(zMaster);
1430 zMaster = 0;
1431 if( rc==SQLITE_DONE ) rc = SQLITE_OK;
1432 goto end_playback;
1433 }
1434 sqlite3OsSeek(pPager->jfd, 0);
1435 pPager->journalOff = 0;
1436
1437 /* This loop terminates either when the readJournalHdr() call returns
1438 ** SQLITE_DONE or an IO error occurs. */
1439 while( 1 ){
1440
1441 /* Read the next journal header from the journal file. If there are
1442 ** not enough bytes left in the journal file for a complete header, or
1443 ** it is corrupted, then a process must of failed while writing it.
1444 ** This indicates nothing more needs to be rolled back.
1445 */
1446 rc = readJournalHdr(pPager, szJ, &nRec, &mxPg);
1447 if( rc!=SQLITE_OK ){
1448 if( rc==SQLITE_DONE ){
1449 rc = SQLITE_OK;
1450 }
1451 goto end_playback;
1452 }
1453
1454 /* If nRec is 0xffffffff, then this journal was created by a process
1455 ** working in no-sync mode. This means that the rest of the journal
1456 ** file consists of pages, there are no more journal headers. Compute
1457 ** the value of nRec based on this assumption.
1458 */
1459 if( nRec==0xffffffff ){
1460 assert( pPager->journalOff==JOURNAL_HDR_SZ(pPager) );
1461 nRec = (szJ - JOURNAL_HDR_SZ(pPager))/JOURNAL_PG_SZ(pPager);
1462 }
1463
1464 /* If nRec is 0 and this rollback is of a transaction created by this
1465 ** process. In this case the rest of the journal file consists of
1466 ** journalled copies of pages that need to be read back into the cache.
1467 */
1468 if( nRec==0 && !isHot ){
1469 nRec = (szJ - pPager->journalOff) / JOURNAL_PG_SZ(pPager);
1470 }
1471
1472 /* If this is the first header read from the journal, truncate the
1473 ** database file back to it's original size.
1474 */
1475 if( pPager->journalOff==JOURNAL_HDR_SZ(pPager) ){
1476 rc = pager_truncate(pPager, mxPg);
1477 if( rc!=SQLITE_OK ){
1478 goto end_playback;
1479 }
1480 }
1481
1482 /* Copy original pages out of the journal and back into the database file.
1483 */
1484 for(i=0; i<nRec; i++){
1485 rc = pager_playback_one_page(pPager, pPager->jfd, 1);
1486 if( rc!=SQLITE_OK ){
1487 if( rc==SQLITE_DONE ){
1488 rc = SQLITE_OK;
1489 pPager->journalOff = szJ;
1490 break;
1491 }else{
1492 goto end_playback;
1493 }
1494 }
1495 }
1496 }
1497 /*NOTREACHED*/
1498 assert( 0 );
1499
1500end_playback:
1501 if( rc==SQLITE_OK ){
1502 rc = pager_end_transaction(pPager);
1503 }
1504 if( zMaster ){
1505 /* If there was a master journal and this routine will return success,
1506 ** see if it is possible to delete the master journal.
1507 */
1508 if( rc==SQLITE_OK ){
1509 rc = pager_delmaster(zMaster);
1510 }
1511 sqliteFree(zMaster);
1512 }
1513
1514 /* The Pager.sectorSize variable may have been updated while rolling
1515 ** back a journal created by a process with a different sector size
1516 ** value. Reset it to the correct value for this process.
1517 */
1518 setSectorSize(pPager);
1519 return rc;
1520}
1521
1522/*
1523** Playback the statement journal.
1524**
1525** This is similar to playing back the transaction journal but with
1526** a few extra twists.
1527**
1528** (1) The number of pages in the database file at the start of
1529** the statement is stored in pPager->stmtSize, not in the
1530** journal file itself.
1531**
1532** (2) In addition to playing back the statement journal, also
1533** playback all pages of the transaction journal beginning
1534** at offset pPager->stmtJSize.
1535*/
1536static int pager_stmt_playback(Pager *pPager){
1537 i64 szJ; /* Size of the full journal */
1538 i64 hdrOff;
1539 int nRec; /* Number of Records */
1540 int i; /* Loop counter */
1541 int rc;
1542
1543 szJ = pPager->journalOff;
1544#ifndef NDEBUG
1545 {
1546 i64 os_szJ;
1547 rc = sqlite3OsFileSize(pPager->jfd, &os_szJ);
1548 if( rc!=SQLITE_OK ) return rc;
1549 assert( szJ==os_szJ );
1550 }
1551#endif
1552
1553 /* Set hdrOff to be the offset just after the end of the last journal
1554 ** page written before the first journal-header for this statement
1555 ** transaction was written, or the end of the file if no journal
1556 ** header was written.
1557 */
1558 hdrOff = pPager->stmtHdrOff;
1559 assert( pPager->fullSync || !hdrOff );
1560 if( !hdrOff ){
1561 hdrOff = szJ;
1562 }
1563
1564 /* Truncate the database back to its original size.
1565 */
1566 rc = pager_truncate(pPager, pPager->stmtSize);
1567 assert( pPager->state>=PAGER_SHARED );
1568
1569 /* Figure out how many records are in the statement journal.
1570 */
1571 assert( pPager->stmtInUse && pPager->journalOpen );
1572 sqlite3OsSeek(pPager->stfd, 0);
1573 nRec = pPager->stmtNRec;
1574
1575 /* Copy original pages out of the statement journal and back into the
1576 ** database file. Note that the statement journal omits checksums from
1577 ** each record since power-failure recovery is not important to statement
1578 ** journals.
1579 */
1580 for(i=nRec-1; i>=0; i--){
1581 rc = pager_playback_one_page(pPager, pPager->stfd, 0);
1582 assert( rc!=SQLITE_DONE );
1583 if( rc!=SQLITE_OK ) goto end_stmt_playback;
1584 }
1585
1586 /* Now roll some pages back from the transaction journal. Pager.stmtJSize
1587 ** was the size of the journal file when this statement was started, so
1588 ** everything after that needs to be rolled back, either into the
1589 ** database, the memory cache, or both.
1590 **
1591 ** If it is not zero, then Pager.stmtHdrOff is the offset to the start
1592 ** of the first journal header written during this statement transaction.
1593 */
1594 rc = sqlite3OsSeek(pPager->jfd, pPager->stmtJSize);
1595 if( rc!=SQLITE_OK ){
1596 goto end_stmt_playback;
1597 }
1598 pPager->journalOff = pPager->stmtJSize;
1599 pPager->cksumInit = pPager->stmtCksum;
1600 while( pPager->journalOff < hdrOff ){
1601 rc = pager_playback_one_page(pPager, pPager->jfd, 1);
1602 assert( rc!=SQLITE_DONE );
1603 if( rc!=SQLITE_OK ) goto end_stmt_playback;
1604 }
1605
1606 while( pPager->journalOff < szJ ){
1607 u32 nJRec; /* Number of Journal Records */
1608 u32 dummy;
1609 rc = readJournalHdr(pPager, szJ, &nJRec, &dummy);
1610 if( rc!=SQLITE_OK ){
1611 assert( rc!=SQLITE_DONE );
1612 goto end_stmt_playback;
1613 }
1614 if( nJRec==0 ){
1615 nJRec = (szJ - pPager->journalOff) / (pPager->pageSize+8);
1616 }
1617 for(i=nJRec-1; i>=0 && pPager->journalOff < szJ; i--){
1618 rc = pager_playback_one_page(pPager, pPager->jfd, 1);
1619 assert( rc!=SQLITE_DONE );
1620 if( rc!=SQLITE_OK ) goto end_stmt_playback;
1621 }
1622 }
1623
1624 pPager->journalOff = szJ;
1625
1626end_stmt_playback:
1627 if( rc==SQLITE_OK) {
1628 pPager->journalOff = szJ;
1629 /* pager_reload_cache(pPager); */
1630 }
1631 return rc;
1632}
1633
1634/*
1635** Change the maximum number of in-memory pages that are allowed.
1636*/
1637void sqlite3PagerSetCachesize(Pager *pPager, int mxPage){
1638 if( mxPage>10 ){
1639 pPager->mxPage = mxPage;
1640 }else{
1641 pPager->mxPage = 10;
1642 }
1643}
1644
1645/*
1646** Adjust the robustness of the database to damage due to OS crashes
1647** or power failures by changing the number of syncs()s when writing
1648** the rollback journal. There are three levels:
1649**
1650** OFF sqlite3OsSync() is never called. This is the default
1651** for temporary and transient files.
1652**
1653** NORMAL The journal is synced once before writes begin on the
1654** database. This is normally adequate protection, but
1655** it is theoretically possible, though very unlikely,
1656** that an inopertune power failure could leave the journal
1657** in a state which would cause damage to the database
1658** when it is rolled back.
1659**
1660** FULL The journal is synced twice before writes begin on the
1661** database (with some additional information - the nRec field
1662** of the journal header - being written in between the two
1663** syncs). If we assume that writing a
1664** single disk sector is atomic, then this mode provides
1665** assurance that the journal will not be corrupted to the
1666** point of causing damage to the database during rollback.
1667**
1668** Numeric values associated with these states are OFF==1, NORMAL=2,
1669** and FULL=3.
1670*/
1671#ifndef SQLITE_OMIT_PAGER_PRAGMAS
1672void sqlite3PagerSetSafetyLevel(Pager *pPager, int level, int full_fsync){
1673 pPager->noSync = level==1 || pPager->tempFile;
1674 pPager->fullSync = level==3 && !pPager->tempFile;
1675 pPager->full_fsync = full_fsync;
1676 if( pPager->noSync ) pPager->needSync = 0;
1677}
1678#endif
1679
1680/*
1681** The following global variable is incremented whenever the library
1682** attempts to open a temporary file. This information is used for
1683** testing and analysis only.
1684*/
1685#ifdef SQLITE_TEST
1686int sqlite3_opentemp_count = 0;
1687#endif
1688
1689/*
1690** Open a temporary file.
1691**
1692** Write the file descriptor into *fd. Return SQLITE_OK on success or some
1693** other error code if we fail.
1694**
1695** The OS will automatically delete the temporary file when it is
1696** closed.
1697*/
1698static int sqlite3PagerOpentemp(OsFile **pFd){
1699 int cnt = 8;
1700 int rc;
1701 char zFile[SQLITE_TEMPNAME_SIZE];
1702
1703#ifdef SQLITE_TEST
1704 sqlite3_opentemp_count++; /* Used for testing and analysis only */
1705#endif
1706 do{
1707 cnt--;
1708 sqlite3OsTempFileName(zFile);
1709 rc = sqlite3OsOpenExclusive(zFile, pFd, 1);
1710 assert( rc!=SQLITE_OK || *pFd );
1711 }while( cnt>0 && rc!=SQLITE_OK && rc!=SQLITE_NOMEM );
1712 return rc;
1713}
1714
1715/*
1716** Create a new page cache and put a pointer to the page cache in *ppPager.
1717** The file to be cached need not exist. The file is not locked until
1718** the first call to sqlite3PagerGet() and is only held open until the
1719** last page is released using sqlite3PagerUnref().
1720**
1721** If zFilename is NULL then a randomly-named temporary file is created
1722** and used as the file to be cached. The file will be deleted
1723** automatically when it is closed.
1724**
1725** If zFilename is ":memory:" then all information is held in cache.
1726** It is never written to disk. This can be used to implement an
1727** in-memory database.
1728*/
1729int sqlite3PagerOpen(
1730 Pager **ppPager, /* Return the Pager structure here */
1731 const char *zFilename, /* Name of the database file to open */
1732 int nExtra, /* Extra bytes append to each in-memory page */
1733 int flags /* flags controlling this file */
1734){
1735 Pager *pPager = 0;
1736 char *zFullPathname = 0;
1737 int nameLen; /* Compiler is wrong. This is always initialized before use */
1738 OsFile *fd = 0;
1739 int rc = SQLITE_OK;
1740 int i;
1741 int tempFile = 0;
1742 int memDb = 0;
1743 int readOnly = 0;
1744 int useJournal = (flags & PAGER_OMIT_JOURNAL)==0;
1745 int noReadlock = (flags & PAGER_NO_READLOCK)!=0;
1746 char zTemp[SQLITE_TEMPNAME_SIZE];
1747#ifdef SQLITE_ENABLE_MEMORY_MANAGEMENT
1748 /* A malloc() cannot fail in sqlite3ThreadData() as one or more calls to
1749 ** malloc() must have already been made by this thread before it gets
1750 ** to this point. This means the ThreadData must have been allocated already
1751 ** so that ThreadData.nAlloc can be set. It would be nice to assert
1752 ** that ThreadData.nAlloc is non-zero, but alas this breaks test cases
1753 ** written to invoke the pager directly.
1754 */
1755 ThreadData *pTsd = sqlite3ThreadData();
1756 assert( pTsd );
1757#endif
1758
1759 /* We used to test if malloc() had already failed before proceeding.
1760 ** But the way this function is used in SQLite means that can never
1761 ** happen. Furthermore, if the malloc-failed flag is already set,
1762 ** either the call to sqliteStrDup() or sqliteMalloc() below will
1763 ** fail shortly and SQLITE_NOMEM returned anyway.
1764 */
1765 *ppPager = 0;
1766
1767 /* Open the pager file and set zFullPathname to point at malloc()ed
1768 ** memory containing the complete filename (i.e. including the directory).
1769 */
1770 if( zFilename && zFilename[0] ){
1771#ifndef SQLITE_OMIT_MEMORYDB
1772 if( strcmp(zFilename,":memory:")==0 ){
1773 memDb = 1;
1774 zFullPathname = sqliteStrDup("");
1775 }else
1776#endif
1777 {
1778 zFullPathname = sqlite3OsFullPathname(zFilename);
1779 if( zFullPathname ){
1780 rc = sqlite3OsOpenReadWrite(zFullPathname, &fd, &readOnly);
1781 assert( rc!=SQLITE_OK || fd );
1782 }
1783 }
1784 }else{
1785 rc = sqlite3PagerOpentemp(&fd);
1786 sqlite3OsTempFileName(zTemp);
1787 zFilename = zTemp;
1788 zFullPathname = sqlite3OsFullPathname(zFilename);
1789 if( rc==SQLITE_OK ){
1790 tempFile = 1;
1791 }
1792 }
1793
1794 /* Allocate the Pager structure. As part of the same allocation, allocate
1795 ** space for the full paths of the file, directory and journal
1796 ** (Pager.zFilename, Pager.zDirectory and Pager.zJournal).
1797 */
1798 if( zFullPathname ){
1799 nameLen = strlen(zFullPathname);
1800 pPager = sqliteMalloc( sizeof(*pPager) + nameLen*3 + 30 );
1801 if( pPager && rc==SQLITE_OK ){
1802 pPager->pTmpSpace = (char *)sqliteMallocRaw(SQLITE_DEFAULT_PAGE_SIZE);
1803 }
1804 }
1805
1806
1807 /* If an error occured in either of the blocks above, free the memory
1808 ** pointed to by zFullPathname, free the Pager structure and close the
1809 ** file. Since the pager is not allocated there is no need to set
1810 ** any Pager.errMask variables.
1811 */
1812 if( !pPager || !zFullPathname || !pPager->pTmpSpace || rc!=SQLITE_OK ){
1813 sqlite3OsClose(&fd);
1814 sqliteFree(zFullPathname);
1815 sqliteFree(pPager);
1816 return ((rc==SQLITE_OK)?SQLITE_NOMEM:rc);
1817 }
1818
1819 PAGERTRACE3("OPEN %d %s\n", FILEHANDLEID(fd), zFullPathname);
1820 IOTRACE(("OPEN %p %s\n", pPager, zFullPathname))
1821 pPager->zFilename = (char*)&pPager[1];
1822 pPager->zDirectory = &pPager->zFilename[nameLen+1];
1823 pPager->zJournal = &pPager->zDirectory[nameLen+1];
1824 memcpy(pPager->zFilename, zFullPathname, nameLen+1);
1825 memcpy(pPager->zDirectory, zFullPathname, nameLen+1);
1826
1827 for(i=nameLen; i>0 && pPager->zDirectory[i-1]!='/'; i--){}
1828 if( i>0 ) pPager->zDirectory[i-1] = 0;
1829 memcpy(pPager->zJournal, zFullPathname,nameLen);
1830 sqliteFree(zFullPathname);
1831 memcpy(&pPager->zJournal[nameLen], "-journal",sizeof("-journal"));
1832 pPager->fd = fd;
1833 /* pPager->journalOpen = 0; */
1834 pPager->useJournal = useJournal && !memDb;
1835 pPager->noReadlock = noReadlock && readOnly;
1836 /* pPager->stmtOpen = 0; */
1837 /* pPager->stmtInUse = 0; */
1838 /* pPager->nRef = 0; */
1839 pPager->dbSize = memDb-1;
1840 pPager->pageSize = SQLITE_DEFAULT_PAGE_SIZE;
1841 /* pPager->stmtSize = 0; */
1842 /* pPager->stmtJSize = 0; */
1843 /* pPager->nPage = 0; */
1844 pPager->mxPage = 100;
1845 pPager->mxPgno = SQLITE_MAX_PAGE_COUNT;
1846 assert( PAGER_UNLOCK==0 );
1847 /* pPager->state = PAGER_UNLOCK; */
1848 /* pPager->errMask = 0; */
1849 pPager->tempFile = tempFile;
1850 assert( tempFile==PAGER_LOCKINGMODE_NORMAL
1851 || tempFile==PAGER_LOCKINGMODE_EXCLUSIVE );
1852 assert( PAGER_LOCKINGMODE_EXCLUSIVE==1 );
1853 pPager->exclusiveMode = tempFile;
1854 pPager->memDb = memDb;
1855 pPager->readOnly = readOnly;
1856 /* pPager->needSync = 0; */
1857 pPager->noSync = pPager->tempFile || !useJournal;
1858 pPager->fullSync = (pPager->noSync?0:1);
1859 /* pPager->pFirst = 0; */
1860 /* pPager->pFirstSynced = 0; */
1861 /* pPager->pLast = 0; */
1862 pPager->nExtra = FORCE_ALIGNMENT(nExtra);
1863 assert(fd||memDb);
1864 if( !memDb ){
1865 setSectorSize(pPager);
1866 }
1867 /* pPager->pBusyHandler = 0; */
1868 /* memset(pPager->aHash, 0, sizeof(pPager->aHash)); */
1869 *ppPager = pPager;
1870#ifdef SQLITE_ENABLE_MEMORY_MANAGEMENT
1871 pPager->pNext = pTsd->pPager;
1872 pTsd->pPager = pPager;
1873#endif
1874 return SQLITE_OK;
1875}
1876
1877/*
1878** Set the busy handler function.
1879*/
1880void sqlite3PagerSetBusyhandler(Pager *pPager, BusyHandler *pBusyHandler){
1881 pPager->pBusyHandler = pBusyHandler;
1882}
1883
1884/*
1885** Set the destructor for this pager. If not NULL, the destructor is called
1886** when the reference count on each page reaches zero. The destructor can
1887** be used to clean up information in the extra segment appended to each page.
1888**
1889** The destructor is not called as a result sqlite3PagerClose().
1890** Destructors are only called by sqlite3PagerUnref().
1891*/
1892void sqlite3PagerSetDestructor(Pager *pPager, void (*xDesc)(DbPage*,int)){
1893 pPager->xDestructor = xDesc;
1894}
1895
1896/*
1897** Set the reinitializer for this pager. If not NULL, the reinitializer
1898** is called when the content of a page in cache is restored to its original
1899** value as a result of a rollback. The callback gives higher-level code
1900** an opportunity to restore the EXTRA section to agree with the restored
1901** page data.
1902*/
1903void sqlite3PagerSetReiniter(Pager *pPager, void (*xReinit)(DbPage*,int)){
1904 pPager->xReiniter = xReinit;
1905}
1906
1907/*
1908** Set the page size. Return the new size. If the suggest new page
1909** size is inappropriate, then an alternative page size is selected
1910** and returned.
1911*/
1912int sqlite3PagerSetPagesize(Pager *pPager, int pageSize){
1913 assert( pageSize>=512 && pageSize<=SQLITE_MAX_PAGE_SIZE );
1914 if( !pPager->memDb && pPager->nRef==0 ){
1915 pager_reset(pPager);
1916 pPager->pageSize = pageSize;
1917 pPager->pTmpSpace = sqlite3ReallocOrFree(pPager->pTmpSpace, pageSize);
1918 }
1919 return pPager->pageSize;
1920}
1921
1922/*
1923** Attempt to set the maximum database page count if mxPage is positive.
1924** Make no changes if mxPage is zero or negative. And never reduce the
1925** maximum page count below the current size of the database.
1926**
1927** Regardless of mxPage, return the current maximum page count.
1928*/
1929int sqlite3PagerMaxPageCount(Pager *pPager, int mxPage){
1930 if( mxPage>0 ){
1931 pPager->mxPgno = mxPage;
1932 }
1933 sqlite3PagerPagecount(pPager);
1934 return pPager->mxPgno;
1935}
1936
1937/*
1938** The following set of routines are used to disable the simulated
1939** I/O error mechanism. These routines are used to avoid simulated
1940** errors in places where we do not care about errors.
1941**
1942** Unless -DSQLITE_TEST=1 is used, these routines are all no-ops
1943** and generate no code.
1944*/
1945#ifdef SQLITE_TEST
1946extern int sqlite3_io_error_pending;
1947extern int sqlite3_io_error_hit;
1948static int saved_cnt;
1949void disable_simulated_io_errors(void){
1950 saved_cnt = sqlite3_io_error_pending;
1951 sqlite3_io_error_pending = -1;
1952}
1953void enable_simulated_io_errors(void){
1954 sqlite3_io_error_pending = saved_cnt;
1955}
1956#else
1957# define disable_simulated_io_errors()
1958# define enable_simulated_io_errors()
1959#endif
1960
1961/*
1962** Read the first N bytes from the beginning of the file into memory
1963** that pDest points to.
1964**
1965** No error checking is done. The rational for this is that this function
1966** may be called even if the file does not exist or contain a header. In
1967** these cases sqlite3OsRead() will return an error, to which the correct
1968** response is to zero the memory at pDest and continue. A real IO error
1969** will presumably recur and be picked up later (Todo: Think about this).
1970*/
1971int sqlite3PagerReadFileheader(Pager *pPager, int N, unsigned char *pDest){
1972 int rc = SQLITE_OK;
1973 memset(pDest, 0, N);
1974 if( MEMDB==0 ){
1975 disable_simulated_io_errors();
1976 sqlite3OsSeek(pPager->fd, 0);
1977 enable_simulated_io_errors();
1978 IOTRACE(("DBHDR %p 0 %d\n", pPager, N))
1979 rc = sqlite3OsRead(pPager->fd, pDest, N);
1980 if( rc==SQLITE_IOERR_SHORT_READ ){
1981 rc = SQLITE_OK;
1982 }
1983 }
1984 return rc;
1985}
1986
1987/*
1988** Return the total number of pages in the disk file associated with
1989** pPager.
1990**
1991** If the PENDING_BYTE lies on the page directly after the end of the
1992** file, then consider this page part of the file too. For example, if
1993** PENDING_BYTE is byte 4096 (the first byte of page 5) and the size of the
1994** file is 4096 bytes, 5 is returned instead of 4.
1995*/
1996int sqlite3PagerPagecount(Pager *pPager){
1997 i64 n;
1998 int rc;
1999 assert( pPager!=0 );
2000 if( pPager->errCode ){
2001 return 0;
2002 }
2003 if( pPager->dbSize>=0 ){
2004 n = pPager->dbSize;
2005 } else {
2006 if( (rc = sqlite3OsFileSize(pPager->fd, &n))!=SQLITE_OK ){
2007 pager_error(pPager, rc);
2008 return 0;
2009 }
2010 if( n>0 && n<pPager->pageSize ){
2011 n = 1;
2012 }else{
2013 n /= pPager->pageSize;
2014 }
2015 if( pPager->state!=PAGER_UNLOCK ){
2016 pPager->dbSize = n;
2017 }
2018 }
2019 if( n==(PENDING_BYTE/pPager->pageSize) ){
2020 n++;
2021 }
2022 if( n>pPager->mxPgno ){
2023 pPager->mxPgno = n;
2024 }
2025 return n;
2026}
2027
2028
2029#ifndef SQLITE_OMIT_MEMORYDB
2030/*
2031** Clear a PgHistory block
2032*/
2033static void clearHistory(PgHistory *pHist){
2034 sqliteFree(pHist->pOrig);
2035 sqliteFree(pHist->pStmt);
2036 pHist->pOrig = 0;
2037 pHist->pStmt = 0;
2038}
2039#else
2040#define clearHistory(x)
2041#endif
2042
2043/*
2044** Forward declaration
2045*/
2046static int syncJournal(Pager*);
2047
2048/*
2049** Unlink pPg from it's hash chain. Also set the page number to 0 to indicate
2050** that the page is not part of any hash chain. This is required because the
2051** sqlite3PagerMovepage() routine can leave a page in the
2052** pNextFree/pPrevFree list that is not a part of any hash-chain.
2053*/
2054static void unlinkHashChain(Pager *pPager, PgHdr *pPg){
2055 if( pPg->pgno==0 ){
2056 assert( pPg->pNextHash==0 && pPg->pPrevHash==0 );
2057 return;
2058 }
2059 if( pPg->pNextHash ){
2060 pPg->pNextHash->pPrevHash = pPg->pPrevHash;
2061 }
2062 if( pPg->pPrevHash ){
2063 assert( pPager->aHash[pPg->pgno & (pPager->nHash-1)]!=pPg );
2064 pPg->pPrevHash->pNextHash = pPg->pNextHash;
2065 }else{
2066 int h = pPg->pgno & (pPager->nHash-1);
2067 pPager->aHash[h] = pPg->pNextHash;
2068 }
2069 if( MEMDB ){
2070 clearHistory(PGHDR_TO_HIST(pPg, pPager));
2071 }
2072 pPg->pgno = 0;
2073 pPg->pNextHash = pPg->pPrevHash = 0;
2074}
2075
2076/*
2077** Unlink a page from the free list (the list of all pages where nRef==0)
2078** and from its hash collision chain.
2079*/
2080static void unlinkPage(PgHdr *pPg){
2081 Pager *pPager = pPg->pPager;
2082
2083 /* Keep the pFirstSynced pointer pointing at the first synchronized page */
2084 if( pPg==pPager->pFirstSynced ){
2085 PgHdr *p = pPg->pNextFree;
2086 while( p && p->needSync ){ p = p->pNextFree; }
2087 pPager->pFirstSynced = p;
2088 }
2089
2090 /* Unlink from the freelist */
2091 if( pPg->pPrevFree ){
2092 pPg->pPrevFree->pNextFree = pPg->pNextFree;
2093 }else{
2094 assert( pPager->pFirst==pPg );
2095 pPager->pFirst = pPg->pNextFree;
2096 }
2097 if( pPg->pNextFree ){
2098 pPg->pNextFree->pPrevFree = pPg->pPrevFree;
2099 }else{
2100 assert( pPager->pLast==pPg );
2101 pPager->pLast = pPg->pPrevFree;
2102 }
2103 pPg->pNextFree = pPg->pPrevFree = 0;
2104
2105 /* Unlink from the pgno hash table */
2106 unlinkHashChain(pPager, pPg);
2107}
2108
2109/*
2110** This routine is used to truncate the cache when a database
2111** is truncated. Drop from the cache all pages whose pgno is
2112** larger than pPager->dbSize and is unreferenced.
2113**
2114** Referenced pages larger than pPager->dbSize are zeroed.
2115**
2116** Actually, at the point this routine is called, it would be
2117** an error to have a referenced page. But rather than delete
2118** that page and guarantee a subsequent segfault, it seems better
2119** to zero it and hope that we error out sanely.
2120*/
2121static void pager_truncate_cache(Pager *pPager){
2122 PgHdr *pPg;
2123 PgHdr **ppPg;
2124 int dbSize = pPager->dbSize;
2125
2126 ppPg = &pPager->pAll;
2127 while( (pPg = *ppPg)!=0 ){
2128 if( pPg->pgno<=dbSize ){
2129 ppPg = &pPg->pNextAll;
2130 }else if( pPg->nRef>0 ){
2131 memset(PGHDR_TO_DATA(pPg), 0, pPager->pageSize);
2132 ppPg = &pPg->pNextAll;
2133 }else{
2134 *ppPg = pPg->pNextAll;
2135 IOTRACE(("PGFREE %p %d\n", pPager, pPg->pgno));
2136 PAGER_INCR(sqlite3_pager_pgfree_count);
2137 unlinkPage(pPg);
2138 makeClean(pPg);
2139 sqliteFree(pPg);
2140 pPager->nPage--;
2141 }
2142 }
2143}
2144
2145/*
2146** Try to obtain a lock on a file. Invoke the busy callback if the lock
2147** is currently not available. Repeat until the busy callback returns
2148** false or until the lock succeeds.
2149**
2150** Return SQLITE_OK on success and an error code if we cannot obtain
2151** the lock.
2152*/
2153static int pager_wait_on_lock(Pager *pPager, int locktype){
2154 int rc;
2155
2156 /* The OS lock values must be the same as the Pager lock values */
2157 assert( PAGER_SHARED==SHARED_LOCK );
2158 assert( PAGER_RESERVED==RESERVED_LOCK );
2159 assert( PAGER_EXCLUSIVE==EXCLUSIVE_LOCK );
2160
2161 /* If the file is currently unlocked then the size must be unknown */
2162 assert( pPager->state>=PAGER_SHARED || pPager->dbSize<0 || MEMDB );
2163
2164 if( pPager->state>=locktype ){
2165 rc = SQLITE_OK;
2166 }else{
2167 do {
2168 rc = sqlite3OsLock(pPager->fd, locktype);
2169 }while( rc==SQLITE_BUSY && sqlite3InvokeBusyHandler(pPager->pBusyHandler) );
2170 if( rc==SQLITE_OK ){
2171 pPager->state = locktype;
2172 IOTRACE(("LOCK %p %d\n", pPager, locktype))
2173 }
2174 }
2175 return rc;
2176}
2177
2178/*
2179** Truncate the file to the number of pages specified.
2180*/
2181int sqlite3PagerTruncate(Pager *pPager, Pgno nPage){
2182 int rc;
2183 assert( pPager->state>=PAGER_SHARED || MEMDB );
2184 sqlite3PagerPagecount(pPager);
2185 if( pPager->errCode ){
2186 rc = pPager->errCode;
2187 return rc;
2188 }
2189 if( nPage>=(unsigned)pPager->dbSize ){
2190 return SQLITE_OK;
2191 }
2192 if( MEMDB ){
2193 pPager->dbSize = nPage;
2194 pager_truncate_cache(pPager);
2195 return SQLITE_OK;
2196 }
2197 rc = syncJournal(pPager);
2198 if( rc!=SQLITE_OK ){
2199 return rc;
2200 }
2201
2202 /* Get an exclusive lock on the database before truncating. */
2203 rc = pager_wait_on_lock(pPager, EXCLUSIVE_LOCK);
2204 if( rc!=SQLITE_OK ){
2205 return rc;
2206 }
2207
2208 rc = pager_truncate(pPager, nPage);
2209 return rc;
2210}
2211
2212/*
2213** Shutdown the page cache. Free all memory and close all files.
2214**
2215** If a transaction was in progress when this routine is called, that
2216** transaction is rolled back. All outstanding pages are invalidated
2217** and their memory is freed. Any attempt to use a page associated
2218** with this page cache after this function returns will likely
2219** result in a coredump.
2220**
2221** This function always succeeds. If a transaction is active an attempt
2222** is made to roll it back. If an error occurs during the rollback
2223** a hot journal may be left in the filesystem but no error is returned
2224** to the caller.
2225*/
2226int sqlite3PagerClose(Pager *pPager){
2227#ifdef SQLITE_ENABLE_MEMORY_MANAGEMENT
2228 /* A malloc() cannot fail in sqlite3ThreadData() as one or more calls to
2229 ** malloc() must have already been made by this thread before it gets
2230 ** to this point. This means the ThreadData must have been allocated already
2231 ** so that ThreadData.nAlloc can be set.
2232 */
2233 ThreadData *pTsd = sqlite3ThreadData();
2234 assert( pPager );
2235 assert( pTsd && pTsd->nAlloc );
2236#endif
2237
2238 disable_simulated_io_errors();
2239 pPager->errCode = 0;
2240 pPager->exclusiveMode = 0;
2241 pager_reset(pPager);
2242 pagerUnlockAndRollback(pPager);
2243 enable_simulated_io_errors();
2244 PAGERTRACE2("CLOSE %d\n", PAGERID(pPager));
2245 IOTRACE(("CLOSE %p\n", pPager))
2246 assert( pPager->errCode || (pPager->journalOpen==0 && pPager->stmtOpen==0) );
2247 if( pPager->journalOpen ){
2248 sqlite3OsClose(&pPager->jfd);
2249 }
2250 sqliteFree(pPager->aInJournal);
2251 if( pPager->stmtOpen ){
2252 sqlite3OsClose(&pPager->stfd);
2253 }
2254 sqlite3OsClose(&pPager->fd);
2255 /* Temp files are automatically deleted by the OS
2256 ** if( pPager->tempFile ){
2257 ** sqlite3OsDelete(pPager->zFilename);
2258 ** }
2259 */
2260
2261#ifdef SQLITE_ENABLE_MEMORY_MANAGEMENT
2262 /* Remove the pager from the linked list of pagers starting at
2263 ** ThreadData.pPager if memory-management is enabled.
2264 */
2265 if( pPager==pTsd->pPager ){
2266 pTsd->pPager = pPager->pNext;
2267 }else{
2268 Pager *pTmp;
2269 for(pTmp = pTsd->pPager; pTmp->pNext!=pPager; pTmp=pTmp->pNext){}
2270 pTmp->pNext = pPager->pNext;
2271 }
2272#endif
2273 sqliteFree(pPager->aHash);
2274 sqliteFree(pPager->pTmpSpace);
2275 sqliteFree(pPager);
2276 return SQLITE_OK;
2277}
2278
2279#if !defined(NDEBUG) || defined(SQLITE_TEST)
2280/*
2281** Return the page number for the given page data.
2282*/
2283Pgno sqlite3PagerPagenumber(DbPage *p){
2284 return p->pgno;
2285}
2286#endif
2287
2288/*
2289** The page_ref() function increments the reference count for a page.
2290** If the page is currently on the freelist (the reference count is zero) then
2291** remove it from the freelist.
2292**
2293** For non-test systems, page_ref() is a macro that calls _page_ref()
2294** online of the reference count is zero. For test systems, page_ref()
2295** is a real function so that we can set breakpoints and trace it.
2296*/
2297static void _page_ref(PgHdr *pPg){
2298 if( pPg->nRef==0 ){
2299 /* The page is currently on the freelist. Remove it. */
2300 if( pPg==pPg->pPager->pFirstSynced ){
2301 PgHdr *p = pPg->pNextFree;
2302 while( p && p->needSync ){ p = p->pNextFree; }
2303 pPg->pPager->pFirstSynced = p;
2304 }
2305 if( pPg->pPrevFree ){
2306 pPg->pPrevFree->pNextFree = pPg->pNextFree;
2307 }else{
2308 pPg->pPager->pFirst = pPg->pNextFree;
2309 }
2310 if( pPg->pNextFree ){
2311 pPg->pNextFree->pPrevFree = pPg->pPrevFree;
2312 }else{
2313 pPg->pPager->pLast = pPg->pPrevFree;
2314 }
2315 pPg->pPager->nRef++;
2316 }
2317 pPg->nRef++;
2318 REFINFO(pPg);
2319}
2320#ifdef SQLITE_DEBUG
2321 static void page_ref(PgHdr *pPg){
2322 if( pPg->nRef==0 ){
2323 _page_ref(pPg);
2324 }else{
2325 pPg->nRef++;
2326 REFINFO(pPg);
2327 }
2328 }
2329#else
2330# define page_ref(P) ((P)->nRef==0?_page_ref(P):(void)(P)->nRef++)
2331#endif
2332
2333/*
2334** Increment the reference count for a page. The input pointer is
2335** a reference to the page data.
2336*/
2337int sqlite3PagerRef(DbPage *pPg){
2338 page_ref(pPg);
2339 return SQLITE_OK;
2340}
2341
2342/*
2343** Sync the journal. In other words, make sure all the pages that have
2344** been written to the journal have actually reached the surface of the
2345** disk. It is not safe to modify the original database file until after
2346** the journal has been synced. If the original database is modified before
2347** the journal is synced and a power failure occurs, the unsynced journal
2348** data would be lost and we would be unable to completely rollback the
2349** database changes. Database corruption would occur.
2350**
2351** This routine also updates the nRec field in the header of the journal.
2352** (See comments on the pager_playback() routine for additional information.)
2353** If the sync mode is FULL, two syncs will occur. First the whole journal
2354** is synced, then the nRec field is updated, then a second sync occurs.
2355**
2356** For temporary databases, we do not care if we are able to rollback
2357** after a power failure, so sync occurs.
2358**
2359** This routine clears the needSync field of every page current held in
2360** memory.
2361*/
2362static int syncJournal(Pager *pPager){
2363 PgHdr *pPg;
2364 int rc = SQLITE_OK;
2365
2366 /* Sync the journal before modifying the main database
2367 ** (assuming there is a journal and it needs to be synced.)
2368 */
2369 if( pPager->needSync ){
2370 if( !pPager->tempFile ){
2371 assert( pPager->journalOpen );
2372 /* assert( !pPager->noSync ); // noSync might be set if synchronous
2373 ** was turned off after the transaction was started. Ticket #615 */
2374#ifndef NDEBUG
2375 {
2376 /* Make sure the pPager->nRec counter we are keeping agrees
2377 ** with the nRec computed from the size of the journal file.
2378 */
2379 i64 jSz;
2380 rc = sqlite3OsFileSize(pPager->jfd, &jSz);
2381 if( rc!=0 ) return rc;
2382 assert( pPager->journalOff==jSz );
2383 }
2384#endif
2385 {
2386 /* Write the nRec value into the journal file header. If in
2387 ** full-synchronous mode, sync the journal first. This ensures that
2388 ** all data has really hit the disk before nRec is updated to mark
2389 ** it as a candidate for rollback.
2390 */
2391 if( pPager->fullSync ){
2392 PAGERTRACE2("SYNC journal of %d\n", PAGERID(pPager));
2393 IOTRACE(("JSYNC %p\n", pPager))
2394 rc = sqlite3OsSync(pPager->jfd, 0);
2395 if( rc!=0 ) return rc;
2396 }
2397 rc = sqlite3OsSeek(pPager->jfd,
2398 pPager->journalHdr + sizeof(aJournalMagic));
2399 if( rc ) return rc;
2400 IOTRACE(("JHDR %p %lld %d\n", pPager,
2401 pPager->journalHdr + sizeof(aJournalMagic), 4))
2402 rc = write32bits(pPager->jfd, pPager->nRec);
2403 if( rc ) return rc;
2404
2405 rc = sqlite3OsSeek(pPager->jfd, pPager->journalOff);
2406 if( rc ) return rc;
2407 }
2408 PAGERTRACE2("SYNC journal of %d\n", PAGERID(pPager));
2409 IOTRACE(("JSYNC %p\n", pPager))
2410 rc = sqlite3OsSync(pPager->jfd, pPager->full_fsync);
2411 if( rc!=0 ) return rc;
2412 pPager->journalStarted = 1;
2413 }
2414 pPager->needSync = 0;
2415
2416 /* Erase the needSync flag from every page.
2417 */
2418 for(pPg=pPager->pAll; pPg; pPg=pPg->pNextAll){
2419 pPg->needSync = 0;
2420 }
2421 pPager->pFirstSynced = pPager->pFirst;
2422 }
2423
2424#ifndef NDEBUG
2425 /* If the Pager.needSync flag is clear then the PgHdr.needSync
2426 ** flag must also be clear for all pages. Verify that this
2427 ** invariant is true.
2428 */
2429 else{
2430 for(pPg=pPager->pAll; pPg; pPg=pPg->pNextAll){
2431 assert( pPg->needSync==0 );
2432 }
2433 assert( pPager->pFirstSynced==pPager->pFirst );
2434 }
2435#endif
2436
2437 return rc;
2438}
2439
2440/*
2441** Merge two lists of pages connected by pDirty and in pgno order.
2442** Do not both fixing the pPrevDirty pointers.
2443*/
2444static PgHdr *merge_pagelist(PgHdr *pA, PgHdr *pB){
2445 PgHdr result, *pTail;
2446 pTail = &result;
2447 while( pA && pB ){
2448 if( pA->pgno<pB->pgno ){
2449 pTail->pDirty = pA;
2450 pTail = pA;
2451 pA = pA->pDirty;
2452 }else{
2453 pTail->pDirty = pB;
2454 pTail = pB;
2455 pB = pB->pDirty;
2456 }
2457 }
2458 if( pA ){
2459 pTail->pDirty = pA;
2460 }else if( pB ){
2461 pTail->pDirty = pB;
2462 }else{
2463 pTail->pDirty = 0;
2464 }
2465 return result.pDirty;
2466}
2467
2468/*
2469** Sort the list of pages in accending order by pgno. Pages are
2470** connected by pDirty pointers. The pPrevDirty pointers are
2471** corrupted by this sort.
2472*/
2473#define N_SORT_BUCKET_ALLOC 25
2474#define N_SORT_BUCKET 25
2475#ifdef SQLITE_TEST
2476 int sqlite3_pager_n_sort_bucket = 0;
2477 #undef N_SORT_BUCKET
2478 #define N_SORT_BUCKET \
2479 (sqlite3_pager_n_sort_bucket?sqlite3_pager_n_sort_bucket:N_SORT_BUCKET_ALLOC)
2480#endif
2481static PgHdr *sort_pagelist(PgHdr *pIn){
2482 PgHdr *a[N_SORT_BUCKET_ALLOC], *p;
2483 int i;
2484 memset(a, 0, sizeof(a));
2485 while( pIn ){
2486 p = pIn;
2487 pIn = p->pDirty;
2488 p->pDirty = 0;
2489 for(i=0; i<N_SORT_BUCKET-1; i++){
2490 if( a[i]==0 ){
2491 a[i] = p;
2492 break;
2493 }else{
2494 p = merge_pagelist(a[i], p);
2495 a[i] = 0;
2496 }
2497 }
2498 if( i==N_SORT_BUCKET-1 ){
2499 /* Coverage: To get here, there need to be 2^(N_SORT_BUCKET)
2500 ** elements in the input list. This is possible, but impractical.
2501 ** Testing this line is the point of global variable
2502 ** sqlite3_pager_n_sort_bucket.
2503 */
2504 a[i] = merge_pagelist(a[i], p);
2505 }
2506 }
2507 p = a[0];
2508 for(i=1; i<N_SORT_BUCKET; i++){
2509 p = merge_pagelist(p, a[i]);
2510 }
2511 return p;
2512}
2513
2514/*
2515** Given a list of pages (connected by the PgHdr.pDirty pointer) write
2516** every one of those pages out to the database file and mark them all
2517** as clean.
2518*/
2519static int pager_write_pagelist(PgHdr *pList){
2520 Pager *pPager;
2521 int rc;
2522
2523 if( pList==0 ) return SQLITE_OK;
2524 pPager = pList->pPager;
2525
2526 /* At this point there may be either a RESERVED or EXCLUSIVE lock on the
2527 ** database file. If there is already an EXCLUSIVE lock, the following
2528 ** calls to sqlite3OsLock() are no-ops.
2529 **
2530 ** Moving the lock from RESERVED to EXCLUSIVE actually involves going
2531 ** through an intermediate state PENDING. A PENDING lock prevents new
2532 ** readers from attaching to the database but is unsufficient for us to
2533 ** write. The idea of a PENDING lock is to prevent new readers from
2534 ** coming in while we wait for existing readers to clear.
2535 **
2536 ** While the pager is in the RESERVED state, the original database file
2537 ** is unchanged and we can rollback without having to playback the
2538 ** journal into the original database file. Once we transition to
2539 ** EXCLUSIVE, it means the database file has been changed and any rollback
2540 ** will require a journal playback.
2541 */
2542 rc = pager_wait_on_lock(pPager, EXCLUSIVE_LOCK);
2543 if( rc!=SQLITE_OK ){
2544 return rc;
2545 }
2546
2547 pList = sort_pagelist(pList);
2548 while( pList ){
2549 assert( pList->dirty );
2550 rc = sqlite3OsSeek(pPager->fd, (pList->pgno-1)*(i64)pPager->pageSize);
2551 if( rc ) return rc;
2552 /* If there are dirty pages in the page cache with page numbers greater
2553 ** than Pager.dbSize, this means sqlite3PagerTruncate() was called to
2554 ** make the file smaller (presumably by auto-vacuum code). Do not write
2555 ** any such pages to the file.
2556 */
2557 if( pList->pgno<=pPager->dbSize ){
2558 char *pData = CODEC2(pPager, PGHDR_TO_DATA(pList), pList->pgno, 6);
2559 PAGERTRACE4("STORE %d page %d hash(%08x)\n",
2560 PAGERID(pPager), pList->pgno, pager_pagehash(pList));
2561 IOTRACE(("PGOUT %p %d\n", pPager, pList->pgno));
2562 rc = sqlite3OsWrite(pPager->fd, pData, pPager->pageSize);
2563 PAGER_INCR(sqlite3_pager_writedb_count);
2564 PAGER_INCR(pPager->nWrite);
2565 if( pList->pgno==1 ){
2566 memcpy(&pPager->dbFileVers, &pData[24], sizeof(pPager->dbFileVers));
2567 }
2568 }
2569#ifndef NDEBUG
2570 else{
2571 PAGERTRACE3("NOSTORE %d page %d\n", PAGERID(pPager), pList->pgno);
2572 }
2573#endif
2574 if( rc ) return rc;
2575 pList->dirty = 0;
2576#ifdef SQLITE_CHECK_PAGES
2577 pList->pageHash = pager_pagehash(pList);
2578#endif
2579 pList = pList->pDirty;
2580 }
2581 return SQLITE_OK;
2582}
2583
2584/*
2585** Collect every dirty page into a dirty list and
2586** return a pointer to the head of that list. All pages are
2587** collected even if they are still in use.
2588*/
2589static PgHdr *pager_get_all_dirty_pages(Pager *pPager){
2590 return pPager->pDirty;
2591}
2592
2593/*
2594** Return TRUE if there is a hot journal on the given pager.
2595** A hot journal is one that needs to be played back.
2596**
2597** If the current size of the database file is 0 but a journal file
2598** exists, that is probably an old journal left over from a prior
2599** database with the same name. Just delete the journal.
2600*/
2601static int hasHotJournal(Pager *pPager){
2602 if( !pPager->useJournal ) return 0;
2603 if( !sqlite3OsFileExists(pPager->zJournal) ){
2604 return 0;
2605 }
2606 if( sqlite3OsCheckReservedLock(pPager->fd) ){
2607 return 0;
2608 }
2609 if( sqlite3PagerPagecount(pPager)==0 ){
2610 sqlite3OsDelete(pPager->zJournal);
2611 return 0;
2612 }else{
2613 return 1;
2614 }
2615}
2616
2617/*
2618** Try to find a page in the cache that can be recycled.
2619**
2620** This routine may return SQLITE_IOERR, SQLITE_FULL or SQLITE_OK. It
2621** does not set the pPager->errCode variable.
2622*/
2623static int pager_recycle(Pager *pPager, int syncOk, PgHdr **ppPg){
2624 PgHdr *pPg;
2625 *ppPg = 0;
2626
2627 assert(!MEMDB);
2628
2629 /* Find a page to recycle. Try to locate a page that does not
2630 ** require us to do an fsync() on the journal.
2631 */
2632 pPg = pPager->pFirstSynced;
2633
2634 /* If we could not find a page that does not require an fsync()
2635 ** on the journal file then fsync the journal file. This is a
2636 ** very slow operation, so we work hard to avoid it. But sometimes
2637 ** it can't be helped.
2638 */
2639 if( pPg==0 && pPager->pFirst && syncOk && !MEMDB){
2640 int rc = syncJournal(pPager);
2641 if( rc!=0 ){
2642 return rc;
2643 }
2644 if( pPager->fullSync ){
2645 /* If in full-sync mode, write a new journal header into the
2646 ** journal file. This is done to avoid ever modifying a journal
2647 ** header that is involved in the rollback of pages that have
2648 ** already been written to the database (in case the header is
2649 ** trashed when the nRec field is updated).
2650 */
2651 pPager->nRec = 0;
2652 assert( pPager->journalOff > 0 );
2653 assert( pPager->doNotSync==0 );
2654 rc = writeJournalHdr(pPager);
2655 if( rc!=0 ){
2656 return rc;
2657 }
2658 }
2659 pPg = pPager->pFirst;
2660 }
2661 if( pPg==0 ){
2662 return SQLITE_OK;
2663 }
2664
2665 assert( pPg->nRef==0 );
2666
2667 /* Write the page to the database file if it is dirty.
2668 */
2669 if( pPg->dirty ){
2670 int rc;
2671 assert( pPg->needSync==0 );
2672 makeClean(pPg);
2673 pPg->dirty = 1;
2674 pPg->pDirty = 0;
2675 rc = pager_write_pagelist( pPg );
2676 if( rc!=SQLITE_OK ){
2677 return rc;
2678 }
2679 }
2680 assert( pPg->dirty==0 );
2681
2682 /* If the page we are recycling is marked as alwaysRollback, then
2683 ** set the global alwaysRollback flag, thus disabling the
2684 ** sqlite3PagerDontRollback() optimization for the rest of this transaction.
2685 ** It is necessary to do this because the page marked alwaysRollback
2686 ** might be reloaded at a later time but at that point we won't remember
2687 ** that is was marked alwaysRollback. This means that all pages must
2688 ** be marked as alwaysRollback from here on out.
2689 */
2690 if( pPg->alwaysRollback ){
2691 IOTRACE(("ALWAYS_ROLLBACK %p\n", pPager))
2692 pPager->alwaysRollback = 1;
2693 }
2694
2695 /* Unlink the old page from the free list and the hash table
2696 */
2697 unlinkPage(pPg);
2698 assert( pPg->pgno==0 );
2699
2700 *ppPg = pPg;
2701 return SQLITE_OK;
2702}
2703
2704/*
2705** This function is called to free superfluous dynamically allocated memory
2706** held by the pager system. Memory in use by any SQLite pager allocated
2707** by the current thread may be sqliteFree()ed.
2708**
2709** nReq is the number of bytes of memory required. Once this much has
2710** been released, the function returns. A negative value for nReq means
2711** free as much memory as possible. The return value is the total number
2712** of bytes of memory released.
2713*/
2714#if defined(SQLITE_ENABLE_MEMORY_MANAGEMENT) && !defined(SQLITE_OMIT_DISKIO)
2715int sqlite3PagerReleaseMemory(int nReq){
2716 const ThreadData *pTsdro = sqlite3ThreadDataReadOnly();
2717 int nReleased = 0;
2718 int i;
2719
2720 /* If the the global mutex is held, this subroutine becomes a
2721 ** o-op; zero bytes of memory are freed. This is because
2722 ** some of the code invoked by this function may also
2723 ** try to obtain the mutex, resulting in a deadlock.
2724 */
2725 if( sqlite3OsInMutex(0) ){
2726 return 0;
2727 }
2728
2729 /* Outermost loop runs for at most two iterations. First iteration we
2730 ** try to find memory that can be released without calling fsync(). Second
2731 ** iteration (which only runs if the first failed to free nReq bytes of
2732 ** memory) is permitted to call fsync(). This is of course much more
2733 ** expensive.
2734 */
2735 for(i=0; i<=1; i++){
2736
2737 /* Loop through all the SQLite pagers opened by the current thread. */
2738 Pager *pPager = pTsdro->pPager;
2739 for( ; pPager && (nReq<0 || nReleased<nReq); pPager=pPager->pNext){
2740 PgHdr *pPg;
2741 int rc;
2742
2743 if( MEMDB ){
2744 continue;
2745 }
2746
2747 /* For each pager, try to free as many pages as possible (without
2748 ** calling fsync() if this is the first iteration of the outermost
2749 ** loop).
2750 */
2751 while( SQLITE_OK==(rc = pager_recycle(pPager, i, &pPg)) && pPg) {
2752 /* We've found a page to free. At this point the page has been
2753 ** removed from the page hash-table, free-list and synced-list
2754 ** (pFirstSynced). It is still in the all pages (pAll) list.
2755 ** Remove it from this list before freeing.
2756 **
2757 ** Todo: Check the Pager.pStmt list to make sure this is Ok. It
2758 ** probably is though.
2759 */
2760 PgHdr *pTmp;
2761 assert( pPg );
2762 if( pPg==pPager->pAll ){
2763 pPager->pAll = pPg->pNextAll;
2764 }else{
2765 for( pTmp=pPager->pAll; pTmp->pNextAll!=pPg; pTmp=pTmp->pNextAll ){}
2766 pTmp->pNextAll = pPg->pNextAll;
2767 }
2768 nReleased += sqliteAllocSize(pPg);
2769 IOTRACE(("PGFREE %p %d\n", pPager, pPg->pgno));
2770 PAGER_INCR(sqlite3_pager_pgfree_count);
2771 sqliteFree(pPg);
2772 }
2773
2774 if( rc!=SQLITE_OK ){
2775 /* An error occured whilst writing to the database file or
2776 ** journal in pager_recycle(). The error is not returned to the
2777 ** caller of this function. Instead, set the Pager.errCode variable.
2778 ** The error will be returned to the user (or users, in the case
2779 ** of a shared pager cache) of the pager for which the error occured.
2780 */
2781 assert( (rc&0xff)==SQLITE_IOERR || rc==SQLITE_FULL );
2782 assert( pPager->state>=PAGER_RESERVED );
2783 pager_error(pPager, rc);
2784 }
2785 }
2786 }
2787
2788 return nReleased;
2789}
2790#endif /* SQLITE_ENABLE_MEMORY_MANAGEMENT && !SQLITE_OMIT_DISKIO */
2791
2792/*
2793** Read the content of page pPg out of the database file.
2794*/
2795static int readDbPage(Pager *pPager, PgHdr *pPg, Pgno pgno){
2796 int rc;
2797 assert( MEMDB==0 );
2798 rc = sqlite3OsSeek(pPager->fd, (pgno-1)*(i64)pPager->pageSize);
2799 if( rc==SQLITE_OK ){
2800 rc = sqlite3OsRead(pPager->fd, PGHDR_TO_DATA(pPg),
2801 pPager->pageSize);
2802 }
2803 PAGER_INCR(sqlite3_pager_readdb_count);
2804 PAGER_INCR(pPager->nRead);
2805 IOTRACE(("PGIN %p %d\n", pPager, pgno));
2806 if( pgno==1 ){
2807 memcpy(&pPager->dbFileVers, &((u8*)PGHDR_TO_DATA(pPg))[24],
2808 sizeof(pPager->dbFileVers));
2809 }
2810 CODEC1(pPager, PGHDR_TO_DATA(pPg), pPg->pgno, 3);
2811 PAGERTRACE4("FETCH %d page %d hash(%08x)\n",
2812 PAGERID(pPager), pPg->pgno, pager_pagehash(pPg));
2813 return rc;
2814}
2815
2816
2817/*
2818** This function is called to obtain the shared lock required before
2819** data may be read from the pager cache. If the shared lock has already
2820** been obtained, this function is a no-op.
2821**
2822** Immediately after obtaining the shared lock (if required), this function
2823** checks for a hot-journal file. If one is found, an emergency rollback
2824** is performed immediately.
2825*/
2826static int pagerSharedLock(Pager *pPager){
2827 int rc = SQLITE_OK;
2828
2829 if( pPager->state==PAGER_UNLOCK ){
2830 if( !MEMDB ){
2831 assert( pPager->nRef==0 );
2832 if( !pPager->noReadlock ){
2833 rc = pager_wait_on_lock(pPager, SHARED_LOCK);
2834 if( rc!=SQLITE_OK ){
2835 return pager_error(pPager, rc);
2836 }
2837 assert( pPager->state>=SHARED_LOCK );
2838 }
2839
2840 /* If a journal file exists, and there is no RESERVED lock on the
2841 ** database file, then it either needs to be played back or deleted.
2842 */
2843 if( hasHotJournal(pPager) ){
2844 /* Get an EXCLUSIVE lock on the database file. At this point it is
2845 ** important that a RESERVED lock is not obtained on the way to the
2846 ** EXCLUSIVE lock. If it were, another process might open the
2847 ** database file, detect the RESERVED lock, and conclude that the
2848 ** database is safe to read while this process is still rolling it
2849 ** back.
2850 **
2851 ** Because the intermediate RESERVED lock is not requested, the
2852 ** second process will get to this point in the code and fail to
2853 ** obtain it's own EXCLUSIVE lock on the database file.
2854 */
2855 rc = sqlite3OsLock(pPager->fd, EXCLUSIVE_LOCK);
2856 if( rc!=SQLITE_OK ){
2857 pager_unlock(pPager);
2858 return pager_error(pPager, rc);
2859 }
2860 pPager->state = PAGER_EXCLUSIVE;
2861
2862 /* Open the journal for reading only. Return SQLITE_BUSY if
2863 ** we are unable to open the journal file.
2864 **
2865 ** The journal file does not need to be locked itself. The
2866 ** journal file is never open unless the main database file holds
2867 ** a write lock, so there is never any chance of two or more
2868 ** processes opening the journal at the same time.
2869 **
2870** Open the journal for read/write access. This is because in
2871** exclusive-access mode the file descriptor will be kept open and
2872 ** possibly used for a transaction later on. On some systems, the
2873 ** OsTruncate() call used in exclusive-access mode also requires
2874 ** a read/write file handle.
2875 */
2876 rc = SQLITE_BUSY;
2877 if( sqlite3OsFileExists(pPager->zJournal) ){
2878 int ro;
2879 assert( !pPager->tempFile );
2880 rc = sqlite3OsOpenReadWrite(pPager->zJournal, &pPager->jfd, &ro);
2881 assert( rc!=SQLITE_OK || pPager->jfd );
2882 if( ro ){
2883 rc = SQLITE_BUSY;
2884 sqlite3OsClose(&pPager->jfd);
2885 }
2886 }
2887 if( rc!=SQLITE_OK ){
2888 pager_unlock(pPager);
2889 return SQLITE_BUSY;
2890 }
2891 pPager->journalOpen = 1;
2892 pPager->journalStarted = 0;
2893 pPager->journalOff = 0;
2894 pPager->setMaster = 0;
2895 pPager->journalHdr = 0;
2896
2897 /* Playback and delete the journal. Drop the database write
2898 ** lock and reacquire the read lock.
2899 */
2900 rc = pager_playback(pPager, 1);
2901 if( rc!=SQLITE_OK ){
2902 return pager_error(pPager, rc);
2903 }
2904 assert(pPager->state==PAGER_SHARED ||
2905 (pPager->exclusiveMode && pPager->state>PAGER_SHARED)
2906 );
2907 }
2908
2909 if( pPager->pAll ){
2910 /* The shared-lock has just been acquired on the database file
2911 ** and there are already pages in the cache (from a previous
2912 ** read or write transaction). Check to see if the database
2913 ** has been modified. If the database has changed, flush the
2914 ** cache.
2915 **
2916 ** Database changes is detected by looking at 15 bytes beginning
2917 ** at offset 24 into the file. The first 4 of these 16 bytes are
2918 ** a 32-bit counter that is incremented with each change. The
2919 ** other bytes change randomly with each file change when
2920 ** a codec is in use.
2921 **
2922 ** There is a vanishingly small chance that a change will not be
2923 ** detected. The chance of an undetected change is so small that
2924 ** it can be neglected.
2925 */
2926 char dbFileVers[sizeof(pPager->dbFileVers)];
2927 sqlite3PagerPagecount(pPager);
2928
2929 if( pPager->errCode ){
2930 return pPager->errCode;
2931 }
2932
2933 if( pPager->dbSize>0 ){
2934 IOTRACE(("CKVERS %p %d\n", pPager, sizeof(dbFileVers)));
2935 rc = sqlite3OsSeek(pPager->fd, 24);
2936 if( rc!=SQLITE_OK ){
2937 return rc;
2938 }
2939 rc = sqlite3OsRead(pPager->fd, &dbFileVers, sizeof(dbFileVers));
2940 if( rc!=SQLITE_OK ){
2941 return rc;
2942 }
2943 }else{
2944 memset(dbFileVers, 0, sizeof(dbFileVers));
2945 }
2946
2947 if( memcmp(pPager->dbFileVers, dbFileVers, sizeof(dbFileVers))!=0 ){
2948 pager_reset(pPager);
2949 }
2950 }
2951 }
2952 assert( pPager->exclusiveMode || pPager->state<=PAGER_SHARED );
2953 if( pPager->state==PAGER_UNLOCK ){
2954 pPager->state = PAGER_SHARED;
2955 }
2956 }
2957
2958 return rc;
2959}
2960
2961/*
2962** Allocate a PgHdr object. Either create a new one or reuse
2963** an existing one that is not otherwise in use.
2964**
2965** A new PgHdr structure is created if any of the following are
2966** true:
2967**
2968** (1) We have not exceeded our maximum allocated cache size
2969** as set by the "PRAGMA cache_size" command.
2970**
2971** (2) There are no unused PgHdr objects available at this time.
2972**
2973** (3) This is an in-memory database.
2974**
2975** (4) There are no PgHdr objects that do not require a journal
2976** file sync and a sync of the journal file is currently
2977** prohibited.
2978**
2979** Otherwise, reuse an existing PgHdr. In other words, reuse an
2980** existing PgHdr if all of the following are true:
2981**
2982** (1) We have reached or exceeded the maximum cache size
2983** allowed by "PRAGMA cache_size".
2984**
2985** (2) There is a PgHdr available with PgHdr->nRef==0
2986**
2987** (3) We are not in an in-memory database
2988**
2989** (4) Either there is an available PgHdr that does not need
2990** to be synced to disk or else disk syncing is currently
2991** allowed.
2992*/
2993static int pagerAllocatePage(Pager *pPager, PgHdr **ppPg){
2994 int rc = SQLITE_OK;
2995 PgHdr *pPg;
2996
2997 /* Create a new PgHdr if any of the four conditions defined
2998 ** above is met: */
2999 if( pPager->nPage<pPager->mxPage
3000 || pPager->pFirst==0
3001 || MEMDB
3002 || (pPager->pFirstSynced==0 && pPager->doNotSync)
3003 ){
3004 if( pPager->nPage>=pPager->nHash ){
3005 pager_resize_hash_table(pPager,
3006 pPager->nHash<256 ? 256 : pPager->nHash*2);
3007 if( pPager->nHash==0 ){
3008 rc = SQLITE_NOMEM;
3009 goto pager_allocate_out;
3010 }
3011 }
3012 pPg = sqliteMallocRaw( sizeof(*pPg) + pPager->pageSize
3013 + sizeof(u32) + pPager->nExtra
3014 + MEMDB*sizeof(PgHistory) );
3015 if( pPg==0 ){
3016 rc = SQLITE_NOMEM;
3017 goto pager_allocate_out;
3018 }
3019 memset(pPg, 0, sizeof(*pPg));
3020 if( MEMDB ){
3021 memset(PGHDR_TO_HIST(pPg, pPager), 0, sizeof(PgHistory));
3022 }
3023 pPg->pPager = pPager;
3024 pPg->pNextAll = pPager->pAll;
3025 pPager->pAll = pPg;
3026 pPager->nPage++;
3027 }else{
3028 /* Recycle an existing page with a zero ref-count. */
3029 rc = pager_recycle(pPager, 1, &pPg);
3030 if( rc==SQLITE_BUSY ){
3031 rc = SQLITE_IOERR_BLOCKED;
3032 }
3033 if( rc!=SQLITE_OK ){
3034 goto pager_allocate_out;
3035 }
3036 assert( pPager->state>=SHARED_LOCK );
3037 assert(pPg);
3038 }
3039 *ppPg = pPg;
3040
3041pager_allocate_out:
3042 return rc;
3043}
3044
3045/*
3046** Make sure we have the content for a page. If the page was
3047** previously acquired with noContent==1, then the content was
3048** just initialized to zeros instead of being read from disk.
3049** But now we need the real data off of disk. So make sure we
3050** have it. Read it in if we do not have it already.
3051*/
3052static int pager_get_content(PgHdr *pPg){
3053 if( pPg->needRead ){
3054 int rc = readDbPage(pPg->pPager, pPg, pPg->pgno);
3055 if( rc==SQLITE_OK ){
3056 pPg->needRead = 0;
3057 }else{
3058 return rc;
3059 }
3060 }
3061 return SQLITE_OK;
3062}
3063
3064/*
3065** Acquire a page.
3066**
3067** A read lock on the disk file is obtained when the first page is acquired.
3068** This read lock is dropped when the last page is released.
3069**
3070** This routine works for any page number greater than 0. If the database
3071** file is smaller than the requested page, then no actual disk
3072** read occurs and the memory image of the page is initialized to
3073** all zeros. The extra data appended to a page is always initialized
3074** to zeros the first time a page is loaded into memory.
3075**
3076** The acquisition might fail for several reasons. In all cases,
3077** an appropriate error code is returned and *ppPage is set to NULL.
3078**
3079** See also sqlite3PagerLookup(). Both this routine and Lookup() attempt
3080** to find a page in the in-memory cache first. If the page is not already
3081** in memory, this routine goes to disk to read it in whereas Lookup()
3082** just returns 0. This routine acquires a read-lock the first time it
3083** has to go to disk, and could also playback an old journal if necessary.
3084** Since Lookup() never goes to disk, it never has to deal with locks
3085** or journal files.
3086**
3087** If noContent is false, the page contents are actually read from disk.
3088** If noContent is true, it means that we do not care about the contents
3089** of the page at this time, so do not do a disk read. Just fill in the
3090** page content with zeros. But mark the fact that we have not read the
3091** content by setting the PgHdr.needRead flag. Later on, if
3092** sqlite3PagerWrite() is called on this page or if this routine is
3093** called again with noContent==0, that means that the content is needed
3094** and the disk read should occur at that point.
3095*/
3096int sqlite3PagerAcquire(
3097 Pager *pPager, /* The pager open on the database file */
3098 Pgno pgno, /* Page number to fetch */
3099 DbPage **ppPage, /* Write a pointer to the page here */
3100 int noContent /* Do not bother reading content from disk if true */
3101){
3102 PgHdr *pPg;
3103 int rc;
3104
3105 assert( pPager->state==PAGER_UNLOCK || pPager->nRef>0 || pgno==1 );
3106
3107 /* The maximum page number is 2^31. Return SQLITE_CORRUPT if a page
3108 ** number greater than this, or zero, is requested.
3109 */
3110 if( pgno>PAGER_MAX_PGNO || pgno==0 || pgno==PAGER_MJ_PGNO(pPager) ){
3111 return SQLITE_CORRUPT_BKPT;
3112 }
3113
3114 /* Make sure we have not hit any critical errors.
3115 */
3116 assert( pPager!=0 );
3117 *ppPage = 0;
3118 if( pPager->errCode && pPager->errCode!=SQLITE_FULL ){
3119 return pPager->errCode;
3120 }
3121
3122 /* If this is the first page accessed, then get a SHARED lock
3123 ** on the database file. pagerSharedLock() is a no-op if
3124 ** a database lock is already held.
3125 */
3126 rc = pagerSharedLock(pPager);
3127 if( rc!=SQLITE_OK ){
3128 return rc;
3129 }
3130 assert( pPager->state!=PAGER_UNLOCK );
3131
3132 pPg = pager_lookup(pPager, pgno);
3133 if( pPg==0 ){
3134 /* The requested page is not in the page cache. */
3135 int nMax;
3136 int h;
3137 PAGER_INCR(pPager->nMiss);
3138 rc = pagerAllocatePage(pPager, &pPg);
3139 if( rc!=SQLITE_OK ){
3140 return rc;
3141 }
3142
3143 pPg->pgno = pgno;
3144 assert( !MEMDB || pgno>pPager->stmtSize );
3145 if( pPager->aInJournal && (int)pgno<=pPager->origDbSize ){
3146 sqlite3CheckMemory(pPager->aInJournal, pgno/8);
3147 assert( pPager->journalOpen );
3148 pPg->inJournal = (pPager->aInJournal[pgno/8] & (1<<(pgno&7)))!=0;
3149 pPg->needSync = 0;
3150 }else{
3151 pPg->inJournal = 0;
3152 pPg->needSync = 0;
3153 }
3154
3155 makeClean(pPg);
3156 pPg->nRef = 1;
3157 REFINFO(pPg);
3158
3159 pPager->nRef++;
3160 if( pPager->nExtra>0 ){
3161 memset(PGHDR_TO_EXTRA(pPg, pPager), 0, pPager->nExtra);
3162 }
3163 nMax = sqlite3PagerPagecount(pPager);
3164 if( pPager->errCode ){
3165 sqlite3PagerUnref(pPg);
3166 rc = pPager->errCode;
3167 return rc;
3168 }
3169
3170 /* Populate the page with data, either by reading from the database
3171 ** file, or by setting the entire page to zero.
3172 */
3173 if( nMax<(int)pgno || MEMDB || (noContent && !pPager->alwaysRollback) ){
3174 if( pgno>pPager->mxPgno ){
3175 sqlite3PagerUnref(pPg);
3176 return SQLITE_FULL;
3177 }
3178 memset(PGHDR_TO_DATA(pPg), 0, pPager->pageSize);
3179 pPg->needRead = noContent && !pPager->alwaysRollback;
3180 IOTRACE(("ZERO %p %d\n", pPager, pgno));
3181 }else{
3182 rc = readDbPage(pPager, pPg, pgno);
3183 if( rc!=SQLITE_OK && rc!=SQLITE_IOERR_SHORT_READ ){
3184 pPg->pgno = 0;
3185 sqlite3PagerUnref(pPg);
3186 return rc;
3187 }
3188 pPg->needRead = 0;
3189 }
3190
3191 /* Link the page into the page hash table */
3192 h = pgno & (pPager->nHash-1);
3193 assert( pgno!=0 );
3194 pPg->pNextHash = pPager->aHash[h];
3195 pPager->aHash[h] = pPg;
3196 if( pPg->pNextHash ){
3197 assert( pPg->pNextHash->pPrevHash==0 );
3198 pPg->pNextHash->pPrevHash = pPg;
3199 }
3200
3201#ifdef SQLITE_CHECK_PAGES
3202 pPg->pageHash = pager_pagehash(pPg);
3203#endif
3204 }else{
3205 /* The requested page is in the page cache. */
3206 assert(pPager->nRef>0 || pgno==1);
3207 PAGER_INCR(pPager->nHit);
3208 if( !noContent ){
3209 rc = pager_get_content(pPg);
3210 if( rc ){
3211 return rc;
3212 }
3213 }
3214 page_ref(pPg);
3215 }
3216 *ppPage = pPg;
3217 return SQLITE_OK;
3218}
3219
3220/*
3221** Acquire a page if it is already in the in-memory cache. Do
3222** not read the page from disk. Return a pointer to the page,
3223** or 0 if the page is not in cache.
3224**
3225** See also sqlite3PagerGet(). The difference between this routine
3226** and sqlite3PagerGet() is that _get() will go to the disk and read
3227** in the page if the page is not already in cache. This routine
3228** returns NULL if the page is not in cache or if a disk I/O error
3229** has ever happened.
3230*/
3231DbPage *sqlite3PagerLookup(Pager *pPager, Pgno pgno){
3232 PgHdr *pPg;
3233
3234 assert( pPager!=0 );
3235 assert( pgno!=0 );
3236
3237 if( pPager->state==PAGER_UNLOCK ){
3238 assert( !pPager->pAll || pPager->exclusiveMode );
3239 return 0;
3240 }
3241 if( pPager->errCode && pPager->errCode!=SQLITE_FULL ){
3242 return 0;
3243 }
3244 pPg = pager_lookup(pPager, pgno);
3245 if( pPg==0 ) return 0;
3246 page_ref(pPg);
3247 return pPg;
3248}
3249
3250/*
3251** Release a page.
3252**
3253** If the number of references to the page drop to zero, then the
3254** page is added to the LRU list. When all references to all pages
3255** are released, a rollback occurs and the lock on the database is
3256** removed.
3257*/
3258int sqlite3PagerUnref(DbPage *pPg){
3259
3260 /* Decrement the reference count for this page
3261 */
3262 assert( pPg->nRef>0 );
3263 pPg->nRef--;
3264 REFINFO(pPg);
3265
3266 CHECK_PAGE(pPg);
3267
3268 /* When the number of references to a page reach 0, call the
3269 ** destructor and add the page to the freelist.
3270 */
3271 if( pPg->nRef==0 ){
3272 Pager *pPager;
3273 pPager = pPg->pPager;
3274 pPg->pNextFree = 0;
3275 pPg->pPrevFree = pPager->pLast;
3276 pPager->pLast = pPg;
3277 if( pPg->pPrevFree ){
3278 pPg->pPrevFree->pNextFree = pPg;
3279 }else{
3280 pPager->pFirst = pPg;
3281 }
3282 if( pPg->needSync==0 && pPager->pFirstSynced==0 ){
3283 pPager->pFirstSynced = pPg;
3284 }
3285 if( pPager->xDestructor ){
3286 pPager->xDestructor(pPg, pPager->pageSize);
3287 }
3288
3289 /* When all pages reach the freelist, drop the read lock from
3290 ** the database file.
3291 */
3292 pPager->nRef--;
3293 assert( pPager->nRef>=0 );
3294 if( pPager->nRef==0 && (!pPager->exclusiveMode || pPager->journalOff>0) ){
3295 pagerUnlockAndRollback(pPager);
3296 }
3297 }
3298 return SQLITE_OK;
3299}
3300
3301/*
3302** Create a journal file for pPager. There should already be a RESERVED
3303** or EXCLUSIVE lock on the database file when this routine is called.
3304**
3305** Return SQLITE_OK if everything. Return an error code and release the
3306** write lock if anything goes wrong.
3307*/
3308static int pager_open_journal(Pager *pPager){
3309 int rc;
3310 assert( !MEMDB );
3311 assert( pPager->state>=PAGER_RESERVED );
3312 assert( pPager->journalOpen==0 );
3313 assert( pPager->useJournal );
3314 assert( pPager->aInJournal==0 );
3315 sqlite3PagerPagecount(pPager);
3316 pPager->aInJournal = sqliteMalloc( pPager->dbSize/8 + 1 );
3317 if( pPager->aInJournal==0 ){
3318 rc = SQLITE_NOMEM;
3319 goto failed_to_open_journal;
3320 }
3321 rc = sqlite3OsOpenExclusive(pPager->zJournal, &pPager->jfd,
3322 pPager->tempFile);
3323 assert( rc!=SQLITE_OK || pPager->jfd );
3324 pPager->journalOff = 0;
3325 pPager->setMaster = 0;
3326 pPager->journalHdr = 0;
3327 if( rc!=SQLITE_OK ){
3328 if( rc==SQLITE_NOMEM ){
3329 sqlite3OsDelete(pPager->zJournal);
3330 }
3331 goto failed_to_open_journal;
3332 }
3333 sqlite3OsSetFullSync(pPager->jfd, pPager->full_fsync);
3334 sqlite3OsSetFullSync(pPager->fd, pPager->full_fsync);
3335 sqlite3OsOpenDirectory(pPager->jfd, pPager->zDirectory);
3336 pPager->journalOpen = 1;
3337 pPager->journalStarted = 0;
3338 pPager->needSync = 0;
3339 pPager->alwaysRollback = 0;
3340 pPager->nRec = 0;
3341 if( pPager->errCode ){
3342 rc = pPager->errCode;
3343 goto failed_to_open_journal;
3344 }
3345 pPager->origDbSize = pPager->dbSize;
3346
3347 rc = writeJournalHdr(pPager);
3348
3349 if( pPager->stmtAutoopen && rc==SQLITE_OK ){
3350 rc = sqlite3PagerStmtBegin(pPager);
3351 }
3352 if( rc!=SQLITE_OK && rc!=SQLITE_NOMEM ){
3353 rc = pager_end_transaction(pPager);
3354 if( rc==SQLITE_OK ){
3355 rc = SQLITE_FULL;
3356 }
3357 }
3358 return rc;
3359
3360failed_to_open_journal:
3361 sqliteFree(pPager->aInJournal);
3362 pPager->aInJournal = 0;
3363 return rc;
3364}
3365
3366/*
3367** Acquire a write-lock on the database. The lock is removed when
3368** the any of the following happen:
3369**
3370** * sqlite3PagerCommitPhaseTwo() is called.
3371** * sqlite3PagerRollback() is called.
3372** * sqlite3PagerClose() is called.
3373** * sqlite3PagerUnref() is called to on every outstanding page.
3374**
3375** The first parameter to this routine is a pointer to any open page of the
3376** database file. Nothing changes about the page - it is used merely to
3377** acquire a pointer to the Pager structure and as proof that there is
3378** already a read-lock on the database.
3379**
3380** The second parameter indicates how much space in bytes to reserve for a
3381** master journal file-name at the start of the journal when it is created.
3382**
3383** A journal file is opened if this is not a temporary file. For temporary
3384** files, the opening of the journal file is deferred until there is an
3385** actual need to write to the journal.
3386**
3387** If the database is already reserved for writing, this routine is a no-op.
3388**
3389** If exFlag is true, go ahead and get an EXCLUSIVE lock on the file
3390** immediately instead of waiting until we try to flush the cache. The
3391** exFlag is ignored if a transaction is already active.
3392*/
3393int sqlite3PagerBegin(DbPage *pPg, int exFlag){
3394 Pager *pPager = pPg->pPager;
3395 int rc = SQLITE_OK;
3396 assert( pPg->nRef>0 );
3397 assert( pPager->state!=PAGER_UNLOCK );
3398 if( pPager->state==PAGER_SHARED ){
3399 assert( pPager->aInJournal==0 );
3400 if( MEMDB ){
3401 pPager->state = PAGER_EXCLUSIVE;
3402 pPager->origDbSize = pPager->dbSize;
3403 }else{
3404 rc = sqlite3OsLock(pPager->fd, RESERVED_LOCK);
3405 if( rc==SQLITE_OK ){
3406 pPager->state = PAGER_RESERVED;
3407 if( exFlag ){
3408 rc = pager_wait_on_lock(pPager, EXCLUSIVE_LOCK);
3409 }
3410 }
3411 if( rc!=SQLITE_OK ){
3412 return rc;
3413 }
3414 pPager->dirtyCache = 0;
3415 PAGERTRACE2("TRANSACTION %d\n", PAGERID(pPager));
3416 if( pPager->useJournal && !pPager->tempFile ){
3417 rc = pager_open_journal(pPager);
3418 }
3419 }
3420 }else if( pPager->journalOpen && pPager->journalOff==0 ){
3421 /* This happens when the pager was in exclusive-access mode last
3422 ** time a (read or write) transaction was successfully concluded
3423 ** by this connection. Instead of deleting the journal file it was
3424 ** kept open and truncated to 0 bytes.
3425 */
3426 assert( pPager->nRec==0 );
3427 assert( pPager->origDbSize==0 );
3428 assert( pPager->aInJournal==0 );
3429 sqlite3PagerPagecount(pPager);
3430 pPager->aInJournal = sqliteMalloc( pPager->dbSize/8 + 1 );
3431 if( !pPager->aInJournal ){
3432 rc = SQLITE_NOMEM;
3433 }else{
3434 pPager->origDbSize = pPager->dbSize;
3435 rc = writeJournalHdr(pPager);
3436 }
3437 }
3438 assert( !pPager->journalOpen || pPager->journalOff>0 || rc!=SQLITE_OK );
3439 return rc;
3440}
3441
3442/*
3443** Make a page dirty. Set its dirty flag and add it to the dirty
3444** page list.
3445*/
3446static void makeDirty(PgHdr *pPg){
3447 if( pPg->dirty==0 ){
3448 Pager *pPager = pPg->pPager;
3449 pPg->dirty = 1;
3450 pPg->pDirty = pPager->pDirty;
3451 if( pPager->pDirty ){
3452 pPager->pDirty->pPrevDirty = pPg;
3453 }
3454 pPg->pPrevDirty = 0;
3455 pPager->pDirty = pPg;
3456 }
3457}
3458
3459/*
3460** Make a page clean. Clear its dirty bit and remove it from the
3461** dirty page list.
3462*/
3463static void makeClean(PgHdr *pPg){
3464 if( pPg->dirty ){
3465 pPg->dirty = 0;
3466 if( pPg->pDirty ){
3467 pPg->pDirty->pPrevDirty = pPg->pPrevDirty;
3468 }
3469 if( pPg->pPrevDirty ){
3470 pPg->pPrevDirty->pDirty = pPg->pDirty;
3471 }else{
3472 pPg->pPager->pDirty = pPg->pDirty;
3473 }
3474 }
3475}
3476
3477
3478/*
3479** Mark a data page as writeable. The page is written into the journal
3480** if it is not there already. This routine must be called before making
3481** changes to a page.
3482**
3483** The first time this routine is called, the pager creates a new
3484** journal and acquires a RESERVED lock on the database. If the RESERVED
3485** lock could not be acquired, this routine returns SQLITE_BUSY. The
3486** calling routine must check for that return value and be careful not to
3487** change any page data until this routine returns SQLITE_OK.
3488**
3489** If the journal file could not be written because the disk is full,
3490** then this routine returns SQLITE_FULL and does an immediate rollback.
3491** All subsequent write attempts also return SQLITE_FULL until there
3492** is a call to sqlite3PagerCommit() or sqlite3PagerRollback() to
3493** reset.
3494*/
3495static int pager_write(PgHdr *pPg){
3496 void *pData = PGHDR_TO_DATA(pPg);
3497 Pager *pPager = pPg->pPager;
3498 int rc = SQLITE_OK;
3499
3500 /* Check for errors
3501 */
3502 if( pPager->errCode ){
3503 return pPager->errCode;
3504 }
3505 if( pPager->readOnly ){
3506 return SQLITE_PERM;
3507 }
3508
3509 assert( !pPager->setMaster );
3510
3511 CHECK_PAGE(pPg);
3512
3513 /* If this page was previously acquired with noContent==1, that means
3514 ** we didn't really read in the content of the page. This can happen
3515 ** (for example) when the page is being moved to the freelist. But
3516 ** now we are (perhaps) moving the page off of the freelist for
3517 ** reuse and we need to know its original content so that content
3518 ** can be stored in the rollback journal. So do the read at this
3519 ** time.
3520 */
3521 rc = pager_get_content(pPg);
3522 if( rc ){
3523 return rc;
3524 }
3525
3526 /* Mark the page as dirty. If the page has already been written
3527 ** to the journal then we can return right away.
3528 */
3529 makeDirty(pPg);
3530 if( pPg->inJournal && (pageInStatement(pPg) || pPager->stmtInUse==0) ){
3531 pPager->dirtyCache = 1;
3532 }else{
3533
3534 /* If we get this far, it means that the page needs to be
3535 ** written to the transaction journal or the ckeckpoint journal
3536 ** or both.
3537 **
3538 ** First check to see that the transaction journal exists and
3539 ** create it if it does not.
3540 */
3541 assert( pPager->state!=PAGER_UNLOCK );
3542 rc = sqlite3PagerBegin(pPg, 0);
3543 if( rc!=SQLITE_OK ){
3544 return rc;
3545 }
3546 assert( pPager->state>=PAGER_RESERVED );
3547 if( !pPager->journalOpen && pPager->useJournal ){
3548 rc = pager_open_journal(pPager);
3549 if( rc!=SQLITE_OK ) return rc;
3550 }
3551 assert( pPager->journalOpen || !pPager->useJournal );
3552 pPager->dirtyCache = 1;
3553
3554 /* The transaction journal now exists and we have a RESERVED or an
3555 ** EXCLUSIVE lock on the main database file. Write the current page to
3556 ** the transaction journal if it is not there already.
3557 */
3558 if( !pPg->inJournal && (pPager->useJournal || MEMDB) ){
3559 if( (int)pPg->pgno <= pPager->origDbSize ){
3560 int szPg;
3561 if( MEMDB ){
3562 PgHistory *pHist = PGHDR_TO_HIST(pPg, pPager);
3563 PAGERTRACE3("JOURNAL %d page %d\n", PAGERID(pPager), pPg->pgno);
3564 assert( pHist->pOrig==0 );
3565 pHist->pOrig = sqliteMallocRaw( pPager->pageSize );
3566 if( pHist->pOrig ){
3567 memcpy(pHist->pOrig, PGHDR_TO_DATA(pPg), pPager->pageSize);
3568 }
3569 }else{
3570 u32 cksum, saved;
3571 char *pData2, *pEnd;
3572 /* We should never write to the journal file the page that
3573 ** contains the database locks. The following assert verifies
3574 ** that we do not. */
3575 assert( pPg->pgno!=PAGER_MJ_PGNO(pPager) );
3576 pData2 = CODEC2(pPager, pData, pPg->pgno, 7);
3577 cksum = pager_cksum(pPager, (u8*)pData2);
3578 pEnd = pData2 + pPager->pageSize;
3579 pData2 -= 4;
3580 saved = *(u32*)pEnd;
3581 put32bits(pEnd, cksum);
3582 szPg = pPager->pageSize+8;
3583 put32bits(pData2, pPg->pgno);
3584 rc = sqlite3OsWrite(pPager->jfd, pData2, szPg);
3585 IOTRACE(("JOUT %p %d %lld %d\n", pPager, pPg->pgno,
3586 pPager->journalOff, szPg));
3587 PAGER_INCR(sqlite3_pager_writej_count);
3588 pPager->journalOff += szPg;
3589 PAGERTRACE5("JOURNAL %d page %d needSync=%d hash(%08x)\n",
3590 PAGERID(pPager), pPg->pgno, pPg->needSync, pager_pagehash(pPg));
3591 *(u32*)pEnd = saved;
3592
3593 /* An error has occured writing to the journal file. The
3594 ** transaction will be rolled back by the layer above.
3595 */
3596 if( rc!=SQLITE_OK ){
3597 return rc;
3598 }
3599
3600 pPager->nRec++;
3601 assert( pPager->aInJournal!=0 );
3602 pPager->aInJournal[pPg->pgno/8] |= 1<<(pPg->pgno&7);
3603 pPg->needSync = !pPager->noSync;
3604 if( pPager->stmtInUse ){
3605 pPager->aInStmt[pPg->pgno/8] |= 1<<(pPg->pgno&7);
3606 }
3607 }
3608 }else{
3609 pPg->needSync = !pPager->journalStarted && !pPager->noSync;
3610 PAGERTRACE4("APPEND %d page %d needSync=%d\n",
3611 PAGERID(pPager), pPg->pgno, pPg->needSync);
3612 }
3613 if( pPg->needSync ){
3614 pPager->needSync = 1;
3615 }
3616 pPg->inJournal = 1;
3617 }
3618
3619 /* If the statement journal is open and the page is not in it,
3620 ** then write the current page to the statement journal. Note that
3621 ** the statement journal format differs from the standard journal format
3622 ** in that it omits the checksums and the header.
3623 */
3624 if( pPager->stmtInUse
3625 && !pageInStatement(pPg)
3626 && (int)pPg->pgno<=pPager->stmtSize
3627 ){
3628 assert( pPg->inJournal || (int)pPg->pgno>pPager->origDbSize );
3629 if( MEMDB ){
3630 PgHistory *pHist = PGHDR_TO_HIST(pPg, pPager);
3631 assert( pHist->pStmt==0 );
3632 pHist->pStmt = sqliteMallocRaw( pPager->pageSize );
3633 if( pHist->pStmt ){
3634 memcpy(pHist->pStmt, PGHDR_TO_DATA(pPg), pPager->pageSize);
3635 }
3636 PAGERTRACE3("STMT-JOURNAL %d page %d\n", PAGERID(pPager), pPg->pgno);
3637 page_add_to_stmt_list(pPg);
3638 }else{
3639 char *pData2 = CODEC2(pPager, pData, pPg->pgno, 7)-4;
3640 put32bits(pData2, pPg->pgno);
3641 rc = sqlite3OsWrite(pPager->stfd, pData2, pPager->pageSize+4);
3642 PAGERTRACE3("STMT-JOURNAL %d page %d\n", PAGERID(pPager), pPg->pgno);
3643 if( rc!=SQLITE_OK ){
3644 return rc;
3645 }
3646 pPager->stmtNRec++;
3647 assert( pPager->aInStmt!=0 );
3648 pPager->aInStmt[pPg->pgno/8] |= 1<<(pPg->pgno&7);
3649 }
3650 }
3651 }
3652
3653 /* Update the database size and return.
3654 */
3655 assert( pPager->state>=PAGER_SHARED );
3656 if( pPager->dbSize<(int)pPg->pgno ){
3657 pPager->dbSize = pPg->pgno;
3658 if( !MEMDB && pPager->dbSize==PENDING_BYTE/pPager->pageSize ){
3659 pPager->dbSize++;
3660 }
3661 }
3662 return rc;
3663}
3664
3665/*
3666** This function is used to mark a data-page as writable. It uses
3667** pager_write() to open a journal file (if it is not already open)
3668** and write the page *pData to the journal.
3669**
3670** The difference between this function and pager_write() is that this
3671** function also deals with the special case where 2 or more pages
3672** fit on a single disk sector. In this case all co-resident pages
3673** must have been written to the journal file before returning.
3674*/
3675int sqlite3PagerWrite(DbPage *pDbPage){
3676 int rc = SQLITE_OK;
3677
3678 PgHdr *pPg = pDbPage;
3679 Pager *pPager = pPg->pPager;
3680 Pgno nPagePerSector = (pPager->sectorSize/pPager->pageSize);
3681
3682 if( !MEMDB && nPagePerSector>1 ){
3683 Pgno nPageCount; /* Total number of pages in database file */
3684 Pgno pg1; /* First page of the sector pPg is located on. */
3685 int nPage; /* Number of pages starting at pg1 to journal */
3686 int ii;
3687
3688 /* Set the doNotSync flag to 1. This is because we cannot allow a journal
3689 ** header to be written between the pages journaled by this function.
3690 */
3691 assert( pPager->doNotSync==0 );
3692 pPager->doNotSync = 1;
3693
3694 /* This trick assumes that both the page-size and sector-size are
3695 ** an integer power of 2. It sets variable pg1 to the identifier
3696 ** of the first page of the sector pPg is located on.
3697 */
3698 pg1 = ((pPg->pgno-1) & ~(nPagePerSector-1)) + 1;
3699
3700 nPageCount = sqlite3PagerPagecount(pPager);
3701 if( pPg->pgno>nPageCount ){
3702 nPage = (pPg->pgno - pg1)+1;
3703 }else if( (pg1+nPagePerSector-1)>nPageCount ){
3704 nPage = nPageCount+1-pg1;
3705 }else{
3706 nPage = nPagePerSector;
3707 }
3708 assert(nPage>0);
3709 assert(pg1<=pPg->pgno);
3710 assert((pg1+nPage)>pPg->pgno);
3711
3712 for(ii=0; ii<nPage && rc==SQLITE_OK; ii++){
3713 Pgno pg = pg1+ii;
3714 if( !pPager->aInJournal || pg==pPg->pgno ||
3715 pg>pPager->origDbSize || !(pPager->aInJournal[pg/8]&(1<<(pg&7)))
3716 ) {
3717 if( pg!=PAGER_MJ_PGNO(pPager) ){
3718 PgHdr *pPage;
3719 rc = sqlite3PagerGet(pPager, pg, &pPage);
3720 if( rc==SQLITE_OK ){
3721 rc = pager_write(pPage);
3722 sqlite3PagerUnref(pPage);
3723 }
3724 }
3725 }
3726 }
3727
3728 assert( pPager->doNotSync==1 );
3729 pPager->doNotSync = 0;
3730 }else{
3731 rc = pager_write(pDbPage);
3732 }
3733 return rc;
3734}
3735
3736/*
3737** Return TRUE if the page given in the argument was previously passed
3738** to sqlite3PagerWrite(). In other words, return TRUE if it is ok
3739** to change the content of the page.
3740*/
3741#ifndef NDEBUG
3742int sqlite3PagerIswriteable(DbPage *pPg){
3743 return pPg->dirty;
3744}
3745#endif
3746
3747#ifndef SQLITE_OMIT_VACUUM
3748/*
3749** Replace the content of a single page with the information in the third
3750** argument.
3751*/
3752int sqlite3PagerOverwrite(Pager *pPager, Pgno pgno, void *pData){
3753 PgHdr *pPg;
3754 int rc;
3755
3756 rc = sqlite3PagerGet(pPager, pgno, &pPg);
3757 if( rc==SQLITE_OK ){
3758 rc = sqlite3PagerWrite(pPg);
3759 if( rc==SQLITE_OK ){
3760 memcpy(sqlite3PagerGetData(pPg), pData, pPager->pageSize);
3761 }
3762 sqlite3PagerUnref(pPg);
3763 }
3764 return rc;
3765}
3766#endif
3767
3768/*
3769** A call to this routine tells the pager that it is not necessary to
3770** write the information on page pPg back to the disk, even though
3771** that page might be marked as dirty.
3772**
3773** The overlying software layer calls this routine when all of the data
3774** on the given page is unused. The pager marks the page as clean so
3775** that it does not get written to disk.
3776**
3777** Tests show that this optimization, together with the
3778** sqlite3PagerDontRollback() below, more than double the speed
3779** of large INSERT operations and quadruple the speed of large DELETEs.
3780**
3781** When this routine is called, set the alwaysRollback flag to true.
3782** Subsequent calls to sqlite3PagerDontRollback() for the same page
3783** will thereafter be ignored. This is necessary to avoid a problem
3784** where a page with data is added to the freelist during one part of
3785** a transaction then removed from the freelist during a later part
3786** of the same transaction and reused for some other purpose. When it
3787** is first added to the freelist, this routine is called. When reused,
3788** the sqlite3PagerDontRollback() routine is called. But because the
3789** page contains critical data, we still need to be sure it gets
3790** rolled back in spite of the sqlite3PagerDontRollback() call.
3791*/
3792void sqlite3PagerDontWrite(DbPage *pDbPage){
3793 PgHdr *pPg = pDbPage;
3794 Pager *pPager = pPg->pPager;
3795
3796 if( MEMDB ) return;
3797 pPg->alwaysRollback = 1;
3798 if( pPg->dirty && !pPager->stmtInUse ){
3799 assert( pPager->state>=PAGER_SHARED );
3800 if( pPager->dbSize==(int)pPg->pgno && pPager->origDbSize<pPager->dbSize ){
3801 /* If this pages is the last page in the file and the file has grown
3802 ** during the current transaction, then do NOT mark the page as clean.
3803 ** When the database file grows, we must make sure that the last page
3804 ** gets written at least once so that the disk file will be the correct
3805 ** size. If you do not write this page and the size of the file
3806 ** on the disk ends up being too small, that can lead to database
3807 ** corruption during the next transaction.
3808 */
3809 }else{
3810 PAGERTRACE3("DONT_WRITE page %d of %d\n", pPg->pgno, PAGERID(pPager));
3811 IOTRACE(("CLEAN %p %d\n", pPager, pPg->pgno))
3812 makeClean(pPg);
3813#ifdef SQLITE_CHECK_PAGES
3814 pPg->pageHash = pager_pagehash(pPg);
3815#endif
3816 }
3817 }
3818}
3819
3820/*
3821** A call to this routine tells the pager that if a rollback occurs,
3822** it is not necessary to restore the data on the given page. This
3823** means that the pager does not have to record the given page in the
3824** rollback journal.
3825**
3826** If we have not yet actually read the content of this page (if
3827** the PgHdr.needRead flag is set) then this routine acts as a promise
3828** that we will never need to read the page content in the future.
3829** so the needRead flag can be cleared at this point.
3830*/
3831void sqlite3PagerDontRollback(DbPage *pPg){
3832 Pager *pPager = pPg->pPager;
3833
3834 assert( pPager->state>=PAGER_RESERVED );
3835 if( pPager->journalOpen==0 ) return;
3836 if( pPg->alwaysRollback || pPager->alwaysRollback || MEMDB ) return;
3837 if( !pPg->inJournal && (int)pPg->pgno <= pPager->origDbSize ){
3838 assert( pPager->aInJournal!=0 );
3839 pPager->aInJournal[pPg->pgno/8] |= 1<<(pPg->pgno&7);
3840 pPg->inJournal = 1;
3841 pPg->needRead = 0;
3842 if( pPager->stmtInUse ){
3843 pPager->aInStmt[pPg->pgno/8] |= 1<<(pPg->pgno&7);
3844 }
3845 PAGERTRACE3("DONT_ROLLBACK page %d of %d\n", pPg->pgno, PAGERID(pPager));
3846 IOTRACE(("GARBAGE %p %d\n", pPager, pPg->pgno))
3847 }
3848 if( pPager->stmtInUse
3849 && !pageInStatement(pPg)
3850 && (int)pPg->pgno<=pPager->stmtSize
3851 ){
3852 assert( pPg->inJournal || (int)pPg->pgno>pPager->origDbSize );
3853 assert( pPager->aInStmt!=0 );
3854 pPager->aInStmt[pPg->pgno/8] |= 1<<(pPg->pgno&7);
3855 }
3856}
3857
3858
3859/*
3860** This routine is called to increment the database file change-counter,
3861** stored at byte 24 of the pager file.
3862*/
3863static int pager_incr_changecounter(Pager *pPager){
3864 PgHdr *pPgHdr;
3865 u32 change_counter;
3866 int rc;
3867
3868 if( !pPager->changeCountDone ){
3869 /* Open page 1 of the file for writing. */
3870 rc = sqlite3PagerGet(pPager, 1, &pPgHdr);
3871 if( rc!=SQLITE_OK ) return rc;
3872 rc = sqlite3PagerWrite(pPgHdr);
3873 if( rc!=SQLITE_OK ) return rc;
3874
3875 /* Read the current value at byte 24. */
3876 change_counter = retrieve32bits(pPgHdr, 24);
3877
3878 /* Increment the value just read and write it back to byte 24. */
3879 change_counter++;
3880 put32bits(((char*)PGHDR_TO_DATA(pPgHdr))+24, change_counter);
3881
3882 /* Release the page reference. */
3883 sqlite3PagerUnref(pPgHdr);
3884 pPager->changeCountDone = 1;
3885 }
3886 return SQLITE_OK;
3887}
3888
3889/*
3890** Sync the database file for the pager pPager. zMaster points to the name
3891** of a master journal file that should be written into the individual
3892** journal file. zMaster may be NULL, which is interpreted as no master
3893** journal (a single database transaction).
3894**
3895** This routine ensures that the journal is synced, all dirty pages written
3896** to the database file and the database file synced. The only thing that
3897** remains to commit the transaction is to delete the journal file (or
3898** master journal file if specified).
3899**
3900** Note that if zMaster==NULL, this does not overwrite a previous value
3901** passed to an sqlite3PagerCommitPhaseOne() call.
3902**
3903** If parameter nTrunc is non-zero, then the pager file is truncated to
3904** nTrunc pages (this is used by auto-vacuum databases).
3905*/
3906int sqlite3PagerCommitPhaseOne(Pager *pPager, const char *zMaster, Pgno nTrunc){
3907 int rc = SQLITE_OK;
3908
3909 PAGERTRACE4("DATABASE SYNC: File=%s zMaster=%s nTrunc=%d\n",
3910 pPager->zFilename, zMaster, nTrunc);
3911
3912 /* If this is an in-memory db, or no pages have been written to, or this
3913 ** function has already been called, it is a no-op.
3914 */
3915 if( pPager->state!=PAGER_SYNCED && !MEMDB && pPager->dirtyCache ){
3916 PgHdr *pPg;
3917 assert( pPager->journalOpen );
3918
3919 /* If a master journal file name has already been written to the
3920 ** journal file, then no sync is required. This happens when it is
3921 ** written, then the process fails to upgrade from a RESERVED to an
3922 ** EXCLUSIVE lock. The next time the process tries to commit the
3923 ** transaction the m-j name will have already been written.
3924 */
3925 if( !pPager->setMaster ){
3926 rc = pager_incr_changecounter(pPager);
3927 if( rc!=SQLITE_OK ) goto sync_exit;
3928#ifndef SQLITE_OMIT_AUTOVACUUM
3929 if( nTrunc!=0 ){
3930 /* If this transaction has made the database smaller, then all pages
3931 ** being discarded by the truncation must be written to the journal
3932 ** file.
3933 */
3934 Pgno i;
3935 int iSkip = PAGER_MJ_PGNO(pPager);
3936 for( i=nTrunc+1; i<=pPager->origDbSize; i++ ){
3937 if( !(pPager->aInJournal[i/8] & (1<<(i&7))) && i!=iSkip ){
3938 rc = sqlite3PagerGet(pPager, i, &pPg);
3939 if( rc!=SQLITE_OK ) goto sync_exit;
3940 rc = sqlite3PagerWrite(pPg);
3941 sqlite3PagerUnref(pPg);
3942 if( rc!=SQLITE_OK ) goto sync_exit;
3943 }
3944 }
3945 }
3946#endif
3947 rc = writeMasterJournal(pPager, zMaster);
3948 if( rc!=SQLITE_OK ) goto sync_exit;
3949 rc = syncJournal(pPager);
3950 if( rc!=SQLITE_OK ) goto sync_exit;
3951 }
3952
3953#ifndef SQLITE_OMIT_AUTOVACUUM
3954 if( nTrunc!=0 ){
3955 rc = sqlite3PagerTruncate(pPager, nTrunc);
3956 if( rc!=SQLITE_OK ) goto sync_exit;
3957 }
3958#endif
3959
3960 /* Write all dirty pages to the database file */
3961 pPg = pager_get_all_dirty_pages(pPager);
3962 rc = pager_write_pagelist(pPg);
3963 if( rc!=SQLITE_OK ) goto sync_exit;
3964 pPager->pDirty = 0;
3965
3966 /* Sync the database file. */
3967 if( !pPager->noSync ){
3968 rc = sqlite3OsSync(pPager->fd, 0);
3969 }
3970 IOTRACE(("DBSYNC %p\n", pPager))
3971
3972 pPager->state = PAGER_SYNCED;
3973 }else if( MEMDB && nTrunc!=0 ){
3974 rc = sqlite3PagerTruncate(pPager, nTrunc);
3975 }
3976
3977sync_exit:
3978 if( rc==SQLITE_IOERR_BLOCKED ){
3979 /* pager_incr_changecounter() may attempt to obtain an exclusive
3980 * lock to spill the cache and return IOERR_BLOCKED. But since
3981 * there is no chance the cache is inconsistent, it's
3982 * better to return SQLITE_BUSY.
3983 */
3984 rc = SQLITE_BUSY;
3985 }
3986 return rc;
3987}
3988
3989
3990/*
3991** Commit all changes to the database and release the write lock.
3992**
3993** If the commit fails for any reason, a rollback attempt is made
3994** and an error code is returned. If the commit worked, SQLITE_OK
3995** is returned.
3996*/
3997int sqlite3PagerCommitPhaseTwo(Pager *pPager){
3998 int rc;
3999 PgHdr *pPg;
4000
4001 if( pPager->errCode ){
4002 return pPager->errCode;
4003 }
4004 if( pPager->state<PAGER_RESERVED ){
4005 return SQLITE_ERROR;
4006 }
4007 PAGERTRACE2("COMMIT %d\n", PAGERID(pPager));
4008 if( MEMDB ){
4009 pPg = pager_get_all_dirty_pages(pPager);
4010 while( pPg ){
4011 PgHistory *pHist = PGHDR_TO_HIST(pPg, pPager);
4012 clearHistory(pHist);
4013 pPg->dirty = 0;
4014 pPg->inJournal = 0;
4015 pHist->inStmt = 0;
4016 pPg->needSync = 0;
4017 pHist->pPrevStmt = pHist->pNextStmt = 0;
4018 pPg = pPg->pDirty;
4019 }
4020 pPager->pDirty = 0;
4021#ifndef NDEBUG
4022 for(pPg=pPager->pAll; pPg; pPg=pPg->pNextAll){
4023 PgHistory *pHist = PGHDR_TO_HIST(pPg, pPager);
4024 assert( !pPg->alwaysRollback );
4025 assert( !pHist->pOrig );
4026 assert( !pHist->pStmt );
4027 }
4028#endif
4029 pPager->pStmt = 0;
4030 pPager->state = PAGER_SHARED;
4031 return SQLITE_OK;
4032 }
4033 assert( pPager->journalOpen || !pPager->dirtyCache );
4034 assert( pPager->state==PAGER_SYNCED || !pPager->dirtyCache );
4035 rc = pager_end_transaction(pPager);
4036 return pager_error(pPager, rc);
4037}
4038
4039/*
4040** Rollback all changes. The database falls back to PAGER_SHARED mode.
4041** All in-memory cache pages revert to their original data contents.
4042** The journal is deleted.
4043**
4044** This routine cannot fail unless some other process is not following
4045** the correct locking protocol or unless some other
4046** process is writing trash into the journal file (SQLITE_CORRUPT) or
4047** unless a prior malloc() failed (SQLITE_NOMEM). Appropriate error
4048** codes are returned for all these occasions. Otherwise,
4049** SQLITE_OK is returned.
4050*/
4051int sqlite3PagerRollback(Pager *pPager){
4052 int rc;
4053 PAGERTRACE2("ROLLBACK %d\n", PAGERID(pPager));
4054 if( MEMDB ){
4055 PgHdr *p;
4056 for(p=pPager->pAll; p; p=p->pNextAll){
4057 PgHistory *pHist;
4058 assert( !p->alwaysRollback );
4059 if( !p->dirty ){
4060 assert( !((PgHistory *)PGHDR_TO_HIST(p, pPager))->pOrig );
4061 assert( !((PgHistory *)PGHDR_TO_HIST(p, pPager))->pStmt );
4062 continue;
4063 }
4064
4065 pHist = PGHDR_TO_HIST(p, pPager);
4066 if( pHist->pOrig ){
4067 memcpy(PGHDR_TO_DATA(p), pHist->pOrig, pPager->pageSize);
4068 PAGERTRACE3("ROLLBACK-PAGE %d of %d\n", p->pgno, PAGERID(pPager));
4069 }else{
4070 PAGERTRACE3("PAGE %d is clean on %d\n", p->pgno, PAGERID(pPager));
4071 }
4072 clearHistory(pHist);
4073 p->dirty = 0;
4074 p->inJournal = 0;
4075 pHist->inStmt = 0;
4076 pHist->pPrevStmt = pHist->pNextStmt = 0;
4077 if( pPager->xReiniter ){
4078 pPager->xReiniter(p, pPager->pageSize);
4079 }
4080 }
4081 pPager->pDirty = 0;
4082 pPager->pStmt = 0;
4083 pPager->dbSize = pPager->origDbSize;
4084 pager_truncate_cache(pPager);
4085 pPager->stmtInUse = 0;
4086 pPager->state = PAGER_SHARED;
4087 return SQLITE_OK;
4088 }
4089
4090 if( !pPager->dirtyCache || !pPager->journalOpen ){
4091 rc = pager_end_transaction(pPager);
4092 return rc;
4093 }
4094
4095 if( pPager->errCode && pPager->errCode!=SQLITE_FULL ){
4096 if( pPager->state>=PAGER_EXCLUSIVE ){
4097 pager_playback(pPager, 0);
4098 }
4099 return pPager->errCode;
4100 }
4101 if( pPager->state==PAGER_RESERVED ){
4102 int rc2;
4103 rc = pager_playback(pPager, 0);
4104 rc2 = pager_end_transaction(pPager);
4105 if( rc==SQLITE_OK ){
4106 rc = rc2;
4107 }
4108 }else{
4109 rc = pager_playback(pPager, 0);
4110 }
4111 /* pager_reset(pPager); */
4112 pPager->dbSize = -1;
4113
4114 /* If an error occurs during a ROLLBACK, we can no longer trust the pager
4115 ** cache. So call pager_error() on the way out to make any error
4116 ** persistent.
4117 */
4118 return pager_error(pPager, rc);
4119}
4120
4121/*
4122** Return TRUE if the database file is opened read-only. Return FALSE
4123** if the database is (in theory) writable.
4124*/
4125int sqlite3PagerIsreadonly(Pager *pPager){
4126 return pPager->readOnly;
4127}
4128
4129/*
4130** Return the number of references to the pager.
4131*/
4132int sqlite3PagerRefcount(Pager *pPager){
4133 return pPager->nRef;
4134}
4135
4136#ifdef SQLITE_TEST
4137/*
4138** This routine is used for testing and analysis only.
4139*/
4140int *sqlite3PagerStats(Pager *pPager){
4141 static int a[11];
4142 a[0] = pPager->nRef;
4143 a[1] = pPager->nPage;
4144 a[2] = pPager->mxPage;
4145 a[3] = pPager->dbSize;
4146 a[4] = pPager->state;
4147 a[5] = pPager->errCode;
4148 a[6] = pPager->nHit;
4149 a[7] = pPager->nMiss;
4150 a[8] = 0; /* Used to be pPager->nOvfl */
4151 a[9] = pPager->nRead;
4152 a[10] = pPager->nWrite;
4153 return a;
4154}
4155#endif
4156
4157/*
4158** Set the statement rollback point.
4159**
4160** This routine should be called with the transaction journal already
4161** open. A new statement journal is created that can be used to rollback
4162** changes of a single SQL command within a larger transaction.
4163*/
4164int sqlite3PagerStmtBegin(Pager *pPager){
4165 int rc;
4166 assert( !pPager->stmtInUse );
4167 assert( pPager->state>=PAGER_SHARED );
4168 assert( pPager->dbSize>=0 );
4169 PAGERTRACE2("STMT-BEGIN %d\n", PAGERID(pPager));
4170 if( MEMDB ){
4171 pPager->stmtInUse = 1;
4172 pPager->stmtSize = pPager->dbSize;
4173 return SQLITE_OK;
4174 }
4175 if( !pPager->journalOpen ){
4176 pPager->stmtAutoopen = 1;
4177 return SQLITE_OK;
4178 }
4179 assert( pPager->journalOpen );
4180 pPager->aInStmt = sqliteMalloc( pPager->dbSize/8 + 1 );
4181 if( pPager->aInStmt==0 ){
4182 /* sqlite3OsLock(pPager->fd, SHARED_LOCK); */
4183 return SQLITE_NOMEM;
4184 }
4185#ifndef NDEBUG
4186 rc = sqlite3OsFileSize(pPager->jfd, &pPager->stmtJSize);
4187 if( rc ) goto stmt_begin_failed;
4188 assert( pPager->stmtJSize == pPager->journalOff );
4189#endif
4190 pPager->stmtJSize = pPager->journalOff;
4191 pPager->stmtSize = pPager->dbSize;
4192 pPager->stmtHdrOff = 0;
4193 pPager->stmtCksum = pPager->cksumInit;
4194 if( !pPager->stmtOpen ){
4195 rc = sqlite3PagerOpentemp(&pPager->stfd);
4196 if( rc ) goto stmt_begin_failed;
4197 pPager->stmtOpen = 1;
4198 pPager->stmtNRec = 0;
4199 }
4200 pPager->stmtInUse = 1;
4201 return SQLITE_OK;
4202
4203stmt_begin_failed:
4204 if( pPager->aInStmt ){
4205 sqliteFree(pPager->aInStmt);
4206 pPager->aInStmt = 0;
4207 }
4208 return rc;
4209}
4210
4211/*
4212** Commit a statement.
4213*/
4214int sqlite3PagerStmtCommit(Pager *pPager){
4215 if( pPager->stmtInUse ){
4216 PgHdr *pPg, *pNext;
4217 PAGERTRACE2("STMT-COMMIT %d\n", PAGERID(pPager));
4218 if( !MEMDB ){
4219 sqlite3OsSeek(pPager->stfd, 0);
4220 /* sqlite3OsTruncate(pPager->stfd, 0); */
4221 sqliteFree( pPager->aInStmt );
4222 pPager->aInStmt = 0;
4223 }else{
4224 for(pPg=pPager->pStmt; pPg; pPg=pNext){
4225 PgHistory *pHist = PGHDR_TO_HIST(pPg, pPager);
4226 pNext = pHist->pNextStmt;
4227 assert( pHist->inStmt );
4228 pHist->inStmt = 0;
4229 pHist->pPrevStmt = pHist->pNextStmt = 0;
4230 sqliteFree(pHist->pStmt);
4231 pHist->pStmt = 0;
4232 }
4233 }
4234 pPager->stmtNRec = 0;
4235 pPager->stmtInUse = 0;
4236 pPager->pStmt = 0;
4237 }
4238 pPager->stmtAutoopen = 0;
4239 return SQLITE_OK;
4240}
4241
4242/*
4243** Rollback a statement.
4244*/
4245int sqlite3PagerStmtRollback(Pager *pPager){
4246 int rc;
4247 if( pPager->stmtInUse ){
4248 PAGERTRACE2("STMT-ROLLBACK %d\n", PAGERID(pPager));
4249 if( MEMDB ){
4250 PgHdr *pPg;
4251 PgHistory *pHist;
4252 for(pPg=pPager->pStmt; pPg; pPg=pHist->pNextStmt){
4253 pHist = PGHDR_TO_HIST(pPg, pPager);
4254 if( pHist->pStmt ){
4255 memcpy(PGHDR_TO_DATA(pPg), pHist->pStmt, pPager->pageSize);
4256 sqliteFree(pHist->pStmt);
4257 pHist->pStmt = 0;
4258 }
4259 }
4260 pPager->dbSize = pPager->stmtSize;
4261 pager_truncate_cache(pPager);
4262 rc = SQLITE_OK;
4263 }else{
4264 rc = pager_stmt_playback(pPager);
4265 }
4266 sqlite3PagerStmtCommit(pPager);
4267 }else{
4268 rc = SQLITE_OK;
4269 }
4270 pPager->stmtAutoopen = 0;
4271 return rc;
4272}
4273
4274/*
4275** Return the full pathname of the database file.
4276*/
4277const char *sqlite3PagerFilename(Pager *pPager){
4278 return pPager->zFilename;
4279}
4280
4281/*
4282** Return the directory of the database file.
4283*/
4284const char *sqlite3PagerDirname(Pager *pPager){
4285 return pPager->zDirectory;
4286}
4287
4288/*
4289** Return the full pathname of the journal file.
4290*/
4291const char *sqlite3PagerJournalname(Pager *pPager){
4292 return pPager->zJournal;
4293}
4294
4295/*
4296** Return true if fsync() calls are disabled for this pager. Return FALSE
4297** if fsync()s are executed normally.
4298*/
4299int sqlite3PagerNosync(Pager *pPager){
4300 return pPager->noSync;
4301}
4302
4303#ifdef SQLITE_HAS_CODEC
4304/*
4305** Set the codec for this pager
4306*/
4307void sqlite3PagerSetCodec(
4308 Pager *pPager,
4309 void *(*xCodec)(void*,void*,Pgno,int),
4310 void *pCodecArg
4311){
4312 pPager->xCodec = xCodec;
4313 pPager->pCodecArg = pCodecArg;
4314}
4315#endif
4316
4317#ifndef SQLITE_OMIT_AUTOVACUUM
4318/*
4319** Move the page pPg to location pgno in the file.
4320**
4321** There must be no references to the page previously located at
4322** pgno (which we call pPgOld) though that page is allowed to be
4323** in cache. If the page previous located at pgno is not already
4324** in the rollback journal, it is not put there by by this routine.
4325**
4326** References to the page pPg remain valid. Updating any
4327** meta-data associated with pPg (i.e. data stored in the nExtra bytes
4328** allocated along with the page) is the responsibility of the caller.
4329**
4330** A transaction must be active when this routine is called. It used to be
4331** required that a statement transaction was not active, but this restriction
4332** has been removed (CREATE INDEX needs to move a page when a statement
4333** transaction is active).
4334*/
4335int sqlite3PagerMovepage(Pager *pPager, DbPage *pPg, Pgno pgno){
4336 PgHdr *pPgOld; /* The page being overwritten. */
4337 int h;
4338 Pgno needSyncPgno = 0;
4339
4340 assert( pPg->nRef>0 );
4341
4342 PAGERTRACE5("MOVE %d page %d (needSync=%d) moves to %d\n",
4343 PAGERID(pPager), pPg->pgno, pPg->needSync, pgno);
4344 IOTRACE(("MOVE %p %d %d\n", pPager, pPg->pgno, pgno))
4345
4346 pager_get_content(pPg);
4347 if( pPg->needSync ){
4348 needSyncPgno = pPg->pgno;
4349 assert( pPg->inJournal || (int)pgno>pPager->origDbSize );
4350 assert( pPg->dirty );
4351 assert( pPager->needSync );
4352 }
4353
4354 /* Unlink pPg from it's hash-chain */
4355 unlinkHashChain(pPager, pPg);
4356
4357 /* If the cache contains a page with page-number pgno, remove it
4358 ** from it's hash chain. Also, if the PgHdr.needSync was set for
4359 ** page pgno before the 'move' operation, it needs to be retained
4360 ** for the page moved there.
4361 */
4362 pPg->needSync = 0;
4363 pPgOld = pager_lookup(pPager, pgno);
4364 if( pPgOld ){
4365 assert( pPgOld->nRef==0 );
4366 unlinkHashChain(pPager, pPgOld);
4367 makeClean(pPgOld);
4368 pPg->needSync = pPgOld->needSync;
4369 }else{
4370 pPg->needSync = 0;
4371 }
4372 if( pPager->aInJournal && (int)pgno<=pPager->origDbSize ){
4373 pPg->inJournal = (pPager->aInJournal[pgno/8] & (1<<(pgno&7)))!=0;
4374 }else{
4375 pPg->inJournal = 0;
4376 assert( pPg->needSync==0 || (int)pgno>pPager->origDbSize );
4377 }
4378
4379 /* Change the page number for pPg and insert it into the new hash-chain. */
4380 assert( pgno!=0 );
4381 pPg->pgno = pgno;
4382 h = pgno & (pPager->nHash-1);
4383 if( pPager->aHash[h] ){
4384 assert( pPager->aHash[h]->pPrevHash==0 );
4385 pPager->aHash[h]->pPrevHash = pPg;
4386 }
4387 pPg->pNextHash = pPager->aHash[h];
4388 pPager->aHash[h] = pPg;
4389 pPg->pPrevHash = 0;
4390
4391 makeDirty(pPg);
4392 pPager->dirtyCache = 1;
4393
4394 if( needSyncPgno ){
4395 /* If needSyncPgno is non-zero, then the journal file needs to be
4396 ** sync()ed before any data is written to database file page needSyncPgno.
4397 ** Currently, no such page exists in the page-cache and the
4398 ** Pager.aInJournal bit has been set. This needs to be remedied by loading
4399 ** the page into the pager-cache and setting the PgHdr.needSync flag.
4400 **
4401 ** The sqlite3PagerGet() call may cause the journal to sync. So make
4402 ** sure the Pager.needSync flag is set too.
4403 */
4404 int rc;
4405 PgHdr *pPgHdr;
4406 assert( pPager->needSync );
4407 rc = sqlite3PagerGet(pPager, needSyncPgno, &pPgHdr);
4408 if( rc!=SQLITE_OK ) return rc;
4409 pPager->needSync = 1;
4410 pPgHdr->needSync = 1;
4411 pPgHdr->inJournal = 1;
4412 makeDirty(pPgHdr);
4413 sqlite3PagerUnref(pPgHdr);
4414 }
4415
4416 return SQLITE_OK;
4417}
4418#endif
4419
4420/*
4421** Return a pointer to the data for the specified page.
4422*/
4423void *sqlite3PagerGetData(DbPage *pPg){
4424 return PGHDR_TO_DATA(pPg);
4425}
4426
4427/*
4428** Return a pointer to the Pager.nExtra bytes of "extra" space
4429** allocated along with the specified page.
4430*/
4431void *sqlite3PagerGetExtra(DbPage *pPg){
4432 Pager *pPager = pPg->pPager;
4433 return (pPager?PGHDR_TO_EXTRA(pPg, pPager):0);
4434}
4435
4436/*
4437** Get/set the locking-mode for this pager. Parameter eMode must be one
4438** of PAGER_LOCKINGMODE_QUERY, PAGER_LOCKINGMODE_NORMAL or
4439** PAGER_LOCKINGMODE_EXCLUSIVE. If the parameter is not _QUERY, then
4440** the locking-mode is set to the value specified.
4441**
4442** The returned value is either PAGER_LOCKINGMODE_NORMAL or
4443** PAGER_LOCKINGMODE_EXCLUSIVE, indicating the current (possibly updated)
4444** locking-mode.
4445*/
4446int sqlite3PagerLockingMode(Pager *pPager, int eMode){
4447 assert( eMode==PAGER_LOCKINGMODE_QUERY
4448 || eMode==PAGER_LOCKINGMODE_NORMAL
4449 || eMode==PAGER_LOCKINGMODE_EXCLUSIVE );
4450 assert( PAGER_LOCKINGMODE_QUERY<0 );
4451 assert( PAGER_LOCKINGMODE_NORMAL>=0 && PAGER_LOCKINGMODE_EXCLUSIVE>=0 );
4452 if( eMode>=0 && !pPager->tempFile ){
4453 pPager->exclusiveMode = eMode;
4454 }
4455 return (int)pPager->exclusiveMode;
4456}
4457
4458#if defined(SQLITE_DEBUG) || defined(SQLITE_TEST)
4459/*
4460** Return the current state of the file lock for the given pager.
4461** The return value is one of NO_LOCK, SHARED_LOCK, RESERVED_LOCK,
4462** PENDING_LOCK, or EXCLUSIVE_LOCK.
4463*/
4464int sqlite3PagerLockstate(Pager *pPager){
4465 return sqlite3OsLockState(pPager->fd);
4466}
4467#endif
4468
4469#ifdef SQLITE_DEBUG
4470/*
4471** Print a listing of all referenced pages and their ref count.
4472*/
4473void sqlite3PagerRefdump(Pager *pPager){
4474 PgHdr *pPg;
4475 for(pPg=pPager->pAll; pPg; pPg=pPg->pNextAll){
4476 if( pPg->nRef<=0 ) continue;
4477 sqlite3DebugPrintf("PAGE %3d addr=%p nRef=%d\n",
4478 pPg->pgno, PGHDR_TO_DATA(pPg), pPg->nRef);
4479 }
4480}
4481#endif
4482
4483#endif /* SQLITE_OMIT_DISKIO */

Archive Download this file

Branches

Tags

Quick Links:     www.monotone.ca    -     Downloads    -     Documentation    -     Wiki    -     Code Forge    -     Build Status