aboutsummaryrefslogtreecommitdiffstatshomepage
path: root/libraries/sqlite/win32/pager.c
diff options
context:
space:
mode:
Diffstat (limited to 'libraries/sqlite/win32/pager.c')
-rwxr-xr-xlibraries/sqlite/win32/pager.c5104
1 files changed, 5104 insertions, 0 deletions
diff --git a/libraries/sqlite/win32/pager.c b/libraries/sqlite/win32/pager.c
new file mode 100755
index 0000000..b0ad715
--- /dev/null
+++ b/libraries/sqlite/win32/pager.c
@@ -0,0 +1,5104 @@
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 1.392 2007/10/03 15:22:26 danielk1977 Exp $
22*/
23#ifndef SQLITE_OMIT_DISKIO
24#include "sqliteInt.h"
25#include <assert.h>
26#include <string.h>
27
28/*
29** Macros for troubleshooting. Normally turned off
30*/
31#if 0
32#define sqlite3DebugPrintf printf
33#define PAGERTRACE1(X) sqlite3DebugPrintf(X)
34#define PAGERTRACE2(X,Y) sqlite3DebugPrintf(X,Y)
35#define PAGERTRACE3(X,Y,Z) sqlite3DebugPrintf(X,Y,Z)
36#define PAGERTRACE4(X,Y,Z,W) sqlite3DebugPrintf(X,Y,Z,W)
37#define PAGERTRACE5(X,Y,Z,W,V) sqlite3DebugPrintf(X,Y,Z,W,V)
38#else
39#define PAGERTRACE1(X)
40#define PAGERTRACE2(X,Y)
41#define PAGERTRACE3(X,Y,Z)
42#define PAGERTRACE4(X,Y,Z,W)
43#define PAGERTRACE5(X,Y,Z,W,V)
44#endif
45
46/*
47** The following two macros are used within the PAGERTRACEX() macros above
48** to print out file-descriptors.
49**
50** PAGERID() takes a pointer to a Pager struct as it's argument. The
51** associated file-descriptor is returned. FILEHANDLEID() takes an sqlite3_file
52** struct as it's argument.
53*/
54#define PAGERID(p) ((int)(p->fd))
55#define FILEHANDLEID(fd) ((int)fd)
56
57/*
58** The page cache as a whole is always in one of the following
59** states:
60**
61** PAGER_UNLOCK The page cache is not currently reading or
62** writing the database file. There is no
63** data held in memory. This is the initial
64** state.
65**
66** PAGER_SHARED The page cache is reading the database.
67** Writing is not permitted. There can be
68** multiple readers accessing the same database
69** file at the same time.
70**
71** PAGER_RESERVED This process has reserved the database for writing
72** but has not yet made any changes. Only one process
73** at a time can reserve the database. The original
74** database file has not been modified so other
75** processes may still be reading the on-disk
76** database file.
77**
78** PAGER_EXCLUSIVE The page cache is writing the database.
79** Access is exclusive. No other processes or
80** threads can be reading or writing while one
81** process is writing.
82**
83** PAGER_SYNCED The pager moves to this state from PAGER_EXCLUSIVE
84** after all dirty pages have been written to the
85** database file and the file has been synced to
86** disk. All that remains to do is to remove or
87** truncate the journal file and the transaction
88** will be committed.
89**
90** The page cache comes up in PAGER_UNLOCK. The first time a
91** sqlite3PagerGet() occurs, the state transitions to PAGER_SHARED.
92** After all pages have been released using sqlite_page_unref(),
93** the state transitions back to PAGER_UNLOCK. The first time
94** that sqlite3PagerWrite() is called, the state transitions to
95** PAGER_RESERVED. (Note that sqlite3PagerWrite() can only be
96** called on an outstanding page which means that the pager must
97** be in PAGER_SHARED before it transitions to PAGER_RESERVED.)
98** PAGER_RESERVED means that there is an open rollback journal.
99** The transition to PAGER_EXCLUSIVE occurs before any changes
100** are made to the database file, though writes to the rollback
101** journal occurs with just PAGER_RESERVED. After an sqlite3PagerRollback()
102** or sqlite3PagerCommitPhaseTwo(), the state can go back to PAGER_SHARED,
103** or it can stay at PAGER_EXCLUSIVE if we are in exclusive access mode.
104*/
105#define PAGER_UNLOCK 0
106#define PAGER_SHARED 1 /* same as SHARED_LOCK */
107#define PAGER_RESERVED 2 /* same as RESERVED_LOCK */
108#define PAGER_EXCLUSIVE 4 /* same as EXCLUSIVE_LOCK */
109#define PAGER_SYNCED 5
110
111/*
112** If the SQLITE_BUSY_RESERVED_LOCK macro is set to true at compile-time,
113** then failed attempts to get a reserved lock will invoke the busy callback.
114** This is off by default. To see why, consider the following scenario:
115**
116** Suppose thread A already has a shared lock and wants a reserved lock.
117** Thread B already has a reserved lock and wants an exclusive lock. If
118** both threads are using their busy callbacks, it might be a long time
119** be for one of the threads give up and allows the other to proceed.
120** But if the thread trying to get the reserved lock gives up quickly
121** (if it never invokes its busy callback) then the contention will be
122** resolved quickly.
123*/
124#ifndef SQLITE_BUSY_RESERVED_LOCK
125# define SQLITE_BUSY_RESERVED_LOCK 0
126#endif
127
128/*
129** This macro rounds values up so that if the value is an address it
130** is guaranteed to be an address that is aligned to an 8-byte boundary.
131*/
132#define FORCE_ALIGNMENT(X) (((X)+7)&~7)
133
134typedef struct PgHdr PgHdr;
135
136/*
137** Each pager stores all currently unreferenced pages in a list sorted
138** in least-recently-used (LRU) order (i.e. the first item on the list has
139** not been referenced in a long time, the last item has been recently
140** used). An instance of this structure is included as part of each
141** pager structure for this purpose (variable Pager.lru).
142**
143** Additionally, if memory-management is enabled, all unreferenced pages
144** are stored in a global LRU list (global variable sqlite3LruPageList).
145**
146** In both cases, the PagerLruList.pFirstSynced variable points to
147** the first page in the corresponding list that does not require an
148** fsync() operation before it's memory can be reclaimed. If no such
149** page exists, PagerLruList.pFirstSynced is set to NULL.
150*/
151typedef struct PagerLruList PagerLruList;
152struct PagerLruList {
153 PgHdr *pFirst; /* First page in LRU list */
154 PgHdr *pLast; /* Last page in LRU list (the most recently used) */
155 PgHdr *pFirstSynced; /* First page in list with PgHdr.needSync==0 */
156};
157
158/*
159** The following structure contains the next and previous pointers used
160** to link a PgHdr structure into a PagerLruList linked list.
161*/
162typedef struct PagerLruLink PagerLruLink;
163struct PagerLruLink {
164 PgHdr *pNext;
165 PgHdr *pPrev;
166};
167
168/*
169** Each in-memory image of a page begins with the following header.
170** This header is only visible to this pager module. The client
171** code that calls pager sees only the data that follows the header.
172**
173** Client code should call sqlite3PagerWrite() on a page prior to making
174** any modifications to that page. The first time sqlite3PagerWrite()
175** is called, the original page contents are written into the rollback
176** journal and PgHdr.inJournal and PgHdr.needSync are set. Later, once
177** the journal page has made it onto the disk surface, PgHdr.needSync
178** is cleared. The modified page cannot be written back into the original
179** database file until the journal pages has been synced to disk and the
180** PgHdr.needSync has been cleared.
181**
182** The PgHdr.dirty flag is set when sqlite3PagerWrite() is called and
183** is cleared again when the page content is written back to the original
184** database file.
185**
186** Details of important structure elements:
187**
188** needSync
189**
190** If this is true, this means that it is not safe to write the page
191** content to the database because the original content needed
192** for rollback has not by synced to the main rollback journal.
193** The original content may have been written to the rollback journal
194** but it has not yet been synced. So we cannot write to the database
195** file because power failure might cause the page in the journal file
196** to never reach the disk. It is as if the write to the journal file
197** does not occur until the journal file is synced.
198**
199** This flag is false if the page content exactly matches what
200** currently exists in the database file. The needSync flag is also
201** false if the original content has been written to the main rollback
202** journal and synced. If the page represents a new page that has
203** been added onto the end of the database during the current
204** transaction, the needSync flag is true until the original database
205** size in the journal header has been synced to disk.
206**
207** inJournal
208**
209** This is true if the original page has been written into the main
210** rollback journal. This is always false for new pages added to
211** the end of the database file during the current transaction.
212** And this flag says nothing about whether or not the journal
213** has been synced to disk. For pages that are in the original
214** database file, the following expression should always be true:
215**
216** inJournal = (pPager->aInJournal[(pgno-1)/8] & (1<<((pgno-1)%8))!=0
217**
218** The pPager->aInJournal[] array is only valid for the original
219** pages of the database, not new pages that are added to the end
220** of the database, so obviously the above expression cannot be
221** valid for new pages. For new pages inJournal is always 0.
222**
223** dirty
224**
225** When true, this means that the content of the page has been
226** modified and needs to be written back to the database file.
227** If false, it means that either the content of the page is
228** unchanged or else the content is unimportant and we do not
229** care whether or not it is preserved.
230**
231** alwaysRollback
232**
233** This means that the sqlite3PagerDontRollback() API should be
234** ignored for this page. The DontRollback() API attempts to say
235** that the content of the page on disk is unimportant (it is an
236** unused page on the freelist) so that it is unnecessary to
237** rollback changes to this page because the content of the page
238** can change without changing the meaning of the database. This
239** flag overrides any DontRollback() attempt. This flag is set
240** when a page that originally contained valid data is added to
241** the freelist. Later in the same transaction, this page might
242** be pulled from the freelist and reused for something different
243** and at that point the DontRollback() API will be called because
244** pages taken from the freelist do not need to be protected by
245** the rollback journal. But this flag says that the page was
246** not originally part of the freelist so that it still needs to
247** be rolled back in spite of any subsequent DontRollback() calls.
248**
249** needRead
250**
251** This flag means (when true) that the content of the page has
252** not yet been loaded from disk. The in-memory content is just
253** garbage. (Actually, we zero the content, but you should not
254** make any assumptions about the content nevertheless.) If the
255** content is needed in the future, it should be read from the
256** original database file.
257*/
258struct PgHdr {
259 Pager *pPager; /* The pager to which this page belongs */
260 Pgno pgno; /* The page number for this page */
261 PgHdr *pNextHash, *pPrevHash; /* Hash collision chain for PgHdr.pgno */
262 PagerLruLink free; /* Next and previous free pages */
263 PgHdr *pNextAll; /* A list of all pages */
264 u8 inJournal; /* TRUE if has been written to journal */
265 u8 dirty; /* TRUE if we need to write back changes */
266 u8 needSync; /* Sync journal before writing this page */
267 u8 alwaysRollback; /* Disable DontRollback() for this page */
268 u8 needRead; /* Read content if PagerWrite() is called */
269 short int nRef; /* Number of users of this page */
270 PgHdr *pDirty, *pPrevDirty; /* Dirty pages */
271#ifdef SQLITE_ENABLE_MEMORY_MANAGEMENT
272 PagerLruLink gfree; /* Global list of nRef==0 pages */
273#endif
274#ifdef SQLITE_CHECK_PAGES
275 u32 pageHash;
276#endif
277 void *pData; /* Page data */
278 /* Pager.nExtra bytes of local data appended to this header */
279};
280
281/*
282** For an in-memory only database, some extra information is recorded about
283** each page so that changes can be rolled back. (Journal files are not
284** used for in-memory databases.) The following information is added to
285** the end of every EXTRA block for in-memory databases.
286**
287** This information could have been added directly to the PgHdr structure.
288** But then it would take up an extra 8 bytes of storage on every PgHdr
289** even for disk-based databases. Splitting it out saves 8 bytes. This
290** is only a savings of 0.8% but those percentages add up.
291*/
292typedef struct PgHistory PgHistory;
293struct PgHistory {
294 u8 *pOrig; /* Original page text. Restore to this on a full rollback */
295 u8 *pStmt; /* Text as it was at the beginning of the current statement */
296 PgHdr *pNextStmt, *pPrevStmt; /* List of pages in the statement journal */
297 u8 inStmt; /* TRUE if in the statement subjournal */
298};
299
300/*
301** A macro used for invoking the codec if there is one
302*/
303#ifdef SQLITE_HAS_CODEC
304# define CODEC1(P,D,N,X) if( P->xCodec!=0 ){ P->xCodec(P->pCodecArg,D,N,X); }
305# define CODEC2(P,D,N,X) ((char*)(P->xCodec!=0?P->xCodec(P->pCodecArg,D,N,X):D))
306#else
307# define CODEC1(P,D,N,X) /* NO-OP */
308# define CODEC2(P,D,N,X) ((char*)D)
309#endif
310
311/*
312** Convert a pointer to a PgHdr into a pointer to its data
313** and back again.
314*/
315#define PGHDR_TO_DATA(P) ((P)->pData)
316#define PGHDR_TO_EXTRA(G,P) ((void*)&((G)[1]))
317#define PGHDR_TO_HIST(P,PGR) \
318 ((PgHistory*)&((char*)(&(P)[1]))[(PGR)->nExtra])
319
320/*
321** A open page cache is an instance of the following structure.
322**
323** Pager.errCode may be set to SQLITE_IOERR, SQLITE_CORRUPT, or
324** or SQLITE_FULL. Once one of the first three errors occurs, it persists
325** and is returned as the result of every major pager API call. The
326** SQLITE_FULL return code is slightly different. It persists only until the
327** next successful rollback is performed on the pager cache. Also,
328** SQLITE_FULL does not affect the sqlite3PagerGet() and sqlite3PagerLookup()
329** APIs, they may still be used successfully.
330*/
331struct Pager {
332 sqlite3_vfs *pVfs; /* OS functions to use for IO */
333 u8 journalOpen; /* True if journal file descriptors is valid */
334 u8 journalStarted; /* True if header of journal is synced */
335 u8 useJournal; /* Use a rollback journal on this file */
336 u8 noReadlock; /* Do not bother to obtain readlocks */
337 u8 stmtOpen; /* True if the statement subjournal is open */
338 u8 stmtInUse; /* True we are in a statement subtransaction */
339 u8 stmtAutoopen; /* Open stmt journal when main journal is opened*/
340 u8 noSync; /* Do not sync the journal if true */
341 u8 fullSync; /* Do extra syncs of the journal for robustness */
342 u8 sync_flags; /* One of SYNC_NORMAL or SYNC_FULL */
343 u8 state; /* PAGER_UNLOCK, _SHARED, _RESERVED, etc. */
344 u8 tempFile; /* zFilename is a temporary file */
345 u8 readOnly; /* True for a read-only database */
346 u8 needSync; /* True if an fsync() is needed on the journal */
347 u8 dirtyCache; /* True if cached pages have changed */
348 u8 alwaysRollback; /* Disable DontRollback() for all pages */
349 u8 memDb; /* True to inhibit all file I/O */
350 u8 setMaster; /* True if a m-j name has been written to jrnl */
351 u8 doNotSync; /* Boolean. While true, do not spill the cache */
352 u8 exclusiveMode; /* Boolean. True if locking_mode==EXCLUSIVE */
353 u8 changeCountDone; /* Set after incrementing the change-counter */
354 u32 vfsFlags; /* Flags for sqlite3_vfs.xOpen() */
355 int errCode; /* One of several kinds of errors */
356 int dbSize; /* Number of pages in the file */
357 int origDbSize; /* dbSize before the current change */
358 int stmtSize; /* Size of database (in pages) at stmt_begin() */
359 int nRec; /* Number of pages written to the journal */
360 u32 cksumInit; /* Quasi-random value added to every checksum */
361 int stmtNRec; /* Number of records in stmt subjournal */
362 int nExtra; /* Add this many bytes to each in-memory page */
363 int pageSize; /* Number of bytes in a page */
364 int nPage; /* Total number of in-memory pages */
365 int nRef; /* Number of in-memory pages with PgHdr.nRef>0 */
366 int mxPage; /* Maximum number of pages to hold in cache */
367 Pgno mxPgno; /* Maximum allowed size of the database */
368 u8 *aInJournal; /* One bit for each page in the database file */
369 u8 *aInStmt; /* One bit for each page in the database */
370 char *zFilename; /* Name of the database file */
371 char *zJournal; /* Name of the journal file */
372 char *zDirectory; /* Directory hold database and journal files */
373 char *zStmtJrnl; /* Name of the statement journal file */
374 sqlite3_file *fd, *jfd; /* File descriptors for database and journal */
375 sqlite3_file *stfd; /* File descriptor for the statement subjournal*/
376 BusyHandler *pBusyHandler; /* Pointer to sqlite.busyHandler */
377 PagerLruList lru; /* LRU list of free pages */
378 PgHdr *pAll; /* List of all pages */
379 PgHdr *pStmt; /* List of pages in the statement subjournal */
380 PgHdr *pDirty; /* List of all dirty pages */
381 i64 journalOff; /* Current byte offset in the journal file */
382 i64 journalHdr; /* Byte offset to previous journal header */
383 i64 stmtHdrOff; /* First journal header written this statement */
384 i64 stmtCksum; /* cksumInit when statement was started */
385 i64 stmtJSize; /* Size of journal at stmt_begin() */
386 int sectorSize; /* Assumed sector size during rollback */
387#ifdef SQLITE_TEST
388 int nHit, nMiss; /* Cache hits and missing */
389 int nRead, nWrite; /* Database pages read/written */
390#endif
391 void (*xDestructor)(DbPage*,int); /* Call this routine when freeing pages */
392 void (*xReiniter)(DbPage*,int); /* Call this routine when reloading pages */
393#ifdef SQLITE_HAS_CODEC
394 void *(*xCodec)(void*,void*,Pgno,int); /* Routine for en/decoding data */
395 void *pCodecArg; /* First argument to xCodec() */
396#endif
397 int nHash; /* Size of the pager hash table */
398 PgHdr **aHash; /* Hash table to map page number to PgHdr */
399#ifdef SQLITE_ENABLE_MEMORY_MANAGEMENT
400 Pager *pNext; /* Doubly linked list of pagers on which */
401 Pager *pPrev; /* sqlite3_release_memory() will work */
402 int iInUseMM; /* Non-zero if unavailable to MM */
403 int iInUseDB; /* Non-zero if in sqlite3_release_memory() */
404#endif
405 char *pTmpSpace; /* Pager.pageSize bytes of space for tmp use */
406 char dbFileVers[16]; /* Changes whenever database file changes */
407};
408
409/*
410** The following global variables hold counters used for
411** testing purposes only. These variables do not exist in
412** a non-testing build. These variables are not thread-safe.
413*/
414#ifdef SQLITE_TEST
415int sqlite3_pager_readdb_count = 0; /* Number of full pages read from DB */
416int sqlite3_pager_writedb_count = 0; /* Number of full pages written to DB */
417int sqlite3_pager_writej_count = 0; /* Number of pages written to journal */
418int sqlite3_pager_pgfree_count = 0; /* Number of cache pages freed */
419# define PAGER_INCR(v) v++
420#else
421# define PAGER_INCR(v)
422#endif
423
424/*
425** The following variable points to the head of a double-linked list
426** of all pagers that are eligible for page stealing by the
427** sqlite3_release_memory() interface. Access to this list is
428** protected by the SQLITE_MUTEX_STATIC_MEM2 mutex.
429*/
430#ifdef SQLITE_ENABLE_MEMORY_MANAGEMENT
431static Pager *sqlite3PagerList = 0;
432static PagerLruList sqlite3LruPageList = {0, 0, 0};
433#endif
434
435
436/*
437** Journal files begin with the following magic string. The data
438** was obtained from /dev/random. It is used only as a sanity check.
439**
440** Since version 2.8.0, the journal format contains additional sanity
441** checking information. If the power fails while the journal is begin
442** written, semi-random garbage data might appear in the journal
443** file after power is restored. If an attempt is then made
444** to roll the journal back, the database could be corrupted. The additional
445** sanity checking data is an attempt to discover the garbage in the
446** journal and ignore it.
447**
448** The sanity checking information for the new journal format consists
449** of a 32-bit checksum on each page of data. The checksum covers both
450** the page number and the pPager->pageSize bytes of data for the page.
451** This cksum is initialized to a 32-bit random value that appears in the
452** journal file right after the header. The random initializer is important,
453** because garbage data that appears at the end of a journal is likely
454** data that was once in other files that have now been deleted. If the
455** garbage data came from an obsolete journal file, the checksums might
456** be correct. But by initializing the checksum to random value which
457** is different for every journal, we minimize that risk.
458*/
459static const unsigned char aJournalMagic[] = {
460 0xd9, 0xd5, 0x05, 0xf9, 0x20, 0xa1, 0x63, 0xd7,
461};
462
463/*
464** The size of the header and of each page in the journal is determined
465** by the following macros.
466*/
467#define JOURNAL_PG_SZ(pPager) ((pPager->pageSize) + 8)
468
469/*
470** The journal header size for this pager. In the future, this could be
471** set to some value read from the disk controller. The important
472** characteristic is that it is the same size as a disk sector.
473*/
474#define JOURNAL_HDR_SZ(pPager) (pPager->sectorSize)
475
476/*
477** The macro MEMDB is true if we are dealing with an in-memory database.
478** We do this as a macro so that if the SQLITE_OMIT_MEMORYDB macro is set,
479** the value of MEMDB will be a constant and the compiler will optimize
480** out code that would never execute.
481*/
482#ifdef SQLITE_OMIT_MEMORYDB
483# define MEMDB 0
484#else
485# define MEMDB pPager->memDb
486#endif
487
488/*
489** Page number PAGER_MJ_PGNO is never used in an SQLite database (it is
490** reserved for working around a windows/posix incompatibility). It is
491** used in the journal to signify that the remainder of the journal file
492** is devoted to storing a master journal name - there are no more pages to
493** roll back. See comments for function writeMasterJournal() for details.
494*/
495/* #define PAGER_MJ_PGNO(x) (PENDING_BYTE/((x)->pageSize)) */
496#define PAGER_MJ_PGNO(x) ((PENDING_BYTE/((x)->pageSize))+1)
497
498/*
499** The maximum legal page number is (2^31 - 1).
500*/
501#define PAGER_MAX_PGNO 2147483647
502
503/*
504** The pagerEnter() and pagerLeave() routines acquire and release
505** a mutex on each pager. The mutex is recursive.
506**
507** This is a special-purpose mutex. It only provides mutual exclusion
508** between the Btree and the Memory Management sqlite3_release_memory()
509** function. It does not prevent, for example, two Btrees from accessing
510** the same pager at the same time. Other general-purpose mutexes in
511** the btree layer handle that chore.
512*/
513#ifdef SQLITE_ENABLE_MEMORY_MANAGEMENT
514 static void pagerEnter(Pager *p){
515 p->iInUseDB++;
516 if( p->iInUseMM && p->iInUseDB==1 ){
517 sqlite3_mutex *mutex;
518 mutex = sqlite3_mutex_alloc(SQLITE_MUTEX_STATIC_MEM2);
519 p->iInUseDB = 0;
520 sqlite3_mutex_enter(mutex);
521 p->iInUseDB = 1;
522 sqlite3_mutex_leave(mutex);
523 }
524 assert( p->iInUseMM==0 );
525 }
526 static void pagerLeave(Pager *p){
527 p->iInUseDB--;
528 assert( p->iInUseDB>=0 );
529 }
530#else
531# define pagerEnter(X)
532# define pagerLeave(X)
533#endif
534
535/*
536** Enable reference count tracking (for debugging) here:
537*/
538#ifdef SQLITE_DEBUG
539 int pager3_refinfo_enable = 0;
540 static void pager_refinfo(PgHdr *p){
541 static int cnt = 0;
542 if( !pager3_refinfo_enable ) return;
543 sqlite3DebugPrintf(
544 "REFCNT: %4d addr=%p nRef=%-3d total=%d\n",
545 p->pgno, PGHDR_TO_DATA(p), p->nRef, p->pPager->nRef
546 );
547 cnt++; /* Something to set a breakpoint on */
548 }
549# define REFINFO(X) pager_refinfo(X)
550#else
551# define REFINFO(X)
552#endif
553
554/*
555** Add page pPg to the end of the linked list managed by structure
556** pList (pPg becomes the last entry in the list - the most recently
557** used). Argument pLink should point to either pPg->free or pPg->gfree,
558** depending on whether pPg is being added to the pager-specific or
559** global LRU list.
560*/
561static void listAdd(PagerLruList *pList, PagerLruLink *pLink, PgHdr *pPg){
562 pLink->pNext = 0;
563 pLink->pPrev = pList->pLast;
564
565#ifdef SQLITE_ENABLE_MEMORY_MANAGEMENT
566 assert(pLink==&pPg->free || pLink==&pPg->gfree);
567 assert(pLink==&pPg->gfree || pList!=&sqlite3LruPageList);
568#endif
569
570 if( pList->pLast ){
571 int iOff = (char *)pLink - (char *)pPg;
572 PagerLruLink *pLastLink = (PagerLruLink *)(&((u8 *)pList->pLast)[iOff]);
573 pLastLink->pNext = pPg;
574 }else{
575 assert(!pList->pFirst);
576 pList->pFirst = pPg;
577 }
578
579 pList->pLast = pPg;
580 if( !pList->pFirstSynced && pPg->needSync==0 ){
581 pList->pFirstSynced = pPg;
582 }
583}
584
585/*
586** Remove pPg from the list managed by the structure pointed to by pList.
587**
588** Argument pLink should point to either pPg->free or pPg->gfree, depending
589** on whether pPg is being added to the pager-specific or global LRU list.
590*/
591static void listRemove(PagerLruList *pList, PagerLruLink *pLink, PgHdr *pPg){
592 int iOff = (char *)pLink - (char *)pPg;
593
594#ifdef SQLITE_ENABLE_MEMORY_MANAGEMENT
595 assert(pLink==&pPg->free || pLink==&pPg->gfree);
596 assert(pLink==&pPg->gfree || pList!=&sqlite3LruPageList);
597#endif
598
599 if( pPg==pList->pFirst ){
600 pList->pFirst = pLink->pNext;
601 }
602 if( pPg==pList->pLast ){
603 pList->pLast = pLink->pPrev;
604 }
605 if( pLink->pPrev ){
606 PagerLruLink *pPrevLink = (PagerLruLink *)(&((u8 *)pLink->pPrev)[iOff]);
607 pPrevLink->pNext = pLink->pNext;
608 }
609 if( pLink->pNext ){
610 PagerLruLink *pNextLink = (PagerLruLink *)(&((u8 *)pLink->pNext)[iOff]);
611 pNextLink->pPrev = pLink->pPrev;
612 }
613 if( pPg==pList->pFirstSynced ){
614 PgHdr *p = pLink->pNext;
615 while( p && p->needSync ){
616 PagerLruLink *pL = (PagerLruLink *)(&((u8 *)p)[iOff]);
617 p = pL->pNext;
618 }
619 pList->pFirstSynced = p;
620 }
621
622 pLink->pNext = pLink->pPrev = 0;
623}
624
625/*
626** Add page pPg to the list of free pages for the pager. If
627** memory-management is enabled, also add the page to the global
628** list of free pages.
629*/
630static void lruListAdd(PgHdr *pPg){
631 listAdd(&pPg->pPager->lru, &pPg->free, pPg);
632#ifdef SQLITE_ENABLE_MEMORY_MANAGEMENT
633 if( !pPg->pPager->memDb ){
634 sqlite3_mutex_enter(sqlite3_mutex_alloc(SQLITE_MUTEX_STATIC_LRU));
635 listAdd(&sqlite3LruPageList, &pPg->gfree, pPg);
636 sqlite3_mutex_leave(sqlite3_mutex_alloc(SQLITE_MUTEX_STATIC_LRU));
637 }
638#endif
639}
640
641/*
642** Remove page pPg from the list of free pages for the associated pager.
643** If memory-management is enabled, also remove pPg from the global list
644** of free pages.
645*/
646static void lruListRemove(PgHdr *pPg){
647 listRemove(&pPg->pPager->lru, &pPg->free, pPg);
648#ifdef SQLITE_ENABLE_MEMORY_MANAGEMENT
649 if( !pPg->pPager->memDb ){
650 sqlite3_mutex_enter(sqlite3_mutex_alloc(SQLITE_MUTEX_STATIC_LRU));
651 listRemove(&sqlite3LruPageList, &pPg->gfree, pPg);
652 sqlite3_mutex_leave(sqlite3_mutex_alloc(SQLITE_MUTEX_STATIC_LRU));
653 }
654#endif
655}
656
657/*
658** This function is called just after the needSync flag has been cleared
659** from all pages managed by pPager (usually because the journal file
660** has just been synced). It updates the pPager->lru.pFirstSynced variable
661** and, if memory-management is enabled, the sqlite3LruPageList.pFirstSynced
662** variable also.
663*/
664static void lruListSetFirstSynced(Pager *pPager){
665 pPager->lru.pFirstSynced = pPager->lru.pFirst;
666#ifdef SQLITE_ENABLE_MEMORY_MANAGEMENT
667 if( !pPager->memDb ){
668 PgHdr *p;
669 sqlite3_mutex_enter(sqlite3_mutex_alloc(SQLITE_MUTEX_STATIC_LRU));
670 for(p=sqlite3LruPageList.pFirst; p && p->needSync; p=p->gfree.pNext);
671 assert(p==pPager->lru.pFirstSynced || p==sqlite3LruPageList.pFirstSynced);
672 sqlite3LruPageList.pFirstSynced = p;
673 sqlite3_mutex_leave(sqlite3_mutex_alloc(SQLITE_MUTEX_STATIC_LRU));
674 }
675#endif
676}
677
678/*
679** Return true if page *pPg has already been written to the statement
680** journal (or statement snapshot has been created, if *pPg is part
681** of an in-memory database).
682*/
683static int pageInStatement(PgHdr *pPg){
684 Pager *pPager = pPg->pPager;
685 if( MEMDB ){
686 return PGHDR_TO_HIST(pPg, pPager)->inStmt;
687 }else{
688 Pgno pgno = pPg->pgno;
689 u8 *a = pPager->aInStmt;
690 return (a && (int)pgno<=pPager->stmtSize && (a[pgno/8] & (1<<(pgno&7))));
691 }
692}
693
694/*
695** Change the size of the pager hash table to N. N must be a power
696** of two.
697*/
698static void pager_resize_hash_table(Pager *pPager, int N){
699 PgHdr **aHash, *pPg;
700 assert( N>0 && (N&(N-1))==0 );
701 pagerLeave(pPager);
702 sqlite3MallocBenignFailure((int)pPager->aHash);
703 aHash = sqlite3MallocZero( sizeof(aHash[0])*N );
704 pagerEnter(pPager);
705 if( aHash==0 ){
706 /* Failure to rehash is not an error. It is only a performance hit. */
707 return;
708 }
709 sqlite3_free(pPager->aHash);
710 pPager->nHash = N;
711 pPager->aHash = aHash;
712 for(pPg=pPager->pAll; pPg; pPg=pPg->pNextAll){
713 int h;
714 if( pPg->pgno==0 ){
715 assert( pPg->pNextHash==0 && pPg->pPrevHash==0 );
716 continue;
717 }
718 h = pPg->pgno & (N-1);
719 pPg->pNextHash = aHash[h];
720 if( aHash[h] ){
721 aHash[h]->pPrevHash = pPg;
722 }
723 aHash[h] = pPg;
724 pPg->pPrevHash = 0;
725 }
726}
727
728/*
729** Read a 32-bit integer from the given file descriptor. Store the integer
730** that is read in *pRes. Return SQLITE_OK if everything worked, or an
731** error code is something goes wrong.
732**
733** All values are stored on disk as big-endian.
734*/
735static int read32bits(sqlite3_file *fd, i64 offset, u32 *pRes){
736 unsigned char ac[4];
737 int rc = sqlite3OsRead(fd, ac, sizeof(ac), offset);
738 if( rc==SQLITE_OK ){
739 *pRes = sqlite3Get4byte(ac);
740 }
741 return rc;
742}
743
744/*
745** Write a 32-bit integer into a string buffer in big-endian byte order.
746*/
747#define put32bits(A,B) sqlite3Put4byte((u8*)A,B)
748
749/*
750** Write a 32-bit integer into the given file descriptor. Return SQLITE_OK
751** on success or an error code is something goes wrong.
752*/
753static int write32bits(sqlite3_file *fd, i64 offset, u32 val){
754 char ac[4];
755 put32bits(ac, val);
756 return sqlite3OsWrite(fd, ac, 4, offset);
757}
758
759/*
760** If file pFd is open, call sqlite3OsUnlock() on it.
761*/
762static int osUnlock(sqlite3_file *pFd, int eLock){
763 if( !pFd->pMethods ){
764 return SQLITE_OK;
765 }
766 return sqlite3OsUnlock(pFd, eLock);
767}
768
769/*
770** This function determines whether or not the atomic-write optimization
771** can be used with this pager. The optimization can be used if:
772**
773** (a) the value returned by OsDeviceCharacteristics() indicates that
774** a database page may be written atomically, and
775** (b) the value returned by OsSectorSize() is less than or equal
776** to the page size.
777**
778** If the optimization cannot be used, 0 is returned. If it can be used,
779** then the value returned is the size of the journal file when it
780** contains rollback data for exactly one page.
781*/
782#ifdef SQLITE_ENABLE_ATOMIC_WRITE
783static int jrnlBufferSize(Pager *pPager){
784 int dc; /* Device characteristics */
785 int nSector; /* Sector size */
786 int nPage; /* Page size */
787 sqlite3_file *fd = pPager->fd;
788
789 if( fd->pMethods ){
790 dc = sqlite3OsDeviceCharacteristics(fd);
791 nSector = sqlite3OsSectorSize(fd);
792 nPage = pPager->pageSize;
793 }
794
795 assert(SQLITE_IOCAP_ATOMIC512==(512>>8));
796 assert(SQLITE_IOCAP_ATOMIC64K==(65536>>8));
797
798 if( !fd->pMethods || (dc&(SQLITE_IOCAP_ATOMIC|(nPage>>8))&&nSector<=nPage) ){
799 return JOURNAL_HDR_SZ(pPager) + JOURNAL_PG_SZ(pPager);
800 }
801 return 0;
802}
803#endif
804
805/*
806** This function should be called when an error occurs within the pager
807** code. The first argument is a pointer to the pager structure, the
808** second the error-code about to be returned by a pager API function.
809** The value returned is a copy of the second argument to this function.
810**
811** If the second argument is SQLITE_IOERR, SQLITE_CORRUPT, or SQLITE_FULL
812** the error becomes persistent. Until the persisten error is cleared,
813** subsequent API calls on this Pager will immediately return the same
814** error code.
815**
816** A persistent error indicates that the contents of the pager-cache
817** cannot be trusted. This state can be cleared by completely discarding
818** the contents of the pager-cache. If a transaction was active when
819** the persistent error occured, then the rollback journal may need
820** to be replayed.
821*/
822static void pager_unlock(Pager *pPager);
823static int pager_error(Pager *pPager, int rc){
824 int rc2 = rc & 0xff;
825 assert(
826 pPager->errCode==SQLITE_FULL ||
827 pPager->errCode==SQLITE_OK ||
828 (pPager->errCode & 0xff)==SQLITE_IOERR
829 );
830 if(
831 rc2==SQLITE_FULL ||
832 rc2==SQLITE_IOERR ||
833 rc2==SQLITE_CORRUPT
834 ){
835 pPager->errCode = rc;
836 if( pPager->state==PAGER_UNLOCK && pPager->nRef==0 ){
837 /* If the pager is already unlocked, call pager_unlock() now to
838 ** clear the error state and ensure that the pager-cache is
839 ** completely empty.
840 */
841 pager_unlock(pPager);
842 }
843 }
844 return rc;
845}
846
847/*
848** If SQLITE_CHECK_PAGES is defined then we do some sanity checking
849** on the cache using a hash function. This is used for testing
850** and debugging only.
851*/
852#ifdef SQLITE_CHECK_PAGES
853/*
854** Return a 32-bit hash of the page data for pPage.
855*/
856static u32 pager_datahash(int nByte, unsigned char *pData){
857 u32 hash = 0;
858 int i;
859 for(i=0; i<nByte; i++){
860 hash = (hash*1039) + pData[i];
861 }
862 return hash;
863}
864static u32 pager_pagehash(PgHdr *pPage){
865 return pager_datahash(pPage->pPager->pageSize,
866 (unsigned char *)PGHDR_TO_DATA(pPage));
867}
868
869/*
870** The CHECK_PAGE macro takes a PgHdr* as an argument. If SQLITE_CHECK_PAGES
871** is defined, and NDEBUG is not defined, an assert() statement checks
872** that the page is either dirty or still matches the calculated page-hash.
873*/
874#define CHECK_PAGE(x) checkPage(x)
875static void checkPage(PgHdr *pPg){
876 Pager *pPager = pPg->pPager;
877 assert( !pPg->pageHash || pPager->errCode || MEMDB || pPg->dirty ||
878 pPg->pageHash==pager_pagehash(pPg) );
879}
880
881#else
882#define pager_datahash(X,Y) 0
883#define pager_pagehash(X) 0
884#define CHECK_PAGE(x)
885#endif
886
887/*
888** When this is called the journal file for pager pPager must be open.
889** The master journal file name is read from the end of the file and
890** written into memory supplied by the caller.
891**
892** zMaster must point to a buffer of at least nMaster bytes allocated by
893** the caller. This should be sqlite3_vfs.mxPathname+1 (to ensure there is
894** enough space to write the master journal name). If the master journal
895** name in the journal is longer than nMaster bytes (including a
896** nul-terminator), then this is handled as if no master journal name
897** were present in the journal.
898**
899** If no master journal file name is present zMaster[0] is set to 0 and
900** SQLITE_OK returned.
901*/
902static int readMasterJournal(sqlite3_file *pJrnl, char *zMaster, int nMaster){
903 int rc;
904 u32 len;
905 i64 szJ;
906 u32 cksum;
907 int i;
908 unsigned char aMagic[8]; /* A buffer to hold the magic header */
909
910 zMaster[0] = '\0';
911
912 rc = sqlite3OsFileSize(pJrnl, &szJ);
913 if( rc!=SQLITE_OK || szJ<16 ) return rc;
914
915 rc = read32bits(pJrnl, szJ-16, &len);
916 if( rc!=SQLITE_OK ) return rc;
917
918 if( len>=nMaster ){
919 return SQLITE_OK;
920 }
921
922 rc = read32bits(pJrnl, szJ-12, &cksum);
923 if( rc!=SQLITE_OK ) return rc;
924
925 rc = sqlite3OsRead(pJrnl, aMagic, 8, szJ-8);
926 if( rc!=SQLITE_OK || memcmp(aMagic, aJournalMagic, 8) ) return rc;
927
928 rc = sqlite3OsRead(pJrnl, zMaster, len, szJ-16-len);
929 if( rc!=SQLITE_OK ){
930 return rc;
931 }
932 zMaster[len] = '\0';
933
934 /* See if the checksum matches the master journal name */
935 for(i=0; i<len; i++){
936 cksum -= zMaster[i];
937 }
938 if( cksum ){
939 /* If the checksum doesn't add up, then one or more of the disk sectors
940 ** containing the master journal filename is corrupted. This means
941 ** definitely roll back, so just return SQLITE_OK and report a (nul)
942 ** master-journal filename.
943 */
944 zMaster[0] = '\0';
945 }
946
947 return SQLITE_OK;
948}
949
950/*
951** Seek the journal file descriptor to the next sector boundary where a
952** journal header may be read or written. Pager.journalOff is updated with
953** the new seek offset.
954**
955** i.e for a sector size of 512:
956**
957** Input Offset Output Offset
958** ---------------------------------------
959** 0 0
960** 512 512
961** 100 512
962** 2000 2048
963**
964*/
965static void seekJournalHdr(Pager *pPager){
966 i64 offset = 0;
967 i64 c = pPager->journalOff;
968 if( c ){
969 offset = ((c-1)/JOURNAL_HDR_SZ(pPager) + 1) * JOURNAL_HDR_SZ(pPager);
970 }
971 assert( offset%JOURNAL_HDR_SZ(pPager)==0 );
972 assert( offset>=c );
973 assert( (offset-c)<JOURNAL_HDR_SZ(pPager) );
974 pPager->journalOff = offset;
975}
976
977/*
978** The journal file must be open when this routine is called. A journal
979** header (JOURNAL_HDR_SZ bytes) is written into the journal file at the
980** current location.
981**
982** The format for the journal header is as follows:
983** - 8 bytes: Magic identifying journal format.
984** - 4 bytes: Number of records in journal, or -1 no-sync mode is on.
985** - 4 bytes: Random number used for page hash.
986** - 4 bytes: Initial database page count.
987** - 4 bytes: Sector size used by the process that wrote this journal.
988**
989** Followed by (JOURNAL_HDR_SZ - 24) bytes of unused space.
990*/
991static int writeJournalHdr(Pager *pPager){
992 char zHeader[sizeof(aJournalMagic)+16];
993 int rc;
994
995 if( pPager->stmtHdrOff==0 ){
996 pPager->stmtHdrOff = pPager->journalOff;
997 }
998
999 seekJournalHdr(pPager);
1000 pPager->journalHdr = pPager->journalOff;
1001
1002 memcpy(zHeader, aJournalMagic, sizeof(aJournalMagic));
1003
1004 /*
1005 ** Write the nRec Field - the number of page records that follow this
1006 ** journal header. Normally, zero is written to this value at this time.
1007 ** After the records are added to the journal (and the journal synced,
1008 ** if in full-sync mode), the zero is overwritten with the true number
1009 ** of records (see syncJournal()).
1010 **
1011 ** A faster alternative is to write 0xFFFFFFFF to the nRec field. When
1012 ** reading the journal this value tells SQLite to assume that the
1013 ** rest of the journal file contains valid page records. This assumption
1014 ** is dangerous, as if a failure occured whilst writing to the journal
1015 ** file it may contain some garbage data. There are two scenarios
1016 ** where this risk can be ignored:
1017 **
1018 ** * When the pager is in no-sync mode. Corruption can follow a
1019 ** power failure in this case anyway.
1020 **
1021 ** * When the SQLITE_IOCAP_SAFE_APPEND flag is set. This guarantees
1022 ** that garbage data is never appended to the journal file.
1023 */
1024 assert(pPager->fd->pMethods||pPager->noSync);
1025 if( (pPager->noSync)
1026 || (sqlite3OsDeviceCharacteristics(pPager->fd)&SQLITE_IOCAP_SAFE_APPEND)
1027 ){
1028 put32bits(&zHeader[sizeof(aJournalMagic)], 0xffffffff);
1029 }else{
1030 put32bits(&zHeader[sizeof(aJournalMagic)], 0);
1031 }
1032
1033 /* The random check-hash initialiser */
1034 sqlite3Randomness(sizeof(pPager->cksumInit), &pPager->cksumInit);
1035 put32bits(&zHeader[sizeof(aJournalMagic)+4], pPager->cksumInit);
1036 /* The initial database size */
1037 put32bits(&zHeader[sizeof(aJournalMagic)+8], pPager->dbSize);
1038 /* The assumed sector size for this process */
1039 put32bits(&zHeader[sizeof(aJournalMagic)+12], pPager->sectorSize);
1040 IOTRACE(("JHDR %p %lld %d\n", pPager, pPager->journalHdr, sizeof(zHeader)))
1041 rc = sqlite3OsWrite(pPager->jfd, zHeader, sizeof(zHeader),pPager->journalOff);
1042 pPager->journalOff += JOURNAL_HDR_SZ(pPager);
1043
1044 /* The journal header has been written successfully. Seek the journal
1045 ** file descriptor to the end of the journal header sector.
1046 */
1047 if( rc==SQLITE_OK ){
1048 IOTRACE(("JTAIL %p %lld\n", pPager, pPager->journalOff-1))
1049 rc = sqlite3OsWrite(pPager->jfd, "\000", 1, pPager->journalOff-1);
1050 }
1051 return rc;
1052}
1053
1054/*
1055** The journal file must be open when this is called. A journal header file
1056** (JOURNAL_HDR_SZ bytes) is read from the current location in the journal
1057** file. See comments above function writeJournalHdr() for a description of
1058** the journal header format.
1059**
1060** If the header is read successfully, *nRec is set to the number of
1061** page records following this header and *dbSize is set to the size of the
1062** database before the transaction began, in pages. Also, pPager->cksumInit
1063** is set to the value read from the journal header. SQLITE_OK is returned
1064** in this case.
1065**
1066** If the journal header file appears to be corrupted, SQLITE_DONE is
1067** returned and *nRec and *dbSize are not set. If JOURNAL_HDR_SZ bytes
1068** cannot be read from the journal file an error code is returned.
1069*/
1070static int readJournalHdr(
1071 Pager *pPager,
1072 i64 journalSize,
1073 u32 *pNRec,
1074 u32 *pDbSize
1075){
1076 int rc;
1077 unsigned char aMagic[8]; /* A buffer to hold the magic header */
1078 i64 jrnlOff;
1079
1080 seekJournalHdr(pPager);
1081 if( pPager->journalOff+JOURNAL_HDR_SZ(pPager) > journalSize ){
1082 return SQLITE_DONE;
1083 }
1084 jrnlOff = pPager->journalOff;
1085
1086 rc = sqlite3OsRead(pPager->jfd, aMagic, sizeof(aMagic), jrnlOff);
1087 if( rc ) return rc;
1088 jrnlOff += sizeof(aMagic);
1089
1090 if( memcmp(aMagic, aJournalMagic, sizeof(aMagic))!=0 ){
1091 return SQLITE_DONE;
1092 }
1093
1094 rc = read32bits(pPager->jfd, jrnlOff, pNRec);
1095 if( rc ) return rc;
1096
1097 rc = read32bits(pPager->jfd, jrnlOff+4, &pPager->cksumInit);
1098 if( rc ) return rc;
1099
1100 rc = read32bits(pPager->jfd, jrnlOff+8, pDbSize);
1101 if( rc ) return rc;
1102
1103 /* Update the assumed sector-size to match the value used by
1104 ** the process that created this journal. If this journal was
1105 ** created by a process other than this one, then this routine
1106 ** is being called from within pager_playback(). The local value
1107 ** of Pager.sectorSize is restored at the end of that routine.
1108 */
1109 rc = read32bits(pPager->jfd, jrnlOff+12, (u32 *)&pPager->sectorSize);
1110 if( rc ) return rc;
1111
1112 pPager->journalOff += JOURNAL_HDR_SZ(pPager);
1113 return SQLITE_OK;
1114}
1115
1116
1117/*
1118** Write the supplied master journal name into the journal file for pager
1119** pPager at the current location. The master journal name must be the last
1120** thing written to a journal file. If the pager is in full-sync mode, the
1121** journal file descriptor is advanced to the next sector boundary before
1122** anything is written. The format is:
1123**
1124** + 4 bytes: PAGER_MJ_PGNO.
1125** + N bytes: length of master journal name.
1126** + 4 bytes: N
1127** + 4 bytes: Master journal name checksum.
1128** + 8 bytes: aJournalMagic[].
1129**
1130** The master journal page checksum is the sum of the bytes in the master
1131** journal name.
1132**
1133** If zMaster is a NULL pointer (occurs for a single database transaction),
1134** this call is a no-op.
1135*/
1136static int writeMasterJournal(Pager *pPager, const char *zMaster){
1137 int rc;
1138 int len;
1139 int i;
1140 i64 jrnlOff;
1141 u32 cksum = 0;
1142 char zBuf[sizeof(aJournalMagic)+2*4];
1143
1144 if( !zMaster || pPager->setMaster) return SQLITE_OK;
1145 pPager->setMaster = 1;
1146
1147 len = strlen(zMaster);
1148 for(i=0; i<len; i++){
1149 cksum += zMaster[i];
1150 }
1151
1152 /* If in full-sync mode, advance to the next disk sector before writing
1153 ** the master journal name. This is in case the previous page written to
1154 ** the journal has already been synced.
1155 */
1156 if( pPager->fullSync ){
1157 seekJournalHdr(pPager);
1158 }
1159 jrnlOff = pPager->journalOff;
1160 pPager->journalOff += (len+20);
1161
1162 rc = write32bits(pPager->jfd, jrnlOff, PAGER_MJ_PGNO(pPager));
1163 if( rc!=SQLITE_OK ) return rc;
1164 jrnlOff += 4;
1165
1166 rc = sqlite3OsWrite(pPager->jfd, zMaster, len, jrnlOff);
1167 if( rc!=SQLITE_OK ) return rc;
1168 jrnlOff += len;
1169
1170 put32bits(zBuf, len);
1171 put32bits(&zBuf[4], cksum);
1172 memcpy(&zBuf[8], aJournalMagic, sizeof(aJournalMagic));
1173 rc = sqlite3OsWrite(pPager->jfd, zBuf, 8+sizeof(aJournalMagic), jrnlOff);
1174 pPager->needSync = !pPager->noSync;
1175 return rc;
1176}
1177
1178/*
1179** Add or remove a page from the list of all pages that are in the
1180** statement journal.
1181**
1182** The Pager keeps a separate list of pages that are currently in
1183** the statement journal. This helps the sqlite3PagerStmtCommit()
1184** routine run MUCH faster for the common case where there are many
1185** pages in memory but only a few are in the statement journal.
1186*/
1187static void page_add_to_stmt_list(PgHdr *pPg){
1188 Pager *pPager = pPg->pPager;
1189 PgHistory *pHist = PGHDR_TO_HIST(pPg, pPager);
1190 assert( MEMDB );
1191 if( !pHist->inStmt ){
1192 assert( pHist->pPrevStmt==0 && pHist->pNextStmt==0 );
1193 if( pPager->pStmt ){
1194 PGHDR_TO_HIST(pPager->pStmt, pPager)->pPrevStmt = pPg;
1195 }
1196 pHist->pNextStmt = pPager->pStmt;
1197 pPager->pStmt = pPg;
1198 pHist->inStmt = 1;
1199 }
1200}
1201
1202/*
1203** Find a page in the hash table given its page number. Return
1204** a pointer to the page or NULL if not found.
1205*/
1206static PgHdr *pager_lookup(Pager *pPager, Pgno pgno){
1207 PgHdr *p;
1208 if( pPager->aHash==0 ) return 0;
1209 p = pPager->aHash[pgno & (pPager->nHash-1)];
1210 while( p && p->pgno!=pgno ){
1211 p = p->pNextHash;
1212 }
1213 return p;
1214}
1215
1216/*
1217** Clear the in-memory cache. This routine
1218** sets the state of the pager back to what it was when it was first
1219** opened. Any outstanding pages are invalidated and subsequent attempts
1220** to access those pages will likely result in a coredump.
1221*/
1222static void pager_reset(Pager *pPager){
1223 PgHdr *pPg, *pNext;
1224 if( pPager->errCode ) return;
1225 for(pPg=pPager->pAll; pPg; pPg=pNext){
1226 IOTRACE(("PGFREE %p %d\n", pPager, pPg->pgno));
1227 PAGER_INCR(sqlite3_pager_pgfree_count);
1228 pNext = pPg->pNextAll;
1229 lruListRemove(pPg);
1230 sqlite3_free(pPg->pData);
1231 sqlite3_free(pPg);
1232 }
1233 assert(pPager->lru.pFirst==0);
1234 assert(pPager->lru.pFirstSynced==0);
1235 assert(pPager->lru.pLast==0);
1236 pPager->pStmt = 0;
1237 pPager->pAll = 0;
1238 pPager->pDirty = 0;
1239 pPager->nHash = 0;
1240 sqlite3_free(pPager->aHash);
1241 pPager->nPage = 0;
1242 pPager->aHash = 0;
1243 pPager->nRef = 0;
1244}
1245
1246/*
1247** Unlock the database file.
1248**
1249** If the pager is currently in error state, discard the contents of
1250** the cache and reset the Pager structure internal state. If there is
1251** an open journal-file, then the next time a shared-lock is obtained
1252** on the pager file (by this or any other process), it will be
1253** treated as a hot-journal and rolled back.
1254*/
1255static void pager_unlock(Pager *pPager){
1256 if( !pPager->exclusiveMode ){
1257 if( !MEMDB ){
1258 if( pPager->fd->pMethods ){
1259 osUnlock(pPager->fd, NO_LOCK);
1260 }
1261 pPager->dbSize = -1;
1262 IOTRACE(("UNLOCK %p\n", pPager))
1263
1264 /* If Pager.errCode is set, the contents of the pager cache cannot be
1265 ** trusted. Now that the pager file is unlocked, the contents of the
1266 ** cache can be discarded and the error code safely cleared.
1267 */
1268 if( pPager->errCode ){
1269 pPager->errCode = SQLITE_OK;
1270 pager_reset(pPager);
1271 if( pPager->stmtOpen ){
1272 sqlite3OsClose(pPager->stfd);
1273 sqlite3_free(pPager->aInStmt);
1274 pPager->aInStmt = 0;
1275 }
1276 if( pPager->journalOpen ){
1277 sqlite3OsClose(pPager->jfd);
1278 pPager->journalOpen = 0;
1279 sqlite3_free(pPager->aInJournal);
1280 pPager->aInJournal = 0;
1281 }
1282 pPager->stmtOpen = 0;
1283 pPager->stmtInUse = 0;
1284 pPager->journalOff = 0;
1285 pPager->journalStarted = 0;
1286 pPager->stmtAutoopen = 0;
1287 pPager->origDbSize = 0;
1288 }
1289 }
1290
1291 if( !MEMDB || pPager->errCode==SQLITE_OK ){
1292 pPager->state = PAGER_UNLOCK;
1293 pPager->changeCountDone = 0;
1294 }
1295 }
1296}
1297
1298/*
1299** Execute a rollback if a transaction is active and unlock the
1300** database file. If the pager has already entered the error state,
1301** do not attempt the rollback.
1302*/
1303static void pagerUnlockAndRollback(Pager *p){
1304 assert( p->state>=PAGER_RESERVED || p->journalOpen==0 );
1305 if( p->errCode==SQLITE_OK && p->state>=PAGER_RESERVED ){
1306 sqlite3PagerRollback(p);
1307 }
1308 pager_unlock(p);
1309 assert( p->errCode || !p->journalOpen || (p->exclusiveMode&&!p->journalOff) );
1310 assert( p->errCode || !p->stmtOpen || p->exclusiveMode );
1311}
1312
1313/*
1314** This routine ends a transaction. A transaction is ended by either
1315** a COMMIT or a ROLLBACK.
1316**
1317** When this routine is called, the pager has the journal file open and
1318** a RESERVED or EXCLUSIVE lock on the database. This routine will release
1319** the database lock and acquires a SHARED lock in its place if that is
1320** the appropriate thing to do. Release locks usually is appropriate,
1321** unless we are in exclusive access mode or unless this is a
1322** COMMIT AND BEGIN or ROLLBACK AND BEGIN operation.
1323**
1324** The journal file is either deleted or truncated.
1325**
1326** TODO: Consider keeping the journal file open for temporary databases.
1327** This might give a performance improvement on windows where opening
1328** a file is an expensive operation.
1329*/
1330static int pager_end_transaction(Pager *pPager){
1331 PgHdr *pPg;
1332 int rc = SQLITE_OK;
1333 int rc2 = SQLITE_OK;
1334 assert( !MEMDB );
1335 if( pPager->state<PAGER_RESERVED ){
1336 return SQLITE_OK;
1337 }
1338 sqlite3PagerStmtCommit(pPager);
1339 if( pPager->stmtOpen && !pPager->exclusiveMode ){
1340 sqlite3OsClose(pPager->stfd);
1341 pPager->stmtOpen = 0;
1342 }
1343 if( pPager->journalOpen ){
1344 if( pPager->exclusiveMode
1345 && (rc = sqlite3OsTruncate(pPager->jfd, 0))==SQLITE_OK ){;
1346 pPager->journalOff = 0;
1347 pPager->journalStarted = 0;
1348 }else{
1349 sqlite3OsClose(pPager->jfd);
1350 pPager->journalOpen = 0;
1351 if( rc==SQLITE_OK ){
1352 rc = sqlite3OsDelete(pPager->pVfs, pPager->zJournal, 0);
1353 }
1354 }
1355 sqlite3_free( pPager->aInJournal );
1356 pPager->aInJournal = 0;
1357 for(pPg=pPager->pAll; pPg; pPg=pPg->pNextAll){
1358 pPg->inJournal = 0;
1359 pPg->dirty = 0;
1360 pPg->needSync = 0;
1361 pPg->alwaysRollback = 0;
1362#ifdef SQLITE_CHECK_PAGES
1363 pPg->pageHash = pager_pagehash(pPg);
1364#endif
1365 }
1366 pPager->pDirty = 0;
1367 pPager->dirtyCache = 0;
1368 pPager->nRec = 0;
1369 }else{
1370 assert( pPager->aInJournal==0 );
1371 assert( pPager->dirtyCache==0 || pPager->useJournal==0 );
1372 }
1373
1374 if( !pPager->exclusiveMode ){
1375 rc2 = osUnlock(pPager->fd, SHARED_LOCK);
1376 pPager->state = PAGER_SHARED;
1377 }else if( pPager->state==PAGER_SYNCED ){
1378 pPager->state = PAGER_EXCLUSIVE;
1379 }
1380 pPager->origDbSize = 0;
1381 pPager->setMaster = 0;
1382 pPager->needSync = 0;
1383 lruListSetFirstSynced(pPager);
1384 pPager->dbSize = -1;
1385
1386 return (rc==SQLITE_OK?rc2:rc);
1387}
1388
1389/*
1390** Compute and return a checksum for the page of data.
1391**
1392** This is not a real checksum. It is really just the sum of the
1393** random initial value and the page number. We experimented with
1394** a checksum of the entire data, but that was found to be too slow.
1395**
1396** Note that the page number is stored at the beginning of data and
1397** the checksum is stored at the end. This is important. If journal
1398** corruption occurs due to a power failure, the most likely scenario
1399** is that one end or the other of the record will be changed. It is
1400** much less likely that the two ends of the journal record will be
1401** correct and the middle be corrupt. Thus, this "checksum" scheme,
1402** though fast and simple, catches the mostly likely kind of corruption.
1403**
1404** FIX ME: Consider adding every 200th (or so) byte of the data to the
1405** checksum. That way if a single page spans 3 or more disk sectors and
1406** only the middle sector is corrupt, we will still have a reasonable
1407** chance of failing the checksum and thus detecting the problem.
1408*/
1409static u32 pager_cksum(Pager *pPager, const u8 *aData){
1410 u32 cksum = pPager->cksumInit;
1411 int i = pPager->pageSize-200;
1412 while( i>0 ){
1413 cksum += aData[i];
1414 i -= 200;
1415 }
1416 return cksum;
1417}
1418
1419/* Forward declaration */
1420static void makeClean(PgHdr*);
1421
1422/*
1423** Read a single page from the journal file opened on file descriptor
1424** jfd. Playback this one page.
1425**
1426** If useCksum==0 it means this journal does not use checksums. Checksums
1427** are not used in statement journals because statement journals do not
1428** need to survive power failures.
1429*/
1430static int pager_playback_one_page(
1431 Pager *pPager,
1432 sqlite3_file *jfd,
1433 i64 offset,
1434 int useCksum
1435){
1436 int rc;
1437 PgHdr *pPg; /* An existing page in the cache */
1438 Pgno pgno; /* The page number of a page in journal */
1439 u32 cksum; /* Checksum used for sanity checking */
1440 u8 *aData = (u8 *)pPager->pTmpSpace; /* Temp storage for a page */
1441
1442 /* useCksum should be true for the main journal and false for
1443 ** statement journals. Verify that this is always the case
1444 */
1445 assert( jfd == (useCksum ? pPager->jfd : pPager->stfd) );
1446 assert( aData );
1447
1448 rc = read32bits(jfd, offset, &pgno);
1449 if( rc!=SQLITE_OK ) return rc;
1450 rc = sqlite3OsRead(jfd, aData, pPager->pageSize, offset+4);
1451 if( rc!=SQLITE_OK ) return rc;
1452 pPager->journalOff += pPager->pageSize + 4;
1453
1454 /* Sanity checking on the page. This is more important that I originally
1455 ** thought. If a power failure occurs while the journal is being written,
1456 ** it could cause invalid data to be written into the journal. We need to
1457 ** detect this invalid data (with high probability) and ignore it.
1458 */
1459 if( pgno==0 || pgno==PAGER_MJ_PGNO(pPager) ){
1460 return SQLITE_DONE;
1461 }
1462 if( pgno>(unsigned)pPager->dbSize ){
1463 return SQLITE_OK;
1464 }
1465 if( useCksum ){
1466 rc = read32bits(jfd, offset+pPager->pageSize+4, &cksum);
1467 if( rc ) return rc;
1468 pPager->journalOff += 4;
1469 if( pager_cksum(pPager, aData)!=cksum ){
1470 return SQLITE_DONE;
1471 }
1472 }
1473
1474 assert( pPager->state==PAGER_RESERVED || pPager->state>=PAGER_EXCLUSIVE );
1475
1476 /* If the pager is in RESERVED state, then there must be a copy of this
1477 ** page in the pager cache. In this case just update the pager cache,
1478 ** not the database file. The page is left marked dirty in this case.
1479 **
1480 ** An exception to the above rule: If the database is in no-sync mode
1481 ** and a page is moved during an incremental vacuum then the page may
1482 ** not be in the pager cache. Later: if a malloc() or IO error occurs
1483 ** during a Movepage() call, then the page may not be in the cache
1484 ** either. So the condition described in the above paragraph is not
1485 ** assert()able.
1486 **
1487 ** If in EXCLUSIVE state, then we update the pager cache if it exists
1488 ** and the main file. The page is then marked not dirty.
1489 **
1490 ** Ticket #1171: The statement journal might contain page content that is
1491 ** different from the page content at the start of the transaction.
1492 ** This occurs when a page is changed prior to the start of a statement
1493 ** then changed again within the statement. When rolling back such a
1494 ** statement we must not write to the original database unless we know
1495 ** for certain that original page contents are synced into the main rollback
1496 ** journal. Otherwise, a power loss might leave modified data in the
1497 ** database file without an entry in the rollback journal that can
1498 ** restore the database to its original form. Two conditions must be
1499 ** met before writing to the database files. (1) the database must be
1500 ** locked. (2) we know that the original page content is fully synced
1501 ** in the main journal either because the page is not in cache or else
1502 ** the page is marked as needSync==0.
1503 */
1504 pPg = pager_lookup(pPager, pgno);
1505 PAGERTRACE4("PLAYBACK %d page %d hash(%08x)\n",
1506 PAGERID(pPager), pgno, pager_datahash(pPager->pageSize, aData));
1507 if( pPager->state>=PAGER_EXCLUSIVE && (pPg==0 || pPg->needSync==0) ){
1508 i64 offset = (pgno-1)*(i64)pPager->pageSize;
1509 rc = sqlite3OsWrite(pPager->fd, aData, pPager->pageSize, offset);
1510 if( pPg ){
1511 makeClean(pPg);
1512 }
1513 }
1514 if( pPg ){
1515 /* No page should ever be explicitly rolled back that is in use, except
1516 ** for page 1 which is held in use in order to keep the lock on the
1517 ** database active. However such a page may be rolled back as a result
1518 ** of an internal error resulting in an automatic call to
1519 ** sqlite3PagerRollback().
1520 */
1521 void *pData;
1522 /* assert( pPg->nRef==0 || pPg->pgno==1 ); */
1523 pData = PGHDR_TO_DATA(pPg);
1524 memcpy(pData, aData, pPager->pageSize);
1525 if( pPager->xReiniter ){
1526 pPager->xReiniter(pPg, pPager->pageSize);
1527 }
1528#ifdef SQLITE_CHECK_PAGES
1529 pPg->pageHash = pager_pagehash(pPg);
1530#endif
1531 /* If this was page 1, then restore the value of Pager.dbFileVers.
1532 ** Do this before any decoding. */
1533 if( pgno==1 ){
1534 memcpy(&pPager->dbFileVers, &((u8*)pData)[24],sizeof(pPager->dbFileVers));
1535 }
1536
1537 /* Decode the page just read from disk */
1538 CODEC1(pPager, pData, pPg->pgno, 3);
1539 }
1540 return rc;
1541}
1542
1543/*
1544** Parameter zMaster is the name of a master journal file. A single journal
1545** file that referred to the master journal file has just been rolled back.
1546** This routine checks if it is possible to delete the master journal file,
1547** and does so if it is.
1548**
1549** Argument zMaster may point to Pager.pTmpSpace. So that buffer is not
1550** available for use within this function.
1551**
1552**
1553** The master journal file contains the names of all child journals.
1554** To tell if a master journal can be deleted, check to each of the
1555** children. If all children are either missing or do not refer to
1556** a different master journal, then this master journal can be deleted.
1557*/
1558static int pager_delmaster(Pager *pPager, const char *zMaster){
1559 sqlite3_vfs *pVfs = pPager->pVfs;
1560 int rc;
1561 int master_open = 0;
1562 sqlite3_file *pMaster;
1563 sqlite3_file *pJournal;
1564 char *zMasterJournal = 0; /* Contents of master journal file */
1565 i64 nMasterJournal; /* Size of master journal file */
1566
1567 /* Open the master journal file exclusively in case some other process
1568 ** is running this routine also. Not that it makes too much difference.
1569 */
1570 pMaster = (sqlite3_file *)sqlite3_malloc(pVfs->szOsFile * 2);
1571 pJournal = (sqlite3_file *)(((u8 *)pMaster) + pVfs->szOsFile);
1572 if( !pMaster ){
1573 rc = SQLITE_NOMEM;
1574 }else{
1575 int flags = (SQLITE_OPEN_READONLY|SQLITE_OPEN_MASTER_JOURNAL);
1576 rc = sqlite3OsOpen(pVfs, zMaster, pMaster, flags, 0);
1577 }
1578 if( rc!=SQLITE_OK ) goto delmaster_out;
1579 master_open = 1;
1580
1581 rc = sqlite3OsFileSize(pMaster, &nMasterJournal);
1582 if( rc!=SQLITE_OK ) goto delmaster_out;
1583
1584 if( nMasterJournal>0 ){
1585 char *zJournal;
1586 char *zMasterPtr = 0;
1587 int nMasterPtr = pPager->pVfs->mxPathname+1;
1588
1589 /* Load the entire master journal file into space obtained from
1590 ** sqlite3_malloc() and pointed to by zMasterJournal.
1591 */
1592 zMasterJournal = (char *)sqlite3_malloc(nMasterJournal + nMasterPtr);
1593 if( !zMasterJournal ){
1594 rc = SQLITE_NOMEM;
1595 goto delmaster_out;
1596 }
1597 zMasterPtr = &zMasterJournal[nMasterJournal];
1598 rc = sqlite3OsRead(pMaster, zMasterJournal, nMasterJournal, 0);
1599 if( rc!=SQLITE_OK ) goto delmaster_out;
1600
1601 zJournal = zMasterJournal;
1602 while( (zJournal-zMasterJournal)<nMasterJournal ){
1603 if( sqlite3OsAccess(pVfs, zJournal, SQLITE_ACCESS_EXISTS) ){
1604 /* One of the journals pointed to by the master journal exists.
1605 ** Open it and check if it points at the master journal. If
1606 ** so, return without deleting the master journal file.
1607 */
1608 int c;
1609 int flags = (SQLITE_OPEN_READONLY|SQLITE_OPEN_MAIN_JOURNAL);
1610 rc = sqlite3OsOpen(pVfs, zJournal, pJournal, flags, 0);
1611 if( rc!=SQLITE_OK ){
1612 goto delmaster_out;
1613 }
1614
1615 rc = readMasterJournal(pJournal, zMasterPtr, nMasterPtr);
1616 sqlite3OsClose(pJournal);
1617 if( rc!=SQLITE_OK ){
1618 goto delmaster_out;
1619 }
1620
1621 c = zMasterPtr[0]!=0 && strcmp(zMasterPtr, zMaster)==0;
1622 if( c ){
1623 /* We have a match. Do not delete the master journal file. */
1624 goto delmaster_out;
1625 }
1626 }
1627 zJournal += (strlen(zJournal)+1);
1628 }
1629 }
1630
1631 rc = sqlite3OsDelete(pVfs, zMaster, 0);
1632
1633delmaster_out:
1634 if( zMasterJournal ){
1635 sqlite3_free(zMasterJournal);
1636 }
1637 if( master_open ){
1638 sqlite3OsClose(pMaster);
1639 }
1640 sqlite3_free(pMaster);
1641 return rc;
1642}
1643
1644
1645static void pager_truncate_cache(Pager *pPager);
1646
1647/*
1648** Truncate the main file of the given pager to the number of pages
1649** indicated. Also truncate the cached representation of the file.
1650*/
1651static int pager_truncate(Pager *pPager, int nPage){
1652 int rc = SQLITE_OK;
1653 if( pPager->state>=PAGER_EXCLUSIVE && pPager->fd->pMethods ){
1654 rc = sqlite3OsTruncate(pPager->fd, pPager->pageSize*(i64)nPage);
1655 }
1656 if( rc==SQLITE_OK ){
1657 pPager->dbSize = nPage;
1658 pager_truncate_cache(pPager);
1659 }
1660 return rc;
1661}
1662
1663/*
1664** Set the sectorSize for the given pager.
1665**
1666** The sector size is the larger of the sector size reported
1667** by sqlite3OsSectorSize() and the pageSize.
1668*/
1669static void setSectorSize(Pager *pPager){
1670 assert(pPager->fd->pMethods||pPager->tempFile);
1671 if( !pPager->tempFile ){
1672 /* Sector size doesn't matter for temporary files. Also, the file
1673 ** may not have been opened yet, in whcih case the OsSectorSize()
1674 ** call will segfault.
1675 */
1676 pPager->sectorSize = sqlite3OsSectorSize(pPager->fd);
1677 }
1678 if( pPager->sectorSize<pPager->pageSize ){
1679 pPager->sectorSize = pPager->pageSize;
1680 }
1681}
1682
1683/*
1684** Playback the journal and thus restore the database file to
1685** the state it was in before we started making changes.
1686**
1687** The journal file format is as follows:
1688**
1689** (1) 8 byte prefix. A copy of aJournalMagic[].
1690** (2) 4 byte big-endian integer which is the number of valid page records
1691** in the journal. If this value is 0xffffffff, then compute the
1692** number of page records from the journal size.
1693** (3) 4 byte big-endian integer which is the initial value for the
1694** sanity checksum.
1695** (4) 4 byte integer which is the number of pages to truncate the
1696** database to during a rollback.
1697** (5) 4 byte integer which is the number of bytes in the master journal
1698** name. The value may be zero (indicate that there is no master
1699** journal.)
1700** (6) N bytes of the master journal name. The name will be nul-terminated
1701** and might be shorter than the value read from (5). If the first byte
1702** of the name is \000 then there is no master journal. The master
1703** journal name is stored in UTF-8.
1704** (7) Zero or more pages instances, each as follows:
1705** + 4 byte page number.
1706** + pPager->pageSize bytes of data.
1707** + 4 byte checksum
1708**
1709** When we speak of the journal header, we mean the first 6 items above.
1710** Each entry in the journal is an instance of the 7th item.
1711**
1712** Call the value from the second bullet "nRec". nRec is the number of
1713** valid page entries in the journal. In most cases, you can compute the
1714** value of nRec from the size of the journal file. But if a power
1715** failure occurred while the journal was being written, it could be the
1716** case that the size of the journal file had already been increased but
1717** the extra entries had not yet made it safely to disk. In such a case,
1718** the value of nRec computed from the file size would be too large. For
1719** that reason, we always use the nRec value in the header.
1720**
1721** If the nRec value is 0xffffffff it means that nRec should be computed
1722** from the file size. This value is used when the user selects the
1723** no-sync option for the journal. A power failure could lead to corruption
1724** in this case. But for things like temporary table (which will be
1725** deleted when the power is restored) we don't care.
1726**
1727** If the file opened as the journal file is not a well-formed
1728** journal file then all pages up to the first corrupted page are rolled
1729** back (or no pages if the journal header is corrupted). The journal file
1730** is then deleted and SQLITE_OK returned, just as if no corruption had
1731** been encountered.
1732**
1733** If an I/O or malloc() error occurs, the journal-file is not deleted
1734** and an error code is returned.
1735*/
1736static int pager_playback(Pager *pPager, int isHot){
1737 sqlite3_vfs *pVfs = pPager->pVfs;
1738 i64 szJ; /* Size of the journal file in bytes */
1739 u32 nRec; /* Number of Records in the journal */
1740 int i; /* Loop counter */
1741 Pgno mxPg = 0; /* Size of the original file in pages */
1742 int rc; /* Result code of a subroutine */
1743 char *zMaster = 0; /* Name of master journal file if any */
1744
1745 /* Figure out how many records are in the journal. Abort early if
1746 ** the journal is empty.
1747 */
1748 assert( pPager->journalOpen );
1749 rc = sqlite3OsFileSize(pPager->jfd, &szJ);
1750 if( rc!=SQLITE_OK || szJ==0 ){
1751 goto end_playback;
1752 }
1753
1754 /* Read the master journal name from the journal, if it is present.
1755 ** If a master journal file name is specified, but the file is not
1756 ** present on disk, then the journal is not hot and does not need to be
1757 ** played back.
1758 */
1759 zMaster = pPager->pTmpSpace;
1760 rc = readMasterJournal(pPager->jfd, zMaster, pPager->pVfs->mxPathname+1);
1761 assert( rc!=SQLITE_DONE );
1762 if( rc!=SQLITE_OK
1763 || (zMaster[0] && !sqlite3OsAccess(pVfs, zMaster, SQLITE_ACCESS_EXISTS))
1764 ){
1765 zMaster = 0;
1766 if( rc==SQLITE_DONE ) rc = SQLITE_OK;
1767 goto end_playback;
1768 }
1769 pPager->journalOff = 0;
1770 zMaster = 0;
1771
1772 /* This loop terminates either when the readJournalHdr() call returns
1773 ** SQLITE_DONE or an IO error occurs. */
1774 while( 1 ){
1775
1776 /* Read the next journal header from the journal file. If there are
1777 ** not enough bytes left in the journal file for a complete header, or
1778 ** it is corrupted, then a process must of failed while writing it.
1779 ** This indicates nothing more needs to be rolled back.
1780 */
1781 rc = readJournalHdr(pPager, szJ, &nRec, &mxPg);
1782 if( rc!=SQLITE_OK ){
1783 if( rc==SQLITE_DONE ){
1784 rc = SQLITE_OK;
1785 }
1786 goto end_playback;
1787 }
1788
1789 /* If nRec is 0xffffffff, then this journal was created by a process
1790 ** working in no-sync mode. This means that the rest of the journal
1791 ** file consists of pages, there are no more journal headers. Compute
1792 ** the value of nRec based on this assumption.
1793 */
1794 if( nRec==0xffffffff ){
1795 assert( pPager->journalOff==JOURNAL_HDR_SZ(pPager) );
1796 nRec = (szJ - JOURNAL_HDR_SZ(pPager))/JOURNAL_PG_SZ(pPager);
1797 }
1798
1799 /* If nRec is 0 and this rollback is of a transaction created by this
1800 ** process and if this is the final header in the journal, then it means
1801 ** that this part of the journal was being filled but has not yet been
1802 ** synced to disk. Compute the number of pages based on the remaining
1803 ** size of the file.
1804 **
1805 ** The third term of the test was added to fix ticket #2565.
1806 */
1807 if( nRec==0 && !isHot &&
1808 pPager->journalHdr+JOURNAL_HDR_SZ(pPager)==pPager->journalOff ){
1809 nRec = (szJ - pPager->journalOff) / JOURNAL_PG_SZ(pPager);
1810 }
1811
1812 /* If this is the first header read from the journal, truncate the
1813 ** database file back to it's original size.
1814 */
1815 if( pPager->journalOff==JOURNAL_HDR_SZ(pPager) ){
1816 rc = pager_truncate(pPager, mxPg);
1817 if( rc!=SQLITE_OK ){
1818 goto end_playback;
1819 }
1820 }
1821
1822 /* Copy original pages out of the journal and back into the database file.
1823 */
1824 for(i=0; i<nRec; i++){
1825 rc = pager_playback_one_page(pPager, pPager->jfd, pPager->journalOff, 1);
1826 if( rc!=SQLITE_OK ){
1827 if( rc==SQLITE_DONE ){
1828 rc = SQLITE_OK;
1829 pPager->journalOff = szJ;
1830 break;
1831 }else{
1832 goto end_playback;
1833 }
1834 }
1835 }
1836 }
1837 /*NOTREACHED*/
1838 assert( 0 );
1839
1840end_playback:
1841 if( rc==SQLITE_OK ){
1842 zMaster = pPager->pTmpSpace;
1843 rc = readMasterJournal(pPager->jfd, zMaster, pPager->pVfs->mxPathname+1);
1844 }
1845 if( rc==SQLITE_OK ){
1846 rc = pager_end_transaction(pPager);
1847 }
1848 if( rc==SQLITE_OK && zMaster[0] ){
1849 /* If there was a master journal and this routine will return success,
1850 ** see if it is possible to delete the master journal.
1851 */
1852 rc = pager_delmaster(pPager, zMaster);
1853 }
1854
1855 /* The Pager.sectorSize variable may have been updated while rolling
1856 ** back a journal created by a process with a different sector size
1857 ** value. Reset it to the correct value for this process.
1858 */
1859 setSectorSize(pPager);
1860 return rc;
1861}
1862
1863/*
1864** Playback the statement journal.
1865**
1866** This is similar to playing back the transaction journal but with
1867** a few extra twists.
1868**
1869** (1) The number of pages in the database file at the start of
1870** the statement is stored in pPager->stmtSize, not in the
1871** journal file itself.
1872**
1873** (2) In addition to playing back the statement journal, also
1874** playback all pages of the transaction journal beginning
1875** at offset pPager->stmtJSize.
1876*/
1877static int pager_stmt_playback(Pager *pPager){
1878 i64 szJ; /* Size of the full journal */
1879 i64 hdrOff;
1880 int nRec; /* Number of Records */
1881 int i; /* Loop counter */
1882 int rc;
1883
1884 szJ = pPager->journalOff;
1885#ifndef NDEBUG
1886 {
1887 i64 os_szJ;
1888 rc = sqlite3OsFileSize(pPager->jfd, &os_szJ);
1889 if( rc!=SQLITE_OK ) return rc;
1890 assert( szJ==os_szJ );
1891 }
1892#endif
1893
1894 /* Set hdrOff to be the offset just after the end of the last journal
1895 ** page written before the first journal-header for this statement
1896 ** transaction was written, or the end of the file if no journal
1897 ** header was written.
1898 */
1899 hdrOff = pPager->stmtHdrOff;
1900 assert( pPager->fullSync || !hdrOff );
1901 if( !hdrOff ){
1902 hdrOff = szJ;
1903 }
1904
1905 /* Truncate the database back to its original size.
1906 */
1907 rc = pager_truncate(pPager, pPager->stmtSize);
1908 assert( pPager->state>=PAGER_SHARED );
1909
1910 /* Figure out how many records are in the statement journal.
1911 */
1912 assert( pPager->stmtInUse && pPager->journalOpen );
1913 nRec = pPager->stmtNRec;
1914
1915 /* Copy original pages out of the statement journal and back into the
1916 ** database file. Note that the statement journal omits checksums from
1917 ** each record since power-failure recovery is not important to statement
1918 ** journals.
1919 */
1920 for(i=0; i<nRec; i++){
1921 i64 offset = i*(4+pPager->pageSize);
1922 rc = pager_playback_one_page(pPager, pPager->stfd, offset, 0);
1923 assert( rc!=SQLITE_DONE );
1924 if( rc!=SQLITE_OK ) goto end_stmt_playback;
1925 }
1926
1927 /* Now roll some pages back from the transaction journal. Pager.stmtJSize
1928 ** was the size of the journal file when this statement was started, so
1929 ** everything after that needs to be rolled back, either into the
1930 ** database, the memory cache, or both.
1931 **
1932 ** If it is not zero, then Pager.stmtHdrOff is the offset to the start
1933 ** of the first journal header written during this statement transaction.
1934 */
1935 pPager->journalOff = pPager->stmtJSize;
1936 pPager->cksumInit = pPager->stmtCksum;
1937 while( pPager->journalOff < hdrOff ){
1938 rc = pager_playback_one_page(pPager, pPager->jfd, pPager->journalOff, 1);
1939 assert( rc!=SQLITE_DONE );
1940 if( rc!=SQLITE_OK ) goto end_stmt_playback;
1941 }
1942
1943 while( pPager->journalOff < szJ ){
1944 u32 nJRec; /* Number of Journal Records */
1945 u32 dummy;
1946 rc = readJournalHdr(pPager, szJ, &nJRec, &dummy);
1947 if( rc!=SQLITE_OK ){
1948 assert( rc!=SQLITE_DONE );
1949 goto end_stmt_playback;
1950 }
1951 if( nJRec==0 ){
1952 nJRec = (szJ - pPager->journalOff) / (pPager->pageSize+8);
1953 }
1954 for(i=nJRec-1; i>=0 && pPager->journalOff < szJ; i--){
1955 rc = pager_playback_one_page(pPager, pPager->jfd, pPager->journalOff, 1);
1956 assert( rc!=SQLITE_DONE );
1957 if( rc!=SQLITE_OK ) goto end_stmt_playback;
1958 }
1959 }
1960
1961 pPager->journalOff = szJ;
1962
1963end_stmt_playback:
1964 if( rc==SQLITE_OK) {
1965 pPager->journalOff = szJ;
1966 /* pager_reload_cache(pPager); */
1967 }
1968 return rc;
1969}
1970
1971/*
1972** Change the maximum number of in-memory pages that are allowed.
1973*/
1974void sqlite3PagerSetCachesize(Pager *pPager, int mxPage){
1975 if( mxPage>10 ){
1976 pPager->mxPage = mxPage;
1977 }else{
1978 pPager->mxPage = 10;
1979 }
1980}
1981
1982/*
1983** Adjust the robustness of the database to damage due to OS crashes
1984** or power failures by changing the number of syncs()s when writing
1985** the rollback journal. There are three levels:
1986**
1987** OFF sqlite3OsSync() is never called. This is the default
1988** for temporary and transient files.
1989**
1990** NORMAL The journal is synced once before writes begin on the
1991** database. This is normally adequate protection, but
1992** it is theoretically possible, though very unlikely,
1993** that an inopertune power failure could leave the journal
1994** in a state which would cause damage to the database
1995** when it is rolled back.
1996**
1997** FULL The journal is synced twice before writes begin on the
1998** database (with some additional information - the nRec field
1999** of the journal header - being written in between the two
2000** syncs). If we assume that writing a
2001** single disk sector is atomic, then this mode provides
2002** assurance that the journal will not be corrupted to the
2003** point of causing damage to the database during rollback.
2004**
2005** Numeric values associated with these states are OFF==1, NORMAL=2,
2006** and FULL=3.
2007*/
2008#ifndef SQLITE_OMIT_PAGER_PRAGMAS
2009void sqlite3PagerSetSafetyLevel(Pager *pPager, int level, int full_fsync){
2010 pPager->noSync = level==1 || pPager->tempFile;
2011 pPager->fullSync = level==3 && !pPager->tempFile;
2012 pPager->sync_flags = (full_fsync?SQLITE_SYNC_FULL:SQLITE_SYNC_NORMAL);
2013 if( pPager->noSync ) pPager->needSync = 0;
2014}
2015#endif
2016
2017/*
2018** The following global variable is incremented whenever the library
2019** attempts to open a temporary file. This information is used for
2020** testing and analysis only.
2021*/
2022#ifdef SQLITE_TEST
2023int sqlite3_opentemp_count = 0;
2024#endif
2025
2026/*
2027** Open a temporary file.
2028**
2029** Write the file descriptor into *fd. Return SQLITE_OK on success or some
2030** other error code if we fail. The OS will automatically delete the temporary
2031** file when it is closed.
2032*/
2033static int sqlite3PagerOpentemp(
2034 sqlite3_vfs *pVfs, /* The virtual file system layer */
2035 sqlite3_file *pFile, /* Write the file descriptor here */
2036 char *zFilename, /* Name of the file. Might be NULL */
2037 int vfsFlags /* Flags passed through to the VFS */
2038){
2039 int rc;
2040 assert( zFilename!=0 );
2041
2042#ifdef SQLITE_TEST
2043 sqlite3_opentemp_count++; /* Used for testing and analysis only */
2044#endif
2045
2046 vfsFlags |= SQLITE_OPEN_READWRITE | SQLITE_OPEN_CREATE |
2047 SQLITE_OPEN_EXCLUSIVE | SQLITE_OPEN_DELETEONCLOSE;
2048 rc = sqlite3OsOpen(pVfs, zFilename, pFile, vfsFlags, 0);
2049 assert( rc!=SQLITE_OK || pFile->pMethods );
2050 return rc;
2051}
2052
2053/*
2054** Create a new page cache and put a pointer to the page cache in *ppPager.
2055** The file to be cached need not exist. The file is not locked until
2056** the first call to sqlite3PagerGet() and is only held open until the
2057** last page is released using sqlite3PagerUnref().
2058**
2059** If zFilename is NULL then a randomly-named temporary file is created
2060** and used as the file to be cached. The file will be deleted
2061** automatically when it is closed.
2062**
2063** If zFilename is ":memory:" then all information is held in cache.
2064** It is never written to disk. This can be used to implement an
2065** in-memory database.
2066*/
2067int sqlite3PagerOpen(
2068 sqlite3_vfs *pVfs, /* The virtual file system to use */
2069 Pager **ppPager, /* Return the Pager structure here */
2070 const char *zFilename, /* Name of the database file to open */
2071 int nExtra, /* Extra bytes append to each in-memory page */
2072 int flags, /* flags controlling this file */
2073 int vfsFlags /* flags passed through to sqlite3_vfs.xOpen() */
2074){
2075 u8 *pPtr;
2076 Pager *pPager = 0;
2077 int rc = SQLITE_OK;
2078 int i;
2079 int tempFile = 0;
2080 int memDb = 0;
2081 int readOnly = 0;
2082 int useJournal = (flags & PAGER_OMIT_JOURNAL)==0;
2083 int noReadlock = (flags & PAGER_NO_READLOCK)!=0;
2084 int journalFileSize = sqlite3JournalSize(pVfs);
2085 int nDefaultPage = SQLITE_DEFAULT_PAGE_SIZE;
2086 char *zPathname;
2087 int nPathname;
2088
2089 /* The default return is a NULL pointer */
2090 *ppPager = 0;
2091
2092 /* Compute the full pathname */
2093 nPathname = pVfs->mxPathname+1;
2094 zPathname = sqlite3_malloc(nPathname);
2095 if( zPathname==0 ){
2096 return SQLITE_NOMEM;
2097 }
2098 if( zFilename && zFilename[0] ){
2099#ifndef SQLITE_OMIT_MEMORYDB
2100 if( strcmp(zFilename,":memory:")==0 ){
2101 memDb = 1;
2102 zPathname[0] = 0;
2103 }else
2104#endif
2105 {
2106 rc = sqlite3OsFullPathname(pVfs, zFilename, nPathname, zPathname);
2107 }
2108 }else{
2109 rc = sqlite3OsGetTempname(pVfs, nPathname, zPathname);
2110 }
2111 if( rc!=SQLITE_OK ){
2112 sqlite3_free(zPathname);
2113 return rc;
2114 }
2115 nPathname = strlen(zPathname);
2116
2117 /* Allocate memory for the pager structure */
2118 pPager = sqlite3MallocZero(
2119 sizeof(*pPager) + /* Pager structure */
2120 journalFileSize + /* The journal file structure */
2121 pVfs->szOsFile * 2 + /* The db and stmt journal files */
2122 4*nPathname + 40 /* zFilename, zDirectory, zJournal, zStmtJrnl */
2123 );
2124 if( !pPager ){
2125 sqlite3_free(zPathname);
2126 return SQLITE_NOMEM;
2127 }
2128 pPtr = (u8 *)&pPager[1];
2129 pPager->vfsFlags = vfsFlags;
2130 pPager->fd = (sqlite3_file*)&pPtr[pVfs->szOsFile*0];
2131 pPager->stfd = (sqlite3_file*)&pPtr[pVfs->szOsFile*1];
2132 pPager->jfd = (sqlite3_file*)&pPtr[pVfs->szOsFile*2];
2133 pPager->zFilename = (char*)&pPtr[pVfs->szOsFile*2+journalFileSize];
2134 pPager->zDirectory = &pPager->zFilename[nPathname+1];
2135 pPager->zJournal = &pPager->zDirectory[nPathname+1];
2136 pPager->zStmtJrnl = &pPager->zJournal[nPathname+10];
2137 pPager->pVfs = pVfs;
2138 memcpy(pPager->zFilename, zPathname, nPathname+1);
2139 sqlite3_free(zPathname);
2140
2141 /* Open the pager file.
2142 */
2143 if( zFilename && zFilename[0] && !memDb ){
2144 if( nPathname>(pVfs->mxPathname - sizeof("-journal")) ){
2145 rc = SQLITE_CANTOPEN;
2146 }else{
2147 int fout = 0;
2148 rc = sqlite3OsOpen(pVfs, pPager->zFilename, pPager->fd,
2149 pPager->vfsFlags, &fout);
2150 readOnly = (fout&SQLITE_OPEN_READONLY);
2151
2152 /* If the file was successfully opened for read/write access,
2153 ** choose a default page size in case we have to create the
2154 ** database file. The default page size is the maximum of:
2155 **
2156 ** + SQLITE_DEFAULT_PAGE_SIZE,
2157 ** + The value returned by sqlite3OsSectorSize()
2158 ** + The largest page size that can be written atomically.
2159 */
2160 if( rc==SQLITE_OK && !readOnly ){
2161 int iSectorSize = sqlite3OsSectorSize(pPager->fd);
2162 if( nDefaultPage<iSectorSize ){
2163 nDefaultPage = iSectorSize;
2164 }
2165#ifdef SQLITE_ENABLE_ATOMIC_WRITE
2166 {
2167 int iDc = sqlite3OsDeviceCharacteristics(pPager->fd);
2168 int ii;
2169 assert(SQLITE_IOCAP_ATOMIC512==(512>>8));
2170 assert(SQLITE_IOCAP_ATOMIC64K==(65536>>8));
2171 assert(SQLITE_MAX_DEFAULT_PAGE_SIZE<=65536);
2172 for(ii=nDefaultPage; ii<=SQLITE_MAX_DEFAULT_PAGE_SIZE; ii=ii*2){
2173 if( iDc&(SQLITE_IOCAP_ATOMIC|(ii>>8)) ) nDefaultPage = ii;
2174 }
2175 }
2176#endif
2177 if( nDefaultPage>SQLITE_MAX_DEFAULT_PAGE_SIZE ){
2178 nDefaultPage = SQLITE_MAX_DEFAULT_PAGE_SIZE;
2179 }
2180 }
2181 }
2182 }else if( !memDb ){
2183 /* If a temporary file is requested, it is not opened immediately.
2184 ** In this case we accept the default page size and delay actually
2185 ** opening the file until the first call to OsWrite().
2186 */
2187 tempFile = 1;
2188 pPager->state = PAGER_EXCLUSIVE;
2189 }
2190
2191 if( pPager && rc==SQLITE_OK ){
2192 pPager->pTmpSpace = (char *)sqlite3_malloc(nDefaultPage);
2193 }
2194
2195 /* If an error occured in either of the blocks above.
2196 ** Free the Pager structure and close the file.
2197 ** Since the pager is not allocated there is no need to set
2198 ** any Pager.errMask variables.
2199 */
2200 if( !pPager || !pPager->pTmpSpace ){
2201 sqlite3OsClose(pPager->fd);
2202 sqlite3_free(pPager);
2203 return ((rc==SQLITE_OK)?SQLITE_NOMEM:rc);
2204 }
2205
2206 PAGERTRACE3("OPEN %d %s\n", FILEHANDLEID(pPager->fd), pPager->zFilename);
2207 IOTRACE(("OPEN %p %s\n", pPager, pPager->zFilename))
2208
2209 /* Fill in Pager.zDirectory[] */
2210 memcpy(pPager->zDirectory, pPager->zFilename, nPathname+1);
2211 for(i=strlen(pPager->zDirectory); i>0 && pPager->zDirectory[i-1]!='/'; i--){}
2212 if( i>0 ) pPager->zDirectory[i-1] = 0;
2213
2214 /* Fill in Pager.zJournal[] and Pager.zStmtJrnl[] */
2215 memcpy(pPager->zJournal, pPager->zFilename, nPathname);
2216 memcpy(&pPager->zJournal[nPathname], "-journal", 9);
2217 memcpy(pPager->zStmtJrnl, pPager->zFilename, nPathname);
2218 memcpy(&pPager->zStmtJrnl[nPathname], "-stmtjrnl", 10);
2219
2220 /* pPager->journalOpen = 0; */
2221 pPager->useJournal = useJournal && !memDb;
2222 pPager->noReadlock = noReadlock && readOnly;
2223 /* pPager->stmtOpen = 0; */
2224 /* pPager->stmtInUse = 0; */
2225 /* pPager->nRef = 0; */
2226 pPager->dbSize = memDb-1;
2227 pPager->pageSize = nDefaultPage;
2228 /* pPager->stmtSize = 0; */
2229 /* pPager->stmtJSize = 0; */
2230 /* pPager->nPage = 0; */
2231 pPager->mxPage = 100;
2232 pPager->mxPgno = SQLITE_MAX_PAGE_COUNT;
2233 /* pPager->state = PAGER_UNLOCK; */
2234 assert( pPager->state == (tempFile ? PAGER_EXCLUSIVE : PAGER_UNLOCK) );
2235 /* pPager->errMask = 0; */
2236 pPager->tempFile = tempFile;
2237 assert( tempFile==PAGER_LOCKINGMODE_NORMAL
2238 || tempFile==PAGER_LOCKINGMODE_EXCLUSIVE );
2239 assert( PAGER_LOCKINGMODE_EXCLUSIVE==1 );
2240 pPager->exclusiveMode = tempFile;
2241 pPager->memDb = memDb;
2242 pPager->readOnly = readOnly;
2243 /* pPager->needSync = 0; */
2244 pPager->noSync = pPager->tempFile || !useJournal;
2245 pPager->fullSync = (pPager->noSync?0:1);
2246 pPager->sync_flags = SQLITE_SYNC_NORMAL;
2247 /* pPager->pFirst = 0; */
2248 /* pPager->pFirstSynced = 0; */
2249 /* pPager->pLast = 0; */
2250 pPager->nExtra = FORCE_ALIGNMENT(nExtra);
2251 assert(pPager->fd->pMethods||memDb||tempFile);
2252 if( !memDb ){
2253 setSectorSize(pPager);
2254 }
2255 /* pPager->pBusyHandler = 0; */
2256 /* memset(pPager->aHash, 0, sizeof(pPager->aHash)); */
2257 *ppPager = pPager;
2258#ifdef SQLITE_ENABLE_MEMORY_MANAGEMENT
2259 pPager->iInUseMM = 0;
2260 pPager->iInUseDB = 0;
2261 if( !memDb ){
2262 sqlite3_mutex *mutex = sqlite3_mutex_alloc(SQLITE_MUTEX_STATIC_MEM2);
2263 sqlite3_mutex_enter(mutex);
2264 pPager->pNext = sqlite3PagerList;
2265 if( sqlite3PagerList ){
2266 assert( sqlite3PagerList->pPrev==0 );
2267 sqlite3PagerList->pPrev = pPager;
2268 }
2269 pPager->pPrev = 0;
2270 sqlite3PagerList = pPager;
2271 sqlite3_mutex_leave(mutex);
2272 }
2273#endif
2274 return SQLITE_OK;
2275}
2276
2277/*
2278** Set the busy handler function.
2279*/
2280void sqlite3PagerSetBusyhandler(Pager *pPager, BusyHandler *pBusyHandler){
2281 pPager->pBusyHandler = pBusyHandler;
2282}
2283
2284/*
2285** Set the destructor for this pager. If not NULL, the destructor is called
2286** when the reference count on each page reaches zero. The destructor can
2287** be used to clean up information in the extra segment appended to each page.
2288**
2289** The destructor is not called as a result sqlite3PagerClose().
2290** Destructors are only called by sqlite3PagerUnref().
2291*/
2292void sqlite3PagerSetDestructor(Pager *pPager, void (*xDesc)(DbPage*,int)){
2293 pPager->xDestructor = xDesc;
2294}
2295
2296/*
2297** Set the reinitializer for this pager. If not NULL, the reinitializer
2298** is called when the content of a page in cache is restored to its original
2299** value as a result of a rollback. The callback gives higher-level code
2300** an opportunity to restore the EXTRA section to agree with the restored
2301** page data.
2302*/
2303void sqlite3PagerSetReiniter(Pager *pPager, void (*xReinit)(DbPage*,int)){
2304 pPager->xReiniter = xReinit;
2305}
2306
2307/*
2308** Set the page size to *pPageSize. If the suggest new page size is
2309** inappropriate, then an alternative page size is set to that
2310** value before returning.
2311*/
2312int sqlite3PagerSetPagesize(Pager *pPager, u16 *pPageSize){
2313 int rc = SQLITE_OK;
2314 u16 pageSize = *pPageSize;
2315 assert( pageSize==0 || (pageSize>=512 && pageSize<=SQLITE_MAX_PAGE_SIZE) );
2316 if( pageSize && pageSize!=pPager->pageSize
2317 && !pPager->memDb && pPager->nRef==0
2318 ){
2319 char *pNew = (char *)sqlite3_malloc(pageSize);
2320 if( !pNew ){
2321 rc = SQLITE_NOMEM;
2322 }else{
2323 pagerEnter(pPager);
2324 pager_reset(pPager);
2325 pPager->pageSize = pageSize;
2326 setSectorSize(pPager);
2327 sqlite3_free(pPager->pTmpSpace);
2328 pPager->pTmpSpace = pNew;
2329 pagerLeave(pPager);
2330 }
2331 }
2332 *pPageSize = pPager->pageSize;
2333 return rc;
2334}
2335
2336/*
2337** Attempt to set the maximum database page count if mxPage is positive.
2338** Make no changes if mxPage is zero or negative. And never reduce the
2339** maximum page count below the current size of the database.
2340**
2341** Regardless of mxPage, return the current maximum page count.
2342*/
2343int sqlite3PagerMaxPageCount(Pager *pPager, int mxPage){
2344 if( mxPage>0 ){
2345 pPager->mxPgno = mxPage;
2346 }
2347 sqlite3PagerPagecount(pPager);
2348 return pPager->mxPgno;
2349}
2350
2351/*
2352** The following set of routines are used to disable the simulated
2353** I/O error mechanism. These routines are used to avoid simulated
2354** errors in places where we do not care about errors.
2355**
2356** Unless -DSQLITE_TEST=1 is used, these routines are all no-ops
2357** and generate no code.
2358*/
2359#ifdef SQLITE_TEST
2360extern int sqlite3_io_error_pending;
2361extern int sqlite3_io_error_hit;
2362static int saved_cnt;
2363void disable_simulated_io_errors(void){
2364 saved_cnt = sqlite3_io_error_pending;
2365 sqlite3_io_error_pending = -1;
2366}
2367void enable_simulated_io_errors(void){
2368 sqlite3_io_error_pending = saved_cnt;
2369}
2370#else
2371# define disable_simulated_io_errors()
2372# define enable_simulated_io_errors()
2373#endif
2374
2375/*
2376** Read the first N bytes from the beginning of the file into memory
2377** that pDest points to.
2378**
2379** No error checking is done. The rational for this is that this function
2380** may be called even if the file does not exist or contain a header. In
2381** these cases sqlite3OsRead() will return an error, to which the correct
2382** response is to zero the memory at pDest and continue. A real IO error
2383** will presumably recur and be picked up later (Todo: Think about this).
2384*/
2385int sqlite3PagerReadFileheader(Pager *pPager, int N, unsigned char *pDest){
2386 int rc = SQLITE_OK;
2387 memset(pDest, 0, N);
2388 assert(MEMDB||pPager->fd->pMethods||pPager->tempFile);
2389 if( pPager->fd->pMethods ){
2390 IOTRACE(("DBHDR %p 0 %d\n", pPager, N))
2391 rc = sqlite3OsRead(pPager->fd, pDest, N, 0);
2392 if( rc==SQLITE_IOERR_SHORT_READ ){
2393 rc = SQLITE_OK;
2394 }
2395 }
2396 return rc;
2397}
2398
2399/*
2400** Return the total number of pages in the disk file associated with
2401** pPager.
2402**
2403** If the PENDING_BYTE lies on the page directly after the end of the
2404** file, then consider this page part of the file too. For example, if
2405** PENDING_BYTE is byte 4096 (the first byte of page 5) and the size of the
2406** file is 4096 bytes, 5 is returned instead of 4.
2407*/
2408int sqlite3PagerPagecount(Pager *pPager){
2409 i64 n = 0;
2410 int rc;
2411 assert( pPager!=0 );
2412 if( pPager->errCode ){
2413 return 0;
2414 }
2415 if( pPager->dbSize>=0 ){
2416 n = pPager->dbSize;
2417 } else {
2418 assert(pPager->fd->pMethods||pPager->tempFile);
2419 if( (pPager->fd->pMethods)
2420 && (rc = sqlite3OsFileSize(pPager->fd, &n))!=SQLITE_OK ){
2421 pPager->nRef++;
2422 pager_error(pPager, rc);
2423 pPager->nRef--;
2424 return 0;
2425 }
2426 if( n>0 && n<pPager->pageSize ){
2427 n = 1;
2428 }else{
2429 n /= pPager->pageSize;
2430 }
2431 if( pPager->state!=PAGER_UNLOCK ){
2432 pPager->dbSize = n;
2433 }
2434 }
2435 if( n==(PENDING_BYTE/pPager->pageSize) ){
2436 n++;
2437 }
2438 if( n>pPager->mxPgno ){
2439 pPager->mxPgno = n;
2440 }
2441 return n;
2442}
2443
2444
2445#ifndef SQLITE_OMIT_MEMORYDB
2446/*
2447** Clear a PgHistory block
2448*/
2449static void clearHistory(PgHistory *pHist){
2450 sqlite3_free(pHist->pOrig);
2451 sqlite3_free(pHist->pStmt);
2452 pHist->pOrig = 0;
2453 pHist->pStmt = 0;
2454}
2455#else
2456#define clearHistory(x)
2457#endif
2458
2459/*
2460** Forward declaration
2461*/
2462static int syncJournal(Pager*);
2463
2464/*
2465** Unlink pPg from it's hash chain. Also set the page number to 0 to indicate
2466** that the page is not part of any hash chain. This is required because the
2467** sqlite3PagerMovepage() routine can leave a page in the
2468** pNextFree/pPrevFree list that is not a part of any hash-chain.
2469*/
2470static void unlinkHashChain(Pager *pPager, PgHdr *pPg){
2471 if( pPg->pgno==0 ){
2472 assert( pPg->pNextHash==0 && pPg->pPrevHash==0 );
2473 return;
2474 }
2475 if( pPg->pNextHash ){
2476 pPg->pNextHash->pPrevHash = pPg->pPrevHash;
2477 }
2478 if( pPg->pPrevHash ){
2479 assert( pPager->aHash[pPg->pgno & (pPager->nHash-1)]!=pPg );
2480 pPg->pPrevHash->pNextHash = pPg->pNextHash;
2481 }else{
2482 int h = pPg->pgno & (pPager->nHash-1);
2483 pPager->aHash[h] = pPg->pNextHash;
2484 }
2485 if( MEMDB ){
2486 clearHistory(PGHDR_TO_HIST(pPg, pPager));
2487 }
2488 pPg->pgno = 0;
2489 pPg->pNextHash = pPg->pPrevHash = 0;
2490}
2491
2492/*
2493** Unlink a page from the free list (the list of all pages where nRef==0)
2494** and from its hash collision chain.
2495*/
2496static void unlinkPage(PgHdr *pPg){
2497 Pager *pPager = pPg->pPager;
2498
2499 /* Unlink from free page list */
2500 lruListRemove(pPg);
2501
2502 /* Unlink from the pgno hash table */
2503 unlinkHashChain(pPager, pPg);
2504}
2505
2506/*
2507** This routine is used to truncate the cache when a database
2508** is truncated. Drop from the cache all pages whose pgno is
2509** larger than pPager->dbSize and is unreferenced.
2510**
2511** Referenced pages larger than pPager->dbSize are zeroed.
2512**
2513** Actually, at the point this routine is called, it would be
2514** an error to have a referenced page. But rather than delete
2515** that page and guarantee a subsequent segfault, it seems better
2516** to zero it and hope that we error out sanely.
2517*/
2518static void pager_truncate_cache(Pager *pPager){
2519 PgHdr *pPg;
2520 PgHdr **ppPg;
2521 int dbSize = pPager->dbSize;
2522
2523 ppPg = &pPager->pAll;
2524 while( (pPg = *ppPg)!=0 ){
2525 if( pPg->pgno<=dbSize ){
2526 ppPg = &pPg->pNextAll;
2527 }else if( pPg->nRef>0 ){
2528 memset(PGHDR_TO_DATA(pPg), 0, pPager->pageSize);
2529 ppPg = &pPg->pNextAll;
2530 }else{
2531 *ppPg = pPg->pNextAll;
2532 IOTRACE(("PGFREE %p %d\n", pPager, pPg->pgno));
2533 PAGER_INCR(sqlite3_pager_pgfree_count);
2534 unlinkPage(pPg);
2535 makeClean(pPg);
2536 sqlite3_free(pPg->pData);
2537 sqlite3_free(pPg);
2538 pPager->nPage--;
2539 }
2540 }
2541}
2542
2543/*
2544** Try to obtain a lock on a file. Invoke the busy callback if the lock
2545** is currently not available. Repeat until the busy callback returns
2546** false or until the lock succeeds.
2547**
2548** Return SQLITE_OK on success and an error code if we cannot obtain
2549** the lock.
2550*/
2551static int pager_wait_on_lock(Pager *pPager, int locktype){
2552 int rc;
2553
2554 /* The OS lock values must be the same as the Pager lock values */
2555 assert( PAGER_SHARED==SHARED_LOCK );
2556 assert( PAGER_RESERVED==RESERVED_LOCK );
2557 assert( PAGER_EXCLUSIVE==EXCLUSIVE_LOCK );
2558
2559 /* If the file is currently unlocked then the size must be unknown */
2560 assert( pPager->state>=PAGER_SHARED || pPager->dbSize<0 || MEMDB );
2561
2562 if( pPager->state>=locktype ){
2563 rc = SQLITE_OK;
2564 }else{
2565 do {
2566 rc = sqlite3OsLock(pPager->fd, locktype);
2567 }while( rc==SQLITE_BUSY && sqlite3InvokeBusyHandler(pPager->pBusyHandler) );
2568 if( rc==SQLITE_OK ){
2569 pPager->state = locktype;
2570 IOTRACE(("LOCK %p %d\n", pPager, locktype))
2571 }
2572 }
2573 return rc;
2574}
2575
2576/*
2577** Truncate the file to the number of pages specified.
2578*/
2579int sqlite3PagerTruncate(Pager *pPager, Pgno nPage){
2580 int rc;
2581 assert( pPager->state>=PAGER_SHARED || MEMDB );
2582 sqlite3PagerPagecount(pPager);
2583 if( pPager->errCode ){
2584 rc = pPager->errCode;
2585 return rc;
2586 }
2587 if( nPage>=(unsigned)pPager->dbSize ){
2588 return SQLITE_OK;
2589 }
2590 if( MEMDB ){
2591 pPager->dbSize = nPage;
2592 pager_truncate_cache(pPager);
2593 return SQLITE_OK;
2594 }
2595 pagerEnter(pPager);
2596 rc = syncJournal(pPager);
2597 pagerLeave(pPager);
2598 if( rc!=SQLITE_OK ){
2599 return rc;
2600 }
2601
2602 /* Get an exclusive lock on the database before truncating. */
2603 pagerEnter(pPager);
2604 rc = pager_wait_on_lock(pPager, EXCLUSIVE_LOCK);
2605 pagerLeave(pPager);
2606 if( rc!=SQLITE_OK ){
2607 return rc;
2608 }
2609
2610 rc = pager_truncate(pPager, nPage);
2611 return rc;
2612}
2613
2614/*
2615** Shutdown the page cache. Free all memory and close all files.
2616**
2617** If a transaction was in progress when this routine is called, that
2618** transaction is rolled back. All outstanding pages are invalidated
2619** and their memory is freed. Any attempt to use a page associated
2620** with this page cache after this function returns will likely
2621** result in a coredump.
2622**
2623** This function always succeeds. If a transaction is active an attempt
2624** is made to roll it back. If an error occurs during the rollback
2625** a hot journal may be left in the filesystem but no error is returned
2626** to the caller.
2627*/
2628int sqlite3PagerClose(Pager *pPager){
2629#ifdef SQLITE_ENABLE_MEMORY_MANAGEMENT
2630 if( !MEMDB ){
2631 sqlite3_mutex *mutex = sqlite3_mutex_alloc(SQLITE_MUTEX_STATIC_MEM2);
2632 sqlite3_mutex_enter(mutex);
2633 if( pPager->pPrev ){
2634 pPager->pPrev->pNext = pPager->pNext;
2635 }else{
2636 sqlite3PagerList = pPager->pNext;
2637 }
2638 if( pPager->pNext ){
2639 pPager->pNext->pPrev = pPager->pPrev;
2640 }
2641 sqlite3_mutex_leave(mutex);
2642 }
2643#endif
2644
2645 disable_simulated_io_errors();
2646 pPager->errCode = 0;
2647 pPager->exclusiveMode = 0;
2648 pager_reset(pPager);
2649 pagerUnlockAndRollback(pPager);
2650 enable_simulated_io_errors();
2651 PAGERTRACE2("CLOSE %d\n", PAGERID(pPager));
2652 IOTRACE(("CLOSE %p\n", pPager))
2653 assert( pPager->errCode || (pPager->journalOpen==0 && pPager->stmtOpen==0) );
2654 if( pPager->journalOpen ){
2655 sqlite3OsClose(pPager->jfd);
2656 }
2657 sqlite3_free(pPager->aInJournal);
2658 if( pPager->stmtOpen ){
2659 sqlite3OsClose(pPager->stfd);
2660 }
2661 sqlite3OsClose(pPager->fd);
2662 /* Temp files are automatically deleted by the OS
2663 ** if( pPager->tempFile ){
2664 ** sqlite3OsDelete(pPager->zFilename);
2665 ** }
2666 */
2667
2668 sqlite3_free(pPager->aHash);
2669 sqlite3_free(pPager->pTmpSpace);
2670 sqlite3_free(pPager);
2671 return SQLITE_OK;
2672}
2673
2674#if !defined(NDEBUG) || defined(SQLITE_TEST)
2675/*
2676** Return the page number for the given page data.
2677*/
2678Pgno sqlite3PagerPagenumber(DbPage *p){
2679 return p->pgno;
2680}
2681#endif
2682
2683/*
2684** The page_ref() function increments the reference count for a page.
2685** If the page is currently on the freelist (the reference count is zero) then
2686** remove it from the freelist.
2687**
2688** For non-test systems, page_ref() is a macro that calls _page_ref()
2689** online of the reference count is zero. For test systems, page_ref()
2690** is a real function so that we can set breakpoints and trace it.
2691*/
2692static void _page_ref(PgHdr *pPg){
2693 if( pPg->nRef==0 ){
2694 /* The page is currently on the freelist. Remove it. */
2695 lruListRemove(pPg);
2696 pPg->pPager->nRef++;
2697 }
2698 pPg->nRef++;
2699 REFINFO(pPg);
2700}
2701#ifdef SQLITE_DEBUG
2702 static void page_ref(PgHdr *pPg){
2703 if( pPg->nRef==0 ){
2704 _page_ref(pPg);
2705 }else{
2706 pPg->nRef++;
2707 REFINFO(pPg);
2708 }
2709 }
2710#else
2711# define page_ref(P) ((P)->nRef==0?_page_ref(P):(void)(P)->nRef++)
2712#endif
2713
2714/*
2715** Increment the reference count for a page. The input pointer is
2716** a reference to the page data.
2717*/
2718int sqlite3PagerRef(DbPage *pPg){
2719 pagerEnter(pPg->pPager);
2720 page_ref(pPg);
2721 pagerLeave(pPg->pPager);
2722 return SQLITE_OK;
2723}
2724
2725/*
2726** Sync the journal. In other words, make sure all the pages that have
2727** been written to the journal have actually reached the surface of the
2728** disk. It is not safe to modify the original database file until after
2729** the journal has been synced. If the original database is modified before
2730** the journal is synced and a power failure occurs, the unsynced journal
2731** data would be lost and we would be unable to completely rollback the
2732** database changes. Database corruption would occur.
2733**
2734** This routine also updates the nRec field in the header of the journal.
2735** (See comments on the pager_playback() routine for additional information.)
2736** If the sync mode is FULL, two syncs will occur. First the whole journal
2737** is synced, then the nRec field is updated, then a second sync occurs.
2738**
2739** For temporary databases, we do not care if we are able to rollback
2740** after a power failure, so no sync occurs.
2741**
2742** If the IOCAP_SEQUENTIAL flag is set for the persistent media on which
2743** the database is stored, then OsSync() is never called on the journal
2744** file. In this case all that is required is to update the nRec field in
2745** the journal header.
2746**
2747** This routine clears the needSync field of every page current held in
2748** memory.
2749*/
2750static int syncJournal(Pager *pPager){
2751 PgHdr *pPg;
2752 int rc = SQLITE_OK;
2753
2754
2755 /* Sync the journal before modifying the main database
2756 ** (assuming there is a journal and it needs to be synced.)
2757 */
2758 if( pPager->needSync ){
2759 if( !pPager->tempFile ){
2760 int iDc = sqlite3OsDeviceCharacteristics(pPager->fd);
2761 assert( pPager->journalOpen );
2762
2763 /* assert( !pPager->noSync ); // noSync might be set if synchronous
2764 ** was turned off after the transaction was started. Ticket #615 */
2765#ifndef NDEBUG
2766 {
2767 /* Make sure the pPager->nRec counter we are keeping agrees
2768 ** with the nRec computed from the size of the journal file.
2769 */
2770 i64 jSz;
2771 rc = sqlite3OsFileSize(pPager->jfd, &jSz);
2772 if( rc!=0 ) return rc;
2773 assert( pPager->journalOff==jSz );
2774 }
2775#endif
2776 if( 0==(iDc&SQLITE_IOCAP_SAFE_APPEND) ){
2777 /* Write the nRec value into the journal file header. If in
2778 ** full-synchronous mode, sync the journal first. This ensures that
2779 ** all data has really hit the disk before nRec is updated to mark
2780 ** it as a candidate for rollback.
2781 **
2782 ** This is not required if the persistent media supports the
2783 ** SAFE_APPEND property. Because in this case it is not possible
2784 ** for garbage data to be appended to the file, the nRec field
2785 ** is populated with 0xFFFFFFFF when the journal header is written
2786 ** and never needs to be updated.
2787 */
2788 i64 jrnlOff;
2789 if( pPager->fullSync && 0==(iDc&SQLITE_IOCAP_SEQUENTIAL) ){
2790 PAGERTRACE2("SYNC journal of %d\n", PAGERID(pPager));
2791 IOTRACE(("JSYNC %p\n", pPager))
2792 rc = sqlite3OsSync(pPager->jfd, pPager->sync_flags);
2793 if( rc!=0 ) return rc;
2794 }
2795
2796 jrnlOff = pPager->journalHdr + sizeof(aJournalMagic);
2797 IOTRACE(("JHDR %p %lld %d\n", pPager, jrnlOff, 4));
2798 rc = write32bits(pPager->jfd, jrnlOff, pPager->nRec);
2799 if( rc ) return rc;
2800 }
2801 if( 0==(iDc&SQLITE_IOCAP_SEQUENTIAL) ){
2802 PAGERTRACE2("SYNC journal of %d\n", PAGERID(pPager));
2803 IOTRACE(("JSYNC %p\n", pPager))
2804 rc = sqlite3OsSync(pPager->jfd, pPager->sync_flags|
2805 (pPager->sync_flags==SQLITE_SYNC_FULL?SQLITE_SYNC_DATAONLY:0)
2806 );
2807 if( rc!=0 ) return rc;
2808 }
2809 pPager->journalStarted = 1;
2810 }
2811 pPager->needSync = 0;
2812
2813 /* Erase the needSync flag from every page.
2814 */
2815 for(pPg=pPager->pAll; pPg; pPg=pPg->pNextAll){
2816 pPg->needSync = 0;
2817 }
2818 lruListSetFirstSynced(pPager);
2819 }
2820
2821#ifndef NDEBUG
2822 /* If the Pager.needSync flag is clear then the PgHdr.needSync
2823 ** flag must also be clear for all pages. Verify that this
2824 ** invariant is true.
2825 */
2826 else{
2827 for(pPg=pPager->pAll; pPg; pPg=pPg->pNextAll){
2828 assert( pPg->needSync==0 );
2829 }
2830 assert( pPager->lru.pFirstSynced==pPager->lru.pFirst );
2831 }
2832#endif
2833
2834 return rc;
2835}
2836
2837/*
2838** Merge two lists of pages connected by pDirty and in pgno order.
2839** Do not both fixing the pPrevDirty pointers.
2840*/
2841static PgHdr *merge_pagelist(PgHdr *pA, PgHdr *pB){
2842 PgHdr result, *pTail;
2843 pTail = &result;
2844 while( pA && pB ){
2845 if( pA->pgno<pB->pgno ){
2846 pTail->pDirty = pA;
2847 pTail = pA;
2848 pA = pA->pDirty;
2849 }else{
2850 pTail->pDirty = pB;
2851 pTail = pB;
2852 pB = pB->pDirty;
2853 }
2854 }
2855 if( pA ){
2856 pTail->pDirty = pA;
2857 }else if( pB ){
2858 pTail->pDirty = pB;
2859 }else{
2860 pTail->pDirty = 0;
2861 }
2862 return result.pDirty;
2863}
2864
2865/*
2866** Sort the list of pages in accending order by pgno. Pages are
2867** connected by pDirty pointers. The pPrevDirty pointers are
2868** corrupted by this sort.
2869*/
2870#define N_SORT_BUCKET_ALLOC 25
2871#define N_SORT_BUCKET 25
2872#ifdef SQLITE_TEST
2873 int sqlite3_pager_n_sort_bucket = 0;
2874 #undef N_SORT_BUCKET
2875 #define N_SORT_BUCKET \
2876 (sqlite3_pager_n_sort_bucket?sqlite3_pager_n_sort_bucket:N_SORT_BUCKET_ALLOC)
2877#endif
2878static PgHdr *sort_pagelist(PgHdr *pIn){
2879 PgHdr *a[N_SORT_BUCKET_ALLOC], *p;
2880 int i;
2881 memset(a, 0, sizeof(a));
2882 while( pIn ){
2883 p = pIn;
2884 pIn = p->pDirty;
2885 p->pDirty = 0;
2886 for(i=0; i<N_SORT_BUCKET-1; i++){
2887 if( a[i]==0 ){
2888 a[i] = p;
2889 break;
2890 }else{
2891 p = merge_pagelist(a[i], p);
2892 a[i] = 0;
2893 }
2894 }
2895 if( i==N_SORT_BUCKET-1 ){
2896 /* Coverage: To get here, there need to be 2^(N_SORT_BUCKET)
2897 ** elements in the input list. This is possible, but impractical.
2898 ** Testing this line is the point of global variable
2899 ** sqlite3_pager_n_sort_bucket.
2900 */
2901 a[i] = merge_pagelist(a[i], p);
2902 }
2903 }
2904 p = a[0];
2905 for(i=1; i<N_SORT_BUCKET; i++){
2906 p = merge_pagelist(p, a[i]);
2907 }
2908 return p;
2909}
2910
2911/*
2912** Given a list of pages (connected by the PgHdr.pDirty pointer) write
2913** every one of those pages out to the database file and mark them all
2914** as clean.
2915*/
2916static int pager_write_pagelist(PgHdr *pList){
2917 Pager *pPager;
2918 PgHdr *p;
2919 int rc;
2920
2921 if( pList==0 ) return SQLITE_OK;
2922 pPager = pList->pPager;
2923
2924 /* At this point there may be either a RESERVED or EXCLUSIVE lock on the
2925 ** database file. If there is already an EXCLUSIVE lock, the following
2926 ** calls to sqlite3OsLock() are no-ops.
2927 **
2928 ** Moving the lock from RESERVED to EXCLUSIVE actually involves going
2929 ** through an intermediate state PENDING. A PENDING lock prevents new
2930 ** readers from attaching to the database but is unsufficient for us to
2931 ** write. The idea of a PENDING lock is to prevent new readers from
2932 ** coming in while we wait for existing readers to clear.
2933 **
2934 ** While the pager is in the RESERVED state, the original database file
2935 ** is unchanged and we can rollback without having to playback the
2936 ** journal into the original database file. Once we transition to
2937 ** EXCLUSIVE, it means the database file has been changed and any rollback
2938 ** will require a journal playback.
2939 */
2940 rc = pager_wait_on_lock(pPager, EXCLUSIVE_LOCK);
2941 if( rc!=SQLITE_OK ){
2942 return rc;
2943 }
2944
2945 pList = sort_pagelist(pList);
2946 for(p=pList; p; p=p->pDirty){
2947 assert( p->dirty );
2948 p->dirty = 0;
2949 }
2950 while( pList ){
2951
2952 /* If the file has not yet been opened, open it now. */
2953 if( !pPager->fd->pMethods ){
2954 assert(pPager->tempFile);
2955 rc = sqlite3PagerOpentemp(pPager->pVfs, pPager->fd, pPager->zFilename,
2956 pPager->vfsFlags);
2957 if( rc ) return rc;
2958 }
2959
2960 /* If there are dirty pages in the page cache with page numbers greater
2961 ** than Pager.dbSize, this means sqlite3PagerTruncate() was called to
2962 ** make the file smaller (presumably by auto-vacuum code). Do not write
2963 ** any such pages to the file.
2964 */
2965 if( pList->pgno<=pPager->dbSize ){
2966 i64 offset = (pList->pgno-1)*(i64)pPager->pageSize;
2967 char *pData = CODEC2(pPager, PGHDR_TO_DATA(pList), pList->pgno, 6);
2968 PAGERTRACE4("STORE %d page %d hash(%08x)\n",
2969 PAGERID(pPager), pList->pgno, pager_pagehash(pList));
2970 IOTRACE(("PGOUT %p %d\n", pPager, pList->pgno));
2971 rc = sqlite3OsWrite(pPager->fd, pData, pPager->pageSize, offset);
2972 PAGER_INCR(sqlite3_pager_writedb_count);
2973 PAGER_INCR(pPager->nWrite);
2974 if( pList->pgno==1 ){
2975 memcpy(&pPager->dbFileVers, &pData[24], sizeof(pPager->dbFileVers));
2976 }
2977 }
2978#ifndef NDEBUG
2979 else{
2980 PAGERTRACE3("NOSTORE %d page %d\n", PAGERID(pPager), pList->pgno);
2981 }
2982#endif
2983 if( rc ) return rc;
2984#ifdef SQLITE_CHECK_PAGES
2985 pList->pageHash = pager_pagehash(pList);
2986#endif
2987 pList = pList->pDirty;
2988 }
2989 return SQLITE_OK;
2990}
2991
2992/*
2993** Collect every dirty page into a dirty list and
2994** return a pointer to the head of that list. All pages are
2995** collected even if they are still in use.
2996*/
2997static PgHdr *pager_get_all_dirty_pages(Pager *pPager){
2998 return pPager->pDirty;
2999}
3000
3001/*
3002** Return TRUE if there is a hot journal on the given pager.
3003** A hot journal is one that needs to be played back.
3004**
3005** If the current size of the database file is 0 but a journal file
3006** exists, that is probably an old journal left over from a prior
3007** database with the same name. Just delete the journal.
3008*/
3009static int hasHotJournal(Pager *pPager){
3010 sqlite3_vfs *pVfs = pPager->pVfs;
3011 if( !pPager->useJournal ) return 0;
3012 if( !sqlite3OsAccess(pVfs, pPager->zJournal, SQLITE_ACCESS_EXISTS) ){
3013 return 0;
3014 }
3015 if( sqlite3OsCheckReservedLock(pPager->fd) ){
3016 return 0;
3017 }
3018 if( sqlite3PagerPagecount(pPager)==0 ){
3019 sqlite3OsDelete(pVfs, pPager->zJournal, 0);
3020 return 0;
3021 }else{
3022 return 1;
3023 }
3024}
3025
3026/*
3027** Try to find a page in the cache that can be recycled.
3028**
3029** This routine may return SQLITE_IOERR, SQLITE_FULL or SQLITE_OK. It
3030** does not set the pPager->errCode variable.
3031*/
3032static int pager_recycle(Pager *pPager, PgHdr **ppPg){
3033 PgHdr *pPg;
3034 *ppPg = 0;
3035
3036 /* It is illegal to call this function unless the pager object
3037 ** pointed to by pPager has at least one free page (page with nRef==0).
3038 */
3039 assert(!MEMDB);
3040 assert(pPager->lru.pFirst);
3041
3042 /* Find a page to recycle. Try to locate a page that does not
3043 ** require us to do an fsync() on the journal.
3044 */
3045 pPg = pPager->lru.pFirstSynced;
3046
3047 /* If we could not find a page that does not require an fsync()
3048 ** on the journal file then fsync the journal file. This is a
3049 ** very slow operation, so we work hard to avoid it. But sometimes
3050 ** it can't be helped.
3051 */
3052 if( pPg==0 && pPager->lru.pFirst){
3053 int iDc = sqlite3OsDeviceCharacteristics(pPager->fd);
3054 int rc = syncJournal(pPager);
3055 if( rc!=0 ){
3056 return rc;
3057 }
3058 if( pPager->fullSync && 0==(iDc&SQLITE_IOCAP_SAFE_APPEND) ){
3059 /* If in full-sync mode, write a new journal header into the
3060 ** journal file. This is done to avoid ever modifying a journal
3061 ** header that is involved in the rollback of pages that have
3062 ** already been written to the database (in case the header is
3063 ** trashed when the nRec field is updated).
3064 */
3065 pPager->nRec = 0;
3066 assert( pPager->journalOff > 0 );
3067 assert( pPager->doNotSync==0 );
3068 rc = writeJournalHdr(pPager);
3069 if( rc!=0 ){
3070 return rc;
3071 }
3072 }
3073 pPg = pPager->lru.pFirst;
3074 }
3075
3076 assert( pPg->nRef==0 );
3077
3078 /* Write the page to the database file if it is dirty.
3079 */
3080 if( pPg->dirty ){
3081 int rc;
3082 assert( pPg->needSync==0 );
3083 makeClean(pPg);
3084 pPg->dirty = 1;
3085 pPg->pDirty = 0;
3086 rc = pager_write_pagelist( pPg );
3087 pPg->dirty = 0;
3088 if( rc!=SQLITE_OK ){
3089 return rc;
3090 }
3091 }
3092 assert( pPg->dirty==0 );
3093
3094 /* If the page we are recycling is marked as alwaysRollback, then
3095 ** set the global alwaysRollback flag, thus disabling the
3096 ** sqlite3PagerDontRollback() optimization for the rest of this transaction.
3097 ** It is necessary to do this because the page marked alwaysRollback
3098 ** might be reloaded at a later time but at that point we won't remember
3099 ** that is was marked alwaysRollback. This means that all pages must
3100 ** be marked as alwaysRollback from here on out.
3101 */
3102 if( pPg->alwaysRollback ){
3103 IOTRACE(("ALWAYS_ROLLBACK %p\n", pPager))
3104 pPager->alwaysRollback = 1;
3105 }
3106
3107 /* Unlink the old page from the free list and the hash table
3108 */
3109 unlinkPage(pPg);
3110 assert( pPg->pgno==0 );
3111
3112 *ppPg = pPg;
3113 return SQLITE_OK;
3114}
3115
3116#ifdef SQLITE_ENABLE_MEMORY_MANAGEMENT
3117/*
3118** This function is called to free superfluous dynamically allocated memory
3119** held by the pager system. Memory in use by any SQLite pager allocated
3120** by the current thread may be sqlite3_free()ed.
3121**
3122** nReq is the number of bytes of memory required. Once this much has
3123** been released, the function returns. The return value is the total number
3124** of bytes of memory released.
3125*/
3126int sqlite3PagerReleaseMemory(int nReq){
3127 int nReleased = 0; /* Bytes of memory released so far */
3128 sqlite3_mutex *mutex; /* The MEM2 mutex */
3129 Pager *pPager; /* For looping over pagers */
3130 int rc = SQLITE_OK;
3131
3132 /* Acquire the memory-management mutex
3133 */
3134 mutex = sqlite3_mutex_alloc(SQLITE_MUTEX_STATIC_MEM2);
3135 sqlite3_mutex_enter(mutex);
3136
3137 /* Signal all database connections that memory management wants
3138 ** to have access to the pagers.
3139 */
3140 for(pPager=sqlite3PagerList; pPager; pPager=pPager->pNext){
3141 pPager->iInUseMM = 1;
3142 }
3143
3144 while( rc==SQLITE_OK && (nReq<0 || nReleased<nReq) ){
3145 PgHdr *pPg;
3146 PgHdr *pRecycled;
3147
3148 /* Try to find a page to recycle that does not require a sync(). If
3149 ** this is not possible, find one that does require a sync().
3150 */
3151 sqlite3_mutex_enter(sqlite3_mutex_alloc(SQLITE_MUTEX_STATIC_LRU));
3152 pPg = sqlite3LruPageList.pFirstSynced;
3153 while( pPg && (pPg->needSync || pPg->pPager->iInUseDB) ){
3154 pPg = pPg->gfree.pNext;
3155 }
3156 if( !pPg ){
3157 pPg = sqlite3LruPageList.pFirst;
3158 while( pPg && pPg->pPager->iInUseDB ){
3159 pPg = pPg->gfree.pNext;
3160 }
3161 }
3162 sqlite3_mutex_leave(sqlite3_mutex_alloc(SQLITE_MUTEX_STATIC_LRU));
3163
3164 /* If pPg==0, then the block above has failed to find a page to
3165 ** recycle. In this case return early - no further memory will
3166 ** be released.
3167 */
3168 if( !pPg ) break;
3169
3170 pPager = pPg->pPager;
3171 assert(!pPg->needSync || pPg==pPager->lru.pFirst);
3172 assert(pPg->needSync || pPg==pPager->lru.pFirstSynced);
3173
3174 rc = pager_recycle(pPager, &pRecycled);
3175 assert(pRecycled==pPg || rc!=SQLITE_OK);
3176 if( rc==SQLITE_OK ){
3177 /* We've found a page to free. At this point the page has been
3178 ** removed from the page hash-table, free-list and synced-list
3179 ** (pFirstSynced). It is still in the all pages (pAll) list.
3180 ** Remove it from this list before freeing.
3181 **
3182 ** Todo: Check the Pager.pStmt list to make sure this is Ok. It
3183 ** probably is though.
3184 */
3185 PgHdr *pTmp;
3186 assert( pPg );
3187 if( pPg==pPager->pAll ){
3188 pPager->pAll = pPg->pNextAll;
3189 }else{
3190 for( pTmp=pPager->pAll; pTmp->pNextAll!=pPg; pTmp=pTmp->pNextAll ){}
3191 pTmp->pNextAll = pPg->pNextAll;
3192 }
3193 nReleased += (
3194 sizeof(*pPg) + pPager->pageSize
3195 + sizeof(u32) + pPager->nExtra
3196 + MEMDB*sizeof(PgHistory)
3197 );
3198 IOTRACE(("PGFREE %p %d *\n", pPager, pPg->pgno));
3199 PAGER_INCR(sqlite3_pager_pgfree_count);
3200 sqlite3_free(pPg->pData);
3201 sqlite3_free(pPg);
3202 pPager->nPage--;
3203 }else{
3204 /* An error occured whilst writing to the database file or
3205 ** journal in pager_recycle(). The error is not returned to the
3206 ** caller of this function. Instead, set the Pager.errCode variable.
3207 ** The error will be returned to the user (or users, in the case
3208 ** of a shared pager cache) of the pager for which the error occured.
3209 */
3210 assert(
3211 (rc&0xff)==SQLITE_IOERR ||
3212 rc==SQLITE_FULL ||
3213 rc==SQLITE_BUSY
3214 );
3215 assert( pPager->state>=PAGER_RESERVED );
3216 pager_error(pPager, rc);
3217 }
3218 }
3219
3220 /* Clear the memory management flags and release the mutex
3221 */
3222 for(pPager=sqlite3PagerList; pPager; pPager=pPager->pNext){
3223 pPager->iInUseMM = 0;
3224 }
3225 sqlite3_mutex_leave(mutex);
3226
3227 /* Return the number of bytes released
3228 */
3229 return nReleased;
3230}
3231#endif /* SQLITE_ENABLE_MEMORY_MANAGEMENT */
3232
3233/*
3234** Read the content of page pPg out of the database file.
3235*/
3236static int readDbPage(Pager *pPager, PgHdr *pPg, Pgno pgno){
3237 int rc;
3238 i64 offset;
3239 assert( MEMDB==0 );
3240 assert(pPager->fd->pMethods||pPager->tempFile);
3241 if( !pPager->fd->pMethods ){
3242 return SQLITE_IOERR_SHORT_READ;
3243 }
3244 offset = (pgno-1)*(i64)pPager->pageSize;
3245 rc = sqlite3OsRead(pPager->fd, PGHDR_TO_DATA(pPg), pPager->pageSize, offset);
3246 PAGER_INCR(sqlite3_pager_readdb_count);
3247 PAGER_INCR(pPager->nRead);
3248 IOTRACE(("PGIN %p %d\n", pPager, pgno));
3249 if( pgno==1 ){
3250 memcpy(&pPager->dbFileVers, &((u8*)PGHDR_TO_DATA(pPg))[24],
3251 sizeof(pPager->dbFileVers));
3252 }
3253 CODEC1(pPager, PGHDR_TO_DATA(pPg), pPg->pgno, 3);
3254 PAGERTRACE4("FETCH %d page %d hash(%08x)\n",
3255 PAGERID(pPager), pPg->pgno, pager_pagehash(pPg));
3256 return rc;
3257}
3258
3259
3260/*
3261** This function is called to obtain the shared lock required before
3262** data may be read from the pager cache. If the shared lock has already
3263** been obtained, this function is a no-op.
3264**
3265** Immediately after obtaining the shared lock (if required), this function
3266** checks for a hot-journal file. If one is found, an emergency rollback
3267** is performed immediately.
3268*/
3269static int pagerSharedLock(Pager *pPager){
3270 int rc = SQLITE_OK;
3271 int isHot = 0;
3272
3273 /* If this database is opened for exclusive access, has no outstanding
3274 ** page references and is in an error-state, now is the chance to clear
3275 ** the error. Discard the contents of the pager-cache and treat any
3276 ** open journal file as a hot-journal.
3277 */
3278 if( !MEMDB && pPager->exclusiveMode && pPager->nRef==0 && pPager->errCode ){
3279 if( pPager->journalOpen ){
3280 isHot = 1;
3281 }
3282 pager_reset(pPager);
3283 pPager->errCode = SQLITE_OK;
3284 }
3285
3286 /* If the pager is still in an error state, do not proceed. The error
3287 ** state will be cleared at some point in the future when all page
3288 ** references are dropped and the cache can be discarded.
3289 */
3290 if( pPager->errCode && pPager->errCode!=SQLITE_FULL ){
3291 return pPager->errCode;
3292 }
3293
3294 if( pPager->state==PAGER_UNLOCK || isHot ){
3295 sqlite3_vfs *pVfs = pPager->pVfs;
3296 if( !MEMDB ){
3297 assert( pPager->nRef==0 );
3298 if( !pPager->noReadlock ){
3299 rc = pager_wait_on_lock(pPager, SHARED_LOCK);
3300 if( rc!=SQLITE_OK ){
3301 return pager_error(pPager, rc);
3302 }
3303 assert( pPager->state>=SHARED_LOCK );
3304 }
3305
3306 /* If a journal file exists, and there is no RESERVED lock on the
3307 ** database file, then it either needs to be played back or deleted.
3308 */
3309 if( hasHotJournal(pPager) || isHot ){
3310 /* Get an EXCLUSIVE lock on the database file. At this point it is
3311 ** important that a RESERVED lock is not obtained on the way to the
3312 ** EXCLUSIVE lock. If it were, another process might open the
3313 ** database file, detect the RESERVED lock, and conclude that the
3314 ** database is safe to read while this process is still rolling it
3315 ** back.
3316 **
3317 ** Because the intermediate RESERVED lock is not requested, the
3318 ** second process will get to this point in the code and fail to
3319 ** obtain it's own EXCLUSIVE lock on the database file.
3320 */
3321 if( pPager->state<EXCLUSIVE_LOCK ){
3322 rc = sqlite3OsLock(pPager->fd, EXCLUSIVE_LOCK);
3323 if( rc!=SQLITE_OK ){
3324 pager_unlock(pPager);
3325 return pager_error(pPager, rc);
3326 }
3327 pPager->state = PAGER_EXCLUSIVE;
3328 }
3329
3330 /* Open the journal for reading only. Return SQLITE_BUSY if
3331 ** we are unable to open the journal file.
3332 **
3333 ** The journal file does not need to be locked itself. The
3334 ** journal file is never open unless the main database file holds
3335 ** a write lock, so there is never any chance of two or more
3336 ** processes opening the journal at the same time.
3337 **
3338 ** Open the journal for read/write access. This is because in
3339 ** exclusive-access mode the file descriptor will be kept open and
3340 ** possibly used for a transaction later on. On some systems, the
3341 ** OsTruncate() call used in exclusive-access mode also requires
3342 ** a read/write file handle.
3343 */
3344 if( !isHot ){
3345 rc = SQLITE_BUSY;
3346 if( sqlite3OsAccess(pVfs, pPager->zJournal, SQLITE_ACCESS_EXISTS) ){
3347 int fout = 0;
3348 int f = SQLITE_OPEN_READWRITE|SQLITE_OPEN_MAIN_JOURNAL;
3349 assert( !pPager->tempFile );
3350 rc = sqlite3OsOpen(pVfs, pPager->zJournal, pPager->jfd, f, &fout);
3351 assert( rc!=SQLITE_OK || pPager->jfd->pMethods );
3352 if( fout&SQLITE_OPEN_READONLY ){
3353 rc = SQLITE_BUSY;
3354 sqlite3OsClose(pPager->jfd);
3355 }
3356 }
3357 }
3358 if( rc!=SQLITE_OK ){
3359 pager_unlock(pPager);
3360 return ((rc==SQLITE_NOMEM||rc==SQLITE_IOERR_NOMEM)?rc:SQLITE_BUSY);
3361 }
3362 pPager->journalOpen = 1;
3363 pPager->journalStarted = 0;
3364 pPager->journalOff = 0;
3365 pPager->setMaster = 0;
3366 pPager->journalHdr = 0;
3367
3368 /* Playback and delete the journal. Drop the database write
3369 ** lock and reacquire the read lock.
3370 */
3371 rc = pager_playback(pPager, 1);
3372 if( rc!=SQLITE_OK ){
3373 return pager_error(pPager, rc);
3374 }
3375 assert(pPager->state==PAGER_SHARED ||
3376 (pPager->exclusiveMode && pPager->state>PAGER_SHARED)
3377 );
3378 }
3379
3380 if( pPager->pAll ){
3381 /* The shared-lock has just been acquired on the database file
3382 ** and there are already pages in the cache (from a previous
3383 ** read or write transaction). Check to see if the database
3384 ** has been modified. If the database has changed, flush the
3385 ** cache.
3386 **
3387 ** Database changes is detected by looking at 15 bytes beginning
3388 ** at offset 24 into the file. The first 4 of these 16 bytes are
3389 ** a 32-bit counter that is incremented with each change. The
3390 ** other bytes change randomly with each file change when
3391 ** a codec is in use.
3392 **
3393 ** There is a vanishingly small chance that a change will not be
3394 ** detected. The chance of an undetected change is so small that
3395 ** it can be neglected.
3396 */
3397 char dbFileVers[sizeof(pPager->dbFileVers)];
3398 sqlite3PagerPagecount(pPager);
3399
3400 if( pPager->errCode ){
3401 return pPager->errCode;
3402 }
3403
3404 if( pPager->dbSize>0 ){
3405 IOTRACE(("CKVERS %p %d\n", pPager, sizeof(dbFileVers)));
3406 rc = sqlite3OsRead(pPager->fd, &dbFileVers, sizeof(dbFileVers), 24);
3407 if( rc!=SQLITE_OK ){
3408 return rc;
3409 }
3410 }else{
3411 memset(dbFileVers, 0, sizeof(dbFileVers));
3412 }
3413
3414 if( memcmp(pPager->dbFileVers, dbFileVers, sizeof(dbFileVers))!=0 ){
3415 pager_reset(pPager);
3416 }
3417 }
3418 }
3419 assert( pPager->exclusiveMode || pPager->state<=PAGER_SHARED );
3420 if( pPager->state==PAGER_UNLOCK ){
3421 pPager->state = PAGER_SHARED;
3422 }
3423 }
3424
3425 return rc;
3426}
3427
3428/*
3429** Allocate a PgHdr object. Either create a new one or reuse
3430** an existing one that is not otherwise in use.
3431**
3432** A new PgHdr structure is created if any of the following are
3433** true:
3434**
3435** (1) We have not exceeded our maximum allocated cache size
3436** as set by the "PRAGMA cache_size" command.
3437**
3438** (2) There are no unused PgHdr objects available at this time.
3439**
3440** (3) This is an in-memory database.
3441**
3442** (4) There are no PgHdr objects that do not require a journal
3443** file sync and a sync of the journal file is currently
3444** prohibited.
3445**
3446** Otherwise, reuse an existing PgHdr. In other words, reuse an
3447** existing PgHdr if all of the following are true:
3448**
3449** (1) We have reached or exceeded the maximum cache size
3450** allowed by "PRAGMA cache_size".
3451**
3452** (2) There is a PgHdr available with PgHdr->nRef==0
3453**
3454** (3) We are not in an in-memory database
3455**
3456** (4) Either there is an available PgHdr that does not need
3457** to be synced to disk or else disk syncing is currently
3458** allowed.
3459*/
3460static int pagerAllocatePage(Pager *pPager, PgHdr **ppPg){
3461 int rc = SQLITE_OK;
3462 PgHdr *pPg;
3463 void *pData;
3464
3465 /* Create a new PgHdr if any of the four conditions defined
3466 ** above are met: */
3467 if( pPager->nPage<pPager->mxPage
3468 || pPager->lru.pFirst==0
3469 || MEMDB
3470 || (pPager->lru.pFirstSynced==0 && pPager->doNotSync)
3471 ){
3472 if( pPager->nPage>=pPager->nHash ){
3473 pager_resize_hash_table(pPager,
3474 pPager->nHash<256 ? 256 : pPager->nHash*2);
3475 if( pPager->nHash==0 ){
3476 rc = SQLITE_NOMEM;
3477 goto pager_allocate_out;
3478 }
3479 }
3480 pagerLeave(pPager);
3481 pPg = sqlite3_malloc( sizeof(*pPg) + sizeof(u32) + pPager->nExtra
3482 + MEMDB*sizeof(PgHistory) );
3483 if( pPg ){
3484 pData = sqlite3_malloc( pPager->pageSize );
3485 if( pData==0 ){
3486 sqlite3_free(pPg);
3487 pPg = 0;
3488 }
3489 }
3490 pagerEnter(pPager);
3491 if( pPg==0 ){
3492 rc = SQLITE_NOMEM;
3493 goto pager_allocate_out;
3494 }
3495 memset(pPg, 0, sizeof(*pPg));
3496 if( MEMDB ){
3497 memset(PGHDR_TO_HIST(pPg, pPager), 0, sizeof(PgHistory));
3498 }
3499 pPg->pData = pData;
3500 pPg->pPager = pPager;
3501 pPg->pNextAll = pPager->pAll;
3502 pPager->pAll = pPg;
3503 pPager->nPage++;
3504 }else{
3505 /* Recycle an existing page with a zero ref-count. */
3506 rc = pager_recycle(pPager, &pPg);
3507 if( rc==SQLITE_BUSY ){
3508 rc = SQLITE_IOERR_BLOCKED;
3509 }
3510 if( rc!=SQLITE_OK ){
3511 goto pager_allocate_out;
3512 }
3513 assert( pPager->state>=SHARED_LOCK );
3514 assert(pPg);
3515 }
3516 *ppPg = pPg;
3517
3518pager_allocate_out:
3519 return rc;
3520}
3521
3522/*
3523** Make sure we have the content for a page. If the page was
3524** previously acquired with noContent==1, then the content was
3525** just initialized to zeros instead of being read from disk.
3526** But now we need the real data off of disk. So make sure we
3527** have it. Read it in if we do not have it already.
3528*/
3529static int pager_get_content(PgHdr *pPg){
3530 if( pPg->needRead ){
3531 int rc = readDbPage(pPg->pPager, pPg, pPg->pgno);
3532 if( rc==SQLITE_OK ){
3533 pPg->needRead = 0;
3534 }else{
3535 return rc;
3536 }
3537 }
3538 return SQLITE_OK;
3539}
3540
3541/*
3542** Acquire a page.
3543**
3544** A read lock on the disk file is obtained when the first page is acquired.
3545** This read lock is dropped when the last page is released.
3546**
3547** This routine works for any page number greater than 0. If the database
3548** file is smaller than the requested page, then no actual disk
3549** read occurs and the memory image of the page is initialized to
3550** all zeros. The extra data appended to a page is always initialized
3551** to zeros the first time a page is loaded into memory.
3552**
3553** The acquisition might fail for several reasons. In all cases,
3554** an appropriate error code is returned and *ppPage is set to NULL.
3555**
3556** See also sqlite3PagerLookup(). Both this routine and Lookup() attempt
3557** to find a page in the in-memory cache first. If the page is not already
3558** in memory, this routine goes to disk to read it in whereas Lookup()
3559** just returns 0. This routine acquires a read-lock the first time it
3560** has to go to disk, and could also playback an old journal if necessary.
3561** Since Lookup() never goes to disk, it never has to deal with locks
3562** or journal files.
3563**
3564** If noContent is false, the page contents are actually read from disk.
3565** If noContent is true, it means that we do not care about the contents
3566** of the page at this time, so do not do a disk read. Just fill in the
3567** page content with zeros. But mark the fact that we have not read the
3568** content by setting the PgHdr.needRead flag. Later on, if
3569** sqlite3PagerWrite() is called on this page or if this routine is
3570** called again with noContent==0, that means that the content is needed
3571** and the disk read should occur at that point.
3572*/
3573static int pagerAcquire(
3574 Pager *pPager, /* The pager open on the database file */
3575 Pgno pgno, /* Page number to fetch */
3576 DbPage **ppPage, /* Write a pointer to the page here */
3577 int noContent /* Do not bother reading content from disk if true */
3578){
3579 PgHdr *pPg;
3580 int rc;
3581
3582 assert( pPager->state==PAGER_UNLOCK || pPager->nRef>0 || pgno==1 );
3583
3584 /* The maximum page number is 2^31. Return SQLITE_CORRUPT if a page
3585 ** number greater than this, or zero, is requested.
3586 */
3587 if( pgno>PAGER_MAX_PGNO || pgno==0 || pgno==PAGER_MJ_PGNO(pPager) ){
3588 return SQLITE_CORRUPT_BKPT;
3589 }
3590
3591 /* Make sure we have not hit any critical errors.
3592 */
3593 assert( pPager!=0 );
3594 *ppPage = 0;
3595
3596 /* If this is the first page accessed, then get a SHARED lock
3597 ** on the database file. pagerSharedLock() is a no-op if
3598 ** a database lock is already held.
3599 */
3600 rc = pagerSharedLock(pPager);
3601 if( rc!=SQLITE_OK ){
3602 return rc;
3603 }
3604 assert( pPager->state!=PAGER_UNLOCK );
3605
3606 pPg = pager_lookup(pPager, pgno);
3607 if( pPg==0 ){
3608 /* The requested page is not in the page cache. */
3609 int nMax;
3610 int h;
3611 PAGER_INCR(pPager->nMiss);
3612 rc = pagerAllocatePage(pPager, &pPg);
3613 if( rc!=SQLITE_OK ){
3614 return rc;
3615 }
3616
3617 pPg->pgno = pgno;
3618 assert( !MEMDB || pgno>pPager->stmtSize );
3619 if( pPager->aInJournal && (int)pgno<=pPager->origDbSize ){
3620#if 0
3621 sqlite3CheckMemory(pPager->aInJournal, pgno/8);
3622#endif
3623 assert( pPager->journalOpen );
3624 pPg->inJournal = (pPager->aInJournal[pgno/8] & (1<<(pgno&7)))!=0;
3625 pPg->needSync = 0;
3626 }else{
3627 pPg->inJournal = 0;
3628 pPg->needSync = 0;
3629 }
3630
3631 makeClean(pPg);
3632 pPg->nRef = 1;
3633 REFINFO(pPg);
3634
3635 pPager->nRef++;
3636 if( pPager->nExtra>0 ){
3637 memset(PGHDR_TO_EXTRA(pPg, pPager), 0, pPager->nExtra);
3638 }
3639 nMax = sqlite3PagerPagecount(pPager);
3640 if( pPager->errCode ){
3641 rc = pPager->errCode;
3642 sqlite3PagerUnref(pPg);
3643 return rc;
3644 }
3645
3646 /* Populate the page with data, either by reading from the database
3647 ** file, or by setting the entire page to zero.
3648 */
3649 if( nMax<(int)pgno || MEMDB || (noContent && !pPager->alwaysRollback) ){
3650 if( pgno>pPager->mxPgno ){
3651 sqlite3PagerUnref(pPg);
3652 return SQLITE_FULL;
3653 }
3654 memset(PGHDR_TO_DATA(pPg), 0, pPager->pageSize);
3655 pPg->needRead = noContent && !pPager->alwaysRollback;
3656 IOTRACE(("ZERO %p %d\n", pPager, pgno));
3657 }else{
3658 rc = readDbPage(pPager, pPg, pgno);
3659 if( rc!=SQLITE_OK && rc!=SQLITE_IOERR_SHORT_READ ){
3660 pPg->pgno = 0;
3661 sqlite3PagerUnref(pPg);
3662 return rc;
3663 }
3664 pPg->needRead = 0;
3665 }
3666
3667 /* Link the page into the page hash table */
3668 h = pgno & (pPager->nHash-1);
3669 assert( pgno!=0 );
3670 pPg->pNextHash = pPager->aHash[h];
3671 pPager->aHash[h] = pPg;
3672 if( pPg->pNextHash ){
3673 assert( pPg->pNextHash->pPrevHash==0 );
3674 pPg->pNextHash->pPrevHash = pPg;
3675 }
3676
3677#ifdef SQLITE_CHECK_PAGES
3678 pPg->pageHash = pager_pagehash(pPg);
3679#endif
3680 }else{
3681 /* The requested page is in the page cache. */
3682 assert(pPager->nRef>0 || pgno==1);
3683 PAGER_INCR(pPager->nHit);
3684 if( !noContent ){
3685 rc = pager_get_content(pPg);
3686 if( rc ){
3687 return rc;
3688 }
3689 }
3690 page_ref(pPg);
3691 }
3692 *ppPage = pPg;
3693 return SQLITE_OK;
3694}
3695int sqlite3PagerAcquire(
3696 Pager *pPager, /* The pager open on the database file */
3697 Pgno pgno, /* Page number to fetch */
3698 DbPage **ppPage, /* Write a pointer to the page here */
3699 int noContent /* Do not bother reading content from disk if true */
3700){
3701 int rc;
3702 pagerEnter(pPager);
3703 rc = pagerAcquire(pPager, pgno, ppPage, noContent);
3704 pagerLeave(pPager);
3705 return rc;
3706}
3707
3708
3709/*
3710** Acquire a page if it is already in the in-memory cache. Do
3711** not read the page from disk. Return a pointer to the page,
3712** or 0 if the page is not in cache.
3713**
3714** See also sqlite3PagerGet(). The difference between this routine
3715** and sqlite3PagerGet() is that _get() will go to the disk and read
3716** in the page if the page is not already in cache. This routine
3717** returns NULL if the page is not in cache or if a disk I/O error
3718** has ever happened.
3719*/
3720DbPage *sqlite3PagerLookup(Pager *pPager, Pgno pgno){
3721 PgHdr *pPg = 0;
3722
3723 assert( pPager!=0 );
3724 assert( pgno!=0 );
3725
3726 pagerEnter(pPager);
3727 if( pPager->state==PAGER_UNLOCK ){
3728 assert( !pPager->pAll || pPager->exclusiveMode );
3729 }else if( pPager->errCode && pPager->errCode!=SQLITE_FULL ){
3730 /* Do nothing */
3731 }else if( (pPg = pager_lookup(pPager, pgno))!=0 ){
3732 page_ref(pPg);
3733 }
3734 pagerLeave(pPager);
3735 return pPg;
3736}
3737
3738/*
3739** Release a page.
3740**
3741** If the number of references to the page drop to zero, then the
3742** page is added to the LRU list. When all references to all pages
3743** are released, a rollback occurs and the lock on the database is
3744** removed.
3745*/
3746int sqlite3PagerUnref(DbPage *pPg){
3747 Pager *pPager = pPg->pPager;
3748
3749 /* Decrement the reference count for this page
3750 */
3751 assert( pPg->nRef>0 );
3752 pagerEnter(pPg->pPager);
3753 pPg->nRef--;
3754 REFINFO(pPg);
3755
3756 CHECK_PAGE(pPg);
3757
3758 /* When the number of references to a page reach 0, call the
3759 ** destructor and add the page to the freelist.
3760 */
3761 if( pPg->nRef==0 ){
3762
3763 lruListAdd(pPg);
3764 if( pPager->xDestructor ){
3765 pPager->xDestructor(pPg, pPager->pageSize);
3766 }
3767
3768 /* When all pages reach the freelist, drop the read lock from
3769 ** the database file.
3770 */
3771 pPager->nRef--;
3772 assert( pPager->nRef>=0 );
3773 if( pPager->nRef==0 && (!pPager->exclusiveMode || pPager->journalOff>0) ){
3774 pagerUnlockAndRollback(pPager);
3775 }
3776 }
3777 pagerLeave(pPager);
3778 return SQLITE_OK;
3779}
3780
3781/*
3782** Create a journal file for pPager. There should already be a RESERVED
3783** or EXCLUSIVE lock on the database file when this routine is called.
3784**
3785** Return SQLITE_OK if everything. Return an error code and release the
3786** write lock if anything goes wrong.
3787*/
3788static int pager_open_journal(Pager *pPager){
3789 sqlite3_vfs *pVfs = pPager->pVfs;
3790 int flags = (SQLITE_OPEN_READWRITE|SQLITE_OPEN_EXCLUSIVE|SQLITE_OPEN_CREATE);
3791
3792 int rc;
3793 assert( !MEMDB );
3794 assert( pPager->state>=PAGER_RESERVED );
3795 assert( pPager->journalOpen==0 );
3796 assert( pPager->useJournal );
3797 assert( pPager->aInJournal==0 );
3798 sqlite3PagerPagecount(pPager);
3799 pagerLeave(pPager);
3800 pPager->aInJournal = sqlite3MallocZero( pPager->dbSize/8 + 1 );
3801 pagerEnter(pPager);
3802 if( pPager->aInJournal==0 ){
3803 rc = SQLITE_NOMEM;
3804 goto failed_to_open_journal;
3805 }
3806
3807 if( pPager->tempFile ){
3808 flags |= (SQLITE_OPEN_DELETEONCLOSE|SQLITE_OPEN_TEMP_JOURNAL);
3809 }else{
3810 flags |= (SQLITE_OPEN_MAIN_JOURNAL);
3811 }
3812#ifdef SQLITE_ENABLE_ATOMIC_WRITE
3813 rc = sqlite3JournalOpen(
3814 pVfs, pPager->zJournal, pPager->jfd, flags, jrnlBufferSize(pPager)
3815 );
3816#else
3817 rc = sqlite3OsOpen(pVfs, pPager->zJournal, pPager->jfd, flags, 0);
3818#endif
3819 assert( rc!=SQLITE_OK || pPager->jfd->pMethods );
3820 pPager->journalOff = 0;
3821 pPager->setMaster = 0;
3822 pPager->journalHdr = 0;
3823 if( rc!=SQLITE_OK ){
3824 if( rc==SQLITE_NOMEM ){
3825 sqlite3OsDelete(pVfs, pPager->zJournal, 0);
3826 }
3827 goto failed_to_open_journal;
3828 }
3829 pPager->journalOpen = 1;
3830 pPager->journalStarted = 0;
3831 pPager->needSync = 0;
3832 pPager->alwaysRollback = 0;
3833 pPager->nRec = 0;
3834 if( pPager->errCode ){
3835 rc = pPager->errCode;
3836 goto failed_to_open_journal;
3837 }
3838 pPager->origDbSize = pPager->dbSize;
3839
3840 rc = writeJournalHdr(pPager);
3841
3842 if( pPager->stmtAutoopen && rc==SQLITE_OK ){
3843 rc = sqlite3PagerStmtBegin(pPager);
3844 }
3845 if( rc!=SQLITE_OK && rc!=SQLITE_NOMEM && rc!=SQLITE_IOERR_NOMEM ){
3846 rc = pager_end_transaction(pPager);
3847 if( rc==SQLITE_OK ){
3848 rc = SQLITE_FULL;
3849 }
3850 }
3851 return rc;
3852
3853failed_to_open_journal:
3854 sqlite3_free(pPager->aInJournal);
3855 pPager->aInJournal = 0;
3856 return rc;
3857}
3858
3859/*
3860** Acquire a write-lock on the database. The lock is removed when
3861** the any of the following happen:
3862**
3863** * sqlite3PagerCommitPhaseTwo() is called.
3864** * sqlite3PagerRollback() is called.
3865** * sqlite3PagerClose() is called.
3866** * sqlite3PagerUnref() is called to on every outstanding page.
3867**
3868** The first parameter to this routine is a pointer to any open page of the
3869** database file. Nothing changes about the page - it is used merely to
3870** acquire a pointer to the Pager structure and as proof that there is
3871** already a read-lock on the database.
3872**
3873** The second parameter indicates how much space in bytes to reserve for a
3874** master journal file-name at the start of the journal when it is created.
3875**
3876** A journal file is opened if this is not a temporary file. For temporary
3877** files, the opening of the journal file is deferred until there is an
3878** actual need to write to the journal.
3879**
3880** If the database is already reserved for writing, this routine is a no-op.
3881**
3882** If exFlag is true, go ahead and get an EXCLUSIVE lock on the file
3883** immediately instead of waiting until we try to flush the cache. The
3884** exFlag is ignored if a transaction is already active.
3885*/
3886int sqlite3PagerBegin(DbPage *pPg, int exFlag){
3887 Pager *pPager = pPg->pPager;
3888 int rc = SQLITE_OK;
3889 pagerEnter(pPager);
3890 assert( pPg->nRef>0 );
3891 assert( pPager->state!=PAGER_UNLOCK );
3892 if( pPager->state==PAGER_SHARED ){
3893 assert( pPager->aInJournal==0 );
3894 if( MEMDB ){
3895 pPager->state = PAGER_EXCLUSIVE;
3896 pPager->origDbSize = pPager->dbSize;
3897 }else{
3898 rc = sqlite3OsLock(pPager->fd, RESERVED_LOCK);
3899 if( rc==SQLITE_OK ){
3900 pPager->state = PAGER_RESERVED;
3901 if( exFlag ){
3902 rc = pager_wait_on_lock(pPager, EXCLUSIVE_LOCK);
3903 }
3904 }
3905 if( rc!=SQLITE_OK ){
3906 pagerLeave(pPager);
3907 return rc;
3908 }
3909 pPager->dirtyCache = 0;
3910 PAGERTRACE2("TRANSACTION %d\n", PAGERID(pPager));
3911 if( pPager->useJournal && !pPager->tempFile ){
3912 rc = pager_open_journal(pPager);
3913 }
3914 }
3915 }else if( pPager->journalOpen && pPager->journalOff==0 ){
3916 /* This happens when the pager was in exclusive-access mode last
3917 ** time a (read or write) transaction was successfully concluded
3918 ** by this connection. Instead of deleting the journal file it was
3919 ** kept open and truncated to 0 bytes.
3920 */
3921 assert( pPager->nRec==0 );
3922 assert( pPager->origDbSize==0 );
3923 assert( pPager->aInJournal==0 );
3924 sqlite3PagerPagecount(pPager);
3925 pagerLeave(pPager);
3926 pPager->aInJournal = sqlite3MallocZero( pPager->dbSize/8 + 1 );
3927 pagerEnter(pPager);
3928 if( !pPager->aInJournal ){
3929 rc = SQLITE_NOMEM;
3930 }else{
3931 pPager->origDbSize = pPager->dbSize;
3932 rc = writeJournalHdr(pPager);
3933 }
3934 }
3935 assert( !pPager->journalOpen || pPager->journalOff>0 || rc!=SQLITE_OK );
3936 pagerLeave(pPager);
3937 return rc;
3938}
3939
3940/*
3941** Make a page dirty. Set its dirty flag and add it to the dirty
3942** page list.
3943*/
3944static void makeDirty(PgHdr *pPg){
3945 if( pPg->dirty==0 ){
3946 Pager *pPager = pPg->pPager;
3947 pPg->dirty = 1;
3948 pPg->pDirty = pPager->pDirty;
3949 if( pPager->pDirty ){
3950 pPager->pDirty->pPrevDirty = pPg;
3951 }
3952 pPg->pPrevDirty = 0;
3953 pPager->pDirty = pPg;
3954 }
3955}
3956
3957/*
3958** Make a page clean. Clear its dirty bit and remove it from the
3959** dirty page list.
3960*/
3961static void makeClean(PgHdr *pPg){
3962 if( pPg->dirty ){
3963 pPg->dirty = 0;
3964 if( pPg->pDirty ){
3965 assert( pPg->pDirty->pPrevDirty==pPg );
3966 pPg->pDirty->pPrevDirty = pPg->pPrevDirty;
3967 }
3968 if( pPg->pPrevDirty ){
3969 assert( pPg->pPrevDirty->pDirty==pPg );
3970 pPg->pPrevDirty->pDirty = pPg->pDirty;
3971 }else{
3972 assert( pPg->pPager->pDirty==pPg );
3973 pPg->pPager->pDirty = pPg->pDirty;
3974 }
3975 }
3976}
3977
3978
3979/*
3980** Mark a data page as writeable. The page is written into the journal
3981** if it is not there already. This routine must be called before making
3982** changes to a page.
3983**
3984** The first time this routine is called, the pager creates a new
3985** journal and acquires a RESERVED lock on the database. If the RESERVED
3986** lock could not be acquired, this routine returns SQLITE_BUSY. The
3987** calling routine must check for that return value and be careful not to
3988** change any page data until this routine returns SQLITE_OK.
3989**
3990** If the journal file could not be written because the disk is full,
3991** then this routine returns SQLITE_FULL and does an immediate rollback.
3992** All subsequent write attempts also return SQLITE_FULL until there
3993** is a call to sqlite3PagerCommit() or sqlite3PagerRollback() to
3994** reset.
3995*/
3996static int pager_write(PgHdr *pPg){
3997 void *pData = PGHDR_TO_DATA(pPg);
3998 Pager *pPager = pPg->pPager;
3999 int rc = SQLITE_OK;
4000
4001 /* Check for errors
4002 */
4003 if( pPager->errCode ){
4004 return pPager->errCode;
4005 }
4006 if( pPager->readOnly ){
4007 return SQLITE_PERM;
4008 }
4009
4010 assert( !pPager->setMaster );
4011
4012 CHECK_PAGE(pPg);
4013
4014 /* If this page was previously acquired with noContent==1, that means
4015 ** we didn't really read in the content of the page. This can happen
4016 ** (for example) when the page is being moved to the freelist. But
4017 ** now we are (perhaps) moving the page off of the freelist for
4018 ** reuse and we need to know its original content so that content
4019 ** can be stored in the rollback journal. So do the read at this
4020 ** time.
4021 */
4022 rc = pager_get_content(pPg);
4023 if( rc ){
4024 return rc;
4025 }
4026
4027 /* Mark the page as dirty. If the page has already been written
4028 ** to the journal then we can return right away.
4029 */
4030 makeDirty(pPg);
4031 if( pPg->inJournal && (pageInStatement(pPg) || pPager->stmtInUse==0) ){
4032 pPager->dirtyCache = 1;
4033 }else{
4034
4035 /* If we get this far, it means that the page needs to be
4036 ** written to the transaction journal or the ckeckpoint journal
4037 ** or both.
4038 **
4039 ** First check to see that the transaction journal exists and
4040 ** create it if it does not.
4041 */
4042 assert( pPager->state!=PAGER_UNLOCK );
4043 rc = sqlite3PagerBegin(pPg, 0);
4044 if( rc!=SQLITE_OK ){
4045 return rc;
4046 }
4047 assert( pPager->state>=PAGER_RESERVED );
4048 if( !pPager->journalOpen && pPager->useJournal ){
4049 rc = pager_open_journal(pPager);
4050 if( rc!=SQLITE_OK ) return rc;
4051 }
4052 assert( pPager->journalOpen || !pPager->useJournal );
4053 pPager->dirtyCache = 1;
4054
4055 /* The transaction journal now exists and we have a RESERVED or an
4056 ** EXCLUSIVE lock on the main database file. Write the current page to
4057 ** the transaction journal if it is not there already.
4058 */
4059 if( !pPg->inJournal && (pPager->useJournal || MEMDB) ){
4060 if( (int)pPg->pgno <= pPager->origDbSize ){
4061 if( MEMDB ){
4062 PgHistory *pHist = PGHDR_TO_HIST(pPg, pPager);
4063 PAGERTRACE3("JOURNAL %d page %d\n", PAGERID(pPager), pPg->pgno);
4064 assert( pHist->pOrig==0 );
4065 pHist->pOrig = sqlite3_malloc( pPager->pageSize );
4066 if( pHist->pOrig ){
4067 memcpy(pHist->pOrig, PGHDR_TO_DATA(pPg), pPager->pageSize);
4068 }
4069 }else{
4070 u32 cksum;
4071 char *pData2;
4072
4073 /* We should never write to the journal file the page that
4074 ** contains the database locks. The following assert verifies
4075 ** that we do not. */
4076 assert( pPg->pgno!=PAGER_MJ_PGNO(pPager) );
4077 pData2 = CODEC2(pPager, pData, pPg->pgno, 7);
4078 cksum = pager_cksum(pPager, (u8*)pData2);
4079 rc = write32bits(pPager->jfd, pPager->journalOff, pPg->pgno);
4080 if( rc==SQLITE_OK ){
4081 rc = sqlite3OsWrite(pPager->jfd, pData2, pPager->pageSize,
4082 pPager->journalOff + 4);
4083 pPager->journalOff += pPager->pageSize+4;
4084 }
4085 if( rc==SQLITE_OK ){
4086 rc = write32bits(pPager->jfd, pPager->journalOff, cksum);
4087 pPager->journalOff += 4;
4088 }
4089 IOTRACE(("JOUT %p %d %lld %d\n", pPager, pPg->pgno,
4090 pPager->journalOff, pPager->pageSize));
4091 PAGER_INCR(sqlite3_pager_writej_count);
4092 PAGERTRACE5("JOURNAL %d page %d needSync=%d hash(%08x)\n",
4093 PAGERID(pPager), pPg->pgno, pPg->needSync, pager_pagehash(pPg));
4094
4095 /* An error has occured writing to the journal file. The
4096 ** transaction will be rolled back by the layer above.
4097 */
4098 if( rc!=SQLITE_OK ){
4099 return rc;
4100 }
4101
4102 pPager->nRec++;
4103 assert( pPager->aInJournal!=0 );
4104 pPager->aInJournal[pPg->pgno/8] |= 1<<(pPg->pgno&7);
4105 pPg->needSync = !pPager->noSync;
4106 if( pPager->stmtInUse ){
4107 pPager->aInStmt[pPg->pgno/8] |= 1<<(pPg->pgno&7);
4108 }
4109 }
4110 }else{
4111 pPg->needSync = !pPager->journalStarted && !pPager->noSync;
4112 PAGERTRACE4("APPEND %d page %d needSync=%d\n",
4113 PAGERID(pPager), pPg->pgno, pPg->needSync);
4114 }
4115 if( pPg->needSync ){
4116 pPager->needSync = 1;
4117 }
4118 pPg->inJournal = 1;
4119 }
4120
4121 /* If the statement journal is open and the page is not in it,
4122 ** then write the current page to the statement journal. Note that
4123 ** the statement journal format differs from the standard journal format
4124 ** in that it omits the checksums and the header.
4125 */
4126 if( pPager->stmtInUse
4127 && !pageInStatement(pPg)
4128 && (int)pPg->pgno<=pPager->stmtSize
4129 ){
4130 assert( pPg->inJournal || (int)pPg->pgno>pPager->origDbSize );
4131 if( MEMDB ){
4132 PgHistory *pHist = PGHDR_TO_HIST(pPg, pPager);
4133 assert( pHist->pStmt==0 );
4134 pHist->pStmt = sqlite3_malloc( pPager->pageSize );
4135 if( pHist->pStmt ){
4136 memcpy(pHist->pStmt, PGHDR_TO_DATA(pPg), pPager->pageSize);
4137 }
4138 PAGERTRACE3("STMT-JOURNAL %d page %d\n", PAGERID(pPager), pPg->pgno);
4139 page_add_to_stmt_list(pPg);
4140 }else{
4141 i64 offset = pPager->stmtNRec*(4+pPager->pageSize);
4142 char *pData2 = CODEC2(pPager, pData, pPg->pgno, 7);
4143 rc = write32bits(pPager->stfd, offset, pPg->pgno);
4144 if( rc==SQLITE_OK ){
4145 rc = sqlite3OsWrite(pPager->stfd, pData2, pPager->pageSize, offset+4);
4146 }
4147 PAGERTRACE3("STMT-JOURNAL %d page %d\n", PAGERID(pPager), pPg->pgno);
4148 if( rc!=SQLITE_OK ){
4149 return rc;
4150 }
4151 pPager->stmtNRec++;
4152 assert( pPager->aInStmt!=0 );
4153 pPager->aInStmt[pPg->pgno/8] |= 1<<(pPg->pgno&7);
4154 }
4155 }
4156 }
4157
4158 /* Update the database size and return.
4159 */
4160 assert( pPager->state>=PAGER_SHARED );
4161 if( pPager->dbSize<(int)pPg->pgno ){
4162 pPager->dbSize = pPg->pgno;
4163 if( !MEMDB && pPager->dbSize==PENDING_BYTE/pPager->pageSize ){
4164 pPager->dbSize++;
4165 }
4166 }
4167 return rc;
4168}
4169
4170/*
4171** This function is used to mark a data-page as writable. It uses
4172** pager_write() to open a journal file (if it is not already open)
4173** and write the page *pData to the journal.
4174**
4175** The difference between this function and pager_write() is that this
4176** function also deals with the special case where 2 or more pages
4177** fit on a single disk sector. In this case all co-resident pages
4178** must have been written to the journal file before returning.
4179*/
4180int sqlite3PagerWrite(DbPage *pDbPage){
4181 int rc = SQLITE_OK;
4182
4183 PgHdr *pPg = pDbPage;
4184 Pager *pPager = pPg->pPager;
4185 Pgno nPagePerSector = (pPager->sectorSize/pPager->pageSize);
4186
4187 pagerEnter(pPager);
4188 if( !MEMDB && nPagePerSector>1 ){
4189 Pgno nPageCount; /* Total number of pages in database file */
4190 Pgno pg1; /* First page of the sector pPg is located on. */
4191 int nPage; /* Number of pages starting at pg1 to journal */
4192 int ii;
4193 int needSync = 0;
4194
4195 /* Set the doNotSync flag to 1. This is because we cannot allow a journal
4196 ** header to be written between the pages journaled by this function.
4197 */
4198 assert( pPager->doNotSync==0 );
4199 pPager->doNotSync = 1;
4200
4201 /* This trick assumes that both the page-size and sector-size are
4202 ** an integer power of 2. It sets variable pg1 to the identifier
4203 ** of the first page of the sector pPg is located on.
4204 */
4205 pg1 = ((pPg->pgno-1) & ~(nPagePerSector-1)) + 1;
4206
4207 nPageCount = sqlite3PagerPagecount(pPager);
4208 if( pPg->pgno>nPageCount ){
4209 nPage = (pPg->pgno - pg1)+1;
4210 }else if( (pg1+nPagePerSector-1)>nPageCount ){
4211 nPage = nPageCount+1-pg1;
4212 }else{
4213 nPage = nPagePerSector;
4214 }
4215 assert(nPage>0);
4216 assert(pg1<=pPg->pgno);
4217 assert((pg1+nPage)>pPg->pgno);
4218
4219 for(ii=0; ii<nPage && rc==SQLITE_OK; ii++){
4220 Pgno pg = pg1+ii;
4221 PgHdr *pPage;
4222 if( !pPager->aInJournal || pg==pPg->pgno ||
4223 pg>pPager->origDbSize || !(pPager->aInJournal[pg/8]&(1<<(pg&7)))
4224 ) {
4225 if( pg!=PAGER_MJ_PGNO(pPager) ){
4226 rc = sqlite3PagerGet(pPager, pg, &pPage);
4227 if( rc==SQLITE_OK ){
4228 rc = pager_write(pPage);
4229 if( pPage->needSync ){
4230 needSync = 1;
4231 }
4232 sqlite3PagerUnref(pPage);
4233 }
4234 }
4235 }else if( (pPage = pager_lookup(pPager, pg)) ){
4236 if( pPage->needSync ){
4237 needSync = 1;
4238 }
4239 }
4240 }
4241
4242 /* If the PgHdr.needSync flag is set for any of the nPage pages
4243 ** starting at pg1, then it needs to be set for all of them. Because
4244 ** writing to any of these nPage pages may damage the others, the
4245 ** journal file must contain sync()ed copies of all of them
4246 ** before any of them can be written out to the database file.
4247 */
4248 if( needSync ){
4249 for(ii=0; ii<nPage && needSync; ii++){
4250 PgHdr *pPage = pager_lookup(pPager, pg1+ii);
4251 if( pPage ) pPage->needSync = 1;
4252 }
4253 assert(pPager->needSync);
4254 }
4255
4256 assert( pPager->doNotSync==1 );
4257 pPager->doNotSync = 0;
4258 }else{
4259 rc = pager_write(pDbPage);
4260 }
4261 pagerLeave(pPager);
4262 return rc;
4263}
4264
4265/*
4266** Return TRUE if the page given in the argument was previously passed
4267** to sqlite3PagerWrite(). In other words, return TRUE if it is ok
4268** to change the content of the page.
4269*/
4270#ifndef NDEBUG
4271int sqlite3PagerIswriteable(DbPage *pPg){
4272 return pPg->dirty;
4273}
4274#endif
4275
4276#ifndef SQLITE_OMIT_VACUUM
4277/*
4278** Replace the content of a single page with the information in the third
4279** argument.
4280*/
4281int sqlite3PagerOverwrite(Pager *pPager, Pgno pgno, void *pData){
4282 PgHdr *pPg;
4283 int rc;
4284
4285 pagerEnter(pPager);
4286 rc = sqlite3PagerGet(pPager, pgno, &pPg);
4287 if( rc==SQLITE_OK ){
4288 rc = sqlite3PagerWrite(pPg);
4289 if( rc==SQLITE_OK ){
4290 memcpy(sqlite3PagerGetData(pPg), pData, pPager->pageSize);
4291 }
4292 sqlite3PagerUnref(pPg);
4293 }
4294 pagerLeave(pPager);
4295 return rc;
4296}
4297#endif
4298
4299/*
4300** A call to this routine tells the pager that it is not necessary to
4301** write the information on page pPg back to the disk, even though
4302** that page might be marked as dirty.
4303**
4304** The overlying software layer calls this routine when all of the data
4305** on the given page is unused. The pager marks the page as clean so
4306** that it does not get written to disk.
4307**
4308** Tests show that this optimization, together with the
4309** sqlite3PagerDontRollback() below, more than double the speed
4310** of large INSERT operations and quadruple the speed of large DELETEs.
4311**
4312** When this routine is called, set the alwaysRollback flag to true.
4313** Subsequent calls to sqlite3PagerDontRollback() for the same page
4314** will thereafter be ignored. This is necessary to avoid a problem
4315** where a page with data is added to the freelist during one part of
4316** a transaction then removed from the freelist during a later part
4317** of the same transaction and reused for some other purpose. When it
4318** is first added to the freelist, this routine is called. When reused,
4319** the sqlite3PagerDontRollback() routine is called. But because the
4320** page contains critical data, we still need to be sure it gets
4321** rolled back in spite of the sqlite3PagerDontRollback() call.
4322*/
4323void sqlite3PagerDontWrite(DbPage *pDbPage){
4324 PgHdr *pPg = pDbPage;
4325 Pager *pPager = pPg->pPager;
4326
4327 if( MEMDB ) return;
4328 pagerEnter(pPager);
4329 pPg->alwaysRollback = 1;
4330 if( pPg->dirty && !pPager->stmtInUse ){
4331 assert( pPager->state>=PAGER_SHARED );
4332 if( pPager->dbSize==(int)pPg->pgno && pPager->origDbSize<pPager->dbSize ){
4333 /* If this pages is the last page in the file and the file has grown
4334 ** during the current transaction, then do NOT mark the page as clean.
4335 ** When the database file grows, we must make sure that the last page
4336 ** gets written at least once so that the disk file will be the correct
4337 ** size. If you do not write this page and the size of the file
4338 ** on the disk ends up being too small, that can lead to database
4339 ** corruption during the next transaction.
4340 */
4341 }else{
4342 PAGERTRACE3("DONT_WRITE page %d of %d\n", pPg->pgno, PAGERID(pPager));
4343 IOTRACE(("CLEAN %p %d\n", pPager, pPg->pgno))
4344 makeClean(pPg);
4345#ifdef SQLITE_CHECK_PAGES
4346 pPg->pageHash = pager_pagehash(pPg);
4347#endif
4348 }
4349 }
4350 pagerLeave(pPager);
4351}
4352
4353/*
4354** A call to this routine tells the pager that if a rollback occurs,
4355** it is not necessary to restore the data on the given page. This
4356** means that the pager does not have to record the given page in the
4357** rollback journal.
4358**
4359** If we have not yet actually read the content of this page (if
4360** the PgHdr.needRead flag is set) then this routine acts as a promise
4361** that we will never need to read the page content in the future.
4362** so the needRead flag can be cleared at this point.
4363*/
4364void sqlite3PagerDontRollback(DbPage *pPg){
4365 Pager *pPager = pPg->pPager;
4366
4367 pagerEnter(pPager);
4368 assert( pPager->state>=PAGER_RESERVED );
4369 if( pPager->journalOpen==0 ) return;
4370 if( pPg->alwaysRollback || pPager->alwaysRollback || MEMDB ) return;
4371 if( !pPg->inJournal && (int)pPg->pgno <= pPager->origDbSize ){
4372 assert( pPager->aInJournal!=0 );
4373 pPager->aInJournal[pPg->pgno/8] |= 1<<(pPg->pgno&7);
4374 pPg->inJournal = 1;
4375 pPg->needRead = 0;
4376 if( pPager->stmtInUse ){
4377 pPager->aInStmt[pPg->pgno/8] |= 1<<(pPg->pgno&7);
4378 }
4379 PAGERTRACE3("DONT_ROLLBACK page %d of %d\n", pPg->pgno, PAGERID(pPager));
4380 IOTRACE(("GARBAGE %p %d\n", pPager, pPg->pgno))
4381 }
4382 if( pPager->stmtInUse
4383 && !pageInStatement(pPg)
4384 && (int)pPg->pgno<=pPager->stmtSize
4385 ){
4386 assert( pPg->inJournal || (int)pPg->pgno>pPager->origDbSize );
4387 assert( pPager->aInStmt!=0 );
4388 pPager->aInStmt[pPg->pgno/8] |= 1<<(pPg->pgno&7);
4389 }
4390 pagerLeave(pPager);
4391}
4392
4393
4394/*
4395** This routine is called to increment the database file change-counter,
4396** stored at byte 24 of the pager file.
4397*/
4398static int pager_incr_changecounter(Pager *pPager, int isDirect){
4399 PgHdr *pPgHdr;
4400 u32 change_counter;
4401 int rc = SQLITE_OK;
4402
4403 if( !pPager->changeCountDone ){
4404 /* Open page 1 of the file for writing. */
4405 rc = sqlite3PagerGet(pPager, 1, &pPgHdr);
4406 if( rc!=SQLITE_OK ) return rc;
4407
4408 if( !isDirect ){
4409 rc = sqlite3PagerWrite(pPgHdr);
4410 if( rc!=SQLITE_OK ){
4411 sqlite3PagerUnref(pPgHdr);
4412 return rc;
4413 }
4414 }
4415
4416 /* Increment the value just read and write it back to byte 24. */
4417 change_counter = sqlite3Get4byte((u8*)pPager->dbFileVers);
4418 change_counter++;
4419 put32bits(((char*)PGHDR_TO_DATA(pPgHdr))+24, change_counter);
4420
4421 if( isDirect && pPager->fd->pMethods ){
4422 const void *zBuf = PGHDR_TO_DATA(pPgHdr);
4423 rc = sqlite3OsWrite(pPager->fd, zBuf, pPager->pageSize, 0);
4424 }
4425
4426 /* Release the page reference. */
4427 sqlite3PagerUnref(pPgHdr);
4428 pPager->changeCountDone = 1;
4429 }
4430 return rc;
4431}
4432
4433/*
4434** Sync the database file for the pager pPager. zMaster points to the name
4435** of a master journal file that should be written into the individual
4436** journal file. zMaster may be NULL, which is interpreted as no master
4437** journal (a single database transaction).
4438**
4439** This routine ensures that the journal is synced, all dirty pages written
4440** to the database file and the database file synced. The only thing that
4441** remains to commit the transaction is to delete the journal file (or
4442** master journal file if specified).
4443**
4444** Note that if zMaster==NULL, this does not overwrite a previous value
4445** passed to an sqlite3PagerCommitPhaseOne() call.
4446**
4447** If parameter nTrunc is non-zero, then the pager file is truncated to
4448** nTrunc pages (this is used by auto-vacuum databases).
4449*/
4450int sqlite3PagerCommitPhaseOne(Pager *pPager, const char *zMaster, Pgno nTrunc){
4451 int rc = SQLITE_OK;
4452
4453 PAGERTRACE4("DATABASE SYNC: File=%s zMaster=%s nTrunc=%d\n",
4454 pPager->zFilename, zMaster, nTrunc);
4455 pagerEnter(pPager);
4456
4457 /* If this is an in-memory db, or no pages have been written to, or this
4458 ** function has already been called, it is a no-op.
4459 */
4460 if( pPager->state!=PAGER_SYNCED && !MEMDB && pPager->dirtyCache ){
4461 PgHdr *pPg;
4462
4463#ifdef SQLITE_ENABLE_ATOMIC_WRITE
4464 /* The atomic-write optimization can be used if all of the
4465 ** following are true:
4466 **
4467 ** + The file-system supports the atomic-write property for
4468 ** blocks of size page-size, and
4469 ** + This commit is not part of a multi-file transaction, and
4470 ** + Exactly one page has been modified and store in the journal file.
4471 **
4472 ** If the optimization can be used, then the journal file will never
4473 ** be created for this transaction.
4474 */
4475 int useAtomicWrite = (
4476 !zMaster &&
4477 pPager->journalOff==jrnlBufferSize(pPager) &&
4478 nTrunc==0 &&
4479 (0==pPager->pDirty || 0==pPager->pDirty->pDirty)
4480 );
4481 if( useAtomicWrite ){
4482 /* Update the nRec field in the journal file. */
4483 int offset = pPager->journalHdr + sizeof(aJournalMagic);
4484 assert(pPager->nRec==1);
4485 rc = write32bits(pPager->jfd, offset, pPager->nRec);
4486
4487 /* Update the db file change counter. The following call will modify
4488 ** the in-memory representation of page 1 to include the updated
4489 ** change counter and then write page 1 directly to the database
4490 ** file. Because of the atomic-write property of the host file-system,
4491 ** this is safe.
4492 */
4493 if( rc==SQLITE_OK ){
4494 rc = pager_incr_changecounter(pPager, 1);
4495 }
4496 }else{
4497 rc = sqlite3JournalCreate(pPager->jfd);
4498 }
4499
4500 if( !useAtomicWrite && rc==SQLITE_OK )
4501#endif
4502
4503 /* If a master journal file name has already been written to the
4504 ** journal file, then no sync is required. This happens when it is
4505 ** written, then the process fails to upgrade from a RESERVED to an
4506 ** EXCLUSIVE lock. The next time the process tries to commit the
4507 ** transaction the m-j name will have already been written.
4508 */
4509 if( !pPager->setMaster ){
4510 assert( pPager->journalOpen );
4511 rc = pager_incr_changecounter(pPager, 0);
4512 if( rc!=SQLITE_OK ) goto sync_exit;
4513#ifndef SQLITE_OMIT_AUTOVACUUM
4514 if( nTrunc!=0 ){
4515 /* If this transaction has made the database smaller, then all pages
4516 ** being discarded by the truncation must be written to the journal
4517 ** file.
4518 */
4519 Pgno i;
4520 int iSkip = PAGER_MJ_PGNO(pPager);
4521 for( i=nTrunc+1; i<=pPager->origDbSize; i++ ){
4522 if( !(pPager->aInJournal[i/8] & (1<<(i&7))) && i!=iSkip ){
4523 rc = sqlite3PagerGet(pPager, i, &pPg);
4524 if( rc!=SQLITE_OK ) goto sync_exit;
4525 rc = sqlite3PagerWrite(pPg);
4526 sqlite3PagerUnref(pPg);
4527 if( rc!=SQLITE_OK ) goto sync_exit;
4528 }
4529 }
4530 }
4531#endif
4532 rc = writeMasterJournal(pPager, zMaster);
4533 if( rc!=SQLITE_OK ) goto sync_exit;
4534 rc = syncJournal(pPager);
4535 }
4536 if( rc!=SQLITE_OK ) goto sync_exit;
4537
4538#ifndef SQLITE_OMIT_AUTOVACUUM
4539 if( nTrunc!=0 ){
4540 rc = sqlite3PagerTruncate(pPager, nTrunc);
4541 if( rc!=SQLITE_OK ) goto sync_exit;
4542 }
4543#endif
4544
4545 /* Write all dirty pages to the database file */
4546 pPg = pager_get_all_dirty_pages(pPager);
4547 rc = pager_write_pagelist(pPg);
4548 if( rc!=SQLITE_OK ){
4549 while( pPg && !pPg->dirty ){ pPg = pPg->pDirty; }
4550 pPager->pDirty = pPg;
4551 goto sync_exit;
4552 }
4553 pPager->pDirty = 0;
4554
4555 /* Sync the database file. */
4556 if( !pPager->noSync ){
4557 rc = sqlite3OsSync(pPager->fd, pPager->sync_flags);
4558 }
4559 IOTRACE(("DBSYNC %p\n", pPager))
4560
4561 pPager->state = PAGER_SYNCED;
4562 }else if( MEMDB && nTrunc!=0 ){
4563 rc = sqlite3PagerTruncate(pPager, nTrunc);
4564 }
4565
4566sync_exit:
4567 if( rc==SQLITE_IOERR_BLOCKED ){
4568 /* pager_incr_changecounter() may attempt to obtain an exclusive
4569 * lock to spill the cache and return IOERR_BLOCKED. But since
4570 * there is no chance the cache is inconsistent, it's
4571 * better to return SQLITE_BUSY.
4572 */
4573 rc = SQLITE_BUSY;
4574 }
4575 pagerLeave(pPager);
4576 return rc;
4577}
4578
4579
4580/*
4581** Commit all changes to the database and release the write lock.
4582**
4583** If the commit fails for any reason, a rollback attempt is made
4584** and an error code is returned. If the commit worked, SQLITE_OK
4585** is returned.
4586*/
4587int sqlite3PagerCommitPhaseTwo(Pager *pPager){
4588 int rc;
4589 PgHdr *pPg;
4590
4591 if( pPager->errCode ){
4592 return pPager->errCode;
4593 }
4594 if( pPager->state<PAGER_RESERVED ){
4595 return SQLITE_ERROR;
4596 }
4597 pagerEnter(pPager);
4598 PAGERTRACE2("COMMIT %d\n", PAGERID(pPager));
4599 if( MEMDB ){
4600 pPg = pager_get_all_dirty_pages(pPager);
4601 while( pPg ){
4602 PgHistory *pHist = PGHDR_TO_HIST(pPg, pPager);
4603 clearHistory(pHist);
4604 pPg->dirty = 0;
4605 pPg->inJournal = 0;
4606 pHist->inStmt = 0;
4607 pPg->needSync = 0;
4608 pHist->pPrevStmt = pHist->pNextStmt = 0;
4609 pPg = pPg->pDirty;
4610 }
4611 pPager->pDirty = 0;
4612#ifndef NDEBUG
4613 for(pPg=pPager->pAll; pPg; pPg=pPg->pNextAll){
4614 PgHistory *pHist = PGHDR_TO_HIST(pPg, pPager);
4615 assert( !pPg->alwaysRollback );
4616 assert( !pHist->pOrig );
4617 assert( !pHist->pStmt );
4618 }
4619#endif
4620 pPager->pStmt = 0;
4621 pPager->state = PAGER_SHARED;
4622 return SQLITE_OK;
4623 }
4624 assert( pPager->journalOpen || !pPager->dirtyCache );
4625 assert( pPager->state==PAGER_SYNCED || !pPager->dirtyCache );
4626 rc = pager_end_transaction(pPager);
4627 rc = pager_error(pPager, rc);
4628 pagerLeave(pPager);
4629 return rc;
4630}
4631
4632/*
4633** Rollback all changes. The database falls back to PAGER_SHARED mode.
4634** All in-memory cache pages revert to their original data contents.
4635** The journal is deleted.
4636**
4637** This routine cannot fail unless some other process is not following
4638** the correct locking protocol or unless some other
4639** process is writing trash into the journal file (SQLITE_CORRUPT) or
4640** unless a prior malloc() failed (SQLITE_NOMEM). Appropriate error
4641** codes are returned for all these occasions. Otherwise,
4642** SQLITE_OK is returned.
4643*/
4644int sqlite3PagerRollback(Pager *pPager){
4645 int rc;
4646 PAGERTRACE2("ROLLBACK %d\n", PAGERID(pPager));
4647 if( MEMDB ){
4648 PgHdr *p;
4649 for(p=pPager->pAll; p; p=p->pNextAll){
4650 PgHistory *pHist;
4651 assert( !p->alwaysRollback );
4652 if( !p->dirty ){
4653 assert( !((PgHistory *)PGHDR_TO_HIST(p, pPager))->pOrig );
4654 assert( !((PgHistory *)PGHDR_TO_HIST(p, pPager))->pStmt );
4655 continue;
4656 }
4657
4658 pHist = PGHDR_TO_HIST(p, pPager);
4659 if( pHist->pOrig ){
4660 memcpy(PGHDR_TO_DATA(p), pHist->pOrig, pPager->pageSize);
4661 PAGERTRACE3("ROLLBACK-PAGE %d of %d\n", p->pgno, PAGERID(pPager));
4662 }else{
4663 PAGERTRACE3("PAGE %d is clean on %d\n", p->pgno, PAGERID(pPager));
4664 }
4665 clearHistory(pHist);
4666 p->dirty = 0;
4667 p->inJournal = 0;
4668 pHist->inStmt = 0;
4669 pHist->pPrevStmt = pHist->pNextStmt = 0;
4670 if( pPager->xReiniter ){
4671 pPager->xReiniter(p, pPager->pageSize);
4672 }
4673 }
4674 pPager->pDirty = 0;
4675 pPager->pStmt = 0;
4676 pPager->dbSize = pPager->origDbSize;
4677 pager_truncate_cache(pPager);
4678 pPager->stmtInUse = 0;
4679 pPager->state = PAGER_SHARED;
4680 return SQLITE_OK;
4681 }
4682
4683 pagerEnter(pPager);
4684 if( !pPager->dirtyCache || !pPager->journalOpen ){
4685 rc = pager_end_transaction(pPager);
4686 pagerLeave(pPager);
4687 return rc;
4688 }
4689
4690 if( pPager->errCode && pPager->errCode!=SQLITE_FULL ){
4691 if( pPager->state>=PAGER_EXCLUSIVE ){
4692 pager_playback(pPager, 0);
4693 }
4694 pagerLeave(pPager);
4695 return pPager->errCode;
4696 }
4697 if( pPager->state==PAGER_RESERVED ){
4698 int rc2;
4699 rc = pager_playback(pPager, 0);
4700 rc2 = pager_end_transaction(pPager);
4701 if( rc==SQLITE_OK ){
4702 rc = rc2;
4703 }
4704 }else{
4705 rc = pager_playback(pPager, 0);
4706 }
4707 /* pager_reset(pPager); */
4708 pPager->dbSize = -1;
4709
4710 /* If an error occurs during a ROLLBACK, we can no longer trust the pager
4711 ** cache. So call pager_error() on the way out to make any error
4712 ** persistent.
4713 */
4714 rc = pager_error(pPager, rc);
4715 pagerLeave(pPager);
4716 return rc;
4717}
4718
4719/*
4720** Return TRUE if the database file is opened read-only. Return FALSE
4721** if the database is (in theory) writable.
4722*/
4723int sqlite3PagerIsreadonly(Pager *pPager){
4724 return pPager->readOnly;
4725}
4726
4727/*
4728** Return the number of references to the pager.
4729*/
4730int sqlite3PagerRefcount(Pager *pPager){
4731 return pPager->nRef;
4732}
4733
4734#ifdef SQLITE_TEST
4735/*
4736** This routine is used for testing and analysis only.
4737*/
4738int *sqlite3PagerStats(Pager *pPager){
4739 static int a[11];
4740 a[0] = pPager->nRef;
4741 a[1] = pPager->nPage;
4742 a[2] = pPager->mxPage;
4743 a[3] = pPager->dbSize;
4744 a[4] = pPager->state;
4745 a[5] = pPager->errCode;
4746 a[6] = pPager->nHit;
4747 a[7] = pPager->nMiss;
4748 a[8] = 0; /* Used to be pPager->nOvfl */
4749 a[9] = pPager->nRead;
4750 a[10] = pPager->nWrite;
4751 return a;
4752}
4753#endif
4754
4755/*
4756** Set the statement rollback point.
4757**
4758** This routine should be called with the transaction journal already
4759** open. A new statement journal is created that can be used to rollback
4760** changes of a single SQL command within a larger transaction.
4761*/
4762static int pagerStmtBegin(Pager *pPager){
4763 int rc;
4764 assert( !pPager->stmtInUse );
4765 assert( pPager->state>=PAGER_SHARED );
4766 assert( pPager->dbSize>=0 );
4767 PAGERTRACE2("STMT-BEGIN %d\n", PAGERID(pPager));
4768 if( MEMDB ){
4769 pPager->stmtInUse = 1;
4770 pPager->stmtSize = pPager->dbSize;
4771 return SQLITE_OK;
4772 }
4773 if( !pPager->journalOpen ){
4774 pPager->stmtAutoopen = 1;
4775 return SQLITE_OK;
4776 }
4777 assert( pPager->journalOpen );
4778 pagerLeave(pPager);
4779 assert( pPager->aInStmt==0 );
4780 pPager->aInStmt = sqlite3MallocZero( pPager->dbSize/8 + 1 );
4781 pagerEnter(pPager);
4782 if( pPager->aInStmt==0 ){
4783 /* sqlite3OsLock(pPager->fd, SHARED_LOCK); */
4784 return SQLITE_NOMEM;
4785 }
4786#ifndef NDEBUG
4787 rc = sqlite3OsFileSize(pPager->jfd, &pPager->stmtJSize);
4788 if( rc ) goto stmt_begin_failed;
4789 assert( pPager->stmtJSize == pPager->journalOff );
4790#endif
4791 pPager->stmtJSize = pPager->journalOff;
4792 pPager->stmtSize = pPager->dbSize;
4793 pPager->stmtHdrOff = 0;
4794 pPager->stmtCksum = pPager->cksumInit;
4795 if( !pPager->stmtOpen ){
4796 rc = sqlite3PagerOpentemp(pPager->pVfs, pPager->stfd, pPager->zStmtJrnl,
4797 SQLITE_OPEN_SUBJOURNAL);
4798 if( rc ){
4799 goto stmt_begin_failed;
4800 }
4801 pPager->stmtOpen = 1;
4802 pPager->stmtNRec = 0;
4803 }
4804 pPager->stmtInUse = 1;
4805 return SQLITE_OK;
4806
4807stmt_begin_failed:
4808 if( pPager->aInStmt ){
4809 sqlite3_free(pPager->aInStmt);
4810 pPager->aInStmt = 0;
4811 }
4812 return rc;
4813}
4814int sqlite3PagerStmtBegin(Pager *pPager){
4815 int rc;
4816 pagerEnter(pPager);
4817 rc = pagerStmtBegin(pPager);
4818 pagerLeave(pPager);
4819 return rc;
4820}
4821
4822/*
4823** Commit a statement.
4824*/
4825int sqlite3PagerStmtCommit(Pager *pPager){
4826 pagerEnter(pPager);
4827 if( pPager->stmtInUse ){
4828 PgHdr *pPg, *pNext;
4829 PAGERTRACE2("STMT-COMMIT %d\n", PAGERID(pPager));
4830 if( !MEMDB ){
4831 /* sqlite3OsTruncate(pPager->stfd, 0); */
4832 sqlite3_free( pPager->aInStmt );
4833 pPager->aInStmt = 0;
4834 }else{
4835 for(pPg=pPager->pStmt; pPg; pPg=pNext){
4836 PgHistory *pHist = PGHDR_TO_HIST(pPg, pPager);
4837 pNext = pHist->pNextStmt;
4838 assert( pHist->inStmt );
4839 pHist->inStmt = 0;
4840 pHist->pPrevStmt = pHist->pNextStmt = 0;
4841 sqlite3_free(pHist->pStmt);
4842 pHist->pStmt = 0;
4843 }
4844 }
4845 pPager->stmtNRec = 0;
4846 pPager->stmtInUse = 0;
4847 pPager->pStmt = 0;
4848 }
4849 pPager->stmtAutoopen = 0;
4850 pagerLeave(pPager);
4851 return SQLITE_OK;
4852}
4853
4854/*
4855** Rollback a statement.
4856*/
4857int sqlite3PagerStmtRollback(Pager *pPager){
4858 int rc;
4859 pagerEnter(pPager);
4860 if( pPager->stmtInUse ){
4861 PAGERTRACE2("STMT-ROLLBACK %d\n", PAGERID(pPager));
4862 if( MEMDB ){
4863 PgHdr *pPg;
4864 PgHistory *pHist;
4865 for(pPg=pPager->pStmt; pPg; pPg=pHist->pNextStmt){
4866 pHist = PGHDR_TO_HIST(pPg, pPager);
4867 if( pHist->pStmt ){
4868 memcpy(PGHDR_TO_DATA(pPg), pHist->pStmt, pPager->pageSize);
4869 sqlite3_free(pHist->pStmt);
4870 pHist->pStmt = 0;
4871 }
4872 }
4873 pPager->dbSize = pPager->stmtSize;
4874 pager_truncate_cache(pPager);
4875 rc = SQLITE_OK;
4876 }else{
4877 rc = pager_stmt_playback(pPager);
4878 }
4879 sqlite3PagerStmtCommit(pPager);
4880 }else{
4881 rc = SQLITE_OK;
4882 }
4883 pPager->stmtAutoopen = 0;
4884 pagerLeave(pPager);
4885 return rc;
4886}
4887
4888/*
4889** Return the full pathname of the database file.
4890*/
4891const char *sqlite3PagerFilename(Pager *pPager){
4892 return pPager->zFilename;
4893}
4894
4895/*
4896** Return the VFS structure for the pager.
4897*/
4898const sqlite3_vfs *sqlite3PagerVfs(Pager *pPager){
4899 return pPager->pVfs;
4900}
4901
4902/*
4903** Return the file handle for the database file associated
4904** with the pager. This might return NULL if the file has
4905** not yet been opened.
4906*/
4907sqlite3_file *sqlite3PagerFile(Pager *pPager){
4908 return pPager->fd;
4909}
4910
4911/*
4912** Return the directory of the database file.
4913*/
4914const char *sqlite3PagerDirname(Pager *pPager){
4915 return pPager->zDirectory;
4916}
4917
4918/*
4919** Return the full pathname of the journal file.
4920*/
4921const char *sqlite3PagerJournalname(Pager *pPager){
4922 return pPager->zJournal;
4923}
4924
4925/*
4926** Return true if fsync() calls are disabled for this pager. Return FALSE
4927** if fsync()s are executed normally.
4928*/
4929int sqlite3PagerNosync(Pager *pPager){
4930 return pPager->noSync;
4931}
4932
4933#ifdef SQLITE_HAS_CODEC
4934/*
4935** Set the codec for this pager
4936*/
4937void sqlite3PagerSetCodec(
4938 Pager *pPager,
4939 void *(*xCodec)(void*,void*,Pgno,int),
4940 void *pCodecArg
4941){
4942 pPager->xCodec = xCodec;
4943 pPager->pCodecArg = pCodecArg;
4944}
4945#endif
4946
4947#ifndef SQLITE_OMIT_AUTOVACUUM
4948/*
4949** Move the page pPg to location pgno in the file.
4950**
4951** There must be no references to the page previously located at
4952** pgno (which we call pPgOld) though that page is allowed to be
4953** in cache. If the page previous located at pgno is not already
4954** in the rollback journal, it is not put there by by this routine.
4955**
4956** References to the page pPg remain valid. Updating any
4957** meta-data associated with pPg (i.e. data stored in the nExtra bytes
4958** allocated along with the page) is the responsibility of the caller.
4959**
4960** A transaction must be active when this routine is called. It used to be
4961** required that a statement transaction was not active, but this restriction
4962** has been removed (CREATE INDEX needs to move a page when a statement
4963** transaction is active).
4964*/
4965int sqlite3PagerMovepage(Pager *pPager, DbPage *pPg, Pgno pgno){
4966 PgHdr *pPgOld; /* The page being overwritten. */
4967 int h;
4968 Pgno needSyncPgno = 0;
4969
4970 pagerEnter(pPager);
4971 assert( pPg->nRef>0 );
4972
4973 PAGERTRACE5("MOVE %d page %d (needSync=%d) moves to %d\n",
4974 PAGERID(pPager), pPg->pgno, pPg->needSync, pgno);
4975 IOTRACE(("MOVE %p %d %d\n", pPager, pPg->pgno, pgno))
4976
4977 pager_get_content(pPg);
4978 if( pPg->needSync ){
4979 needSyncPgno = pPg->pgno;
4980 assert( pPg->inJournal || (int)pgno>pPager->origDbSize );
4981 assert( pPg->dirty );
4982 assert( pPager->needSync );
4983 }
4984
4985 /* Unlink pPg from it's hash-chain */
4986 unlinkHashChain(pPager, pPg);
4987
4988 /* If the cache contains a page with page-number pgno, remove it
4989 ** from it's hash chain. Also, if the PgHdr.needSync was set for
4990 ** page pgno before the 'move' operation, it needs to be retained
4991 ** for the page moved there.
4992 */
4993 pPg->needSync = 0;
4994 pPgOld = pager_lookup(pPager, pgno);
4995 if( pPgOld ){
4996 assert( pPgOld->nRef==0 );
4997 unlinkHashChain(pPager, pPgOld);
4998 makeClean(pPgOld);
4999 pPg->needSync = pPgOld->needSync;
5000 }else{
5001 pPg->needSync = 0;
5002 }
5003 if( pPager->aInJournal && (int)pgno<=pPager->origDbSize ){
5004 pPg->inJournal = (pPager->aInJournal[pgno/8] & (1<<(pgno&7)))!=0;
5005 }else{
5006 pPg->inJournal = 0;
5007 assert( pPg->needSync==0 || (int)pgno>pPager->origDbSize );
5008 }
5009
5010 /* Change the page number for pPg and insert it into the new hash-chain. */
5011 assert( pgno!=0 );
5012 pPg->pgno = pgno;
5013 h = pgno & (pPager->nHash-1);
5014 if( pPager->aHash[h] ){
5015 assert( pPager->aHash[h]->pPrevHash==0 );
5016 pPager->aHash[h]->pPrevHash = pPg;
5017 }
5018 pPg->pNextHash = pPager->aHash[h];
5019 pPager->aHash[h] = pPg;
5020 pPg->pPrevHash = 0;
5021
5022 makeDirty(pPg);
5023 pPager->dirtyCache = 1;
5024
5025 if( needSyncPgno ){
5026 /* If needSyncPgno is non-zero, then the journal file needs to be
5027 ** sync()ed before any data is written to database file page needSyncPgno.
5028 ** Currently, no such page exists in the page-cache and the
5029 ** Pager.aInJournal bit has been set. This needs to be remedied by loading
5030 ** the page into the pager-cache and setting the PgHdr.needSync flag.
5031 **
5032 ** The sqlite3PagerGet() call may cause the journal to sync. So make
5033 ** sure the Pager.needSync flag is set too.
5034 */
5035 int rc;
5036 PgHdr *pPgHdr;
5037 assert( pPager->needSync );
5038 rc = sqlite3PagerGet(pPager, needSyncPgno, &pPgHdr);
5039 if( rc!=SQLITE_OK ) return rc;
5040 pPager->needSync = 1;
5041 pPgHdr->needSync = 1;
5042 pPgHdr->inJournal = 1;
5043 makeDirty(pPgHdr);
5044 sqlite3PagerUnref(pPgHdr);
5045 }
5046
5047 pagerLeave(pPager);
5048 return SQLITE_OK;
5049}
5050#endif
5051
5052/*
5053** Return a pointer to the data for the specified page.
5054*/
5055void *sqlite3PagerGetData(DbPage *pPg){
5056 return PGHDR_TO_DATA(pPg);
5057}
5058
5059/*
5060** Return a pointer to the Pager.nExtra bytes of "extra" space
5061** allocated along with the specified page.
5062*/
5063void *sqlite3PagerGetExtra(DbPage *pPg){
5064 Pager *pPager = pPg->pPager;
5065 return (pPager?PGHDR_TO_EXTRA(pPg, pPager):0);
5066}
5067
5068/*
5069** Get/set the locking-mode for this pager. Parameter eMode must be one
5070** of PAGER_LOCKINGMODE_QUERY, PAGER_LOCKINGMODE_NORMAL or
5071** PAGER_LOCKINGMODE_EXCLUSIVE. If the parameter is not _QUERY, then
5072** the locking-mode is set to the value specified.
5073**
5074** The returned value is either PAGER_LOCKINGMODE_NORMAL or
5075** PAGER_LOCKINGMODE_EXCLUSIVE, indicating the current (possibly updated)
5076** locking-mode.
5077*/
5078int sqlite3PagerLockingMode(Pager *pPager, int eMode){
5079 assert( eMode==PAGER_LOCKINGMODE_QUERY
5080 || eMode==PAGER_LOCKINGMODE_NORMAL
5081 || eMode==PAGER_LOCKINGMODE_EXCLUSIVE );
5082 assert( PAGER_LOCKINGMODE_QUERY<0 );
5083 assert( PAGER_LOCKINGMODE_NORMAL>=0 && PAGER_LOCKINGMODE_EXCLUSIVE>=0 );
5084 if( eMode>=0 && !pPager->tempFile ){
5085 pPager->exclusiveMode = eMode;
5086 }
5087 return (int)pPager->exclusiveMode;
5088}
5089
5090#ifdef SQLITE_DEBUG
5091/*
5092** Print a listing of all referenced pages and their ref count.
5093*/
5094void sqlite3PagerRefdump(Pager *pPager){
5095 PgHdr *pPg;
5096 for(pPg=pPager->pAll; pPg; pPg=pPg->pNextAll){
5097 if( pPg->nRef<=0 ) continue;
5098 sqlite3DebugPrintf("PAGE %3d addr=%p nRef=%d\n",
5099 pPg->pgno, PGHDR_TO_DATA(pPg), pPg->nRef);
5100 }
5101}
5102#endif
5103
5104#endif /* SQLITE_OMIT_DISKIO */
generated by cgit v1.1 at 2024-11-10 20:44:16 +0000