数据库系统

开发平台：

C/C++

pager.c：源码内容

/*
** 2001 September 15
**
** The author disclaims copyright to this source code. In place of
** a legal notice, here is a blessing:
**
** May you do good and not evil.
** May you find forgiveness for yourself and forgive others.
** May you share freely, never taking more than you give.
**
*************************************************************************
** This is the implementation of the page cache subsystem or "pager".
**
** The pager is used to access a database disk file. It implements
** atomic commit and rollback through the use of a journal file that
** is separate from the database file. The pager also implements file
** locking to prevent two processes from writing the same database
** file simultaneously, or one process from reading the database while
** another is writing.
**
** @(#) $Id: pager.c,v 1.426 2008/04/14 23:13:46 drh Exp $
*/
#ifndef SQLITE_OMIT_DISKIO
#include "sqliteInt.h"
#include <assert.h>
#include <string.h>
/*
** Macros for troubleshooting. Normally turned off
*/
#if 0
#define sqlite3DebugPrintf printf
#define PAGERTRACE1(X) sqlite3DebugPrintf(X)
#define PAGERTRACE2(X,Y) sqlite3DebugPrintf(X,Y)
#define PAGERTRACE3(X,Y,Z) sqlite3DebugPrintf(X,Y,Z)
#define PAGERTRACE4(X,Y,Z,W) sqlite3DebugPrintf(X,Y,Z,W)
#define PAGERTRACE5(X,Y,Z,W,V) sqlite3DebugPrintf(X,Y,Z,W,V)
#else
#define PAGERTRACE1(X)
#define PAGERTRACE2(X,Y)
#define PAGERTRACE3(X,Y,Z)
#define PAGERTRACE4(X,Y,Z,W)
#define PAGERTRACE5(X,Y,Z,W,V)
#endif
/*
** The following two macros are used within the PAGERTRACEX() macros above
** to print out file-descriptors.
**
** PAGERID() takes a pointer to a Pager struct as its argument. The
** associated file-descriptor is returned. FILEHANDLEID() takes an sqlite3_file
** struct as its argument.
*/
#define PAGERID(p) ((int)(p->fd))
#define FILEHANDLEID(fd) ((int)fd)
/*
** The page cache as a whole is always in one of the following
** states:
**
** PAGER_UNLOCK The page cache is not currently reading or
** writing the database file. There is no
** data held in memory. This is the initial
** state.
**
** PAGER_SHARED The page cache is reading the database.
** Writing is not permitted. There can be
** multiple readers accessing the same database
** file at the same time.
**
** PAGER_RESERVED This process has reserved the database for writing
** but has not yet made any changes. Only one process
** at a time can reserve the database. The original
** database file has not been modified so other
** processes may still be reading the on-disk
** database file.
**
** PAGER_EXCLUSIVE The page cache is writing the database.
** Access is exclusive. No other processes or
** threads can be reading or writing while one
** process is writing.
**
** PAGER_SYNCED The pager moves to this state from PAGER_EXCLUSIVE
** after all dirty pages have been written to the
** database file and the file has been synced to
** disk. All that remains to do is to remove or
** truncate the journal file and the transaction
** will be committed.
**
** The page cache comes up in PAGER_UNLOCK. The first time a
** sqlite3PagerGet() occurs, the state transitions to PAGER_SHARED.
** After all pages have been released using sqlite_page_unref(),
** the state transitions back to PAGER_UNLOCK. The first time
** that sqlite3PagerWrite() is called, the state transitions to
** PAGER_RESERVED. (Note that sqlite3PagerWrite() can only be
** called on an outstanding page which means that the pager must
** be in PAGER_SHARED before it transitions to PAGER_RESERVED.)
** PAGER_RESERVED means that there is an open rollback journal.
** The transition to PAGER_EXCLUSIVE occurs before any changes
** are made to the database file, though writes to the rollback
** journal occurs with just PAGER_RESERVED. After an sqlite3PagerRollback()
** or sqlite3PagerCommitPhaseTwo(), the state can go back to PAGER_SHARED,
** or it can stay at PAGER_EXCLUSIVE if we are in exclusive access mode.
*/
#define PAGER_UNLOCK 0
#define PAGER_SHARED 1 /* same as SHARED_LOCK */
#define PAGER_RESERVED 2 /* same as RESERVED_LOCK */
#define PAGER_EXCLUSIVE 4 /* same as EXCLUSIVE_LOCK */
#define PAGER_SYNCED 5
/*
** If the SQLITE_BUSY_RESERVED_LOCK macro is set to true at compile-time,
** then failed attempts to get a reserved lock will invoke the busy callback.
** This is off by default. To see why, consider the following scenario:
**
** Suppose thread A already has a shared lock and wants a reserved lock.
** Thread B already has a reserved lock and wants an exclusive lock. If
** both threads are using their busy callbacks, it might be a long time
** be for one of the threads give up and allows the other to proceed.
** But if the thread trying to get the reserved lock gives up quickly
** (if it never invokes its busy callback) then the contention will be
** resolved quickly.
*/
#ifndef SQLITE_BUSY_RESERVED_LOCK
# define SQLITE_BUSY_RESERVED_LOCK 0
#endif
/*
** This macro rounds values up so that if the value is an address it
** is guaranteed to be an address that is aligned to an 8-byte boundary.
*/
#define FORCE_ALIGNMENT(X) (((X)+7)&~7)
typedef struct PgHdr PgHdr;
/*
** Each pager stores all currently unreferenced pages in a list sorted
** in least-recently-used (LRU) order (i.e. the first item on the list has
** not been referenced in a long time, the last item has been recently
** used). An instance of this structure is included as part of each
** pager structure for this purpose (variable Pager.lru).
**
** Additionally, if memory-management is enabled, all unreferenced pages
** are stored in a global LRU list (global variable sqlite3LruPageList).
**
** In both cases, the PagerLruList.pFirstSynced variable points to
** the first page in the corresponding list that does not require an
** fsync() operation before its memory can be reclaimed. If no such
** page exists, PagerLruList.pFirstSynced is set to NULL.
*/
typedef struct PagerLruList PagerLruList;
struct PagerLruList {
PgHdr *pFirst; /* First page in LRU list */
PgHdr *pLast; /* Last page in LRU list (the most recently used) */
PgHdr *pFirstSynced; /* First page in list with PgHdr.needSync==0 */
};
/*
** The following structure contains the next and previous pointers used
** to link a PgHdr structure into a PagerLruList linked list.
*/
typedef struct PagerLruLink PagerLruLink;
struct PagerLruLink {
PgHdr *pNext;
PgHdr *pPrev;
};
/*
** Each in-memory image of a page begins with the following header.
** This header is only visible to this pager module. The client
** code that calls pager sees only the data that follows the header.
**
** Client code should call sqlite3PagerWrite() on a page prior to making
** any modifications to that page. The first time sqlite3PagerWrite()
** is called, the original page contents are written into the rollback
** journal and PgHdr.inJournal and PgHdr.needSync are set. Later, once
** the journal page has made it onto the disk surface, PgHdr.needSync
** is cleared. The modified page cannot be written back into the original
** database file until the journal pages has been synced to disk and the
** PgHdr.needSync has been cleared.
**
** The PgHdr.dirty flag is set when sqlite3PagerWrite() is called and
** is cleared again when the page content is written back to the original
** database file.
**
** Details of important structure elements:
**
** needSync
**
** If this is true, this means that it is not safe to write the page
** content to the database because the original content needed
** for rollback has not by synced to the main rollback journal.
** The original content may have been written to the rollback journal
** but it has not yet been synced. So we cannot write to the database
** file because power failure might cause the page in the journal file
** to never reach the disk. It is as if the write to the journal file
** does not occur until the journal file is synced.
**
** This flag is false if the page content exactly matches what
** currently exists in the database file. The needSync flag is also
** false if the original content has been written to the main rollback
** journal and synced. If the page represents a new page that has
** been added onto the end of the database during the current
** transaction, the needSync flag is true until the original database
** size in the journal header has been synced to disk.
**
** inJournal
**
** This is true if the original page has been written into the main
** rollback journal. This is always false for new pages added to
** the end of the database file during the current transaction.
** And this flag says nothing about whether or not the journal
** has been synced to disk. For pages that are in the original
** database file, the following expression should always be true:
**
** inJournal = sqlite3BitvecTest(pPager->pInJournal, pgno)
**
** The pPager->pInJournal object is only valid for the original
** pages of the database, not new pages that are added to the end
** of the database, so obviously the above expression cannot be
** valid for new pages. For new pages inJournal is always 0.
**
** dirty
**
** When true, this means that the content of the page has been
** modified and needs to be written back to the database file.
** If false, it means that either the content of the page is
** unchanged or else the content is unimportant and we do not
** care whether or not it is preserved.
**
** alwaysRollback
**
** This means that the sqlite3PagerDontRollback() API should be
** ignored for this page. The DontRollback() API attempts to say
** that the content of the page on disk is unimportant (it is an
** unused page on the freelist) so that it is unnecessary to
** rollback changes to this page because the content of the page
** can change without changing the meaning of the database. This
** flag overrides any DontRollback() attempt. This flag is set
** when a page that originally contained valid data is added to
** the freelist. Later in the same transaction, this page might
** be pulled from the freelist and reused for something different
** and at that point the DontRollback() API will be called because
** pages taken from the freelist do not need to be protected by
** the rollback journal. But this flag says that the page was
** not originally part of the freelist so that it still needs to
** be rolled back in spite of any subsequent DontRollback() calls.
**
** needRead
**
** This flag means (when true) that the content of the page has
** not yet been loaded from disk. The in-memory content is just
** garbage. (Actually, we zero the content, but you should not
** make any assumptions about the content nevertheless.) If the
** content is needed in the future, it should be read from the
** original database file.
*/
struct PgHdr {
Pager *pPager; /* The pager to which this page belongs */
Pgno pgno; /* The page number for this page */
PgHdr *pNextHash, *pPrevHash; /* Hash collision chain for PgHdr.pgno */
PagerLruLink free; /* Next and previous free pages */
PgHdr *pNextAll; /* A list of all pages */
u8 inJournal; /* TRUE if has been written to journal */
u8 dirty; /* TRUE if we need to write back changes */
u8 needSync; /* Sync journal before writing this page */
u8 alwaysRollback; /* Disable DontRollback() for this page */
u8 needRead; /* Read content if PagerWrite() is called */
short int nRef; /* Number of users of this page */
PgHdr *pDirty, *pPrevDirty; /* Dirty pages */
#ifdef SQLITE_ENABLE_MEMORY_MANAGEMENT
PagerLruLink gfree; /* Global list of nRef==0 pages */
#endif
#ifdef SQLITE_CHECK_PAGES
u32 pageHash;
#endif
void *pData; /* Page data */
/* Pager.nExtra bytes of local data appended to this header */
};
/*
** For an in-memory only database, some extra information is recorded about
** each page so that changes can be rolled back. (Journal files are not
** used for in-memory databases.) The following information is added to
** the end of every EXTRA block for in-memory databases.
**
** This information could have been added directly to the PgHdr structure.
** But then it would take up an extra 8 bytes of storage on every PgHdr
** even for disk-based databases. Splitting it out saves 8 bytes. This
** is only a savings of 0.8% but those percentages add up.
*/
typedef struct PgHistory PgHistory;
struct PgHistory {
u8 *pOrig; /* Original page text. Restore to this on a full rollback */
u8 *pStmt; /* Text as it was at the beginning of the current statement */
PgHdr *pNextStmt, *pPrevStmt; /* List of pages in the statement journal */
u8 inStmt; /* TRUE if in the statement subjournal */
};
/*
** A macro used for invoking the codec if there is one
*/
#ifdef SQLITE_HAS_CODEC
# define CODEC1(P,D,N,X) if( P->xCodec!=0 ){ P->xCodec(P->pCodecArg,D,N,X); }
# define CODEC2(P,D,N,X) ((char*)(P->xCodec!=0?P->xCodec(P->pCodecArg,D,N,X):D))
#else
# define CODEC1(P,D,N,X) /* NO-OP */
# define CODEC2(P,D,N,X) ((char*)D)
#endif
/*
** Convert a pointer to a PgHdr into a pointer to its data
** and back again.
*/
#define PGHDR_TO_DATA(P) ((P)->pData)
#define PGHDR_TO_EXTRA(G,P) ((void*)&((G)[1]))
#define PGHDR_TO_HIST(P,PGR)
((PgHistory*)&((char*)(&(P)[1]))[(PGR)->nExtra])
/*
** A open page cache is an instance of the following structure.
**
** Pager.errCode may be set to SQLITE_IOERR, SQLITE_CORRUPT, or
** or SQLITE_FULL. Once one of the first three errors occurs, it persists
** and is returned as the result of every major pager API call. The
** SQLITE_FULL return code is slightly different. It persists only until the
** next successful rollback is performed on the pager cache. Also,
** SQLITE_FULL does not affect the sqlite3PagerGet() and sqlite3PagerLookup()
** APIs, they may still be used successfully.
*/
struct Pager {
sqlite3_vfs *pVfs; /* OS functions to use for IO */
u8 journalOpen; /* True if journal file descriptors is valid */
u8 journalStarted; /* True if header of journal is synced */
u8 useJournal; /* Use a rollback journal on this file */
u8 noReadlock; /* Do not bother to obtain readlocks */
u8 stmtOpen; /* True if the statement subjournal is open */
u8 stmtInUse; /* True we are in a statement subtransaction */
u8 stmtAutoopen; /* Open stmt journal when main journal is opened*/
u8 noSync; /* Do not sync the journal if true */
u8 fullSync; /* Do extra syncs of the journal for robustness */
u8 sync_flags; /* One of SYNC_NORMAL or SYNC_FULL */
u8 state; /* PAGER_UNLOCK, _SHARED, _RESERVED, etc. */
u8 tempFile; /* zFilename is a temporary file */
u8 readOnly; /* True for a read-only database */
u8 needSync; /* True if an fsync() is needed on the journal */
u8 dirtyCache; /* True if cached pages have changed */
u8 alwaysRollback; /* Disable DontRollback() for all pages */
u8 memDb; /* True to inhibit all file I/O */
u8 setMaster; /* True if a m-j name has been written to jrnl */
u8 doNotSync; /* Boolean. While true, do not spill the cache */
u8 exclusiveMode; /* Boolean. True if locking_mode==EXCLUSIVE */
u8 changeCountDone; /* Set after incrementing the change-counter */
u32 vfsFlags; /* Flags for sqlite3_vfs.xOpen() */
int errCode; /* One of several kinds of errors */
int dbSize; /* Number of pages in the file */
int origDbSize; /* dbSize before the current change */
int stmtSize; /* Size of database (in pages) at stmt_begin() */
int nRec; /* Number of pages written to the journal */
u32 cksumInit; /* Quasi-random value added to every checksum */
int stmtNRec; /* Number of records in stmt subjournal */
int nExtra; /* Add this many bytes to each in-memory page */
int pageSize; /* Number of bytes in a page */
int nPage; /* Total number of in-memory pages */
int nRef; /* Number of in-memory pages with PgHdr.nRef>0 */
int mxPage; /* Maximum number of pages to hold in cache */
Pgno mxPgno; /* Maximum allowed size of the database */
Bitvec *pInJournal; /* One bit for each page in the database file */
Bitvec *pInStmt; /* One bit for each page in the database */
char *zFilename; /* Name of the database file */
char *zJournal; /* Name of the journal file */
char *zDirectory; /* Directory hold database and journal files */
char *zStmtJrnl; /* Name of the statement journal file */
sqlite3_file *fd, *jfd; /* File descriptors for database and journal */
sqlite3_file *stfd; /* File descriptor for the statement subjournal*/
BusyHandler *pBusyHandler; /* Pointer to sqlite.busyHandler */
PagerLruList lru; /* LRU list of free pages */
PgHdr *pAll; /* List of all pages */
PgHdr *pStmt; /* List of pages in the statement subjournal */
PgHdr *pDirty; /* List of all dirty pages */
i64 journalOff; /* Current byte offset in the journal file */
i64 journalHdr; /* Byte offset to previous journal header */
i64 stmtHdrOff; /* First journal header written this statement */
i64 stmtCksum; /* cksumInit when statement was started */
i64 stmtJSize; /* Size of journal at stmt_begin() */
int sectorSize; /* Assumed sector size during rollback */
#ifdef SQLITE_TEST
int nHit, nMiss; /* Cache hits and missing */
int nRead, nWrite; /* Database pages read/written */
#endif
void (*xDestructor)(DbPage*,int); /* Call this routine when freeing pages */
void (*xReiniter)(DbPage*,int); /* Call this routine when reloading pages */
#ifdef SQLITE_HAS_CODEC
void *(*xCodec)(void*,void*,Pgno,int); /* Routine for en/decoding data */
void *pCodecArg; /* First argument to xCodec() */
#endif
int nHash; /* Size of the pager hash table */
PgHdr **aHash; /* Hash table to map page number to PgHdr */
#ifdef SQLITE_ENABLE_MEMORY_MANAGEMENT
Pager *pNext; /* Doubly linked list of pagers on which */
Pager *pPrev; /* sqlite3_release_memory() will work */
int iInUseMM; /* Non-zero if unavailable to MM */
int iInUseDB; /* Non-zero if in sqlite3_release_memory() */
#endif
char *pTmpSpace; /* Pager.pageSize bytes of space for tmp use */
char dbFileVers[16]; /* Changes whenever database file changes */
};
/*
** The following global variables hold counters used for
** testing purposes only. These variables do not exist in
** a non-testing build. These variables are not thread-safe.
*/
#ifdef SQLITE_TEST
int sqlite3_pager_readdb_count = 0; /* Number of full pages read from DB */
int sqlite3_pager_writedb_count = 0; /* Number of full pages written to DB */
int sqlite3_pager_writej_count = 0; /* Number of pages written to journal */
int sqlite3_pager_pgfree_count = 0; /* Number of cache pages freed */
# define PAGER_INCR(v) v++
#else
# define PAGER_INCR(v)
#endif
/*
** The following variable points to the head of a double-linked list
** of all pagers that are eligible for page stealing by the
** sqlite3_release_memory() interface. Access to this list is
** protected by the SQLITE_MUTEX_STATIC_MEM2 mutex.
*/
#ifdef SQLITE_ENABLE_MEMORY_MANAGEMENT
static Pager *sqlite3PagerList = 0;
static PagerLruList sqlite3LruPageList = {0, 0, 0};
#endif
/*
** Journal files begin with the following magic string. The data
** was obtained from /dev/random. It is used only as a sanity check.
**
** Since version 2.8.0, the journal format contains additional sanity
** checking information. If the power fails while the journal is begin
** written, semi-random garbage data might appear in the journal
** file after power is restored. If an attempt is then made
** to roll the journal back, the database could be corrupted. The additional
** sanity checking data is an attempt to discover the garbage in the
** journal and ignore it.
**
** The sanity checking information for the new journal format consists
** of a 32-bit checksum on each page of data. The checksum covers both
** the page number and the pPager->pageSize bytes of data for the page.
** This cksum is initialized to a 32-bit random value that appears in the
** journal file right after the header. The random initializer is important,
** because garbage data that appears at the end of a journal is likely
** data that was once in other files that have now been deleted. If the
** garbage data came from an obsolete journal file, the checksums might
** be correct. But by initializing the checksum to random value which
** is different for every journal, we minimize that risk.
*/
static const unsigned char aJournalMagic[] = {
0xd9, 0xd5, 0x05, 0xf9, 0x20, 0xa1, 0x63, 0xd7,
};
/*
** The size of the header and of each page in the journal is determined
** by the following macros.
*/
#define JOURNAL_PG_SZ(pPager) ((pPager->pageSize) + 8)
/*
** The journal header size for this pager. In the future, this could be
** set to some value read from the disk controller. The important
** characteristic is that it is the same size as a disk sector.
*/
#define JOURNAL_HDR_SZ(pPager) (pPager->sectorSize)
/*
** The macro MEMDB is true if we are dealing with an in-memory database.
** We do this as a macro so that if the SQLITE_OMIT_MEMORYDB macro is set,
** the value of MEMDB will be a constant and the compiler will optimize
** out code that would never execute.
*/
#ifdef SQLITE_OMIT_MEMORYDB
# define MEMDB 0
#else
# define MEMDB pPager->memDb
#endif
/*
** Page number PAGER_MJ_PGNO is never used in an SQLite database (it is
** reserved for working around a windows/posix incompatibility). It is
** used in the journal to signify that the remainder of the journal file
** is devoted to storing a master journal name - there are no more pages to
** roll back. See comments for function writeMasterJournal() for details.
*/
/* #define PAGER_MJ_PGNO(x) (PENDING_BYTE/((x)->pageSize)) */
#define PAGER_MJ_PGNO(x) ((PENDING_BYTE/((x)->pageSize))+1)
/*
** The maximum legal page number is (2^31 - 1).
*/
#define PAGER_MAX_PGNO 2147483647
/*
** The pagerEnter() and pagerLeave() routines acquire and release
** a mutex on each pager. The mutex is recursive.
**
** This is a special-purpose mutex. It only provides mutual exclusion
** between the Btree and the Memory Management sqlite3_release_memory()
** function. It does not prevent, for example, two Btrees from accessing
** the same pager at the same time. Other general-purpose mutexes in
** the btree layer handle that chore.
*/
#ifdef SQLITE_ENABLE_MEMORY_MANAGEMENT
static void pagerEnter(Pager *p){
p->iInUseDB++;
if( p->iInUseMM && p->iInUseDB==1 ){
sqlite3_mutex *mutex;
mutex = sqlite3_mutex_alloc(SQLITE_MUTEX_STATIC_MEM2);
p->iInUseDB = 0;
sqlite3_mutex_enter(mutex);
p->iInUseDB = 1;
sqlite3_mutex_leave(mutex);
}
assert( p->iInUseMM==0 );
}
static void pagerLeave(Pager *p){
p->iInUseDB--;
assert( p->iInUseDB>=0 );
}
#else
# define pagerEnter(X)
# define pagerLeave(X)
#endif
/*
** Add page pPg to the end of the linked list managed by structure
** pList (pPg becomes the last entry in the list - the most recently
** used). Argument pLink should point to either pPg->free or pPg->gfree,
** depending on whether pPg is being added to the pager-specific or
** global LRU list.
*/
static void listAdd(PagerLruList *pList, PagerLruLink *pLink, PgHdr *pPg){
pLink->pNext = 0;
pLink->pPrev = pList->pLast;
#ifdef SQLITE_ENABLE_MEMORY_MANAGEMENT
assert(pLink==&pPg->free || pLink==&pPg->gfree);
assert(pLink==&pPg->gfree || pList!=&sqlite3LruPageList);
#endif
if( pList->pLast ){
int iOff = (char *)pLink - (char *)pPg;
PagerLruLink *pLastLink = (PagerLruLink *)(&((u8 *)pList->pLast)[iOff]);
pLastLink->pNext = pPg;
}else{
assert(!pList->pFirst);
pList->pFirst = pPg;
}
pList->pLast = pPg;
if( !pList->pFirstSynced && pPg->needSync==0 ){
pList->pFirstSynced = pPg;
}
}
/*
** Remove pPg from the list managed by the structure pointed to by pList.
**
** Argument pLink should point to either pPg->free or pPg->gfree, depending
** on whether pPg is being added to the pager-specific or global LRU list.
*/
static void listRemove(PagerLruList *pList, PagerLruLink *pLink, PgHdr *pPg){
int iOff = (char *)pLink - (char *)pPg;
#ifdef SQLITE_ENABLE_MEMORY_MANAGEMENT
assert(pLink==&pPg->free || pLink==&pPg->gfree);
assert(pLink==&pPg->gfree || pList!=&sqlite3LruPageList);
#endif
if( pPg==pList->pFirst ){
pList->pFirst = pLink->pNext;
}
if( pPg==pList->pLast ){
pList->pLast = pLink->pPrev;
}
if( pLink->pPrev ){
PagerLruLink *pPrevLink = (PagerLruLink *)(&((u8 *)pLink->pPrev)[iOff]);
pPrevLink->pNext = pLink->pNext;
}
if( pLink->pNext ){
PagerLruLink *pNextLink = (PagerLruLink *)(&((u8 *)pLink->pNext)[iOff]);
pNextLink->pPrev = pLink->pPrev;
}
if( pPg==pList->pFirstSynced ){
PgHdr *p = pLink->pNext;
while( p && p->needSync ){
PagerLruLink *pL = (PagerLruLink *)(&((u8 *)p)[iOff]);
p = pL->pNext;
}
pList->pFirstSynced = p;
}
pLink->pNext = pLink->pPrev = 0;
}
/*
** Add page pPg to the list of free pages for the pager. If
** memory-management is enabled, also add the page to the global
** list of free pages.
*/
static void lruListAdd(PgHdr *pPg){
listAdd(&pPg->pPager->lru, &pPg->free, pPg);
#ifdef SQLITE_ENABLE_MEMORY_MANAGEMENT
if( !pPg->pPager->memDb ){
sqlite3_mutex_enter(sqlite3_mutex_alloc(SQLITE_MUTEX_STATIC_LRU));
listAdd(&sqlite3LruPageList, &pPg->gfree, pPg);
sqlite3_mutex_leave(sqlite3_mutex_alloc(SQLITE_MUTEX_STATIC_LRU));
}
#endif
}
/*
** Remove page pPg from the list of free pages for the associated pager.
** If memory-management is enabled, also remove pPg from the global list
** of free pages.
*/
static void lruListRemove(PgHdr *pPg){
listRemove(&pPg->pPager->lru, &pPg->free, pPg);
#ifdef SQLITE_ENABLE_MEMORY_MANAGEMENT
if( !pPg->pPager->memDb ){
sqlite3_mutex_enter(sqlite3_mutex_alloc(SQLITE_MUTEX_STATIC_LRU));
listRemove(&sqlite3LruPageList, &pPg->gfree, pPg);
sqlite3_mutex_leave(sqlite3_mutex_alloc(SQLITE_MUTEX_STATIC_LRU));
}
#endif
}
/*
** This function is called just after the needSync flag has been cleared
** from all pages managed by pPager (usually because the journal file
** has just been synced). It updates the pPager->lru.pFirstSynced variable
** and, if memory-management is enabled, the sqlite3LruPageList.pFirstSynced
** variable also.
*/
static void lruListSetFirstSynced(Pager *pPager){
pPager->lru.pFirstSynced = pPager->lru.pFirst;
#ifdef SQLITE_ENABLE_MEMORY_MANAGEMENT
if( !pPager->memDb ){
PgHdr *p;
sqlite3_mutex_enter(sqlite3_mutex_alloc(SQLITE_MUTEX_STATIC_LRU));
for(p=sqlite3LruPageList.pFirst; p && p->needSync; p=p->gfree.pNext);
assert(p==pPager->lru.pFirstSynced || p==sqlite3LruPageList.pFirstSynced);
sqlite3LruPageList.pFirstSynced = p;
sqlite3_mutex_leave(sqlite3_mutex_alloc(SQLITE_MUTEX_STATIC_LRU));
}
#endif
}
/*
** Return true if page *pPg has already been written to the statement
** journal (or statement snapshot has been created, if *pPg is part
** of an in-memory database).
*/
static int pageInStatement(PgHdr *pPg){
Pager *pPager = pPg->pPager;
if( MEMDB ){
return PGHDR_TO_HIST(pPg, pPager)->inStmt;
}else{
return sqlite3BitvecTest(pPager->pInStmt, pPg->pgno);
}
}
/*
** Change the size of the pager hash table to N. N must be a power
** of two.
*/
static void pager_resize_hash_table(Pager *pPager, int N){
PgHdr **aHash, *pPg;
assert( N>0 && (N&(N-1))==0 );
#ifdef SQLITE_MALLOC_SOFT_LIMIT
if( N*sizeof(aHash[0])>SQLITE_MALLOC_SOFT_LIMIT ){
N = SQLITE_MALLOC_SOFT_LIMIT/sizeof(aHash[0]);
}
if( N==pPager->nHash ) return;
#endif
pagerLeave(pPager);
sqlite3FaultBenign(SQLITE_FAULTINJECTOR_MALLOC, pPager->aHash!=0);
aHash = sqlite3MallocZero( sizeof(aHash[0])*N );
sqlite3FaultBenign(SQLITE_FAULTINJECTOR_MALLOC, 0);
pagerEnter(pPager);
if( aHash==0 ){
/* Failure to rehash is not an error. It is only a performance hit. */
return;
}
sqlite3_free(pPager->aHash);
pPager->nHash = N;
pPager->aHash = aHash;
for(pPg=pPager->pAll; pPg; pPg=pPg->pNextAll){
int h;
if( pPg->pgno==0 ){
assert( pPg->pNextHash==0 && pPg->pPrevHash==0 );
continue;
}
h = pPg->pgno & (N-1);
pPg->pNextHash = aHash[h];
if( aHash[h] ){
aHash[h]->pPrevHash = pPg;
}
aHash[h] = pPg;
pPg->pPrevHash = 0;
}
}
/*
** Read a 32-bit integer from the given file descriptor. Store the integer
** that is read in *pRes. Return SQLITE_OK if everything worked, or an
** error code is something goes wrong.
**
** All values are stored on disk as big-endian.
*/
static int read32bits(sqlite3_file *fd, i64 offset, u32 *pRes){
unsigned char ac[4];
int rc = sqlite3OsRead(fd, ac, sizeof(ac), offset);
if( rc==SQLITE_OK ){
*pRes = sqlite3Get4byte(ac);
}
return rc;
}
/*
** Write a 32-bit integer into a string buffer in big-endian byte order.
*/
#define put32bits(A,B) sqlite3Put4byte((u8*)A,B)
/*
** Write a 32-bit integer into the given file descriptor. Return SQLITE_OK
** on success or an error code is something goes wrong.
*/
static int write32bits(sqlite3_file *fd, i64 offset, u32 val){
char ac[4];
put32bits(ac, val);
return sqlite3OsWrite(fd, ac, 4, offset);
}
/*
** If file pFd is open, call sqlite3OsUnlock() on it.
*/
static int osUnlock(sqlite3_file *pFd, int eLock){
if( !pFd->pMethods ){
return SQLITE_OK;
}
return sqlite3OsUnlock(pFd, eLock);
}
/*
** This function determines whether or not the atomic-write optimization
** can be used with this pager. The optimization can be used if:
**
** (a) the value returned by OsDeviceCharacteristics() indicates that
** a database page may be written atomically, and
** (b) the value returned by OsSectorSize() is less than or equal
** to the page size.
**
** If the optimization cannot be used, 0 is returned. If it can be used,
** then the value returned is the size of the journal file when it
** contains rollback data for exactly one page.
*/
#ifdef SQLITE_ENABLE_ATOMIC_WRITE
static int jrnlBufferSize(Pager *pPager){
int dc; /* Device characteristics */
int nSector; /* Sector size */
int nPage; /* Page size */
sqlite3_file *fd = pPager->fd;
if( fd->pMethods ){
dc = sqlite3OsDeviceCharacteristics(fd);
nSector = sqlite3OsSectorSize(fd);
nPage = pPager->pageSize;
}
assert(SQLITE_IOCAP_ATOMIC512==(512>>8));
assert(SQLITE_IOCAP_ATOMIC64K==(65536>>8));
if( !fd->pMethods || (dc&(SQLITE_IOCAP_ATOMIC|(nPage>>8))&&nSector<=nPage) ){
return JOURNAL_HDR_SZ(pPager) + JOURNAL_PG_SZ(pPager);
}
return 0;
}
#endif
/*
** This function should be called when an error occurs within the pager
** code. The first argument is a pointer to the pager structure, the
** second the error-code about to be returned by a pager API function.
** The value returned is a copy of the second argument to this function.
**
** If the second argument is SQLITE_IOERR, SQLITE_CORRUPT, or SQLITE_FULL
** the error becomes persistent. Until the persisten error is cleared,
** subsequent API calls on this Pager will immediately return the same
** error code.
**
** A persistent error indicates that the contents of the pager-cache
** cannot be trusted. This state can be cleared by completely discarding
** the contents of the pager-cache. If a transaction was active when
** the persistent error occured, then the rollback journal may need
** to be replayed.
*/
static void pager_unlock(Pager *pPager);
static int pager_error(Pager *pPager, int rc){
int rc2 = rc & 0xff;
assert(
pPager->errCode==SQLITE_FULL ||
pPager->errCode==SQLITE_OK ||
(pPager->errCode & 0xff)==SQLITE_IOERR
);
if(
rc2==SQLITE_FULL ||
rc2==SQLITE_IOERR ||
rc2==SQLITE_CORRUPT
){
pPager->errCode = rc;
if( pPager->state==PAGER_UNLOCK && pPager->nRef==0 ){
/* If the pager is already unlocked, call pager_unlock() now to
** clear the error state and ensure that the pager-cache is
** completely empty.
*/
pager_unlock(pPager);
}
}
return rc;
}
/*
** If SQLITE_CHECK_PAGES is defined then we do some sanity checking
** on the cache using a hash function. This is used for testing
** and debugging only.
*/
#ifdef SQLITE_CHECK_PAGES
/*
** Return a 32-bit hash of the page data for pPage.
*/
static u32 pager_datahash(int nByte, unsigned char *pData){
u32 hash = 0;
int i;
for(i=0; i<nByte; i++){
hash = (hash*1039) + pData[i];
}
return hash;
}
static u32 pager_pagehash(PgHdr *pPage){
return pager_datahash(pPage->pPager->pageSize,
(unsigned char *)PGHDR_TO_DATA(pPage));
}
/*
** The CHECK_PAGE macro takes a PgHdr* as an argument. If SQLITE_CHECK_PAGES
** is defined, and NDEBUG is not defined, an assert() statement checks
** that the page is either dirty or still matches the calculated page-hash.
*/
#define CHECK_PAGE(x) checkPage(x)
static void checkPage(PgHdr *pPg){
Pager *pPager = pPg->pPager;
assert( !pPg->pageHash || pPager->errCode || MEMDB || pPg->dirty ||
pPg->pageHash==pager_pagehash(pPg) );
}
#else
#define pager_datahash(X,Y) 0
#define pager_pagehash(X) 0
#define CHECK_PAGE(x)
#endif
/*
** When this is called the journal file for pager pPager must be open.
** The master journal file name is read from the end of the file and
** written into memory supplied by the caller.
**
** zMaster must point to a buffer of at least nMaster bytes allocated by
** the caller. This should be sqlite3_vfs.mxPathname+1 (to ensure there is
** enough space to write the master journal name). If the master journal
** name in the journal is longer than nMaster bytes (including a
** nul-terminator), then this is handled as if no master journal name
** were present in the journal.
**
** If no master journal file name is present zMaster[0] is set to 0 and
** SQLITE_OK returned.
*/
static int readMasterJournal(sqlite3_file *pJrnl, char *zMaster, int nMaster){
int rc;
u32 len;
i64 szJ;
u32 cksum;
int i;
unsigned char aMagic[8]; /* A buffer to hold the magic header */
zMaster[0] = '';
rc = sqlite3OsFileSize(pJrnl, &szJ);
if( rc!=SQLITE_OK || szJ<16 ) return rc;
rc = read32bits(pJrnl, szJ-16, &len);
if( rc!=SQLITE_OK ) return rc;
if( len>=nMaster ){
return SQLITE_OK;
}
rc = read32bits(pJrnl, szJ-12, &cksum);
if( rc!=SQLITE_OK ) return rc;
rc = sqlite3OsRead(pJrnl, aMagic, 8, szJ-8);
if( rc!=SQLITE_OK || memcmp(aMagic, aJournalMagic, 8) ) return rc;
rc = sqlite3OsRead(pJrnl, zMaster, len, szJ-16-len);
if( rc!=SQLITE_OK ){
return rc;
}
zMaster[len] = '';
/* See if the checksum matches the master journal name */
for(i=0; i<len; i++){
cksum -= zMaster[i];
}
if( cksum ){
/* If the checksum doesn't add up, then one or more of the disk sectors
** containing the master journal filename is corrupted. This means
** definitely roll back, so just return SQLITE_OK and report a (nul)
** master-journal filename.
*/
zMaster[0] = '';
}
return SQLITE_OK;
}
/*
** Seek the journal file descriptor to the next sector boundary where a
** journal header may be read or written. Pager.journalOff is updated with
** the new seek offset.
**
** i.e for a sector size of 512:
**
** Input Offset Output Offset
** ---------------------------------------
** 0 0
** 512 512
** 100 512
** 2000 2048
**
*/
static void seekJournalHdr(Pager *pPager){
i64 offset = 0;
i64 c = pPager->journalOff;
if( c ){
offset = ((c-1)/JOURNAL_HDR_SZ(pPager) + 1) * JOURNAL_HDR_SZ(pPager);
}
assert( offset%JOURNAL_HDR_SZ(pPager)==0 );
assert( offset>=c );
assert( (offset-c)<JOURNAL_HDR_SZ(pPager) );
pPager->journalOff = offset;
}
/*
** The journal file must be open when this routine is called. A journal
** header (JOURNAL_HDR_SZ bytes) is written into the journal file at the
** current location.
**
** The format for the journal header is as follows:
** - 8 bytes: Magic identifying journal format.
** - 4 bytes: Number of records in journal, or -1 no-sync mode is on.
** - 4 bytes: Random number used for page hash.
** - 4 bytes: Initial database page count.
** - 4 bytes: Sector size used by the process that wrote this journal.
** - 4 bytes: Database page size.
**
** Followed by (JOURNAL_HDR_SZ - 28) bytes of unused space.
*/
static int writeJournalHdr(Pager *pPager){
char zHeader[sizeof(aJournalMagic)+20];
int rc;
if( pPager->stmtHdrOff==0 ){
pPager->stmtHdrOff = pPager->journalOff;
}
seekJournalHdr(pPager);
pPager->journalHdr = pPager->journalOff;
memcpy(zHeader, aJournalMagic, sizeof(aJournalMagic));
/*
** Write the nRec Field - the number of page records that follow this
** journal header. Normally, zero is written to this value at this time.
** After the records are added to the journal (and the journal synced,
** if in full-sync mode), the zero is overwritten with the true number
** of records (see syncJournal()).
**
** A faster alternative is to write 0xFFFFFFFF to the nRec field. When
** reading the journal this value tells SQLite to assume that the
** rest of the journal file contains valid page records. This assumption
** is dangerous, as if a failure occured whilst writing to the journal
** file it may contain some garbage data. There are two scenarios
** where this risk can be ignored:
**
** * When the pager is in no-sync mode. Corruption can follow a
** power failure in this case anyway.
**
** * When the SQLITE_IOCAP_SAFE_APPEND flag is set. This guarantees
** that garbage data is never appended to the journal file.
*/
assert(pPager->fd->pMethods||pPager->noSync);
if( (pPager->noSync)
|| (sqlite3OsDeviceCharacteristics(pPager->fd)&SQLITE_IOCAP_SAFE_APPEND)
){
put32bits(&zHeader[sizeof(aJournalMagic)], 0xffffffff);
}else{
put32bits(&zHeader[sizeof(aJournalMagic)], 0);
}
/* The random check-hash initialiser */
sqlite3_randomness(sizeof(pPager->cksumInit), &pPager->cksumInit);
put32bits(&zHeader[sizeof(aJournalMagic)+4], pPager->cksumInit);
/* The initial database size */
put32bits(&zHeader[sizeof(aJournalMagic)+8], pPager->dbSize);
/* The assumed sector size for this process */
put32bits(&zHeader[sizeof(aJournalMagic)+12], pPager->sectorSize);
if( pPager->journalHdr==0 ){
/* The page size */
put32bits(&zHeader[sizeof(aJournalMagic)+16], pPager->pageSize);
}
IOTRACE(("JHDR %p %lld %dn", pPager, pPager->journalHdr, sizeof(zHeader)))
rc = sqlite3OsWrite(pPager->jfd, zHeader, sizeof(zHeader),pPager->journalOff);
pPager->journalOff += JOURNAL_HDR_SZ(pPager);
/* The journal header has been written successfully. Seek the journal
** file descriptor to the end of the journal header sector.
*/
if( rc==SQLITE_OK ){
IOTRACE(("JTAIL %p %lldn", pPager, pPager->journalOff-1))
rc = sqlite3OsWrite(pPager->jfd, "00", 1, pPager->journalOff-1);
}
return rc;
}
/*
** The journal file must be open when this is called. A journal header file
** (JOURNAL_HDR_SZ bytes) is read from the current location in the journal
** file. See comments above function writeJournalHdr() for a description of
** the journal header format.
**
** If the header is read successfully, *nRec is set to the number of
** page records following this header and *dbSize is set to the size of the
** database before the transaction began, in pages. Also, pPager->cksumInit
** is set to the value read from the journal header. SQLITE_OK is returned
** in this case.
**
** If the journal header file appears to be corrupted, SQLITE_DONE is
** returned and *nRec and *dbSize are not set. If JOURNAL_HDR_SZ bytes
** cannot be read from the journal file an error code is returned.
*/
static int readJournalHdr(
Pager *pPager,
i64 journalSize,
u32 *pNRec,
u32 *pDbSize
){
int rc;
unsigned char aMagic[8]; /* A buffer to hold the magic header */
i64 jrnlOff;
int iPageSize;
seekJournalHdr(pPager);
if( pPager->journalOff+JOURNAL_HDR_SZ(pPager) > journalSize ){
return SQLITE_DONE;
}
jrnlOff = pPager->journalOff;
rc = sqlite3OsRead(pPager->jfd, aMagic, sizeof(aMagic), jrnlOff);
if( rc ) return rc;
jrnlOff += sizeof(aMagic);
if( memcmp(aMagic, aJournalMagic, sizeof(aMagic))!=0 ){
return SQLITE_DONE;
}
rc = read32bits(pPager->jfd, jrnlOff, pNRec);
if( rc ) return rc;
rc = read32bits(pPager->jfd, jrnlOff+4, &pPager->cksumInit);
if( rc ) return rc;
rc = read32bits(pPager->jfd, jrnlOff+8, pDbSize);
if( rc ) return rc;
rc = read32bits(pPager->jfd, jrnlOff+16, (u32 *)&iPageSize);
if( rc==SQLITE_OK
&& iPageSize>=512
&& iPageSize<=SQLITE_MAX_PAGE_SIZE
&& ((iPageSize-1)&iPageSize)==0
){
u16 pagesize = iPageSize;
rc = sqlite3PagerSetPagesize(pPager, &pagesize);
}
if( rc ) return rc;
/* Update the assumed sector-size to match the value used by
** the process that created this journal. If this journal was
** created by a process other than this one, then this routine
** is being called from within pager_playback(). The local value
** of Pager.sectorSize is restored at the end of that routine.
*/
rc = read32bits(pPager->jfd, jrnlOff+12, (u32 *)&pPager->sectorSize);
if( rc ) return rc;
pPager->journalOff += JOURNAL_HDR_SZ(pPager);
return SQLITE_OK;
}
/*
** Write the supplied master journal name into the journal file for pager
** pPager at the current location. The master journal name must be the last
** thing written to a journal file. If the pager is in full-sync mode, the
** journal file descriptor is advanced to the next sector boundary before
** anything is written. The format is:
**
** + 4 bytes: PAGER_MJ_PGNO.
** + N bytes: length of master journal name.
** + 4 bytes: N
** + 4 bytes: Master journal name checksum.
** + 8 bytes: aJournalMagic[].
**
** The master journal page checksum is the sum of the bytes in the master
** journal name.
**
** If zMaster is a NULL pointer (occurs for a single database transaction),
** this call is a no-op.
*/
static int writeMasterJournal(Pager *pPager, const char *zMaster){
int rc;
int len;
int i;
i64 jrnlOff;
u32 cksum = 0;
char zBuf[sizeof(aJournalMagic)+2*4];
if( !zMaster || pPager->setMaster) return SQLITE_OK;
pPager->setMaster = 1;
len = strlen(zMaster);
for(i=0; i<len; i++){
cksum += zMaster[i];
}
/* If in full-sync mode, advance to the next disk sector before writing
** the master journal name. This is in case the previous page written to
** the journal has already been synced.
*/
if( pPager->fullSync ){
seekJournalHdr(pPager);
}
jrnlOff = pPager->journalOff;
pPager->journalOff += (len+20);
rc = write32bits(pPager->jfd, jrnlOff, PAGER_MJ_PGNO(pPager));
if( rc!=SQLITE_OK ) return rc;
jrnlOff += 4;
rc = sqlite3OsWrite(pPager->jfd, zMaster, len, jrnlOff);
if( rc!=SQLITE_OK ) return rc;
jrnlOff += len;
put32bits(zBuf, len);
put32bits(&zBuf[4], cksum);
memcpy(&zBuf[8], aJournalMagic, sizeof(aJournalMagic));
rc = sqlite3OsWrite(pPager->jfd, zBuf, 8+sizeof(aJournalMagic), jrnlOff);
pPager->needSync = !pPager->noSync;
return rc;
}
/*
** Add or remove a page from the list of all pages that are in the
** statement journal.
**
** The Pager keeps a separate list of pages that are currently in
** the statement journal. This helps the sqlite3PagerStmtCommit()
** routine run MUCH faster for the common case where there are many
** pages in memory but only a few are in the statement journal.
*/
static void page_add_to_stmt_list(PgHdr *pPg){
Pager *pPager = pPg->pPager;
PgHistory *pHist = PGHDR_TO_HIST(pPg, pPager);
assert( MEMDB );
if( !pHist->inStmt ){
assert( pHist->pPrevStmt==0 && pHist->pNextStmt==0 );
if( pPager->pStmt ){
PGHDR_TO_HIST(pPager->pStmt, pPager)->pPrevStmt = pPg;
}
pHist->pNextStmt = pPager->pStmt;
pPager->pStmt = pPg;
pHist->inStmt = 1;
}
}
/*
** Find a page in the hash table given its page number. Return
** a pointer to the page or NULL if not found.
*/
static PgHdr *pager_lookup(Pager *pPager, Pgno pgno){
PgHdr *p;
if( pPager->aHash==0 ) return 0;
p = pPager->aHash[pgno & (pPager->nHash-1)];
while( p && p->pgno!=pgno ){
p = p->pNextHash;
}
return p;
}
/*
** Clear the in-memory cache. This routine
** sets the state of the pager back to what it was when it was first
** opened. Any outstanding pages are invalidated and subsequent attempts
** to access those pages will likely result in a coredump.
*/
static void pager_reset(Pager *pPager){
PgHdr *pPg, *pNext;
if( pPager->errCode ) return;
for(pPg=pPager->pAll; pPg; pPg=pNext){
IOTRACE(("PGFREE %p %dn", pPager, pPg->pgno));
PAGER_INCR(sqlite3_pager_pgfree_count);
pNext = pPg->pNextAll;
lruListRemove(pPg);
sqlite3_free(pPg->pData);
sqlite3_free(pPg);
}
assert(pPager->lru.pFirst==0);
assert(pPager->lru.pFirstSynced==0);
assert(pPager->lru.pLast==0);
pPager->pStmt = 0;
pPager->pAll = 0;
pPager->pDirty = 0;
pPager->nHash = 0;
sqlite3_free(pPager->aHash);
pPager->nPage = 0;
pPager->aHash = 0;
pPager->nRef = 0;
}
/*
** Unlock the database file.
**
** If the pager is currently in error state, discard the contents of
** the cache and reset the Pager structure internal state. If there is
** an open journal-file, then the next time a shared-lock is obtained
** on the pager file (by this or any other process), it will be
** treated as a hot-journal and rolled back.
*/
static void pager_unlock(Pager *pPager){
if( !pPager->exclusiveMode ){
if( !MEMDB ){
int rc = osUnlock(pPager->fd, NO_LOCK);
if( rc ) pPager->errCode = rc;
pPager->dbSize = -1;
IOTRACE(("UNLOCK %pn", pPager))
/* If Pager.errCode is set, the contents of the pager cache cannot be
** trusted. Now that the pager file is unlocked, the contents of the
** cache can be discarded and the error code safely cleared.
*/
if( pPager->errCode ){
if( rc==SQLITE_OK ) pPager->errCode = SQLITE_OK;
pager_reset(pPager);
if( pPager->stmtOpen ){
sqlite3OsClose(pPager->stfd);
sqlite3BitvecDestroy(pPager->pInStmt);
pPager->pInStmt = 0;
}
if( pPager->journalOpen ){
sqlite3OsClose(pPager->jfd);
pPager->journalOpen = 0;
sqlite3BitvecDestroy(pPager->pInJournal);
pPager->pInJournal = 0;
}
pPager->stmtOpen = 0;
pPager->stmtInUse = 0;
pPager->journalOff = 0;
pPager->journalStarted = 0;
pPager->stmtAutoopen = 0;
pPager->origDbSize = 0;
}
}
if( !MEMDB || pPager->errCode==SQLITE_OK ){
pPager->state = PAGER_UNLOCK;
pPager->changeCountDone = 0;
}
}
}
/*
** Execute a rollback if a transaction is active and unlock the
** database file. If the pager has already entered the error state,
** do not attempt the rollback.
*/
static void pagerUnlockAndRollback(Pager *p){
assert( p->state>=PAGER_RESERVED || p->journalOpen==0 );
if( p->errCode==SQLITE_OK && p->state>=PAGER_RESERVED ){
sqlite3PagerRollback(p);
}
pager_unlock(p);
assert( p->errCode || !p->journalOpen || (p->exclusiveMode&&!p->journalOff) );
assert( p->errCode || !p->stmtOpen || p->exclusiveMode );
}
/*
** This routine ends a transaction. A transaction is ended by either
** a COMMIT or a ROLLBACK.
**
** When this routine is called, the pager has the journal file open and
** a RESERVED or EXCLUSIVE lock on the database. This routine will release
** the database lock and acquires a SHARED lock in its place if that is
** the appropriate thing to do. Release locks usually is appropriate,
** unless we are in exclusive access mode or unless this is a
** COMMIT AND BEGIN or ROLLBACK AND BEGIN operation.
**
** The journal file is either deleted or truncated.
**
** TODO: Consider keeping the journal file open for temporary databases.
** This might give a performance improvement on windows where opening
** a file is an expensive operation.
*/
static int pager_end_transaction(Pager *pPager){
PgHdr *pPg;
int rc = SQLITE_OK;
int rc2 = SQLITE_OK;
assert( !MEMDB );
if( pPager->state<PAGER_RESERVED ){
return SQLITE_OK;
}
sqlite3PagerStmtCommit(pPager);
if( pPager->stmtOpen && !pPager->exclusiveMode ){
sqlite3OsClose(pPager->stfd);
pPager->stmtOpen = 0;
}
if( pPager->journalOpen ){
if( pPager->exclusiveMode
&& (rc = sqlite3OsTruncate(pPager->jfd, 0))==SQLITE_OK ){;
pPager->journalOff = 0;
pPager->journalStarted = 0;
}else{
sqlite3OsClose(pPager->jfd);
pPager->journalOpen = 0;
if( rc==SQLITE_OK ){
rc = sqlite3OsDelete(pPager->pVfs, pPager->zJournal, 0);
}
}
sqlite3BitvecDestroy(pPager->pInJournal);
pPager->pInJournal = 0;
for(pPg=pPager->pAll; pPg; pPg=pPg->pNextAll){
pPg->inJournal = 0;
pPg->dirty = 0;
pPg->needSync = 0;
pPg->alwaysRollback = 0;
#ifdef SQLITE_CHECK_PAGES
pPg->pageHash = pager_pagehash(pPg);
#endif
}
pPager->pDirty = 0;
pPager->dirtyCache = 0;
pPager->nRec = 0;
}else{
assert( pPager->pInJournal==0 );
assert( pPager->dirtyCache==0 || pPager->useJournal==0 );
}
if( !pPager->exclusiveMode ){
rc2 = osUnlock(pPager->fd, SHARED_LOCK);
pPager->state = PAGER_SHARED;
}else if( pPager->state==PAGER_SYNCED ){
pPager->state = PAGER_EXCLUSIVE;
}
pPager->origDbSize = 0;
pPager->setMaster = 0;
pPager->needSync = 0;
lruListSetFirstSynced(pPager);
pPager->dbSize = -1;
return (rc==SQLITE_OK?rc2:rc);
}
/*
** Compute and return a checksum for the page of data.
**
** This is not a real checksum. It is really just the sum of the
** random initial value and the page number. We experimented with
** a checksum of the entire data, but that was found to be too slow.
**
** Note that the page number is stored at the beginning of data and
** the checksum is stored at the end. This is important. If journal
** corruption occurs due to a power failure, the most likely scenario
** is that one end or the other of the record will be changed. It is
** much less likely that the two ends of the journal record will be
** correct and the middle be corrupt. Thus, this "checksum" scheme,
** though fast and simple, catches the mostly likely kind of corruption.
**
** FIX ME: Consider adding every 200th (or so) byte of the data to the
** checksum. That way if a single page spans 3 or more disk sectors and
** only the middle sector is corrupt, we will still have a reasonable
** chance of failing the checksum and thus detecting the problem.
*/
static u32 pager_cksum(Pager *pPager, const u8 *aData){
u32 cksum = pPager->cksumInit;
int i = pPager->pageSize-200;
while( i>0 ){
cksum += aData[i];
i -= 200;
}
return cksum;
}
/* Forward declaration */
static void makeClean(PgHdr*);
/*
** Read a single page from the journal file opened on file descriptor
** jfd. Playback this one page.
**
** If useCksum==0 it means this journal does not use checksums. Checksums
** are not used in statement journals because statement journals do not
** need to survive power failures.
*/
static int pager_playback_one_page(
Pager *pPager,
sqlite3_file *jfd,
i64 offset,
int useCksum
){
int rc;
PgHdr *pPg; /* An existing page in the cache */
Pgno pgno; /* The page number of a page in journal */
u32 cksum; /* Checksum used for sanity checking */
u8 *aData = (u8 *)pPager->pTmpSpace; /* Temp storage for a page */
/* useCksum should be true for the main journal and false for
** statement journals. Verify that this is always the case
*/
assert( jfd == (useCksum ? pPager->jfd : pPager->stfd) );
assert( aData );
rc = read32bits(jfd, offset, &pgno);
if( rc!=SQLITE_OK ) return rc;
rc = sqlite3OsRead(jfd, aData, pPager->pageSize, offset+4);
if( rc!=SQLITE_OK ) return rc;
pPager->journalOff += pPager->pageSize + 4;
/* Sanity checking on the page. This is more important that I originally
** thought. If a power failure occurs while the journal is being written,
** it could cause invalid data to be written into the journal. We need to
** detect this invalid data (with high probability) and ignore it.
*/
if( pgno==0 || pgno==PAGER_MJ_PGNO(pPager) ){
return SQLITE_DONE;
}
if( pgno>(unsigned)pPager->dbSize ){
return SQLITE_OK;
}
if( useCksum ){
rc = read32bits(jfd, offset+pPager->pageSize+4, &cksum);
if( rc ) return rc;
pPager->journalOff += 4;
if( pager_cksum(pPager, aData)!=cksum ){
return SQLITE_DONE;
}
}
assert( pPager->state==PAGER_RESERVED || pPager->state>=PAGER_EXCLUSIVE );
/* If the pager is in RESERVED state, then there must be a copy of this
** page in the pager cache. In this case just update the pager cache,
** not the database file. The page is left marked dirty in this case.
**
** An exception to the above rule: If the database is in no-sync mode
** and a page is moved during an incremental vacuum then the page may
** not be in the pager cache. Later: if a malloc() or IO error occurs
** during a Movepage() call, then the page may not be in the cache
** either. So the condition described in the above paragraph is not
** assert()able.
**
** If in EXCLUSIVE state, then we update the pager cache if it exists
** and the main file. The page is then marked not dirty.
**
** Ticket #1171: The statement journal might contain page content that is
** different from the page content at the start of the transaction.
** This occurs when a page is changed prior to the start of a statement
** then changed again within the statement. When rolling back such a
** statement we must not write to the original database unless we know
** for certain that original page contents are synced into the main rollback
** journal. Otherwise, a power loss might leave modified data in the
** database file without an entry in the rollback journal that can
** restore the database to its original form. Two conditions must be
** met before writing to the database files. (1) the database must be
** locked. (2) we know that the original page content is fully synced
** in the main journal either because the page is not in cache or else
** the page is marked as needSync==0.
**
** 2008-04-14: When attempting to vacuum a corrupt database file, it
** is possible to fail a statement on a database that does not yet exist.
** Do not attempt to write if database file has never been opened.
*/
pPg = pager_lookup(pPager, pgno);
PAGERTRACE4("PLAYBACK %d page %d hash(%08x)n",
PAGERID(pPager), pgno, pager_datahash(pPager->pageSize, aData));
if( pPager->state>=PAGER_EXCLUSIVE && (pPg==0 || pPg->needSync==0)
&& pPager->fd->pMethods ){
i64 offset = (pgno-1)*(i64)pPager->pageSize;
rc = sqlite3OsWrite(pPager->fd, aData, pPager->pageSize, offset);
if( pPg ){
makeClean(pPg);
}
}
if( pPg ){
/* No page should ever be explicitly rolled back that is in use, except
** for page 1 which is held in use in order to keep the lock on the
** database active. However such a page may be rolled back as a result
** of an internal error resulting in an automatic call to
** sqlite3PagerRollback().
*/
void *pData;
/* assert( pPg->nRef==0 || pPg->pgno==1 ); */
pData = PGHDR_TO_DATA(pPg);
memcpy(pData, aData, pPager->pageSize);
if( pPager->xReiniter ){
pPager->xReiniter(pPg, pPager->pageSize);
}
#ifdef SQLITE_CHECK_PAGES
pPg->pageHash = pager_pagehash(pPg);
#endif
/* If this was page 1, then restore the value of Pager.dbFileVers.
** Do this before any decoding. */
if( pgno==1 ){
memcpy(&pPager->dbFileVers, &((u8*)pData)[24],sizeof(pPager->dbFileVers));
}
/* Decode the page just read from disk */
CODEC1(pPager, pData, pPg->pgno, 3);
}
return rc;
}
/*
** Parameter zMaster is the name of a master journal file. A single journal
** file that referred to the master journal file has just been rolled back.
** This routine checks if it is possible to delete the master journal file,
** and does so if it is.
**
** Argument zMaster may point to Pager.pTmpSpace. So that buffer is not
** available for use within this function.
**
**
** The master journal file contains the names of all child journals.
** To tell if a master journal can be deleted, check to each of the
** children. If all children are either missing or do not refer to
** a different master journal, then this master journal can be deleted.
*/
static int pager_delmaster(Pager *pPager, const char *zMaster){
sqlite3_vfs *pVfs = pPager->pVfs;
int rc;
int master_open = 0;
sqlite3_file *pMaster;
sqlite3_file *pJournal;
char *zMasterJournal = 0; /* Contents of master journal file */
i64 nMasterJournal; /* Size of master journal file */
/* Open the master journal file exclusively in case some other process
** is running this routine also. Not that it makes too much difference.
*/
pMaster = (sqlite3_file *)sqlite3_malloc(pVfs->szOsFile * 2);
pJournal = (sqlite3_file *)(((u8 *)pMaster) + pVfs->szOsFile);
if( !pMaster ){
rc = SQLITE_NOMEM;
}else{
int flags = (SQLITE_OPEN_READONLY|SQLITE_OPEN_MASTER_JOURNAL);
rc = sqlite3OsOpen(pVfs, zMaster, pMaster, flags, 0);
}
if( rc!=SQLITE_OK ) goto delmaster_out;
master_open = 1;
rc = sqlite3OsFileSize(pMaster, &nMasterJournal);
if( rc!=SQLITE_OK ) goto delmaster_out;
if( nMasterJournal>0 ){
char *zJournal;
char *zMasterPtr = 0;
int nMasterPtr = pPager->pVfs->mxPathname+1;
/* Load the entire master journal file into space obtained from
** sqlite3_malloc() and pointed to by zMasterJournal.
*/
zMasterJournal = (char *)sqlite3_malloc(nMasterJournal + nMasterPtr);
if( !zMasterJournal ){
rc = SQLITE_NOMEM;
goto delmaster_out;
}
zMasterPtr = &zMasterJournal[nMasterJournal];
rc = sqlite3OsRead(pMaster, zMasterJournal, nMasterJournal, 0);
if( rc!=SQLITE_OK ) goto delmaster_out;
zJournal = zMasterJournal;
while( (zJournal-zMasterJournal)<nMasterJournal ){
rc = sqlite3OsAccess(pVfs, zJournal, SQLITE_ACCESS_EXISTS);
if( rc!=0 && rc!=1 ){
rc = SQLITE_IOERR_NOMEM;
goto delmaster_out;
}
if( rc==1 ){
/* One of the journals pointed to by the master journal exists.
** Open it and check if it points at the master journal. If
** so, return without deleting the master journal file.
*/
int c;
int flags = (SQLITE_OPEN_READONLY|SQLITE_OPEN_MAIN_JOURNAL);
rc = sqlite3OsOpen(pVfs, zJournal, pJournal, flags, 0);
if( rc!=SQLITE_OK ){
goto delmaster_out;
}
rc = readMasterJournal(pJournal, zMasterPtr, nMasterPtr);
sqlite3OsClose(pJournal);
if( rc!=SQLITE_OK ){
goto delmaster_out;
}
c = zMasterPtr[0]!=0 && strcmp(zMasterPtr, zMaster)==0;
if( c ){
/* We have a match. Do not delete the master journal file. */
goto delmaster_out;
}
}
zJournal += (strlen(zJournal)+1);
}
}
rc = sqlite3OsDelete(pVfs, zMaster, 0);
delmaster_out:
if( zMasterJournal ){
sqlite3_free(zMasterJournal);
}
if( master_open ){
sqlite3OsClose(pMaster);
}
sqlite3_free(pMaster);
return rc;
}
static void pager_truncate_cache(Pager *pPager);
/*
** Truncate the main file of the given pager to the number of pages
** indicated. Also truncate the cached representation of the file.
**
** Might might be the case that the file on disk is smaller than nPage.
** This can happen, for example, if we are in the middle of a transaction
** which has extended the file size and the new pages are still all held
** in cache, then an INSERT or UPDATE does a statement rollback. Some
** operating system implementations can get confused if you try to
** truncate a file to some size that is larger than it currently is,
** so detect this case and do not do the truncation.
*/
static int pager_truncate(Pager *pPager, int nPage){
int rc = SQLITE_OK;
if( pPager->state>=PAGER_EXCLUSIVE && pPager->fd->pMethods ){
i64 currentSize, newSize;
rc = sqlite3OsFileSize(pPager->fd, &currentSize);
newSize = pPager->pageSize*(i64)nPage;
if( rc==SQLITE_OK && currentSize>newSize ){
rc = sqlite3OsTruncate(pPager->fd, newSize);
}
}
if( rc==SQLITE_OK ){
pPager->dbSize = nPage;
pager_truncate_cache(pPager);
}
return rc;
}
/*
** Set the sectorSize for the given pager.
**
** The sector size is at least as big as the sector size reported
** by sqlite3OsSectorSize(). The minimum sector size is 512.
*/
static void setSectorSize(Pager *pPager){
assert(pPager->fd->pMethods||pPager->tempFile);
if( !pPager->tempFile ){
/* Sector size doesn't matter for temporary files. Also, the file
** may not have been opened yet, in whcih case the OsSectorSize()
** call will segfault.
*/
pPager->sectorSize = sqlite3OsSectorSize(pPager->fd);
}
if( pPager->sectorSize<512 ){
pPager->sectorSize = 512;
}
}
/*
** Playback the journal and thus restore the database file to
** the state it was in before we started making changes.
**
** The journal file format is as follows:
**
** (1) 8 byte prefix. A copy of aJournalMagic[].
** (2) 4 byte big-endian integer which is the number of valid page records
** in the journal. If this value is 0xffffffff, then compute the
** number of page records from the journal size.
** (3) 4 byte big-endian integer which is the initial value for the
** sanity checksum.
** (4) 4 byte integer which is the number of pages to truncate the
** database to during a rollback.
** (5) 4 byte big-endian integer which is the sector size. The header
** is this many bytes in size.
** (6) 4 byte big-endian integer which is the page case.
** (7) 4 byte integer which is the number of bytes in the master journal
** name. The value may be zero (indicate that there is no master
** journal.)
** (8) N bytes of the master journal name. The name will be nul-terminated
** and might be shorter than the value read from (5). If the first byte
** of the name is 00 then there is no master journal. The master
** journal name is stored in UTF-8.
** (9) Zero or more pages instances, each as follows:
** + 4 byte page number.
** + pPager->pageSize bytes of data.
** + 4 byte checksum
**
** When we speak of the journal header, we mean the first 8 items above.
** Each entry in the journal is an instance of the 9th item.
**
** Call the value from the second bullet "nRec". nRec is the number of
** valid page entries in the journal. In most cases, you can compute the
** value of nRec from the size of the journal file. But if a power
** failure occurred while the journal was being written, it could be the
** case that the size of the journal file had already been increased but
** the extra entries had not yet made it safely to disk. In such a case,
** the value of nRec computed from the file size would be too large. For
** that reason, we always use the nRec value in the header.
**
** If the nRec value is 0xffffffff it means that nRec should be computed
** from the file size. This value is used when the user selects the
** no-sync option for the journal. A power failure could lead to corruption
** in this case. But for things like temporary table (which will be
** deleted when the power is restored) we don't care.
**
** If the file opened as the journal file is not a well-formed
** journal file then all pages up to the first corrupted page are rolled
** back (or no pages if the journal header is corrupted). The journal file
** is then deleted and SQLITE_OK returned, just as if no corruption had
** been encountered.
**
** If an I/O or malloc() error occurs, the journal-file is not deleted
** and an error code is returned.
*/
static int pager_playback(Pager *pPager, int isHot){
sqlite3_vfs *pVfs = pPager->pVfs;
i64 szJ; /* Size of the journal file in bytes */
u32 nRec; /* Number of Records in the journal */
int i; /* Loop counter */
Pgno mxPg = 0; /* Size of the original file in pages */
int rc; /* Result code of a subroutine */
int res = 0; /* Value returned by sqlite3OsAccess() */
char *zMaster = 0; /* Name of master journal file if any */
/* Figure out how many records are in the journal. Abort early if
** the journal is empty.
*/
assert( pPager->journalOpen );
rc = sqlite3OsFileSize(pPager->jfd, &szJ);
if( rc!=SQLITE_OK || szJ==0 ){
goto end_playback;
}
/* Read the master journal name from the journal, if it is present.
** If a master journal file name is specified, but the file is not
** present on disk, then the journal is not hot and does not need to be
** played back.
*/
zMaster = pPager->pTmpSpace;
rc = readMasterJournal(pPager->jfd, zMaster, pPager->pVfs->mxPathname+1);
if( rc!=SQLITE_OK || (zMaster[0]
&& (res=sqlite3OsAccess(pVfs, zMaster, SQLITE_ACCESS_EXISTS))==0 )
){
zMaster = 0;
goto end_playback;
}
zMaster = 0;
if( res<0 ){
rc = SQLITE_IOERR_NOMEM;
goto end_playback;
}
pPager->journalOff = 0;
/* This loop terminates either when the readJournalHdr() call returns
** SQLITE_DONE or an IO error occurs. */
while( 1 ){
/* Read the next journal header from the journal file. If there are
** not enough bytes left in the journal file for a complete header, or
** it is corrupted, then a process must of failed while writing it.
** This indicates nothing more needs to be rolled back.
*/
rc = readJournalHdr(pPager, szJ, &nRec, &mxPg);
if( rc!=SQLITE_OK ){
if( rc==SQLITE_DONE ){
rc = SQLITE_OK;
}
goto end_playback;
}
/* If nRec is 0xffffffff, then this journal was created by a process
** working in no-sync mode. This means that the rest of the journal
** file consists of pages, there are no more journal headers. Compute
** the value of nRec based on this assumption.
*/
if( nRec==0xffffffff ){
assert( pPager->journalOff==JOURNAL_HDR_SZ(pPager) );
nRec = (szJ - JOURNAL_HDR_SZ(pPager))/JOURNAL_PG_SZ(pPager);
}
/* If nRec is 0 and this rollback is of a transaction created by this
** process and if this is the final header in the journal, then it means
** that this part of the journal was being filled but has not yet been
** synced to disk. Compute the number of pages based on the remaining
** size of the file.
**
** The third term of the test was added to fix ticket #2565.
*/
if( nRec==0 && !isHot &&
pPager->journalHdr+JOURNAL_HDR_SZ(pPager)==pPager->journalOff ){
nRec = (szJ - pPager->journalOff) / JOURNAL_PG_SZ(pPager);
}
/* If this is the first header read from the journal, truncate the
** database file back to its original size.
*/
if( pPager->journalOff==JOURNAL_HDR_SZ(pPager) ){
rc = pager_truncate(pPager, mxPg);
if( rc!=SQLITE_OK ){
goto end_playback;
}
}
/* Copy original pages out of the journal and back into the database file.
*/
for(i=0; i<nRec; i++){
rc = pager_playback_one_page(pPager, pPager->jfd, pPager->journalOff, 1);
if( rc!=SQLITE_OK ){
if( rc==SQLITE_DONE ){
rc = SQLITE_OK;
pPager->journalOff = szJ;
break;
}else{
goto end_playback;
}
}
}
}
/*NOTREACHED*/
assert( 0 );
end_playback:
if( rc==SQLITE_OK ){
zMaster = pPager->pTmpSpace;
rc = readMasterJournal(pPager->jfd, zMaster, pPager->pVfs->mxPathname+1);
}
if( rc==SQLITE_OK ){
rc = pager_end_transaction(pPager);
}
if( rc==SQLITE_OK && zMaster[0] ){
/* If there was a master journal and this routine will return success,
** see if it is possible to delete the master journal.
*/
rc = pager_delmaster(pPager, zMaster);
}
/* The Pager.sectorSize variable may have been updated while rolling
** back a journal created by a process with a different sector size
** value. Reset it to the correct value for this process.
*/
setSectorSize(pPager);
return rc;
}
/*
** Playback the statement journal.
**
** This is similar to playing back the transaction journal but with
** a few extra twists.
**
** (1) The number of pages in the database file at the start of
** the statement is stored in pPager->stmtSize, not in the
** journal file itself.
**
** (2) In addition to playing back the statement journal, also
** playback all pages of the transaction journal beginning
** at offset pPager->stmtJSize.
*/
static int pager_stmt_playback(Pager *pPager){
i64 szJ; /* Size of the full journal */
i64 hdrOff;
int nRec; /* Number of Records */
int i; /* Loop counter */
int rc;
szJ = pPager->journalOff;
#ifndef NDEBUG
{
i64 os_szJ;
rc = sqlite3OsFileSize(pPager->jfd, &os_szJ);
if( rc!=SQLITE_OK ) return rc;
assert( szJ==os_szJ );
}
#endif
/* Set hdrOff to be the offset just after the end of the last journal
** page written before the first journal-header for this statement
** transaction was written, or the end of the file if no journal
** header was written.
*/
hdrOff = pPager->stmtHdrOff;
assert( pPager->fullSync || !hdrOff );
if( !hdrOff ){
hdrOff = szJ;
}
/* Truncate the database back to its original size.
*/
rc = pager_truncate(pPager, pPager->stmtSize);
assert( pPager->state>=PAGER_SHARED );
/* Figure out how many records are in the statement journal.
*/
assert( pPager->stmtInUse && pPager->journalOpen );
nRec = pPager->stmtNRec;
/* Copy original pages out of the statement journal and back into the
** database file. Note that the statement journal omits checksums from
** each record since power-failure recovery is not important to statement
** journals.
*/
for(i=0; i<nRec; i++){
i64 offset = i*(4+pPager->pageSize);
rc = pager_playback_one_page(pPager, pPager->stfd, offset, 0);
assert( rc!=SQLITE_DONE );
if( rc!=SQLITE_OK ) goto end_stmt_playback;
}
/* Now roll some pages back from the transaction journal. Pager.stmtJSize
** was the size of the journal file when this statement was started, so
** everything after that needs to be rolled back, either into the
** database, the memory cache, or both.
**
** If it is not zero, then Pager.stmtHdrOff is the offset to the start
** of the first journal header written during this statement transaction.
*/
pPager->journalOff = pPager->stmtJSize;
pPager->cksumInit = pPager->stmtCksum;
while( pPager->journalOff < hdrOff ){
rc = pager_playback_one_page(pPager, pPager->jfd, pPager->journalOff, 1);
assert( rc!=SQLITE_DONE );
if( rc!=SQLITE_OK ) goto end_stmt_playback;
}
while( pPager->journalOff < szJ ){
u32 nJRec; /* Number of Journal Records */
u32 dummy;
rc = readJournalHdr(pPager, szJ, &nJRec, &dummy);
if( rc!=SQLITE_OK ){
assert( rc!=SQLITE_DONE );
goto end_stmt_playback;
}
if( nJRec==0 ){
nJRec = (szJ - pPager->journalOff) / (pPager->pageSize+8);
}
for(i=nJRec-1; i>=0 && pPager->journalOff < szJ; i--){
rc = pager_playback_one_page(pPager, pPager->jfd, pPager->journalOff, 1);
assert( rc!=SQLITE_DONE );
if( rc!=SQLITE_OK ) goto end_stmt_playback;
}
}
pPager->journalOff = szJ;
end_stmt_playback:
if( rc==SQLITE_OK) {
pPager->journalOff = szJ;
/* pager_reload_cache(pPager); */
}
return rc;
}
/*
** Change the maximum number of in-memory pages that are allowed.
*/
void sqlite3PagerSetCachesize(Pager *pPager, int mxPage){
if( mxPage>10 ){
pPager->mxPage = mxPage;
}else{
pPager->mxPage = 10;
}
}
/*
** Adjust the robustness of the database to damage due to OS crashes
** or power failures by changing the number of syncs()s when writing
** the rollback journal. There are three levels:
**
** OFF sqlite3OsSync() is never called. This is the default
** for temporary and transient files.
**
** NORMAL The journal is synced once before writes begin on the
** database. This is normally adequate protection, but
** it is theoretically possible, though very unlikely,
** that an inopertune power failure could leave the journal
** in a state which would cause damage to the database
** when it is rolled back.
**
** FULL The journal is synced twice before writes begin on the
** database (with some additional information - the nRec field
** of the journal header - being written in between the two
** syncs). If we assume that writing a
** single disk sector is atomic, then this mode provides
** assurance that the journal will not be corrupted to the
** point of causing damage to the database during rollback.
**
** Numeric values associated with these states are OFF==1, NORMAL=2,
** and FULL=3.
*/
#ifndef SQLITE_OMIT_PAGER_PRAGMAS
void sqlite3PagerSetSafetyLevel(Pager *pPager, int level, int full_fsync){
pPager->noSync = level==1 || pPager->tempFile;
pPager->fullSync = level==3 && !pPager->tempFile;
pPager->sync_flags = (full_fsync?SQLITE_SYNC_FULL:SQLITE_SYNC_NORMAL);
if( pPager->noSync ) pPager->needSync = 0;
}
#endif
/*
** The following global variable is incremented whenever the library
** attempts to open a temporary file. This information is used for
** testing and analysis only.
*/
#ifdef SQLITE_TEST
int sqlite3_opentemp_count = 0;
#endif
/*
** Open a temporary file.
**
** Write the file descriptor into *fd. Return SQLITE_OK on success or some
** other error code if we fail. The OS will automatically delete the temporary
** file when it is closed.
*/
static int sqlite3PagerOpentemp(
sqlite3_vfs *pVfs, /* The virtual file system layer */
sqlite3_file *pFile, /* Write the file descriptor here */
char *zFilename, /* Name of the file. Might be NULL */
int vfsFlags /* Flags passed through to the VFS */
){
int rc;
assert( zFilename!=0 );
#ifdef SQLITE_TEST
sqlite3_opentemp_count++; /* Used for testing and analysis only */
#endif
vfsFlags |= SQLITE_OPEN_READWRITE | SQLITE_OPEN_CREATE |
SQLITE_OPEN_EXCLUSIVE | SQLITE_OPEN_DELETEONCLOSE;
rc = sqlite3OsOpen(pVfs, zFilename, pFile, vfsFlags, 0);
assert( rc!=SQLITE_OK || pFile->pMethods );
return rc;
}
/*
** Create a new page cache and put a pointer to the page cache in *ppPager.
** The file to be cached need not exist. The file is not locked until
** the first call to sqlite3PagerGet() and is only held open until the
** last page is released using sqlite3PagerUnref().
**
** If zFilename is NULL then a randomly-named temporary file is created
** and used as the file to be cached. The file will be deleted
** automatically when it is closed.
**
** If zFilename is ":memory:" then all information is held in cache.
** It is never written to disk. This can be used to implement an
** in-memory database.
*/
int sqlite3PagerOpen(
sqlite3_vfs *pVfs, /* The virtual file system to use */
Pager **ppPager, /* Return the Pager structure here */
const char *zFilename, /* Name of the database file to open */
int nExtra, /* Extra bytes append to each in-memory page */
int flags, /* flags controlling this file */
int vfsFlags /* flags passed through to sqlite3_vfs.xOpen() */
){
u8 *pPtr;
Pager *pPager = 0;
int rc = SQLITE_OK;
int i;
int tempFile = 0;
int memDb = 0;
int readOnly = 0;
int useJournal = (flags & PAGER_OMIT_JOURNAL)==0;
int noReadlock = (flags & PAGER_NO_READLOCK)!=0;
int journalFileSize = sqlite3JournalSize(pVfs);
int nDefaultPage = SQLITE_DEFAULT_PAGE_SIZE;
char *zPathname;
int nPathname;
char *zStmtJrnl;
int nStmtJrnl;
/* The default return is a NULL pointer */
*ppPager = 0;
/* Compute the full pathname */
nPathname = pVfs->mxPathname+1;
zPathname = sqlite3_malloc(nPathname*2);
if( zPathname==0 ){
return SQLITE_NOMEM;
}
if( zFilename && zFilename[0] ){
#ifndef SQLITE_OMIT_MEMORYDB
if( strcmp(zFilename,":memory:")==0 ){
memDb = 1;
zPathname[0] = 0;
}else
#endif
{
rc = sqlite3OsFullPathname(pVfs, zFilename, nPathname, zPathname);
}
}else{
rc = sqlite3OsGetTempname(pVfs, nPathname, zPathname);
}
if( rc!=SQLITE_OK ){
sqlite3_free(zPathname);
return rc;
}
nPathname = strlen(zPathname);
/* Put the statement journal in temporary disk space since this is
** sometimes RAM disk or other optimized storage. Unlikely the main
** main journal file, the statement journal does not need to be
** colocated with the database nor does it need to be persistent.
*/
zStmtJrnl = &zPathname[nPathname+1];
rc = sqlite3OsGetTempname(pVfs, pVfs->mxPathname+1, zStmtJrnl);
if( rc!=SQLITE_OK ){
sqlite3_free(zPathname);
return rc;
}
nStmtJrnl = strlen(zStmtJrnl);
/* Allocate memory for the pager structure */
pPager = sqlite3MallocZero(
sizeof(*pPager) + /* Pager structure */
journalFileSize + /* The journal file structure */
pVfs->szOsFile * 3 + /* The main db and two journal files */
3*nPathname + 40 + /* zFilename, zDirectory, zJournal */
nStmtJrnl /* zStmtJrnl */
);
if( !pPager ){
sqlite3_free(zPathname);
return SQLITE_NOMEM;
}
pPtr = (u8 *)&pPager[1];
pPager->vfsFlags = vfsFlags;
pPager->fd = (sqlite3_file*)&pPtr[pVfs->szOsFile*0];
pPager->stfd = (sqlite3_file*)&pPtr[pVfs->szOsFile*1];
pPager->jfd = (sqlite3_file*)&pPtr[pVfs->szOsFile*2];
pPager->zFilename = (char*)&pPtr[pVfs->szOsFile*2+journalFileSize];
pPager->zDirectory = &pPager->zFilename[nPathname+1];
pPager->zJournal = &pPager->zDirectory[nPathname+1];
pPager->zStmtJrnl = &pPager->zJournal[nPathname+10];
pPager->pVfs = pVfs;
memcpy(pPager->zFilename, zPathname, nPathname+1);
memcpy(pPager->zStmtJrnl, zStmtJrnl, nStmtJrnl+1);
sqlite3_free(zPathname);
/* Open the pager file.
*/
if( zFilename && zFilename[0] && !memDb ){
if( nPathname>(pVfs->mxPathname - sizeof("-journal")) ){
rc = SQLITE_CANTOPEN;
}else{
int fout = 0;
rc = sqlite3OsOpen(pVfs, pPager->zFilename, pPager->fd,
pPager->vfsFlags, &fout);
readOnly = (fout&SQLITE_OPEN_READONLY);
/* If the file was successfully opened for read/write access,
** choose a default page size in case we have to create the
** database file. The default page size is the maximum of:
**
** + SQLITE_DEFAULT_PAGE_SIZE,
** + The value returned by sqlite3OsSectorSize()
** + The largest page size that can be written atomically.
*/
if( rc==SQLITE_OK && !readOnly ){
int iSectorSize = sqlite3OsSectorSize(pPager->fd);
if( nDefaultPage<iSectorSize ){
nDefaultPage = iSectorSize;
}
#ifdef SQLITE_ENABLE_ATOMIC_WRITE
{
int iDc = sqlite3OsDeviceCharacteristics(pPager->fd);
int ii;
assert(SQLITE_IOCAP_ATOMIC512==(512>>8));
assert(SQLITE_IOCAP_ATOMIC64K==(65536>>8));
assert(SQLITE_MAX_DEFAULT_PAGE_SIZE<=65536);
for(ii=nDefaultPage; ii<=SQLITE_MAX_DEFAULT_PAGE_SIZE; ii=ii*2){
if( iDc&(SQLITE_IOCAP_ATOMIC|(ii>>8)) ) nDefaultPage = ii;
}
}
#endif
if( nDefaultPage>SQLITE_MAX_DEFAULT_PAGE_SIZE ){
nDefaultPage = SQLITE_MAX_DEFAULT_PAGE_SIZE;
}
}
}
}else if( !memDb ){
/* If a temporary file is requested, it is not opened immediately.
** In this case we accept the default page size and delay actually
** opening the file until the first call to OsWrite().
*/
tempFile = 1;
pPager->state = PAGER_EXCLUSIVE;
}
if( pPager && rc==SQLITE_OK ){
pPager->pTmpSpace = (char *)sqlite3_malloc(nDefaultPage);
}
/* If an error occured in either of the blocks above.
** Free the Pager structure and close the file.
** Since the pager is not allocated there is no need to set
** any Pager.errMask variables.
*/
if( !pPager || !pPager->pTmpSpace ){
sqlite3OsClose(pPager->fd);
sqlite3_free(pPager);
return ((rc==SQLITE_OK)?SQLITE_NOMEM:rc);
}
PAGERTRACE3("OPEN %d %sn", FILEHANDLEID(pPager->fd), pPager->zFilename);
IOTRACE(("OPEN %p %sn", pPager, pPager->zFilename))
/* Fill in Pager.zDirectory[] */
memcpy(pPager->zDirectory, pPager->zFilename, nPathname+1);
for(i=strlen(pPager->zDirectory); i>0 && pPager->zDirectory[i-1]!='/'; i--){}
if( i>0 ) pPager->zDirectory[i-1] = 0;
/* Fill in Pager.zJournal[] */
memcpy(pPager->zJournal, pPager->zFilename, nPathname);
memcpy(&pPager->zJournal[nPathname], "-journal", 9);
/* pPager->journalOpen = 0; */
pPager->useJournal = useJournal && !memDb;
pPager->noReadlock = noReadlock && readOnly;
/* pPager->stmtOpen = 0; */
/* pPager->stmtInUse = 0; */
/* pPager->nRef = 0; */
pPager->dbSize = memDb-1;
pPager->pageSize = nDefaultPage;
/* pPager->stmtSize = 0; */
/* pPager->stmtJSize = 0; */
/* pPager->nPage = 0; */
pPager->mxPage = 100;
pPager->mxPgno = SQLITE_MAX_PAGE_COUNT;
/* pPager->state = PAGER_UNLOCK; */
assert( pPager->state == (tempFile ? PAGER_EXCLUSIVE : PAGER_UNLOCK) );
/* pPager->errMask = 0; */
pPager->tempFile = tempFile;
assert( tempFile==PAGER_LOCKINGMODE_NORMAL
|| tempFile==PAGER_LOCKINGMODE_EXCLUSIVE );
assert( PAGER_LOCKINGMODE_EXCLUSIVE==1 );
pPager->exclusiveMode = tempFile;
pPager->memDb = memDb;
pPager->readOnly = readOnly;
/* pPager->needSync = 0; */
pPager->noSync = pPager->tempFile || !useJournal;
pPager->fullSync = (pPager->noSync?0:1);
pPager->sync_flags = SQLITE_SYNC_NORMAL;
/* pPager->pFirst = 0; */
/* pPager->pFirstSynced = 0; */
/* pPager->pLast = 0; */
pPager->nExtra = FORCE_ALIGNMENT(nExtra);
assert(pPager->fd->pMethods||memDb||tempFile);
if( !memDb ){
setSectorSize(pPager);
}
/* pPager->pBusyHandler = 0; */
/* memset(pPager->aHash, 0, sizeof(pPager->aHash)); */
*ppPager = pPager;
#ifdef SQLITE_ENABLE_MEMORY_MANAGEMENT
pPager->iInUseMM = 0;
pPager->iInUseDB = 0;
if( !memDb ){
sqlite3_mutex *mutex = sqlite3_mutex_alloc(SQLITE_MUTEX_STATIC_MEM2);
sqlite3_mutex_enter(mutex);
pPager->pNext = sqlite3PagerList;
if( sqlite3PagerList ){
assert( sqlite3PagerList->pPrev==0 );
sqlite3PagerList->pPrev = pPager;
}
pPager->pPrev = 0;
sqlite3PagerList = pPager;
sqlite3_mutex_leave(mutex);
}
#endif
return SQLITE_OK;
}
/*
** Set the busy handler function.
*/
void sqlite3PagerSetBusyhandler(Pager *pPager, BusyHandler *pBusyHandler){
pPager->pBusyHandler = pBusyHandler;
}
/*
** Set the destructor for this pager. If not NULL, the destructor is called
** when the reference count on each page reaches zero. The destructor can
** be used to clean up information in the extra segment appended to each page.
**
** The destructor is not called as a result sqlite3PagerClose().
** Destructors are only called by sqlite3PagerUnref().
*/
void sqlite3PagerSetDestructor(Pager *pPager, void (*xDesc)(DbPage*,int)){
pPager->xDestructor = xDesc;
}
/*
** Set the reinitializer for this pager. If not NULL, the reinitializer
** is called when the content of a page in cache is restored to its original
** value as a result of a rollback. The callback gives higher-level code
** an opportunity to restore the EXTRA section to agree with the restored
** page data.
*/
void sqlite3PagerSetReiniter(Pager *pPager, void (*xReinit)(DbPage*,int)){
pPager->xReiniter = xReinit;
}
/*
** Set the page size to *pPageSize. If the suggest new page size is
** inappropriate, then an alternative page size is set to that
** value before returning.
*/
int sqlite3PagerSetPagesize(Pager *pPager, u16 *pPageSize){
int rc = SQLITE_OK;
u16 pageSize = *pPageSize;
assert( pageSize==0 || (pageSize>=512 && pageSize<=SQLITE_MAX_PAGE_SIZE) );
if( pageSize && pageSize!=pPager->pageSize
&& !pPager->memDb && pPager->nRef==0
){
char *pNew = (char *)sqlite3_malloc(pageSize);
if( !pNew ){
rc = SQLITE_NOMEM;
}else{
pagerEnter(pPager);
pager_reset(pPager);
pPager->pageSize = pageSize;
setSectorSize(pPager);
sqlite3_free(pPager->pTmpSpace);
pPager->pTmpSpace = pNew;
pagerLeave(pPager);
}
}
*pPageSize = pPager->pageSize;
return rc;
}
/*
** Return a pointer to the "temporary page" buffer held internally
** by the pager. This is a buffer that is big enough to hold the
** entire content of a database page. This buffer is used internally
** during rollback and will be overwritten whenever a rollback
** occurs. But other modules are free to use it too, as long as
** no rollbacks are happening.
*/
void *sqlite3PagerTempSpace(Pager *pPager){
return pPager->pTmpSpace;
}
/*
** Attempt to set the maximum database page count if mxPage is positive.
** Make no changes if mxPage is zero or negative. And never reduce the
** maximum page count below the current size of the database.
**
** Regardless of mxPage, return the current maximum page count.
*/
int sqlite3PagerMaxPageCount(Pager *pPager, int mxPage){
if( mxPage>0 ){
pPager->mxPgno = mxPage;
}
sqlite3PagerPagecount(pPager);
return pPager->mxPgno;
}
/*
** The following set of routines are used to disable the simulated
** I/O error mechanism. These routines are used to avoid simulated
** errors in places where we do not care about errors.
**
** Unless -DSQLITE_TEST=1 is used, these routines are all no-ops
** and generate no code.
*/
#ifdef SQLITE_TEST
extern int sqlite3_io_error_pending;
extern int sqlite3_io_error_hit;
static int saved_cnt;
void disable_simulated_io_errors(void){
saved_cnt = sqlite3_io_error_pending;
sqlite3_io_error_pending = -1;
}
void enable_simulated_io_errors(void){
sqlite3_io_error_pending = saved_cnt;
}
#else
# define disable_simulated_io_errors()
# define enable_simulated_io_errors()
#endif
/*
** Read the first N bytes from the beginning of the file into memory
** that pDest points to.
**
** No error checking is done. The rational for this is that this function
** may be called even if the file does not exist or contain a header. In
** these cases sqlite3OsRead() will return an error, to which the correct
** response is to zero the memory at pDest and continue. A real IO error
** will presumably recur and be picked up later (Todo: Think about this).
*/
int sqlite3PagerReadFileheader(Pager *pPager, int N, unsigned char *pDest){
int rc = SQLITE_OK;
memset(pDest, 0, N);
assert(MEMDB||pPager->fd->pMethods||pPager->tempFile);
if( pPager->fd->pMethods ){
IOTRACE(("DBHDR %p 0 %dn", pPager, N))
rc = sqlite3OsRead(pPager->fd, pDest, N, 0);
if( rc==SQLITE_IOERR_SHORT_READ ){
rc = SQLITE_OK;
}
}
return rc;
}
/*
** Return the total number of pages in the disk file associated with
** pPager.
**
** If the PENDING_BYTE lies on the page directly after the end of the
** file, then consider this page part of the file too. For example, if
** PENDING_BYTE is byte 4096 (the first byte of page 5) and the size of the
** file is 4096 bytes, 5 is returned instead of 4.
*/
int sqlite3PagerPagecount(Pager *pPager){
i64 n = 0;
int rc;
assert( pPager!=0 );
if( pPager->errCode ){
return -1;
}
if( pPager->dbSize>=0 ){
n = pPager->dbSize;
} else {
assert(pPager->fd->pMethods||pPager->tempFile);
if( (pPager->fd->pMethods)
&& (rc = sqlite3OsFileSize(pPager->fd, &n))!=SQLITE_OK ){
pPager->nRef++;
pager_error(pPager, rc);
pPager->nRef--;
return -1;
}
if( n>0 && n<pPager->pageSize ){
n = 1;
}else{
n /= pPager->pageSize;
}
if( pPager->state!=PAGER_UNLOCK ){
pPager->dbSize = n;
}
}
if( n==(PENDING_BYTE/pPager->pageSize) ){
n++;
}
if( n>pPager->mxPgno ){
pPager->mxPgno = n;
}
return n;
}
#ifndef SQLITE_OMIT_MEMORYDB
/*
** Clear a PgHistory block
*/
static void clearHistory(PgHistory *pHist){
sqlite3_free(pHist->pOrig);
sqlite3_free(pHist->pStmt);
pHist->pOrig = 0;
pHist->pStmt = 0;
}
#else
#define clearHistory(x)
#endif
/*
** Forward declaration
*/
static int syncJournal(Pager*);
/*
** Unlink pPg from its hash chain. Also set the page number to 0 to indicate
** that the page is not part of any hash chain. This is required because the
** sqlite3PagerMovepage() routine can leave a page in the
** pNextFree/pPrevFree list that is not a part of any hash-chain.
*/
static void unlinkHashChain(Pager *pPager, PgHdr *pPg){
if( pPg->pgno==0 ){
assert( pPg->pNextHash==0 && pPg->pPrevHash==0 );
return;
}
if( pPg->pNextHash ){
pPg->pNextHash->pPrevHash = pPg->pPrevHash;
}
if( pPg->pPrevHash ){
assert( pPager->aHash[pPg->pgno & (pPager->nHash-1)]!=pPg );
pPg->pPrevHash->pNextHash = pPg->pNextHash;
}else{
int h = pPg->pgno & (pPager->nHash-1);
pPager->aHash[h] = pPg->pNextHash;
}
if( MEMDB ){
clearHistory(PGHDR_TO_HIST(pPg, pPager));
}
pPg->pgno = 0;
pPg->pNextHash = pPg->pPrevHash = 0;
}
/*
** Unlink a page from the free list (the list of all pages where nRef==0)
** and from its hash collision chain.
*/
static void unlinkPage(PgHdr *pPg){
Pager *pPager = pPg->pPager;
/* Unlink from free page list */
lruListRemove(pPg);
/* Unlink from the pgno hash table */
unlinkHashChain(pPager, pPg);
}
/*
** This routine is used to truncate the cache when a database
** is truncated. Drop from the cache all pages whose pgno is
** larger than pPager->dbSize and is unreferenced.
**
** Referenced pages larger than pPager->dbSize are zeroed.
**
** Actually, at the point this routine is called, it would be
** an error to have a referenced page. But rather than delete
** that page and guarantee a subsequent segfault, it seems better
** to zero it and hope that we error out sanely.
*/
static void pager_truncate_cache(Pager *pPager){
PgHdr *pPg;
PgHdr **ppPg;
int dbSize = pPager->dbSize;
ppPg = &pPager->pAll;
while( (pPg = *ppPg)!=0 ){
if( pPg->pgno<=dbSize ){
ppPg = &pPg->pNextAll;
}else if( pPg->nRef>0 ){
memset(PGHDR_TO_DATA(pPg), 0, pPager->pageSize);
ppPg = &pPg->pNextAll;
}else{
*ppPg = pPg->pNextAll;
IOTRACE(("PGFREE %p %dn", pPager, pPg->pgno));
PAGER_INCR(sqlite3_pager_pgfree_count);
unlinkPage(pPg);
makeClean(pPg);
sqlite3_free(pPg->pData);
sqlite3_free(pPg);
pPager->nPage--;
}
}
}
/*
** Try to obtain a lock on a file. Invoke the busy callback if the lock
** is currently not available. Repeat until the busy callback returns
** false or until the lock succeeds.
**
** Return SQLITE_OK on success and an error code if we cannot obtain
** the lock.
*/
static int pager_wait_on_lock(Pager *pPager, int locktype){
int rc;
/* The OS lock values must be the same as the Pager lock values */
assert( PAGER_SHARED==SHARED_LOCK );
assert( PAGER_RESERVED==RESERVED_LOCK );
assert( PAGER_EXCLUSIVE==EXCLUSIVE_LOCK );
/* If the file is currently unlocked then the size must be unknown */