pager.c

最新的sqlite3.6.2源代码

  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(u=0; u<nRec; u++){
      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, zMaster[0]!='\0');
  }
  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;

  /* 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){
  sqlite3PcacheSetCachesize(pPager->pPCache, mxPage);
}

/*
** 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 bFullFsync){
  pPager->noSync =  level==1 || pPager->tempFile;
  pPager->fullSync = level==3 && !pPager->tempFile;
  pPager->sync_flags = (bFullFsync?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(
  Pager *pPager,        /* The pager object */
  sqlite3_file *pFile,  /* Write the file descriptor here */
  int vfsFlags          /* Flags passed through to the VFS */
){
  int rc;

#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(pPager->pVfs, 0, pFile, vfsFlags, 0);
  assert( rc!=SQLITE_OK || pFile->pMethods );
  return rc;
}

static int pagerStress(void *,PgHdr *);

/*
** 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 */
  void (*xDesc)(DbPage*),  /* Page destructor function */
  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 pcacheSize = sqlite3PcacheSize();
  int szPageDflt = SQLITE_DEFAULT_PAGE_SIZE;
  char *zPathname = 0;
  int nPathname = 0;

  /* The default return is a NULL pointer */
  *ppPager = 0;

  /* Compute and store the full pathname in an allocated buffer pointed
  ** to by zPathname, length nPathname. Or, if this is a temporary file,
  ** leave both nPathname and zPathname set to 0.
  */
  if( zFilename && zFilename[0] ){
    nPathname = pVfs->mxPathname+1;
    zPathname = sqlite3Malloc(nPathname*2);
    if( zPathname==0 ){
      return SQLITE_NOMEM;
    }
#ifndef SQLITE_OMIT_MEMORYDB
    if( strcmp(zFilename,":memory:")==0 ){
      memDb = 1;
      zPathname[0] = 0;
    }else
#endif
    {
      rc = sqlite3OsFullPathname(pVfs, zFilename, nPathname, zPathname);
    }
    if( rc!=SQLITE_OK ){
      sqlite3_free(zPathname);
      return rc;
    }
    nPathname = strlen(zPathname);
  }

  /* Allocate memory for the pager structure */
  pPager = sqlite3MallocZero(
    sizeof(*pPager) +           /* Pager structure */
    pcacheSize      +           /* PCache object */
    journalFileSize +           /* The journal file structure */ 
    pVfs->szOsFile * 3 +        /* The main db and two journal files */ 
    3*nPathname + 40            /* zFilename, zDirectory, zJournal */
  );
  if( !pPager ){
    sqlite3_free(zPathname);
    return SQLITE_NOMEM;
  }
  pPager->pPCache = (PCache *)&pPager[1];
  pPtr = ((u8 *)&pPager[1]) + pcacheSize;
  pPager->vfsFlags = vfsFlags;
  pPager->fd = (sqlite3_file*)&pPtr[pVfs->szOsFile*0];
  pPager->stfd = (sqlite3_file*)&pPtr[pVfs->szOsFile*1];
  pPager->jfd =