pager.c

一个小型的嵌入式数据库

**       name.  The value may be zero (indicate that there is no master
**       journal.)
**  (6)  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 \000 then there is no master journal.  The master
**       journal name is stored in UTF-8.
**  (7)  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 6 items above.
** Each entry in the journal is an instance of the 7th 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){
  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 */
  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 ){
    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.
  */
  rc = readMasterJournal(&pPager->jfd, &zMaster);
  assert( rc!=SQLITE_DONE );
  if( rc!=SQLITE_OK || (zMaster && !sqlite3OsFileExists(zMaster)) ){
    sqliteFree(zMaster);
    zMaster = 0;
    if( rc==SQLITE_DONE ) rc = SQLITE_OK;
    goto end_playback;
  }
  sqlite3OsSeek(&pPager->jfd, 0);
  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 this is the first header read from the journal, truncate the
    ** database file back to it's original size.
    */
    if( pPager->state>=PAGER_EXCLUSIVE && 
        pPager->journalOff==JOURNAL_HDR_SZ(pPager) ){
      assert( pPager->origDbSize==0 || pPager->origDbSize==mxPg );
      rc = pager_truncate(pPager, mxPg);
      if( rc!=SQLITE_OK ){
        goto end_playback;
      }
      pPager->dbSize = mxPg;
    }

    /* rc = sqlite3OsSeek(&pPager->jfd, JOURNAL_HDR_SZ(pPager)); */
    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, 1);
      if( rc!=SQLITE_OK ){
        if( rc==SQLITE_DONE ){
          rc = SQLITE_OK;
          pPager->journalOff = szJ;
          break;
        }else{
          goto end_playback;
        }
      }
    }
  }

  /* Pages that have been written to the journal but never synced
  ** where not restored by the loop above.  We have to restore those
  ** pages by reading them back from the original database.
  */
  assert( rc==SQLITE_OK );
  pager_reload_cache(pPager);

end_playback:
  if( rc==SQLITE_OK ){
    rc = pager_unwritelock(pPager);
  }
  if( zMaster ){
    /* If there was a master journal and this routine will return true,
    ** see if it is possible to delete the master journal.
    */
    if( rc==SQLITE_OK ){
      rc = pager_delmaster(zMaster);
    }
    sqliteFree(zMaster);
  }

  /* The Pager.sectorSize variable may have been updated while rolling
  ** back a journal created by a process with a different PAGER_SECTOR_SIZE
  ** value. Reset it to the correct value for this process.
  */
  pPager->sectorSize = PAGER_SECTOR_SIZE;
  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 to the first journal header written
  ** this statement transaction, 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.
  */
  if( pPager->state>=PAGER_EXCLUSIVE ){
    rc = pager_truncate(pPager, pPager->stmtSize);
  }
  pPager->dbSize = pPager->stmtSize;

  /* Figure out how many records are in the statement journal.
  */
  assert( pPager->stmtInUse && pPager->journalOpen );
  sqlite3OsSeek(&pPager->stfd, 0);
  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=nRec-1; i>=0; i--){
    rc = pager_playback_one_page(pPager, &pPager->stfd, 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.
  */
  rc = sqlite3OsSeek(&pPager->jfd, pPager->stmtJSize);
  if( rc!=SQLITE_OK ){
    goto end_stmt_playback;
  }
  pPager->journalOff = pPager->stmtJSize;
  pPager->cksumInit = pPager->stmtCksum;
  assert( JOURNAL_HDR_SZ(pPager)<(pPager->pageSize+8) );
  while( pPager->journalOff <= (hdrOff-(pPager->pageSize+8)) ){
    rc = pager_playback_one_page(pPager, &pPager->jfd, 1);
    assert( rc!=SQLITE_DONE );
    if( rc!=SQLITE_OK ) goto end_stmt_playback;
  }

  while( pPager->journalOff < szJ ){
    u32 nRec;
    u32 dummy;
    rc = readJournalHdr(pPager, szJ, &nRec, &dummy);
    if( rc!=SQLITE_OK ){
      assert( rc!=SQLITE_DONE );
      goto end_stmt_playback;
    }
    if( nRec==0 ){
      nRec = (szJ - pPager->journalOff) / (pPager->pageSize+8);
    }
    for(i=nRec-1; i>=0 && pPager->journalOff < szJ; i--){
      rc = pager_playback_one_page(pPager, &pPager->jfd, 1);
      assert( rc!=SQLITE_DONE );
      if( rc!=SQLITE_OK ) goto end_stmt_playback;
    }
  }

  pPager->journalOff = szJ;
  
end_stmt_playback:
  if( rc!=SQLITE_OK ){
    pPager->errMask |= PAGER_ERR_CORRUPT;
    rc = SQLITE_CORRUPT;  /* bkpt-CORRUPT */
  }else{
    pPager->journalOff = szJ;
    /* pager_reload_cache(pPager); */
  }
  return rc;
}

/*
** Change the maximum number of in-memory pages that are allowed.
**
** The maximum number is the absolute value of the mxPage parameter.
** If mxPage is negative, the noSync flag is also set.  noSync bypasses
** calls to sqlite3OsSync().  The pager runs much faster with noSync on,
** but if the operating system crashes or there is an abrupt power 
** failure, the database file might be left in an inconsistent and
** unrepairable state.  
*/
void sqlite3pager_set_cachesize(Pager *pPager, int mxPage){
  if( mxPage>=0 ){
    pPager->noSync = pPager->tempFile;
    if( pPager->noSync ) pPager->needSync = 0; 
  }else{
    pPager->noSync = 1;
    mxPage = -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 sqlite3pager_set_safety_level(Pager *pPager, int level){
  pPager->noSync =  level==1 || pPager->tempFile;
  pPager->fullSync = level==3 && !pPager->tempFile;
  if( pPager->noSync ) pPager->needSync = 0;
}
#endif

/*
** Open a temporary file.  Write the name of the file into zName
** (zName must be at least SQLITE_TEMPNAME_SIZE bytes long.)  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 sqlite3pager_opentemp(char *zFile, OsFile *fd){
  int cnt = 8;
  int rc;
  do{
    cnt--;
    sqlite3OsTempFileName(zFile);
    rc = sqlite3OsOpenExclusive(zFile, fd, 1);
  }while( cnt>0 && rc!=SQLITE_OK && rc!=SQLITE_NOMEM );
  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 sqlite3pager_get() and is only held open until the
** last page is released using sqlite3pager_unref().
**
** 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 sqlite3pager_open(
  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 */
){
  Pager *pPager;
  char *zFullPathname = 0;
  int nameLen;
  OsFile fd;
  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;