pager.c

sqlite database for embed linux

  rc = seekJournalHdr(pPager);
  if( rc ) return rc;

  if( pPager->journalOff+JOURNAL_HDR_SZ(pPager) > journalSize ){
    return SQLITE_DONE;
  }

  rc = sqlite3OsRead(pPager->jfd, aMagic, sizeof(aMagic));
  if( rc ) return rc;

  if( memcmp(aMagic, aJournalMagic, sizeof(aMagic))!=0 ){
    return SQLITE_DONE;
  }

  rc = read32bits(pPager->jfd, pNRec);
  if( rc ) return rc;

  rc = read32bits(pPager->jfd, &pPager->cksumInit);
  if( rc ) return rc;

  rc = read32bits(pPager->jfd, pDbSize);
  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, (u32 *)&pPager->sectorSize);
  if( rc ) return rc;

  pPager->journalOff += JOURNAL_HDR_SZ(pPager);
  rc = sqlite3OsSeek(pPager->jfd, pPager->journalOff);
  return rc;
}


/*
** 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; 
  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 ){
    rc = seekJournalHdr(pPager);
    if( rc!=SQLITE_OK ) return rc;
  }
  pPager->journalOff += (len+20);

  rc = write32bits(pPager->jfd, PAGER_MJ_PGNO(pPager));
  if( rc!=SQLITE_OK ) return rc;

  rc = sqlite3OsWrite(pPager->jfd, zMaster, len);
  if( rc!=SQLITE_OK ) return rc;

  put32bits(zBuf, len);
  put32bits(&zBuf[4], cksum);
  memcpy(&zBuf[8], aJournalMagic, sizeof(aJournalMagic));
  rc = sqlite3OsWrite(pPager->jfd, zBuf, 8+sizeof(aJournalMagic));
  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;
}

/*
** Unlock the database file.
*/
static void pager_unlock(Pager *pPager){
  if( !pPager->exclusiveMode ){
    if( !MEMDB ){
      sqlite3OsUnlock(pPager->fd, NO_LOCK);
      pPager->dbSize = -1;
      IOTRACE(("UNLOCK %p\n", pPager))
    }
    pPager->state = PAGER_UNLOCK;
    pPager->changeCountDone = 0;
  }
}

/*
** Execute a rollback if a transaction is active and unlock the 
** database file. This is a no-op if the pager has already entered
** the error-state.
*/
static void pagerUnlockAndRollback(Pager *p){
  if( p->errCode ) return;
  assert( p->state>=PAGER_RESERVED || p->journalOpen==0 );
  if( 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 );
}


/*
** 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 %d\n", pPager, pPg->pgno));
    PAGER_INCR(sqlite3_pager_pgfree_count);
    pNext = pPg->pNextAll;
    sqliteFree(pPg);
  }
  pPager->pStmt = 0;
  pPager->pFirst = 0;
  pPager->pFirstSynced = 0;
  pPager->pLast = 0;
  pPager->pAll = 0;
  pPager->nHash = 0;
  sqliteFree(pPager->aHash);
  pPager->nPage = 0;
  pPager->aHash = 0;
  pPager->nRef = 0;
}

/*
** 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 ){;
      sqlite3OsSeek(pPager->jfd, 0);
      pPager->journalOff = 0;
      pPager->journalStarted = 0;
    }else{
      sqlite3OsClose(&pPager->jfd);
      pPager->journalOpen = 0;
      if( rc==SQLITE_OK ){
        rc = sqlite3OsDelete(pPager->zJournal);
      }
    }
    sqliteFree( pPager->aInJournal );
    pPager->aInJournal = 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->aInJournal==0 );
    assert( pPager->dirtyCache==0 || pPager->useJournal==0 );
  }

  if( !pPager->exclusiveMode ){
    rc2 = sqlite3OsUnlock(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;
  pPager->pFirstSynced = pPager->pFirst;
  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, OsFile *jfd, 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, &pgno);
  if( rc!=SQLITE_OK ) return rc;
  rc = sqlite3OsRead(jfd, aData, pPager->pageSize);
  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, &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.
  **
  ** 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 in the main rollback
  ** journal.  Otherwise, if a full ROLLBACK occurs after the statement