pager.c

来自「SQLite 2.8.6 源代码,用来在Linux/Unix/Windows上编」· C语言 代码 · 共 2,008 行 · 第 1/5 页

  int rc;

  /* Make sure we have not hit any critical errors.
  */ 
  assert( pPager!=0 );
  assert( pgno!=0 );
  *ppPage = 0;
  if( pPager->errMask & ~(PAGER_ERR_FULL) ){
    return pager_errcode(pPager);
  }

  /* If this is the first page accessed, then get a read lock
  ** on the database file.
  */
  if( pPager->nRef==0 ){
    rc = sqliteOsReadLock(&pPager->fd);
    if( rc!=SQLITE_OK ){
      return rc;
    }
    pPager->state = SQLITE_READLOCK;

    /* If a journal file exists, try to play it back.
    */
    if( pPager->useJournal && sqliteOsFileExists(pPager->zJournal) ){
       int rc;

       /* Get a write lock on the database
       */
       rc = sqliteOsWriteLock(&pPager->fd);
       if( rc!=SQLITE_OK ){
         if( sqliteOsUnlock(&pPager->fd)!=SQLITE_OK ){
           /* This should never happen! */
           rc = SQLITE_INTERNAL;
         }
         return rc;
       }
       pPager->state = SQLITE_WRITELOCK;

       /* Open the journal for reading only.  Return SQLITE_BUSY if
       ** we are unable to open the journal file. 
       **
       ** The journal file does not need to be locked itself.  The
       ** journal file is never open unless the main database file holds
       ** a write lock, so there is never any chance of two or more
       ** processes opening the journal at the same time.
       */
       rc = sqliteOsOpenReadOnly(pPager->zJournal, &pPager->jfd);
       if( rc!=SQLITE_OK ){
         rc = sqliteOsUnlock(&pPager->fd);
         assert( rc==SQLITE_OK );
         return SQLITE_BUSY;
       }
       pPager->journalOpen = 1;
       pPager->journalStarted = 0;

       /* Playback and delete the journal.  Drop the database write
       ** lock and reacquire the read lock.
       */
       rc = pager_playback(pPager, 0);
       if( rc!=SQLITE_OK ){
         return rc;
       }
    }
    pPg = 0;
  }else{
    /* Search for page in cache */
    pPg = pager_lookup(pPager, pgno);
  }
  if( pPg==0 ){
    /* The requested page is not in the page cache. */
    int h;
    pPager->nMiss++;
    if( pPager->nPage<pPager->mxPage || pPager->pFirst==0 ){
      /* Create a new page */
      pPg = sqliteMallocRaw( sizeof(*pPg) + SQLITE_PAGE_SIZE 
                              + sizeof(u32) + pPager->nExtra );
      if( pPg==0 ){
        pager_unwritelock(pPager);
        pPager->errMask |= PAGER_ERR_MEM;
        return SQLITE_NOMEM;
      }
      memset(pPg, 0, sizeof(*pPg));
      pPg->pPager = pPager;
      pPg->pNextAll = pPager->pAll;
      if( pPager->pAll ){
        pPager->pAll->pPrevAll = pPg;
      }
      pPg->pPrevAll = 0;
      pPager->pAll = pPg;
      pPager->nPage++;
    }else{
      /* Find a page to recycle.  Try to locate a page that does not
      ** require us to do an fsync() on the journal.
      */
      pPg = pPager->pFirstSynced;

      /* If we could not find a page that does not require an fsync()
      ** on the journal file then fsync the journal file.  This is a
      ** very slow operation, so we work hard to avoid it.  But sometimes
      ** it can't be helped.
      */
      if( pPg==0 ){
        int rc = syncAllPages(pPager);
        if( rc!=0 ){
          sqlitepager_rollback(pPager);
          return SQLITE_IOERR;
        }
        pPg = pPager->pFirst;
      }
      assert( pPg->nRef==0 );

      /* Write the page to the database file if it is dirty.
      */
      if( pPg->dirty ){
        assert( pPg->needSync==0 );
        pPg->pDirty = 0;
        rc = pager_write_pagelist( pPg );
        if( rc!=SQLITE_OK ){
          sqlitepager_rollback(pPager);
          return SQLITE_IOERR;
        }
      }
      assert( pPg->dirty==0 );

      /* If the page we are recycling is marked as alwaysRollback, then
      ** set the global alwaysRollback flag, thus disabling the
      ** sqlite_dont_rollback() optimization for the rest of this transaction.
      ** It is necessary to do this because the page marked alwaysRollback
      ** might be reloaded at a later time but at that point we won't remember
      ** that is was marked alwaysRollback.  This means that all pages must
      ** be marked as alwaysRollback from here on out.
      */
      if( pPg->alwaysRollback ){
        pPager->alwaysRollback = 1;
      }

      /* Unlink the old page from the free list and the hash table
      */
      if( pPg==pPager->pFirstSynced ){
        PgHdr *p = pPg->pNextFree;
        while( p && p->needSync ){ p = p->pNextFree; }
        pPager->pFirstSynced = p;
      }
      if( pPg->pPrevFree ){
        pPg->pPrevFree->pNextFree = pPg->pNextFree;
      }else{
        assert( pPager->pFirst==pPg );
        pPager->pFirst = pPg->pNextFree;
      }
      if( pPg->pNextFree ){
        pPg->pNextFree->pPrevFree = pPg->pPrevFree;
      }else{
        assert( pPager->pLast==pPg );
        pPager->pLast = pPg->pPrevFree;
      }
      pPg->pNextFree = pPg->pPrevFree = 0;
      if( pPg->pNextHash ){
        pPg->pNextHash->pPrevHash = pPg->pPrevHash;
      }
      if( pPg->pPrevHash ){
        pPg->pPrevHash->pNextHash = pPg->pNextHash;
      }else{
        h = pager_hash(pPg->pgno);
        assert( pPager->aHash[h]==pPg );
        pPager->aHash[h] = pPg->pNextHash;
      }
      pPg->pNextHash = pPg->pPrevHash = 0;
      pPager->nOvfl++;
    }
    pPg->pgno = pgno;
    if( pPager->aInJournal && (int)pgno<=pPager->origDbSize ){
      sqliteCheckMemory(pPager->aInJournal, pgno/8);
      assert( pPager->journalOpen );
      pPg->inJournal = (pPager->aInJournal[pgno/8] & (1<<(pgno&7)))!=0;
      pPg->needSync = 0;
    }else{
      pPg->inJournal = 0;
      pPg->needSync = 0;
    }
    if( pPager->aInCkpt && (int)pgno<=pPager->ckptSize
             && (pPager->aInCkpt[pgno/8] & (1<<(pgno&7)))!=0 ){
      page_add_to_ckpt_list(pPg);
    }else{
      page_remove_from_ckpt_list(pPg);
    }
    pPg->dirty = 0;
    pPg->nRef = 1;
    REFINFO(pPg);
    pPager->nRef++;
    h = pager_hash(pgno);
    pPg->pNextHash = pPager->aHash[h];
    pPager->aHash[h] = pPg;
    if( pPg->pNextHash ){
      assert( pPg->pNextHash->pPrevHash==0 );
      pPg->pNextHash->pPrevHash = pPg;
    }
    if( pPager->nExtra>0 ){
      memset(PGHDR_TO_EXTRA(pPg), 0, pPager->nExtra);
    }
    if( pPager->dbSize<0 ) sqlitepager_pagecount(pPager);
    if( pPager->errMask!=0 ){
      sqlitepager_unref(PGHDR_TO_DATA(pPg));
      rc = pager_errcode(pPager);
      return rc;
    }
    if( pPager->dbSize<(int)pgno ){
      memset(PGHDR_TO_DATA(pPg), 0, SQLITE_PAGE_SIZE);
    }else{
      int rc;
      sqliteOsSeek(&pPager->fd, (pgno-1)*(off_t)SQLITE_PAGE_SIZE);
      rc = sqliteOsRead(&pPager->fd, PGHDR_TO_DATA(pPg), SQLITE_PAGE_SIZE);
      if( rc!=SQLITE_OK ){
        off_t fileSize;
        if( sqliteOsFileSize(&pPager->fd,&fileSize)!=SQLITE_OK
               || fileSize>=pgno*SQLITE_PAGE_SIZE ){
          sqlitepager_unref(PGHDR_TO_DATA(pPg));
          return rc;
        }else{
          memset(PGHDR_TO_DATA(pPg), 0, SQLITE_PAGE_SIZE);
        }
      }
    }
  }else{
    /* The requested page is in the page cache. */
    pPager->nHit++;
    page_ref(pPg);
  }
  *ppPage = PGHDR_TO_DATA(pPg);
  return SQLITE_OK;
}

/*
** Acquire a page if it is already in the in-memory cache.  Do
** not read the page from disk.  Return a pointer to the page,
** or 0 if the page is not in cache.
**
** See also sqlitepager_get().  The difference between this routine
** and sqlitepager_get() is that _get() will go to the disk and read
** in the page if the page is not already in cache.  This routine
** returns NULL if the page is not in cache or if a disk I/O error 
** has ever happened.
*/
void *sqlitepager_lookup(Pager *pPager, Pgno pgno){
  PgHdr *pPg;

  assert( pPager!=0 );
  assert( pgno!=0 );
  if( pPager->errMask & ~(PAGER_ERR_FULL) ){
    return 0;
  }
  /* if( pPager->nRef==0 ){
  **  return 0;
  ** }
  */
  pPg = pager_lookup(pPager, pgno);
  if( pPg==0 ) return 0;
  page_ref(pPg);
  return PGHDR_TO_DATA(pPg);
}

/*
** Release a page.
**
** If the number of references to the page drop to zero, then the
** page is added to the LRU list.  When all references to all pages
** are released, a rollback occurs and the lock on the database is
** removed.
*/
int sqlitepager_unref(void *pData){
  PgHdr *pPg;

  /* Decrement the reference count for this page
  */
  pPg = DATA_TO_PGHDR(pData);
  assert( pPg->nRef>0 );
  pPg->nRef--;
  REFINFO(pPg);

  /* When the number of references to a page reach 0, call the
  ** destructor and add the page to the freelist.
  */
  if( pPg->nRef==0 ){
    Pager *pPager;
    pPager = pPg->pPager;
    pPg->pNextFree = 0;
    pPg->pPrevFree = pPager->pLast;
    pPager->pLast = pPg;
    if( pPg->pPrevFree ){
      pPg->pPrevFree->pNextFree = pPg;
    }else{
      pPager->pFirst = pPg;
    }
    if( pPg->needSync==0 && pPager->pFirstSynced==0 ){
      pPager->pFirstSynced = pPg;
    }
    if( pPager->xDestructor ){
      pPager->xDestructor(pData);
    }
  
    /* When all pages reach the freelist, drop the read lock from
    ** the database file.
    */
    pPager->nRef--;
    assert( pPager->nRef>=0 );
    if( pPager->nRef==0 ){
      pager_reset(pPager);
    }
  }
  return SQLITE_OK;
}

/*
** Create a journal file for pPager.  There should already be a write
** lock on the database file when this routine is called.
**
** Return SQLITE_OK if everything.  Return an error code and release the
** write lock if anything goes wrong.
*/
static int pager_open_journal(Pager *pPager){
  int rc;
  assert( pPager->state==SQLITE_WRITELOCK );
  assert( pPager->journalOpen==0 );
  assert( pPager->useJournal );
  sqlitepager_pagecount(pPager);
  pPager->aInJournal = sqliteMalloc( pPager->dbSize/8 + 1 );
  if( pPager->aInJournal==0 ){
    sqliteOsReadLock(&pPager->fd);
    pPager->state = SQLITE_READLOCK;
    return SQLITE_NOMEM;
  }
  rc = sqliteOsOpenExclusive(pPager->zJournal, &pPager->jfd,pPager->tempFile);
  if( rc!=SQLITE_OK ){
    sqliteFree(pPager->aInJournal);
    pPager->aInJournal = 0;
    sqliteOsReadLock(&pPager->fd);
    pPager->state = SQLITE_READLOCK;
    return SQLITE_CANTOPEN;
  }
  sqliteOsOpenDirectory(pPager->zDirectory, &pPager->jfd);
  pPager->journalOpen = 1;
  pPager->journalStarted = 0;
  pPager->needSync = 0;
  pPager->alwaysRollback = 0;
  pPager->nRec = 0;
  if( pPager->errMask!=0 ){
    rc = pager_errcode(pPager);
    return rc;
  }
  pPager->origDbSize = pPager->dbSize;
  if( journal_format==JOURNAL_FORMAT_3 ){
    rc = sqliteOsWrite(&pPager->jfd, aJournalMagic3, sizeof(aJournalMagic3));
    if( rc==SQLITE_OK ){
      rc = write32bits(&pPager->jfd, pPager->noSync ? 0xffffffff : 0);
    }
    if( rc==SQLITE_OK ){
      pPager->cksumInit = (u32)sqliteRandomInteger();
      rc = write32bits(&pPager->jfd, pPager->cksumInit);
    }
  }else if( journal_format==JOURNAL_FORMAT_2 ){
    rc = sqliteOsWrite(&pPager->jfd, aJournalMagic2, sizeof(aJournalMagic2));
  }else{
    assert( journal_format==JOURNAL_FORMAT_1 );
    rc = sqliteOsWrite(&pPager->jfd, aJournalMagic1, sizeof(aJournalMagic1));
  }
  if( rc==SQLITE_OK ){
    rc = write32bits(&pPager->jfd, pPager->dbSize);
  }
  if( pPager->ckptAutoopen && rc==SQLITE_OK ){
    rc = sqlitepager_ckpt_begin(pPager);
  }
  if( rc!=SQLITE_OK ){
    rc = pager_unwritelock(pPager);
    if( rc==SQLITE_OK ){
      rc = SQLITE_FULL;
    }
  }
  return rc;  
}

/*
** Acquire a write-lock on the database.  The lock is removed when
** the any of the following happen:
**
**   *  sqlitepager_commit() is called.
**   *  sqlitepager_rollback() is called.
**   *  sqlitepager_close() is called.
**   *  sqlitepager_unref() is called to on every outstanding page.
**
** The parameter to this routine is a pointer to any open page of the
** database file.  Nothing changes about the page - it is used merely
** to acquire a pointer to the Pager structure and as proof that there
** is already a read-lock on the database.
**
** A journal file is opened if this is not a temporary file.  For
** temporary files, the opening of the journal file is deferred until
** there is an actual need to write to the journal.
**
** If the database is already write-locked, this routine is a no-op.
*/
int sqlitepager_begin(void *pData){
  PgHdr *pPg = DATA_TO_PGHDR(pData);
  Pager *pPager = pPg->pPager;