btree.c

来自「sqlite-3.4.1,嵌入式数据库.是一个功能强大的开源数据库,给学习和研发」· C语言 代码 · 共 2,088 行 · 第 1/5 页

  for(i=0; i<nCell; i++){
    u8 *pCell = findCell(pPage, i);

    rc = ptrmapPutOvflPtr(pPage, pCell);
    if( rc!=SQLITE_OK ){
      goto set_child_ptrmaps_out;
    }

    if( !pPage->leaf ){
      Pgno childPgno = get4byte(pCell);
      rc = ptrmapPut(pBt, childPgno, PTRMAP_BTREE, pgno);
      if( rc!=SQLITE_OK ) goto set_child_ptrmaps_out;
    }
  }

  if( !pPage->leaf ){
    Pgno childPgno = get4byte(&pPage->aData[pPage->hdrOffset+8]);
    rc = ptrmapPut(pBt, childPgno, PTRMAP_BTREE, pgno);
  }

set_child_ptrmaps_out:
  pPage->isInit = isInitOrig;
  return rc;
}

/*
** Somewhere on pPage, which is guarenteed to be a btree page, not an overflow
** page, is a pointer to page iFrom. Modify this pointer so that it points to
** iTo. Parameter eType describes the type of pointer to be modified, as 
** follows:
**
** PTRMAP_BTREE:     pPage is a btree-page. The pointer points at a child 
**                   page of pPage.
**
** PTRMAP_OVERFLOW1: pPage is a btree-page. The pointer points at an overflow
**                   page pointed to by one of the cells on pPage.
**
** PTRMAP_OVERFLOW2: pPage is an overflow-page. The pointer points at the next
**                   overflow page in the list.
*/
static int modifyPagePointer(MemPage *pPage, Pgno iFrom, Pgno iTo, u8 eType){
  if( eType==PTRMAP_OVERFLOW2 ){
    /* The pointer is always the first 4 bytes of the page in this case.  */
    if( get4byte(pPage->aData)!=iFrom ){
      return SQLITE_CORRUPT_BKPT;
    }
    put4byte(pPage->aData, iTo);
  }else{
    int isInitOrig = pPage->isInit;
    int i;
    int nCell;

    sqlite3BtreeInitPage(pPage, 0);
    nCell = pPage->nCell;

    for(i=0; i<nCell; i++){
      u8 *pCell = findCell(pPage, i);
      if( eType==PTRMAP_OVERFLOW1 ){
        CellInfo info;
        sqlite3BtreeParseCellPtr(pPage, pCell, &info);
        if( info.iOverflow ){
          if( iFrom==get4byte(&pCell[info.iOverflow]) ){
            put4byte(&pCell[info.iOverflow], iTo);
            break;
          }
        }
      }else{
        if( get4byte(pCell)==iFrom ){
          put4byte(pCell, iTo);
          break;
        }
      }
    }
  
    if( i==nCell ){
      if( eType!=PTRMAP_BTREE || 
          get4byte(&pPage->aData[pPage->hdrOffset+8])!=iFrom ){
        return SQLITE_CORRUPT_BKPT;
      }
      put4byte(&pPage->aData[pPage->hdrOffset+8], iTo);
    }

    pPage->isInit = isInitOrig;
  }
  return SQLITE_OK;
}


/*
** Move the open database page pDbPage to location iFreePage in the 
** database. The pDbPage reference remains valid.
*/
static int relocatePage(
  BtShared *pBt,           /* Btree */
  MemPage *pDbPage,        /* Open page to move */
  u8 eType,                /* Pointer map 'type' entry for pDbPage */
  Pgno iPtrPage,           /* Pointer map 'page-no' entry for pDbPage */
  Pgno iFreePage           /* The location to move pDbPage to */
){
  MemPage *pPtrPage;   /* The page that contains a pointer to pDbPage */
  Pgno iDbPage = pDbPage->pgno;
  Pager *pPager = pBt->pPager;
  int rc;

  assert( eType==PTRMAP_OVERFLOW2 || eType==PTRMAP_OVERFLOW1 || 
      eType==PTRMAP_BTREE || eType==PTRMAP_ROOTPAGE );

  /* Move page iDbPage from it's current location to page number iFreePage */
  TRACE(("AUTOVACUUM: Moving %d to free page %d (ptr page %d type %d)\n", 
      iDbPage, iFreePage, iPtrPage, eType));
  rc = sqlite3PagerMovepage(pPager, pDbPage->pDbPage, iFreePage);
  if( rc!=SQLITE_OK ){
    return rc;
  }
  pDbPage->pgno = iFreePage;

  /* If pDbPage was a btree-page, then it may have child pages and/or cells
  ** that point to overflow pages. The pointer map entries for all these
  ** pages need to be changed.
  **
  ** If pDbPage is an overflow page, then the first 4 bytes may store a
  ** pointer to a subsequent overflow page. If this is the case, then
  ** the pointer map needs to be updated for the subsequent overflow page.
  */
  if( eType==PTRMAP_BTREE || eType==PTRMAP_ROOTPAGE ){
    rc = setChildPtrmaps(pDbPage);
    if( rc!=SQLITE_OK ){
      return rc;
    }
  }else{
    Pgno nextOvfl = get4byte(pDbPage->aData);
    if( nextOvfl!=0 ){
      rc = ptrmapPut(pBt, nextOvfl, PTRMAP_OVERFLOW2, iFreePage);
      if( rc!=SQLITE_OK ){
        return rc;
      }
    }
  }

  /* Fix the database pointer on page iPtrPage that pointed at iDbPage so
  ** that it points at iFreePage. Also fix the pointer map entry for
  ** iPtrPage.
  */
  if( eType!=PTRMAP_ROOTPAGE ){
    rc = sqlite3BtreeGetPage(pBt, iPtrPage, &pPtrPage, 0);
    if( rc!=SQLITE_OK ){
      return rc;
    }
    rc = sqlite3PagerWrite(pPtrPage->pDbPage);
    if( rc!=SQLITE_OK ){
      releasePage(pPtrPage);
      return rc;
    }
    rc = modifyPagePointer(pPtrPage, iDbPage, iFreePage, eType);
    releasePage(pPtrPage);
    if( rc==SQLITE_OK ){
      rc = ptrmapPut(pBt, iFreePage, eType, iPtrPage);
    }
  }
  return rc;
}

/* Forward declaration required by incrVacuumStep(). */
static int allocateBtreePage(BtShared *, MemPage **, Pgno *, Pgno, u8);

/*
** Perform a single step of an incremental-vacuum. If successful,
** return SQLITE_OK. If there is no work to do (and therefore no
** point in calling this function again), return SQLITE_DONE.
**
** More specificly, this function attempts to re-organize the 
** database so that the last page of the file currently in use
** is no longer in use.
**
** If the nFin parameter is non-zero, the implementation assumes
** that the caller will keep calling incrVacuumStep() until
** it returns SQLITE_DONE or an error, and that nFin is the
** number of pages the database file will contain after this 
** process is complete.
*/
static int incrVacuumStep(BtShared *pBt, Pgno nFin){
  Pgno iLastPg;             /* Last page in the database */
  Pgno nFreeList;           /* Number of pages still on the free-list */

  iLastPg = pBt->nTrunc;
  if( iLastPg==0 ){
    iLastPg = sqlite3PagerPagecount(pBt->pPager);
  }

  if( !PTRMAP_ISPAGE(pBt, iLastPg) && iLastPg!=PENDING_BYTE_PAGE(pBt) ){
    int rc;
    u8 eType;
    Pgno iPtrPage;

    nFreeList = get4byte(&pBt->pPage1->aData[36]);
    if( nFreeList==0 || nFin==iLastPg ){
      return SQLITE_DONE;
    }

    rc = ptrmapGet(pBt, iLastPg, &eType, &iPtrPage);
    if( rc!=SQLITE_OK ){
      return rc;
    }
    if( eType==PTRMAP_ROOTPAGE ){
      return SQLITE_CORRUPT_BKPT;
    }

    if( eType==PTRMAP_FREEPAGE ){
      if( nFin==0 ){
        /* Remove the page from the files free-list. This is not required
        ** if nFin is non-zero. In that case, the free-list will be
        ** truncated to zero after this function returns, so it doesn't 
        ** matter if it still contains some garbage entries.
        */
        Pgno iFreePg;
        MemPage *pFreePg;
        rc = allocateBtreePage(pBt, &pFreePg, &iFreePg, iLastPg, 1);
        if( rc!=SQLITE_OK ){
          return rc;
        }
        assert( iFreePg==iLastPg );
        releasePage(pFreePg);
      }
    } else {
      Pgno iFreePg;             /* Index of free page to move pLastPg to */
      MemPage *pLastPg;

      rc = sqlite3BtreeGetPage(pBt, iLastPg, &pLastPg, 0);
      if( rc!=SQLITE_OK ){
        return rc;
      }

      /* If nFin is zero, this loop runs exactly once and page pLastPg
      ** is swapped with the first free page pulled off the free list.
      **
      ** On the other hand, if nFin is greater than zero, then keep
      ** looping until a free-page located within the first nFin pages
      ** of the file is found.
      */
      do {
        MemPage *pFreePg;
        rc = allocateBtreePage(pBt, &pFreePg, &iFreePg, 0, 0);
        if( rc!=SQLITE_OK ){
          releasePage(pLastPg);
          return rc;
        }
        releasePage(pFreePg);
      }while( nFin!=0 && iFreePg>nFin );
      assert( iFreePg<iLastPg );
      
      rc = sqlite3PagerWrite(pLastPg->pDbPage);
      if( rc!=SQLITE_OK ){
        return rc;
      } 
      rc = relocatePage(pBt, pLastPg, eType, iPtrPage, iFreePg);
      releasePage(pLastPg);
      if( rc!=SQLITE_OK ){
        return rc;
      } 
    }
  }

  pBt->nTrunc = iLastPg - 1;
  while( pBt->nTrunc==PENDING_BYTE_PAGE(pBt)||PTRMAP_ISPAGE(pBt, pBt->nTrunc) ){
    pBt->nTrunc--;
  }
  return SQLITE_OK;
}

/*
** A write-transaction must be opened before calling this function.
** It performs a single unit of work towards an incremental vacuum.
**
** If the incremental vacuum is finished after this function has run,
** SQLITE_DONE is returned. If it is not finished, but no error occured,
** SQLITE_OK is returned. Otherwise an SQLite error code. 
*/
int sqlite3BtreeIncrVacuum(Btree *p){
  BtShared *pBt = p->pBt;
  assert( pBt->inTransaction==TRANS_WRITE && p->inTrans==TRANS_WRITE );
  if( !pBt->autoVacuum ){
    return SQLITE_DONE;
  }
  invalidateAllOverflowCache(pBt);
  return incrVacuumStep(pBt, 0);
}

/*
** This routine is called prior to sqlite3PagerCommit when a transaction
** is commited for an auto-vacuum database.
**
** If SQLITE_OK is returned, then *pnTrunc is set to the number of pages
** the database file should be truncated to during the commit process. 
** i.e. the database has been reorganized so that only the first *pnTrunc
** pages are in use.
*/
static int autoVacuumCommit(BtShared *pBt, Pgno *pnTrunc){
  int rc = SQLITE_OK;
  Pager *pPager = pBt->pPager;
#ifndef NDEBUG
  int nRef = sqlite3PagerRefcount(pPager);
#endif

  invalidateAllOverflowCache(pBt);
  assert(pBt->autoVacuum);
  if( !pBt->incrVacuum ){
    Pgno nFin = 0;

    if( pBt->nTrunc==0 ){
      Pgno nFree;
      Pgno nPtrmap;
      const int pgsz = pBt->pageSize;
      Pgno nOrig = sqlite3PagerPagecount(pBt->pPager);

      if( PTRMAP_ISPAGE(pBt, nOrig) ){
        return SQLITE_CORRUPT_BKPT;
      }
      if( nOrig==PENDING_BYTE_PAGE(pBt) ){
        nOrig--;
      }
      nFree = get4byte(&pBt->pPage1->aData[36]);
      nPtrmap = (nFree-nOrig+PTRMAP_PAGENO(pBt, nOrig)+pgsz/5)/(pgsz/5);
      nFin = nOrig - nFree - nPtrmap;
      if( nOrig>PENDING_BYTE_PAGE(pBt) && nFin<=PENDING_BYTE_PAGE(pBt) ){
        nFin--;
      }
      while( PTRMAP_ISPAGE(pBt, nFin) || nFin==PENDING_BYTE_PAGE(pBt) ){
        nFin--;
      }
    }

    while( rc==SQLITE_OK ){
      rc = incrVacuumStep(pBt, nFin);
    }
    if( rc==SQLITE_DONE ){
      assert(nFin==0 || pBt->nTrunc==0 || nFin<=pBt->nTrunc);
      rc = SQLITE_OK;
      if( pBt->nTrunc ){
        sqlite3PagerWrite(pBt->pPage1->pDbPage);
        put4byte(&pBt->pPage1->aData[32], 0);
        put4byte(&pBt->pPage1->aData[36], 0);
        pBt->nTrunc = nFin;
      }
    }
    if( rc!=SQLITE_OK ){
      sqlite3PagerRollback(pPager);
    }
  }

  if( rc==SQLITE_OK ){
    *pnTrunc = pBt->nTrunc;
    pBt->nTrunc = 0;
  }
  assert( nRef==sqlite3PagerRefcount(pPager) );
  return rc;
}

#endif

/*
** This routine does the first phase of a two-phase commit.  This routine
** causes a rollback journal to be created (if it does not already exist)
** and populated with enough information so that if a power loss occurs
** the database can be restored to its original state by playing back
** the journal.  Then the contents of the journal are flushed out to
** the disk.  After the journal is safely on oxide, the changes to the
** database are written into the database file and flushed to oxide.
** At the end of this call, the rollback journal still exists on the
** disk and we are still holding all locks, so the transaction has not
** committed.  See sqlite3BtreeCommit() for the second phase of the
** commit process.
**
** This call is a no-op if no write-transaction is currently active on pBt.
**
** Otherwise, sync the database file for the btree pBt. zMaster points to
** the name of a master journal file that should be written into the
** individual journal file, or is NULL, indicating no master journal file 
** (single database transaction).
**
** When this is called, the master journal should already have been
** created, populated with this journal pointer and synced to disk.
**
** Once this is routine has returned, the only thing required to commit
** the write-transaction for this database file is to delete the journal.
*/
int sqlite3BtreeCommitPhaseOne(Btree *p, const char *zMaster){
  int rc = SQLITE_OK;
  if( p->inTrans==TRANS_WRITE ){
    BtShared *pBt = p->pBt;
    Pgno nTrunc = 0;
#ifndef SQLITE_OMIT_AUTOVACUUM
    if( pBt->autoVacuum ){
      rc = autoVacuumCommit(pBt, &nTrunc); 
      if( rc!=SQLITE_OK ){
        return rc;
      }
    }
#endif
    rc = sqlite3PagerCommitPhaseOne(pBt->pPager, zMaster, nTrunc);
  }
  return rc;
}

/*
** Commit the transaction currently in progress.
**
** This routine implements the second phase of a 2-phase commit.  The
** sqlite3BtreeSync() routine does the first phase and should be invoked
** prior to calling this routine.  The sqlite3BtreeSync() routine did
** all the work of writing information out to disk and flushing the
** contents so that they are written onto the disk platter.  All this
** routine has to do is delete or truncate the rollback journal
** (which causes the transaction to commit) and drop locks.
**
** This will re