SQLite

/*
** This function is used to read or overwrite payload information
** for the entry that the pCur cursor is pointing to. The eOp
** argument is interpreted as follows:
**
**   0: The operation is a read. Populate the overflow cache.
**   1: The operation is a write. Populate the overflow cache.
**   2: The operation is a read. Do not populate the overflow cache.
**
** A total of "amt" bytes are read or written beginning at "offset".
** Data is read to or from the buffer pBuf.
**
** The content being read or written might appear on the main page
** or be scattered out on multiple overflow pages.
**
** If the current cursor entry uses one or more overflow pages and the
** eOp argument is not 2, this function may allocate space for and lazily 
** populates the overflow page-list cache array (BtCursor.aOverflow). 
** Subsequent calls use this cache to make seeking to the supplied offset 
** more efficient.
**
** Once an overflow page-list cache has been allocated, it may be
** invalidated if some other cursor writes to the same table, or if
** the cursor is moved to a different row. Additionally, in auto-vacuum
** mode, the following events may invalidate an overflow page-list cache.
**
**   * An incremental vacuum,
**   * A commit in auto_vacuum="full" mode,
**   * Creating a table (may require moving an overflow page).

  MemPage *pPage = pCur->apPage[pCur->iPage]; /* Btree page of current entry */
  BtShared *pBt = pCur->pBt;                  /* Btree this cursor belongs to */
#ifdef SQLITE_DIRECT_OVERFLOW_READ
  unsigned char * const pBufStart = pBuf;     /* Start of original out buffer */
#endif

  assert( pPage );

  assert( pCur->eState==CURSOR_VALID );
  assert( pCur->aiIdx[pCur->iPage]<pPage->nCell );
  assert( cursorHoldsMutex(pCur) );
  assert( eOp!=2 || offset==0 );    /* Always start from beginning for eOp==2 */

  getCellInfo(pCur);
  aPayload = pCur->info.pPayload;
  assert( offset+amt <= pCur->info.nPayload );

  assert( aPayload > pPage->aData );
  if( (uptr)(aPayload - pPage->aData) > (pBt->usableSize - pCur->info.nLocal) ){

  /* Check if data must be read/written to/from the btree page itself. */
  if( offset<pCur->info.nLocal ){
    int a = amt;
    if( a+offset>pCur->info.nLocal ){
      a = pCur->info.nLocal - offset;
    }
    rc = copyPayload(&aPayload[offset], pBuf, a, (eOp & 0x01), pPage->pDbPage);
    offset = 0;
    pBuf += a;
    amt -= a;
  }else{
    offset -= pCur->info.nLocal;
  }


  if( rc==SQLITE_OK && amt>0 ){
    const u32 ovflSize = pBt->usableSize - 4;  /* Bytes content per ovfl page */
    Pgno nextPage;

    nextPage = get4byte(&aPayload[pCur->info.nLocal]);

    /* If the BtCursor.aOverflow[] has not been allocated, allocate it now.
    ** Except, do not allocate aOverflow[] for eOp==2.
    **
    ** The aOverflow[] array is sized at one entry for each overflow page
    ** in the overflow chain. The page number of the first overflow page is
    ** stored in aOverflow[0], etc. A value of 0 in the aOverflow[] array
    ** means "not yet known" (the cache is lazily populated).
    */
    if( eOp!=2 && (pCur->curFlags & BTCF_ValidOvfl)==0 ){
      int nOvfl = (pCur->info.nPayload-pCur->info.nLocal+ovflSize-1)/ovflSize;
      if( nOvfl>pCur->nOvflAlloc ){
        Pgno *aNew = (Pgno*)sqlite3Realloc(
            pCur->aOverflow, nOvfl*2*sizeof(Pgno)
        );
        if( aNew==0 ){
          return SQLITE_NOMEM_BKPT;
        }else{
          pCur->nOvflAlloc = nOvfl*2;
          pCur->aOverflow = aNew;
        }
      }
      memset(pCur->aOverflow, 0, nOvfl*sizeof(Pgno));
      pCur->curFlags |= BTCF_ValidOvfl;
    }

    /* If the overflow page-list cache has been allocated and the
    ** entry for the first required overflow page is valid, skip
    ** directly to it.
    */
    if( (pCur->curFlags & BTCF_ValidOvfl)!=0
     && pCur->aOverflow[offset/ovflSize]
    ){
      iIdx = (offset/ovflSize);
      nextPage = pCur->aOverflow[iIdx];
      offset = (offset%ovflSize);
    }

    assert( rc==SQLITE_OK && amt>0 );
    while( nextPage ){
      /* If required, populate the overflow page-list cache. */
      if( (pCur->curFlags & BTCF_ValidOvfl)!=0 ){
        assert( pCur->aOverflow[iIdx]==0
                || pCur->aOverflow[iIdx]==nextPage
                || CORRUPT_DB );
        pCur->aOverflow[iIdx] = nextPage;
      }

      if( offset>=ovflSize ){
        /* The only reason to read this page is to obtain the page
        ** number for the next page in the overflow chain. The page
        ** data is not required. So first try to lookup the overflow
        ** page-list cache, if any, then fall back to the getOverflowPage()
        ** function.
        **
        ** Note that the aOverflow[] array must be allocated because eOp!=2
        ** here.  If eOp==2, then offset==0 and this branch is never taken.
        */
        assert( eOp!=2 );
        assert( pCur->curFlags & BTCF_ValidOvfl );
        assert( pCur->pBtree->db==pBt->db );
        if( pCur->aOverflow[iIdx+1] ){
          nextPage = pCur->aOverflow[iIdx+1];
        }else{
          rc = getOverflowPage(pBt, nextPage, 0, &nextPage);
        }

        **   5) the page is not in the WAL file
        **   6) at least 4 bytes have already been read into the output buffer 
        **
        ** then data can be read directly from the database file into the
        ** output buffer, bypassing the page-cache altogether. This speeds
        ** up loading large records that span many overflow pages.
        */
        if( (eOp&0x01)==0                                      /* (1) */
         && offset==0                                          /* (2) */
         && pBt->inTransaction==TRANS_READ                     /* (3) */
         && (fd = sqlite3PagerFile(pBt->pPager))->pMethods     /* (4) */
         && 0==sqlite3PagerUseWal(pBt->pPager, nextPage)       /* (5) */
         && &pBuf[-4]>=pBufStart                               /* (6) */
        ){
          u8 aSave[4];
          u8 *aWrite = &pBuf[-4];
          assert( aWrite>=pBufStart );                         /* due to (6) */
          memcpy(aSave, aWrite, 4);
          rc = sqlite3OsRead(fd, aWrite, a+4, (i64)pBt->pageSize*(nextPage-1));
          nextPage = get4byte(aWrite);
          memcpy(aWrite, aSave, 4);
        }else
#endif

        {
          DbPage *pDbPage;
          rc = sqlite3PagerGet(pBt->pPager, nextPage, &pDbPage,
              ((eOp&0x01)==0 ? PAGER_GET_READONLY : 0)
          );
          if( rc==SQLITE_OK ){
            aPayload = sqlite3PagerGetData(pDbPage);
            nextPage = get4byte(aPayload);
            rc = copyPayload(&aPayload[offset+4], pBuf, a, (eOp&0x01), pDbPage);
            sqlite3PagerUnref(pDbPage);
            offset = 0;
          }
        }
        amt -= a;
        pBuf += a;
      }

          }
          pCellKey = sqlite3Malloc( nCell+18 );
          if( pCellKey==0 ){
            rc = SQLITE_NOMEM_BKPT;
            goto moveto_finish;
          }
          pCur->aiIdx[pCur->iPage] = (u16)idx;
          rc = accessPayload(pCur, 0, nCell, (unsigned char*)pCellKey, 2);

          if( rc ){
            sqlite3_free(pCellKey);
            goto moveto_finish;
          }
          c = xRecordCompare(nCell, pCellKey, pIdxKey);
          sqlite3_free(pCellKey);
        }