SQLite

      assert( pLock->eLock==READ_LOCK || pLock->pBtree==p );
      pLock->eLock = READ_LOCK;
    }
  }
}

#endif /* SQLITE_OMIT_SHARED_CACHE */

/*
** The following structure stores the in-memory pointer map used for newly
** allocated pages in UNLOCKED transactions. Such pages are always allocated 
** in a contiguous block (from the end of the file) starting with page
** BtreePtrmap.iFirst.
** 
** The page number for the parent page iFirst is stored in aPtr[0]. For
** (iFirst+1), aPtr[1]. A zero value indicates that the page has not
** been allocated.
**
**
*/

typedef struct RollbackEntry RollbackEntry;
typedef struct PtrmapEntry PtrmapEntry;
struct PtrmapEntry {
  Pgno parent;
  u8 eType;
};
struct RollbackEntry {
  Pgno pgno;
  Pgno parent;
  u8 eType;
};

struct BtreePtrmap {
  Pgno iFirst;                    /* First new page number aPtr[0] */

  int nPtrAlloc;                  /* Allocated size of aPtr[] array */
  PtrmapEntry *aPtr;              /* Array of parent page numbers */

  int nSvpt;                      /* Used size of aSvpt[] array */
  int nSvptAlloc;                 /* Allocated size of aSvpt[] */
  int *aSvpt;                     /* First aRollback[] entry for savepoint i */

  int nRollback;                  /* Used size of aRollback[] array */
  int nRollbackAlloc;             /* Allocated size of aRollback[] array */
  RollbackEntry *aRollback;       /* Array of rollback entries */
};

static int btreePtrmapStore(
  BtreePtrmap *pMap, 
  Pgno pgno, 
  u8 eType, 
  Pgno parent
){
  if( pgno>=pMap->iFirst ){
    int iEntry = pgno - pMap->iFirst;

    /* Grow the aPtr[] array if required */
    if( iEntry>=pMap->nPtrAlloc ){
      int nNew = pMap->nPtrAlloc ? pMap->nPtrAlloc*2 : 16;
      PtrmapEntry *aNew = (PtrmapEntry*)sqlite3_realloc(
          pMap->aPtr, nNew*sizeof(PtrmapEntry)
      );
      if( aNew==0 ){
        return SQLITE_NOMEM;
      }else{
        int nByte = (nNew-pMap->nPtrAlloc)*sizeof(PtrmapEntry);
        memset(&aNew[pMap->nPtrAlloc], 0, nByte);
        pMap->aPtr = aNew;
        pMap->nPtrAlloc = nNew;
      }
    }

    /* Add an entry to the rollback log if required */
    if( pMap->nSvpt>0 && pMap->aPtr[iEntry].parent ){
      if( pMap->nRollback>=pMap->nRollbackAlloc ){
        int nNew = pMap->nRollback ? pMap->nRollback*2 : 16;
        RollbackEntry *aNew = (RollbackEntry*)sqlite3_realloc(
            pMap->aRollback, nNew*sizeof(RollbackEntry)
        );
        if( aNew==0 ){
          return SQLITE_NOMEM;
        }else{
          pMap->aRollback = aNew;
          pMap->nRollbackAlloc = nNew;
        }
      }

      pMap->aRollback[pMap->nRollback].pgno = pgno;
      pMap->aRollback[pMap->nRollback].parent = pMap->aPtr[iEntry].parent;
      pMap->aRollback[pMap->nRollback].eType = pMap->aPtr[iEntry].eType;
    }

    /* Update the aPtr[] array */
    pMap->aPtr[iEntry].parent = parent;
    pMap->aPtr[iEntry].eType = eType;
  }

  return SQLITE_OK;
}

/*
** Open savepoint iSavepoint, if it is not already open.
*/
static int btreePtrmapBegin(BtreePtrmap *pMap, int nSvpt){
  if( nSvpt<pMap->nSvpt ){
    int i;
    if( nSvpt>=pMap->nSvptAlloc ){
      int nNew = pMap->nSvptAlloc ? pMap->nSvptAlloc*2 : 16;
      int *aNew = sqlite3_realloc(pMap->aSvpt, sizeof(int) * nNew);
      if( aNew==0 ){
        return SQLITE_NOMEM;
      }else{
        pMap->aSvpt = aNew;
        pMap->nSvptAlloc = nNew;
      }
    }

    for(i=pMap->nSvpt; i<nSvpt; i++){
      pMap->aSvpt[i] = pMap->nRollback;
    }
    pMap->nSvpt = nSvpt;
  }

  return SQLITE_OK;
}

/*
** Rollback (if op==SAVEPOINT_ROLLBACK) or release (if op==SAVEPOINT_RELEASE)
** savepoint iSvpt.
*/
static void btreePtrmapEnd(BtreePtrmap *pMap, int op, int iSvpt){
  assert( op==SAVEPOINT_ROLLBACK || op==SAVEPOINT_RELEASE );
  assert( iSvpt>=0 || (iSvpt==-1 && op==SAVEPOINT_ROLLBACK) );
  if( iSvpt<0 ){
    pMap->nSvpt = 0;
    pMap->nRollback = 0;
    memset(pMap->aPtr, 0, sizeof(Pgno) * pMap->nPtrAlloc);
  }else if( iSvpt<pMap->nSvpt ){
    if( op==SAVEPOINT_ROLLBACK ){
      int ii;
      for(ii=pMap->nRollback-1; ii>=pMap->aSvpt[iSvpt]; ii--){
        RollbackEntry *p = &pMap->aRollback[ii];
        PtrmapEntry *pEntry = &pMap->aPtr[p->pgno - pMap->iFirst];
        pEntry->parent = p->parent;
        pEntry->eType = p->eType;
      }
    }
    pMap->nSvpt = iSvpt + (op==SAVEPOINT_ROLLBACK);
    pMap->nRollback = pMap->aSvpt[iSvpt];
  }
}


static void releasePage(MemPage *pPage);  /* Forward reference */

/*
***** This routine is used inside of assert() only ****
**
** Verify that the cursor holds the mutex on its BtShared

static void ptrmapPut(BtShared *pBt, Pgno key, u8 eType, Pgno parent, int *pRC){
  DbPage *pDbPage;  /* The pointer map page */
  u8 *pPtrmap;      /* The pointer map data */
  Pgno iPtrmap;     /* The pointer map page number */
  int offset;       /* Offset in pointer map page */
  int rc;           /* Return code from subfunctions */

  assert( sqlite3_mutex_held(pBt->mutex) );
  if( *pRC ) return;

  if( pBt->pMap ){
    *pRC = btreePtrmapStore(pBt->pMap, key, eType, parent);
  }else{
    /* The master-journal page number must never be used as a ptr map page */
    assert( 0==PTRMAP_ISPAGE(pBt, PENDING_BYTE_PAGE(pBt)) );

    assert( pBt->autoVacuum );
    if( key==0 ){
      *pRC = SQLITE_CORRUPT_BKPT;
      return;
    }
    iPtrmap = PTRMAP_PAGENO(pBt, key);
    rc = sqlite3PagerGet(pBt->pPager, iPtrmap, &pDbPage);
    if( rc!=SQLITE_OK ){
      *pRC = rc;
      return;
    }
    offset = PTRMAP_PTROFFSET(iPtrmap, key);
    if( offset<0 ){
      *pRC = SQLITE_CORRUPT_BKPT;

    }else{
      assert( offset <= (int)pBt->usableSize-5 );
      pPtrmap = (u8 *)sqlite3PagerGetData(pDbPage);

      if( eType!=pPtrmap[offset] || get4byte(&pPtrmap[offset+1])!=parent ){
        TRACE(("PTRMAP_UPDATE: %d->(%d,%d)\n", key, eType, parent));
        *pRC= rc = sqlite3PagerWrite(pDbPage);
        if( rc==SQLITE_OK ){
          pPtrmap[offset] = eType;
          put4byte(&pPtrmap[offset+1], parent);
        }
      }
    }

    sqlite3PagerUnref(pDbPage);
  }
}

/*
** Read an entry from the pointer map.
**
** This routine retrieves the pointer map entry for page 'key', writing
** the type and parent page number to *pEType and *pPgno respectively.

  return pBt->nPage;
}
u32 sqlite3BtreeLastPage(Btree *p){
  assert( sqlite3BtreeHoldsMutex(p) );
  assert( ((p->pBt->nPage)&0x8000000)==0 );
  return btreePagecount(p->pBt);
}

#ifdef SQLITE_ENABLE_UNLOCKED
/*
** This function is called before allocating or freeing a b-tree page. If
** the current transaction is UNLOCKED, it allocates the BtreePtrmap 
** structure and zeroes the nFree/iTrunk fields in the database header
** on page 1.
*/
static int allocatePtrmap(BtShared *pBt){
  int rc = SQLITE_OK;
  if( pBt->pMap==0 && sqlite3PagerIsUnlocked(pBt->pPager) ){
    /* If this is an unlocked transaction, set the header values
    ** identifying the size of the free-list and the page number
    ** of the first trunk page to zero. */
    BtreePtrmap *pMap = sqlite3_malloc(sizeof(BtreePtrmap));
    if( pMap==0 ){
      rc = SQLITE_NOMEM;
    }else{
      memset(&pBt->pPage1->aData[32], 0, sizeof(u32)*2);
      memset(pMap, 0, sizeof(BtreePtrmap));
      pMap->iFirst = btreePagecount(pBt) + 1;
      pBt->pMap = pMap;
    }
  }
  return rc;
}

/*
** Free a pointer-map allocated by allocatePtrmap.
*/
static void deletePtrmap(BtShared *pBt){
  BtreePtrmap *pMap = pBt->pMap;
  if( pMap ){
    sqlite3_free(pMap->aRollback);
    sqlite3_free(pMap->aPtr);
    sqlite3_free(pMap->aSvpt);
    sqlite3_free(pMap);
    pBt->pMap = 0;
  }
}
#else
#define allocatePtrmap(x) SQLITE_OK
#define deletePtrmap(x) 
#endif

/*
** Get a page from the pager and initialize it.
**
** If pCur!=0 then the page is being fetched as part of a moveToChild()
** call.  Do additional sanity checking on the page in this case.
** And if the fetch fails, this routine must decrement pCur->iPage.

        int exFlag = (p->db->bUnlocked && !ISAUTOVACUUM) ? -1 : (wrflag>1);
        int bSubjInMem = sqlite3TempInMemory(p->db);
        assert( p->db->bUnlocked==0 || wrflag==1 );
        rc = sqlite3PagerBegin(pBt->pPager, exFlag, bSubjInMem);
        if( rc==SQLITE_OK ){
          rc = newDatabase(pBt);
        }
        if( rc==SQLITE_OK ){
          rc = allocatePtrmap(pBt);
        }
      }
    }
  
    if( rc!=SQLITE_OK ){
      unlockBtreeIfUnused(pBt);
    }
  }while( (rc&0xFF)==SQLITE_BUSY && pBt->inTransaction==TRANS_NONE &&

trans_begun:
  if( rc==SQLITE_OK && wrflag ){
    /* This call makes sure that the pager has the correct number of
    ** open savepoints. If the second parameter is greater than 0 and
    ** the sub-journal is not already open, then it will be opened here.
    */
    int nSavepoint = p->db->nSavepoint;
    rc = sqlite3PagerOpenSavepoint(pBt->pPager, nSavepoint);
    if( pBt->pMap && rc==SQLITE_OK && nSavepoint ){
      rc = btreePtrmapBegin(pBt->pMap, nSavepoint);
    }
  }

  btreeIntegrity(p);
  sqlite3BtreeLeave(p);
  return rc;
}

  assert( nRef>=sqlite3PagerRefcount(pPager) );
  return rc;
}

#else /* ifndef SQLITE_OMIT_AUTOVACUUM */
# define setChildPtrmaps(x) SQLITE_OK
#endif

/*
** The b-tree handle passed as the only argument is about to commit an
** UNLOCKED transaction. At this point it is guaranteed that this is 
** possible - the wal WRITER lock is held and it is known that there are 
** no conflicts with committed transactions.
*/
static int btreeFixUnlocked(Btree *p){
  BtShared *pBt = p->pBt;
  MemPage *pPage1 = pBt->pPage1;
  u8 *p1 = pPage1->aData;
  Pager *pPager = pBt->pPager;
  int rc = SQLITE_OK;

  /* If page 1 of the database is not writable, then no pages were allocated
  ** or freed by this transaction. In this case no special handling is 
  ** required. Otherwise, if page 1 is dirty, proceed.  */
  BtreePtrmap *pMap = pBt->pMap;
  Pgno iTrunk = get4byte(&p1[32]);
  Pgno nPage = btreePagecount(pBt);
  Pgno nOrig = pMap->iFirst-1;
  u32 nFree = get4byte(&p1[36]);

  assert( sqlite3PagerIsUnlocked(pPager) );
  assert( pBt->pMap );
  rc = sqlite3PagerUpgradeSnapshot(pPager, pPage1->pDbPage);
  assert( p1==pPage1->aData );

  if( rc==SQLITE_OK ){
    Pgno nHPage = get4byte(&p1[28]);
    Pgno nFinal = nHPage;

    if( sqlite3PagerIswriteable(pPage1->pDbPage) ){
      Pgno iHTrunk = get4byte(&p1[32]);
      u32 nHFree = get4byte(&p1[36]);

      /* Attach the head database free list to the end of the current
      ** transactions free-list (if any).  */
      if( iTrunk!=0 ){
        put4byte(&p1[36], nHFree + nFree);
        put4byte(&p1[32], iTrunk);
        while( iTrunk ){
          DbPage *pTrunk = sqlite3PagerLookup(pPager, iTrunk);
          iTrunk = get4byte((u8*)pTrunk->pData);
          if( iTrunk==0 ){
            put4byte((u8*)pTrunk->pData, iHTrunk);
          }
          sqlite3PagerUnref(pTrunk);
        };
      }

      if( nHPage<nOrig ){
        rc = SQLITE_CORRUPT_BKPT;
      }else{
        /* The current transaction allocated pages pMap->iFirst through
        ** nPage (inclusive) at the end of the database file. Meanwhile,
        ** other transactions have allocated (iFirst..nHPage). So move
        ** pages (iFirst..MIN(nPage,nHPage)) to (MAX(nPage,nHPage)+1).
        */
        Pgno iLast = MIN(nPage, nHPage);    /* Last page to move */
        Pgno iPg;

        nFinal = MAX(nPage, nHPage);   /* Final size of database */
        for(iPg=pMap->iFirst; iPg<=iLast && rc==SQLITE_OK; iPg++){
          MemPage *pPg = 0;
          Pgno iNew;              /* New page number for pPg */
          PtrmapEntry *pEntry;    /* Pointer map entry for page iPg */

          btreeGetPage(pBt, iPg, &pPg, 0);
          assert( sqlite3PagerIswriteable(pPg->pDbPage) );
          assert( sqlite3PagerPageRefcount(pPg->pDbPage)==1 );
          pEntry = &pMap->aPtr[iPg - pMap->iFirst];
          iNew = ++nFinal;

          rc = relocatePage(pBt, pPg, pEntry->eType, pEntry->parent, iNew, 1);
          releasePageNotNull(pPg);
        }

        put4byte(&p1[28], nFinal);
      }
    }
    sqlite3PagerSetDbsize(pPager, nFinal);
  }

  return rc;
}

/*
** 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

    BtShared *pBt = p->pBt;
    sqlite3BtreeEnter(p);

#ifndef SQLITE_OMIT_AUTOVACUUM
    /* Figure out if this is a commit of an UNLOCKED transaction that 
    ** requires a snapshot upgrade. If so, skip any auto-vacuum 
    ** processing.  */
    if( pBt->autoVacuum ){

      assert( sqlite3PagerIsUnlocked(pBt->pPager)==0 );

      rc = autoVacuumCommit(pBt);
      if( rc!=SQLITE_OK ){
        sqlite3BtreeLeave(p);
        return rc;
      }
    }
    if( pBt->bDoTruncate ){
      sqlite3PagerTruncateImage(pBt->pPager, pBt->nPage);
    }
#endif
    if( rc==SQLITE_OK && sqlite3PagerIsUnlocked(pBt->pPager) ){
      rc = btreeFixUnlocked(p);
    }
    if( rc==SQLITE_OK ){
      rc = sqlite3PagerCommitPhaseOne(pBt->pPager, zMaster, 0);
    }
    sqlite3BtreeLeave(p);
  }
  return rc;
}

    /* Set the current transaction state to TRANS_NONE and unlock the 
    ** pager if this call closed the only read or write transaction.  */
    p->inTrans = TRANS_NONE;
    unlockBtreeIfUnused(pBt);
  }

  /* If this was an UNLOCKED transaction, delete the pBt->pMap object */
  deletePtrmap(pBt);
  btreeIntegrity(p);
}

/*
** Commit the transaction currently in progress.
**
** This routine implements the second phase of a 2-phase commit.  The

  assert( pBt->inTransaction==TRANS_WRITE );
  /* At the pager level, a statement transaction is a savepoint with
  ** an index greater than all savepoints created explicitly using
  ** SQL statements. It is illegal to open, release or rollback any
  ** such savepoints while the statement transaction savepoint is active.
  */
  rc = sqlite3PagerOpenSavepoint(pBt->pPager, iStatement);
  if( rc==SQLITE_OK && pBt->pMap ){
    rc = btreePtrmapBegin(pBt->pMap, iStatement);
  }
  sqlite3BtreeLeave(p);
  return rc;
}

/*
** The second argument to this function, op, is always SAVEPOINT_ROLLBACK
** or SAVEPOINT_RELEASE. This function either releases or rolls back the

int sqlite3BtreeSavepoint(Btree *p, int op, int iSavepoint){
  int rc = SQLITE_OK;
  if( p && p->inTrans==TRANS_WRITE ){
    BtShared *pBt = p->pBt;
    assert( op==SAVEPOINT_RELEASE || op==SAVEPOINT_ROLLBACK );
    assert( iSavepoint>=0 || (iSavepoint==-1 && op==SAVEPOINT_ROLLBACK) );
    sqlite3BtreeEnter(p);
    if( pBt->pMap ) btreePtrmapEnd(pBt->pMap, op, iSavepoint);
    rc = sqlite3PagerSavepoint(pBt->pPager, op, iSavepoint);
    if( rc==SQLITE_OK ){
      if( iSavepoint<0 && (pBt->btsFlags & BTS_INITIALLY_EMPTY)!=0 ){
        pBt->nPage = 0;
      }
      rc = newDatabase(pBt);
      pBt->nPage = get4byte(28 + pBt->pPage1->aData);

  /* EVIDENCE-OF: R-05119-02637 The 4-byte big-endian integer at offset 36
  ** stores stores the total number of pages on the freelist. */
  n = get4byte(&pPage1->aData[36]);
  testcase( n==mxPage-1 );
  if( n>=mxPage ){
    return SQLITE_CORRUPT_BKPT;
  }

  /* Ensure page 1 is writable. This function will either change the number
  ** of pages in the free-list or the size of the database file. Since both
  ** of these operations involve modifying page 1 header fields, page 1
  ** will definitely be written by this transaction. If this is an UNLOCKED
  ** transaction, ensure the BtreePtrmap structure has been allocated.  */
  assert( eMode!=BTALLOC_EXACT || sqlite3PagerIsUnlocked(pBt->pPager)==0 );
  rc = sqlite3PagerWrite(pPage1->pDbPage);
  if( rc ) return rc;

  if( n>0 ){
    /* There are pages on the freelist.  Reuse one of those pages. */
    Pgno iTrunk;
    u8 searchList = 0; /* If the free-list must be searched for 'nearby' */
    u32 nSearch = 0;   /* Count of the number of search attempts */
    
    /* If eMode==BTALLOC_EXACT and a query of the pointer-map

      searchList = 1;
    }
#endif

    /* Decrement the free-list count by 1. Set iTrunk to the index of the
    ** first free-list trunk page. iPrevTrunk is initially 1.
    */


    put4byte(&pPage1->aData[36], n-1);

    /* The code within this loop is run only once if the 'searchList' variable
    ** is not true. Otherwise, it runs once for each trunk-page on the
    ** free-list until the page 'nearby' is located (eMode==BTALLOC_EXACT)
    ** or until a page less than 'nearby' is located (eMode==BTALLOC_LT)
    */

    }
    memset(pPage->aData, 0, pPage->pBt->pageSize);
  }

  /* If the database supports auto-vacuum, write an entry in the pointer-map
  ** to indicate that the page is free.
  */
  if( REQUIRE_PTRMAP ){
    ptrmapPut(pBt, iPage, PTRMAP_FREEPAGE, 0, &rc);
    if( rc ) goto freepage_out;
  }

  /* Now manipulate the actual database free-list structure. There are two
  ** possibilities. If the free-list is currently empty, or if the first
  ** trunk page in the free-list is full, then this page will become a

          pgnoOvfl++;
        } while( 
          PTRMAP_ISPAGE(pBt, pgnoOvfl) || pgnoOvfl==PENDING_BYTE_PAGE(pBt) 
        );
      }
#endif
      rc = allocateBtreePage(pBt, &pOvfl, &pgnoOvfl, pgnoOvfl, 0);

      /* If the database supports auto-vacuum, and the second or subsequent
      ** overflow page is being allocated, add an entry to the pointer-map
      ** for that page now. 
      **
      ** If this is the first overflow page, then write a partial entry 
      ** to the pointer-map. If we write nothing to this pointer-map slot,
      ** then the optimistic overflow chain processing in clearCell()
      ** may misinterpret the uninitialized values and delete the
      ** wrong pages from the database.
      */
      if( REQUIRE_PTRMAP && rc==SQLITE_OK ){
        u8 eType = (pgnoPtrmap?PTRMAP_OVERFLOW2:PTRMAP_OVERFLOW1);
        ptrmapPut(pBt, pgnoOvfl, eType, pgnoPtrmap, &rc);
        if( rc ){
          releasePage(pOvfl);
        }
      }

      if( rc ){
        releasePage(pToRelease);
        return rc;
      }

      /* If pToRelease is not zero than pPrior points into the data area
      ** of pToRelease.  Make sure pToRelease is still writeable. */

    ** sorted order.  This invariants arise because multiple overflows can
    ** only occur when inserting divider cells into the parent page during
    ** balancing, and the dividers are adjacent and sorted.
    */
    assert( j==0 || pPage->aiOvfl[j-1]<(u16)i ); /* Overflows in sorted order */
    assert( j==0 || i==pPage->aiOvfl[j-1]+1 );   /* Overflows are sequential */
  }else{
    BtShared *pBt = pPage->pBt;
    int rc = sqlite3PagerWrite(pPage->pDbPage);
    if( rc!=SQLITE_OK ){
      *pRC = rc;
      return;
    }
    assert( sqlite3PagerIswriteable(pPage->pDbPage) );
    data = pPage->aData;
    assert( &data[pPage->cellOffset]==pPage->aCellIdx );
    rc = allocateSpace(pPage, sz, &idx);
    if( rc ){ *pRC = rc; return; }
    /* The allocateSpace() routine guarantees the following properties
    ** if it returns successfully */
    assert( idx >= 0 );
    assert( idx >= pPage->cellOffset+2*pPage->nCell+2 || CORRUPT_DB );
    assert( idx+sz <= (int)pBt->usableSize );
    pPage->nFree -= (u16)(2 + sz);
    memcpy(&data[idx], pCell, sz);
    if( iChild ){
      put4byte(&data[idx], iChild);
    }
    pIns = pPage->aCellIdx + i*2;
    memmove(pIns+2, pIns, 2*(pPage->nCell - i));
    put2byte(pIns, idx);
    pPage->nCell++;
    /* increment the cell count */
    if( (++data[pPage->hdrOffset+4])==0 ) data[pPage->hdrOffset+3]++;
    assert( get2byte(&data[pPage->hdrOffset+3])==pPage->nCell );

    if( REQUIRE_PTRMAP ){
      /* The cell may contain a pointer to an overflow page. If so, write
      ** the entry for the overflow page into the pointer map.
      */
      ptrmapPutOvflPtr(pPage, pCell, pRC);
    }

  }
}

/*
** A CellArray object contains a cache of pointers and sizes for a
** consecutive sequence of cells that might be held multiple pages.
*/

    ** cell on the page to an overflow page. If either of these
    ** operations fails, the return code is set, but the contents
    ** of the parent page are still manipulated by thh code below.
    ** That is Ok, at this point the parent page is guaranteed to
    ** be marked as dirty. Returning an error code will cause a
    ** rollback, undoing any changes made to the parent page.
    */
    if( REQUIRE_PTRMAP ){
      ptrmapPut(pBt, pgnoNew, PTRMAP_BTREE, pParent->pgno, &rc);
      if( szCell>pNew->minLocal ){
        ptrmapPutOvflPtr(pNew, pCell, &rc);
      }
    }
  
    /* Create a divider cell to insert into pParent. The divider cell

      *pRC = rc;
      return;
    }
  
    /* If this is an auto-vacuum database, update the pointer-map entries
    ** for any b-tree or overflow pages that pTo now contains the pointers to.
    */
    if( REQUIRE_PTRMAP ){
      *pRC = setChildPtrmaps(pTo);
    }
  }
}

/*
** This routine redistributes cells on the iParentIdx'th child of pParent

      if( rc ) goto balance_cleanup;
      zeroPage(pNew, pageFlags);
      apNew[i] = pNew;
      nNew++;
      cntOld[i] = b.nCell;

      /* Set the pointer-map entry for the new sibling page. */
      if( REQUIRE_PTRMAP ){
        ptrmapPut(pBt, pNew->pgno, PTRMAP_BTREE, pParent->pgno, &rc);
        if( rc!=SQLITE_OK ){
          goto balance_cleanup;
        }
      }
    }
  }

  **      with the cell.
  **
  ** If the sibling pages are not leaves, then the pointer map entry 
  ** associated with the right-child of each sibling may also need to be 
  ** updated. This happens below, after the sibling pages have been 
  ** populated, not here.
  */
  if( REQUIRE_PTRMAP ){
    MemPage *pNew = apNew[0];
    u8 *aOld = pNew->aData;
    int cntOldNext = pNew->nCell + pNew->nOverflow;
    int usableSize = pBt->usableSize;
    int iNew = 0;
    int iOld = 0;

    testcase( rc!=SQLITE_OK );
    assert( apNew[0]->nFree == 
        (get2byte(&apNew[0]->aData[5])-apNew[0]->cellOffset-apNew[0]->nCell*2)
      || rc!=SQLITE_OK
    );
    copyNodeContent(apNew[0], pParent, &rc);
    freePage(apNew[0], &rc);
  }else if( REQUIRE_PTRMAP && !leafCorrection ){
    /* Fix the pointer map entries associated with the right-child of each
    ** sibling page. All other pointer map entries have already been taken
    ** care of.  */
    for(i=0; i<nNew; i++){
      u32 key = get4byte(&apNew[i]->aData[8]);
      ptrmapPut(pBt, key, PTRMAP_BTREE, apNew[i]->pgno, &rc);
    }

  ** page that will become the new right-child of pPage. Copy the contents
  ** of the node stored on pRoot into the new child page.
  */
  rc = sqlite3PagerWrite(pRoot->pDbPage);
  if( rc==SQLITE_OK ){
    rc = allocateBtreePage(pBt,&pChild,&pgnoChild,pRoot->pgno,0);
    copyNodeContent(pRoot, pChild, &rc);
    if( REQUIRE_PTRMAP ){
      ptrmapPut(pBt, pgnoChild, PTRMAP_BTREE, pRoot->pgno, &rc);
    }
  }
  if( rc ){
    *ppChild = 0;
    releasePage(pChild);
    return rc;

  return (p->pBt->btsFlags & BTS_READ_ONLY)!=0;
}

/*
** Return the size of the header added to each page by this module.
*/
int sqlite3HeaderSizeBtree(void){ return ROUND8(sizeof(MemPage)); }

int sqlite3BtreeExclusiveLock(Btree *p){
  int rc;
  BtShared *pBt = p->pBt;
  assert( p->inTrans==TRANS_WRITE && pBt->pPage1 );
  sqlite3BtreeEnter(p);
  rc = sqlite3PagerExclusiveLock(pBt->pPager, pBt->pPage1->pDbPage);
  sqlite3BtreeLeave(p);
  return rc;
}

int sqlite3BtreeSetVersion(Btree *pBt, int iVersion);
void sqlite3BtreeCursorHints(BtCursor *, unsigned int mask);
#ifdef SQLITE_DEBUG
int sqlite3BtreeCursorHasHint(BtCursor*, unsigned int mask);
#endif
int sqlite3BtreeIsReadonly(Btree *pBt);
int sqlite3HeaderSizeBtree(void);

int sqlite3BtreeExclusiveLock(Btree *pBt);

#ifndef NDEBUG
int sqlite3BtreeCursorIsValid(BtCursor*);
#endif

#ifndef SQLITE_OMIT_BTREECOUNT
int sqlite3BtreeCount(BtCursor *, i64 *);

*/
#define MX_CELL(pBt) ((pBt->pageSize-8)/6)

/* Forward declarations */
typedef struct MemPage MemPage;
typedef struct BtLock BtLock;
typedef struct CellInfo CellInfo;
typedef struct BtreePtrmap BtreePtrmap;

/*
** This is a magic string that appears at the beginning of every
** SQLite database in order to identify the file as a real database.
**
** You can change this value at compile-time by specifying a
** -DSQLITE_FILE_HEADER="..." on the compiler command-line.  The

#ifndef SQLITE_OMIT_SHARED_CACHE
  int nRef;             /* Number of references to this structure */
  BtShared *pNext;      /* Next on a list of sharable BtShared structs */
  BtLock *pLock;        /* List of locks held on this shared-btree struct */
  Btree *pWriter;       /* Btree with currently open write transaction */
#endif
  u8 *pTmpSpace;        /* Temp space sufficient to hold a single cell */
#ifdef SQLITE_ENABLE_UNLOCKED
  BtreePtrmap *pMap;
#endif
};

/*
** Allowed values for BtShared.btsFlags
*/
#define BTS_READ_ONLY        0x0001   /* Underlying file is readonly */
#define BTS_PAGESIZE_FIXED   0x0002   /* Page size can no longer be changed */

** So, this macro is defined instead.
*/
#ifndef SQLITE_OMIT_AUTOVACUUM
#define ISAUTOVACUUM (pBt->autoVacuum)
#else
#define ISAUTOVACUUM 0
#endif

#ifdef SQLITE_ENABLE_UNLOCKED
# define REQUIRE_PTRMAP (ISAUTOVACUUM || pBt->pMap)
#else
# define REQUIRE_PTRMAP ISAUTOVACUUM
#endif


/*
** This structure is passed around through all the sanity checking routines
** in order to keep track of some global state information.
**
** The aRef[] array is allocated so that there is 1 bit for each page in

  Pgno dbFileSize;            /* Number of pages in the database file */
  Pgno dbHintSize;            /* Value passed to FCNTL_SIZE_HINT call */
  int errCode;                /* One of several kinds of errors */
  int nRec;                   /* Pages journalled since last j-header written */
  u32 cksumInit;              /* Quasi-random value added to every checksum */
  u32 nSubRec;                /* Number of records written to sub-journal */
  Bitvec *pInJournal;         /* One bit for each page in the database file */
#ifdef SQLITE_ENABLE_UNLOCKED
  Bitvec *pAllRead;           /* Pages read within current UNLOCKED trans. */
#endif
  sqlite3_file *fd;           /* File descriptor for database */
  sqlite3_file *jfd;          /* File descriptor for main journal */
  sqlite3_file *sjfd;         /* File descriptor for sub-journal */
  i64 journalOff;             /* Current write offset in the journal file */
  i64 journalHdr;             /* Byte offset to previous journal header */
  sqlite3_backup *pBackup;    /* Pointer to list of ongoing backup processes */
  PagerSavepoint *aSavepoint; /* Array of active savepoints */

  assert( pPager->eState==PAGER_WRITER_CACHEMOD
       || pPager->eState==PAGER_WRITER_DBMOD
  );
  assert( assert_pager_state(pPager) );
  assert( !pagerUseWal(pPager) );

  rc = sqlite3PagerExclusiveLock(pPager, 0);
  if( rc!=SQLITE_OK ) return rc;

  if( !pPager->noSync ){
    assert( !pPager->tempFile );
    if( isOpen(pPager->jfd) && pPager->journalMode!=PAGER_JOURNALMODE_MEMORY ){
      const int iDc = sqlite3OsDeviceCharacteristics(pPager->fd);
      assert( isOpen(pPager->jfd) );

}

/*
** Return TRUE if the page given in the argument was previously passed
** to sqlite3PagerWrite().  In other words, return TRUE if it is ok
** to change the content of the page.
*/

int sqlite3PagerIswriteable(DbPage *pPg){
  return pPg->flags & PGHDR_WRITEABLE;
}


/*
** A call to this routine tells the pager that it is not necessary to
** write the information on page pPg back to the disk, even though
** that page might be marked as dirty.  This happens, for example, when
** the page has been added as a leaf of the freelist and so its
** content no longer matters.

** the database file, an attempt is made to obtain one.
**
** If the EXCLUSIVE lock is already held or the attempt to obtain it is
** successful, or the connection is in WAL mode, SQLITE_OK is returned.
** Otherwise, either SQLITE_BUSY or an SQLITE_IOERR_XXX error code is 
** returned.
*/
int sqlite3PagerExclusiveLock(Pager *pPager, PgHdr *pPage1){
  int rc = SQLITE_OK;
  assert( pPager->eState==PAGER_WRITER_CACHEMOD 
       || pPager->eState==PAGER_WRITER_DBMOD 
       || pPager->eState==PAGER_WRITER_LOCKED 
  );
  assert( assert_pager_state(pPager) );
  if( 0==pagerUseWal(pPager) ){
    rc = pager_wait_on_lock(pPager, EXCLUSIVE_LOCK);
  }else{
    if( pPager->pAllRead ){
      /* This is an UNLOCKED transaction. Attempt to lock the wal database
      ** here. If SQLITE_BUSY (but not SQLITE_BUSY_SNAPSHOT) is returned,
      ** invoke the busy-handler and try again for as long as it returns
      ** non-zero.  */
      do {
        /* rc = sqlite3WalBeginWriteTransaction(pWal); */
        rc = sqlite3WalLockForCommit(pPager->pWal, pPage1, pPager->pAllRead);
      }while( rc==SQLITE_BUSY 
           && pPager->xBusyHandler(pPager->pBusyHandlerArg) 
      );
    }
  }
  return rc;
}

int sqlite3PagerUpgradeSnapshot(Pager *pPager, DbPage *pPage1){
  int rc;
  u32 iFrame = 0;

  assert( pPager->pWal && pPager->pAllRead );
  sqlite3WalUpgradeSnapshot(pPager->pWal);
  rc = sqlite3WalFindFrame(pPager->pWal, 1, &iFrame);
  if( rc==SQLITE_OK ){
    rc = readDbPage(pPage1, iFrame);
  }

  return rc;
}

void sqlite3PagerSetDbsize(Pager *pPager, Pgno nFinal){
  pPager->dbSize = nFinal;
}

/*
** If this is a WAL mode connection and the WRITER lock is currently held,
** relinquish it.
*/
void sqlite3PagerDropExclusiveLock(Pager *pPager){
  if( pagerUseWal(pPager) ){
    sqlite3WalEndWriteTransaction(pPager->pWal);
  }
}

/*
** Return true if this is a WAL database and snapshot upgrade is required
** before the current transaction can be committed.
*/
int sqlite3PagerIsUnlocked(Pager *pPager){

  return pPager->pAllRead!=0;

}



/*
** Sync the database file for the pager pPager. zMaster points to the name
** of a master journal file that should be written into the individual
** journal file. zMaster may be NULL, which is interpreted as no master
** journal (a single database transaction).
**

/* Operations on page references. */
int sqlite3PagerWrite(DbPage*);
void sqlite3PagerDontWrite(DbPage*);
int sqlite3PagerMovepage(Pager*,DbPage*,Pgno,int);
int sqlite3PagerPageRefcount(DbPage*);
void *sqlite3PagerGetData(DbPage *); 
void *sqlite3PagerGetExtra(DbPage *); 
int sqlite3PagerIsDirty(DbPage*);

/* Functions used to manage pager transactions and savepoints. */
void sqlite3PagerPagecount(Pager*, int*);
int sqlite3PagerBegin(Pager*, int exFlag, int);
int sqlite3PagerCommitPhaseOne(Pager*,const char *zMaster, int);
int sqlite3PagerExclusiveLock(Pager*, DbPage *pPage1);
int sqlite3PagerSync(Pager *pPager, const char *zMaster);
int sqlite3PagerCommitPhaseTwo(Pager*);
int sqlite3PagerRollback(Pager*);
int sqlite3PagerOpenSavepoint(Pager *pPager, int n);
int sqlite3PagerSavepoint(Pager *pPager, int op, int iSavepoint);
int sqlite3PagerSharedLock(Pager *pPager);

void sqlite3PagerDropExclusiveLock(Pager*);
int sqlite3PagerIsUnlocked(Pager*);

#ifndef SQLITE_OMIT_WAL
  int sqlite3PagerCheckpoint(Pager *pPager, int, int*, int*);
  int sqlite3PagerWalSupported(Pager *pPager);
  int sqlite3PagerWalCallback(Pager *pPager);
  int sqlite3PagerOpenWal(Pager *pPager, int *pisOpen);
  int sqlite3PagerCloseWal(Pager *pPager);

int sqlite3SectorSize(sqlite3_file *);

/* Functions used to truncate the database file. */
void sqlite3PagerTruncateImage(Pager*,Pgno);

void sqlite3PagerRekey(DbPage*, Pgno, u16);

int sqlite3PagerIswriteable(DbPage*);
int sqlite3PagerUpgradeSnapshot(Pager *pPager, DbPage*);
void sqlite3PagerSetDbsize(Pager *pPager, Pgno);

#if defined(SQLITE_HAS_CODEC) && !defined(SQLITE_OMIT_WAL)
void *sqlite3PagerCodec(DbPage *);
#endif

/* Functions to support testing and debugging. */
#if !defined(NDEBUG) || defined(SQLITE_TEST)
  Pgno sqlite3PagerPagenumber(DbPage*);

#endif
#ifdef SQLITE_TEST
  int *sqlite3PagerStats(Pager*);
  void sqlite3PagerRefdump(Pager*);
  void disable_simulated_io_errors(void);
  void enable_simulated_io_errors(void);
#else
# define disable_simulated_io_errors()
# define enable_simulated_io_errors()
#endif

#endif /* _PAGER_H_ */

  ** file is required for an atomic commit.
  */ 
  for(i=0; rc==SQLITE_OK && i<db->nDb; i++){ 
    Btree *pBt = db->aDb[i].pBt;
    if( sqlite3BtreeIsInTrans(pBt) ){
      needXcommit = 1;
      if( i!=1 ) nTrans++;
      rc = sqlite3BtreeExclusiveLock(pBt);


    }
  }

  if( db->bUnlocked && (rc & 0xFF)==SQLITE_BUSY ){
    /* An SQLITE_BUSY or SQLITE_BUSY_SNAPSHOT was encountered while 
    ** attempting to take the WRITER lock on a wal file. Release the
    ** WRITER locks on all wal files and return early.  */

/* 
** TODO: Combine some code with BeginWriteTransaction()
**
** This function is only ever called when committing a "BEGIN UNLOCKED"
** transaction. It may be assumed that no frames have been written to
** the wal file.
*/
int sqlite3WalLockForCommit(Wal *pWal, PgHdr *pPage1, Bitvec *pAllRead){
  Pager *pPager = pPage1->pPager;
  int rc = walWriteLock(pWal);

  /* If the database has been modified since this transaction was started,
  ** check if it is still possible to commit. The transaction can be 
  ** committed if:
  **
  **   a) None of the pages in pList have been modified since the 
  **      transaction opened, and
  **
  **   b) The database schema cookie has not been modified since the
  **      transaction was started.
  */
  if( rc==SQLITE_OK ){
    volatile WalIndexHdr *pHead;    /* Head of the wal file */
    pHead = walIndexHdr(pWal);

    /* TODO: Check header checksum is good here. */

    if( memcmp(&pWal->hdr, (void*)pHead, sizeof(WalIndexHdr))!=0 ){
      /* TODO: Is this safe? Because it holds the WRITER lock this thread
      ** has exclusive access to the live header, but might it be corrupt? 
      ** This code should check that the wal-index-header is Ok, and return
      ** SQLITE_BUSY_SNAPSHOT if it is not. */
      int iHash;
      int iLastHash = walFramePage(pHead->mxFrame);
      for(iHash=walFramePage(pWal->hdr.mxFrame+1); iHash<=iLastHash; iHash++){
        volatile ht_slot *aHash;
        volatile u32 *aPgno;
        u32 iZero;

        rc = walHashGet(pWal, iHash, &aHash, &aPgno, &iZero);
        if( rc==SQLITE_OK ){
          int i;
          int iMin = (pWal->hdr.mxFrame+1 - iZero);
          int iMax = (iHash==0) ? HASHTABLE_NPAGE_ONE : HASHTABLE_NPAGE;
          if( iMin<1 ) iMin = 1;
          if( iMax>pHead->mxFrame ) iMax = pHead->mxFrame;
          for(i=iMin; i<=iMax; i++){
            PgHdr *pPg;
            if( aPgno[i]==1 ){
              /* Check that the schema cookie has not been modified. If
              ** it has not, the commit can proceed. */
              u8 aNew[4];
              u8 *aOld = &((u8*)pPage1->pData)[40];
              int sz;
              i64 iOffset;
              sz = pWal->hdr.szPage;
              sz = (sz&0xfe00) + ((sz&0x0001)<<16);
              iOffset = walFrameOffset(i+iZero, sz) + WAL_FRAME_HDRSIZE + 40;
              rc = sqlite3OsRead(pWal->pWalFd, aNew, sizeof(aNew), iOffset);
              if( rc==SQLITE_OK && memcmp(aOld, aNew, sizeof(aNew)) ){
                rc = SQLITE_BUSY_SNAPSHOT;
              }
            }else if( sqlite3BitvecTestNotNull(pAllRead, aPgno[i]) ){
              sqlite3_log(SQLITE_OK,
                  "cannot commit UNLOCKED transaction (conflict at page %d)",
                  (int)aPgno[i]
              );
              rc = SQLITE_BUSY_SNAPSHOT;
            }else if( (pPg = sqlite3PagerLookup(pPager, aPgno[i])) ){
              if( sqlite3PagerIswriteable(pPg)==0 ){
                sqlite3PcacheDrop(pPg);
              }else{
                sqlite3PagerUnref(pPg);
              }
            }
          }
        }
        if( rc!=SQLITE_OK ) break;
      }
    }
  }

  return rc;
}

void sqlite3WalUpgradeSnapshot(Wal *pWal){
  assert( pWal->writeLock );
  memcpy(&pWal->hdr, (void*)walIndexHdr(pWal), sizeof(WalIndexHdr));
}

/*
** This function is only ever called while committing an UNLOCKED 
** transaction, after the caller has already obtained the WRITER lock
** (by calling the sqlite3WalLockForCommit() routine). This function 
** returns true if the transaction was prepared against a database 
** snapshot older than the current head of the wal file.

  u32 iFrame;                     /* Next frame address */
  PgHdr *p;                       /* Iterator to run through pList with. */
  PgHdr *pLast = 0;               /* Last frame in list */
  int nExtra = 0;                 /* Number of extra copies of last page */
  int szFrame;                    /* The size of a single frame */
  i64 iOffset;                    /* Next byte to write in WAL file */
  WalWriter w;                    /* The writer */


  assert( pList );
  assert( pWal->writeLock );

  /* If this frame set completes a transaction, then nTruncate>0.  If
  ** nTruncate==0 then this frame set does not complete the transaction. */
  assert( (isCommit!=0)==(nTruncate!=0) );

#if defined(SQLITE_TEST) && defined(SQLITE_DEBUG)
  { int cnt; for(cnt=0, p=pList; p; p=p->pDirty, cnt++){}
    WALTRACE(("WAL%p: frame write begin. %d frames. mxFrame=%d. %s\n",
              pWal, cnt, pWal->hdr.mxFrame, isCommit ? "Commit" : "Spill"));
  }
#endif

















  /* See if it is possible to write these frames into the start of the
  ** log file, instead of appending to it at pWal->hdr.mxFrame.
  */
  if( SQLITE_OK!=(rc = walRestartLog(pWal)) ){
    return rc;
  }

    /* If this is a commit, update the wal-index header too. */
    if( isCommit ){
      walIndexWriteHdr(pWal);
      pWal->iCallback = iFrame;
    }
  }









  WALTRACE(("WAL%p: frame write %s\n", pWal, rc ? "failed" : "ok"));
  return rc;
}

/* 
** This routine is called to implement sqlite3_wal_checkpoint() and
** related interfaces.

/* Return true if the argument is non-NULL and the WAL module is using
** heap-memory for the wal-index. Otherwise, if the argument is NULL or the
** WAL module is using shared-memory, return false. 
*/
int sqlite3WalHeapMemory(Wal *pWal);

/* Return true if the WRITER lock is held. False otherwise. */
int sqlite3WalLockForCommit(Wal *pWal, PgHdr *pPg, Bitvec *pRead);
int sqlite3WalCommitRequiresUpgrade(Wal *pWal);
void sqlite3WalUpgradeSnapshot(Wal *pWal);

#ifdef SQLITE_ENABLE_ZIPVFS
/* If the WAL file is not empty, return the number of bytes of content
** stored in each frame (i.e. the db page-size when the WAL was created).
*/
int sqlite3WalFramesize(Wal *pWal);
#endif

        VALUES(1,1) UNION ALL SELECT a+1,b+1 FROM src WHERE a<10000
      ) INSERT INTO t2 SELECT * FROM src;
    }
  } {}

  do_test 2.$tn.7.3 {
    list [catch { sql1 { COMMIT } } msg] $msg [sqlite3_errcode db]
  } {0 {} SQLITE_OK}


  do_test 2.$tn.7.4 { sql3 { PRAGMA integrity_check } } ok
}



finish_test

# 2015 July 26
#
# The author disclaims copyright to this source code.  In place of
# a legal notice, here is a blessing:
#
#    May you do good and not evil.
#    May you find forgiveness for yourself and forgive others.
#    May you share freely, never taking more than you give.
#
#***********************************************************************
#

set testdir [file dirname $argv0]
source $testdir/tester.tcl
source $testdir/lock_common.tcl
set ::testprefix unlocked2


do_multiclient_test tn {

  do_test 1.$tn.1 {
    sql1 {
      PRAGMA journal_mode = wal;
      CREATE TABLE t1(x);
      CREATE TABLE t2(y);
    }
  } {wal}

  # Test that an UNLOCKED transaction that allocates/frees no pages does
  # not conflict with a transaction that does allocate pages.
  do_test 1.$tn.2  {
    sql1 { 
      BEGIN UNLOCKED;
        INSERT INTO t1 VALUES(4);
    }
    sql2 {
      INSERT INTO t2 VALUES(randomblob(1500));
    }
    sql1 {
      COMMIT;
    }
  } {}
  
  # But that an UNLOCKED transaction does conflict with a transaction
  # that modifies the db schema.
  do_test 1.$tn.3  {
    sql1 {
      BEGIN UNLOCKED;
        INSERT INTO t1 VALUES(5);
    }
    sql2 {
      CREATE TABLE t3(z);
    }
    list [catch { sql1 COMMIT } msg] $msg
  } {1 {database is locked}}
  
  # Test that an UNLOCKED transaction that allocates at least one page 
  # does not conflict with a transaction that allocates no pages.
  do_test 1.$tn.4  {
    sql1 { 
      ROLLBACK;
      BEGIN UNLOCKED;
        INSERT INTO t1 VALUES(randomblob(1500));
    }
    sql2 {
      INSERT INTO t2 VALUES(8);
    }
    sql1 {
      COMMIT;
    }
  } {}

  do_test 1.$tn.5 { sql3 { PRAGMA integrity_check } } {ok}
}

do_multiclient_test tn {
  do_test 2.$tn.1 {
    sql1 {
      PRAGMA journal_mode = wal;
      CREATE TABLE t1(x UNIQUE);
      CREATE TABLE t2(y UNIQUE);
    }
  } {wal}

  do_test 2.$tn.2  {
    sql1 { 
      BEGIN UNLOCKED;
        INSERT INTO t1 VALUES(randomblob(1500));
    }
    sql2 {
      INSERT INTO t2 VALUES(randomblob(1500));
    }
    sql1 {
      COMMIT;
    }
  } {}

  do_test 2.$tn.3 { sql3 { PRAGMA integrity_check } } {ok}
}

finish_test