SQLite

/*
** 2001 September 15
**
** 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.
**
*************************************************************************
** $Id: btree.c,v 1.103 2004/03/10 13:42:38 drh Exp $
**
** This file implements a external (disk-based) database using BTrees.
** For a detailed discussion of BTrees, refer to
**
**     Donald E. Knuth, THE ART OF COMPUTER PROGRAMMING, Volume 3:
**     "Sorting And Searching", pages 473-480. Addison-Wesley
**     Publishing Company, Reading, Massachusetts.

** so forth.
**
** Finding a particular key requires reading O(log(M)) pages from the 
** disk where M is the number of entries in the tree.
**
** In this implementation, a single file can hold one or more separate 
** BTrees.  Each BTree is identified by the index of its root page.  The
** key and data for any entry are combined to form the "payload".  Up to
** MX_LOCAL_PAYLOAD bytes of payload can be carried directly on the
** database page.  If the payload is larger than MX_LOCAL_PAYLOAD bytes
** then surplus bytes are stored on overflow pages.  The payload for an
** entry and the preceding pointer are combined to form a "Cell".  Each 
** page has a small header which contains the Ptr(N+1) pointer.

**








** The first page of the file contains a magic string used to verify that








** the file really is a valid BTree database, a pointer to a list of unused



** pages in the file, and some meta information.  The root of the first









** BTree begins on page 2 of the file.  (Pages are numbered beginning with































** 1, not 0.)  Thus a minimum database contains 2 pages.






































*/
#include "sqliteInt.h"
#include "pager.h"
#include "btree.h"
#include <assert.h>

/* Forward declarations */
static BtOps sqliteBtreeOps;
static BtCursorOps sqliteBtreeCursorOps;

/*
** Macros used for byteswapping.  B is a pointer to the Btree
** structure.  This is needed to access the Btree.needSwab boolean
** in order to tell if byte swapping is needed or not.
** X is an unsigned integer.  SWAB16 byte swaps a 16-bit integer.
** SWAB32 byteswaps a 32-bit integer.
*/
#define SWAB16(B,X)   ((B)->needSwab? swab16((u16)X) : ((u16)X))
#define SWAB32(B,X)   ((B)->needSwab? swab32(X) : (X))
#define SWAB_ADD(B,X,A) \
   if((B)->needSwab){ X=swab32(swab32(X)+A); }else{ X += (A); }

/*
** The following global variable - available only if SQLITE_TEST is
** defined - is used to determine whether new databases are created in
** native byte order or in non-native byte order.  Non-native byte order
** databases are created for testing purposes only.  Under normal operation,
** only native byte-order databases should be created, but we should be
** able to read or write existing databases regardless of the byteorder.
*/
#ifdef SQLITE_TEST
int btree_native_byte_order = 1;
#else
# define btree_native_byte_order 1
#endif

/*
** Forward declarations of structures used only in this file.
*/
typedef struct PageOne PageOne;
typedef struct MemPage MemPage;
typedef struct PageHdr PageHdr;
typedef struct Cell Cell;
typedef struct CellHdr CellHdr;
typedef struct FreeBlk FreeBlk;
typedef struct OverflowPage OverflowPage;
typedef struct FreelistInfo FreelistInfo;

/*
** All structures on a database page are aligned to 4-byte boundries.
** This routine rounds up a number of bytes to the next multiple of 4.
**
** This might need to change for computer architectures that require
** and 8-byte alignment boundry for structures.
*/
#define ROUNDUP(X)  ((X+3) & ~3)

/*
** This is a magic string that appears at the beginning of every
** SQLite database in order to identify the file as a real database.
*/
static const char zMagicHeader[] = 
   "** This file contains an SQLite 2.1 database **";
#define MAGIC_SIZE (sizeof(zMagicHeader))

/*
** This is a magic integer also used to test the integrity of the database
** file.  This integer is used in addition to the string above so that
** if the file is written on a little-endian architecture and read
** on a big-endian architectures (or vice versa) we can detect the
** problem.
**
** The number used was obtained at random and has no special
** significance other than the fact that it represents a different
** integer on little-endian and big-endian machines.
*/
#define MAGIC 0xdae37528

/*
** The first page of the database file contains a magic header string
** to identify the file as an SQLite database file.  It also contains
** a pointer to the first free page of the file.  Page 2 contains the
** root of the principle BTree.  The file might contain other BTrees
** rooted on pages above 2.
**
** The first page also contains SQLITE_N_BTREE_META integers that
** can be used by higher-level routines.
**
** Remember that pages are numbered beginning with 1.  (See pager.c
** for additional information.)  Page 0 does not exist and a page
** number of 0 is used to mean "no such page".
*/
struct PageOne {
  char zMagic[MAGIC_SIZE]; /* String that identifies the file as a database */
  int iMagic;              /* Integer to verify correct byte order */
  Pgno freeList;           /* First free page in a list of all free pages */
  int nFree;               /* Number of pages on the free list */
  int aMeta[SQLITE_N_BTREE_META-1];  /* User defined integers */
};

/*
** Each database page has a header that is an instance of this
** structure.
**
** PageHdr.firstFree is 0 if there is no free space on this page.
** Otherwise, PageHdr.firstFree is the index in MemPage.u.aDisk[] of a 
** FreeBlk structure that describes the first block of free space.  
** All free space is defined by a linked list of FreeBlk structures.
**
** Data is stored in a linked list of Cell structures.  PageHdr.firstCell
** is the index into MemPage.u.aDisk[] of the first cell on the page.  The
** Cells are kept in sorted order.
**
** A Cell contains all information about a database entry and a pointer
** to a child page that contains other entries less than itself.  In
** other words, the i-th Cell contains both Ptr(i) and Key(i).  The
** right-most pointer of the page is contained in PageHdr.rightChild.
*/
struct PageHdr {
  Pgno rightChild;  /* Child page that comes after all cells on this page */
  u16 firstCell;    /* Index in MemPage.u.aDisk[] of the first cell */
  u16 firstFree;    /* Index in MemPage.u.aDisk[] of the first free block */
};

/*
** Entries on a page of the database are called "Cells".  Each Cell
** has a header and data.  This structure defines the header.  The
** key and data (collectively the "payload") follow this header on
** the database page.
**
** A definition of the complete Cell structure is given below.  The
** header for the cell must be defined first in order to do some
** of the sizing #defines that follow.
*/
struct CellHdr {
  Pgno leftChild; /* Child page that comes before this cell */
  u16 nKey;       /* Number of bytes in the key */
  u16 iNext;      /* Index in MemPage.u.aDisk[] of next cell in sorted order */
  u8 nKeyHi;      /* Upper 8 bits of key size for keys larger than 64K bytes */
  u8 nDataHi;     /* Upper 8 bits of data size when the size is more than 64K */
  u16 nData;      /* Number of bytes of data */
};

/*
** The key and data size are split into a lower 16-bit segment and an
** upper 8-bit segment in order to pack them together into a smaller
** space.  The following macros reassembly a key or data size back
** into an integer.
*/
#define NKEY(b,h)  (SWAB16(b,h.nKey) + h.nKeyHi*65536)
#define NDATA(b,h) (SWAB16(b,h.nData) + h.nDataHi*65536)

/*
** The minimum size of a complete Cell.  The Cell must contain a header
** and at least 4 bytes of payload.
*/
#define MIN_CELL_SIZE  (sizeof(CellHdr)+4)

/*
** The maximum number of database entries that can be held in a single
** page of the database. 
*/
#define MX_CELL ((SQLITE_USABLE_SIZE-sizeof(PageHdr))/MIN_CELL_SIZE)

/*
** The amount of usable space on a single page of the BTree.  This is the
** page size minus the overhead of the page header.
*/
#define USABLE_SPACE  (SQLITE_USABLE_SIZE - sizeof(PageHdr))

/*
** The maximum amount of payload (in bytes) that can be stored locally for
** a database entry.  If the entry contains more data than this, the
** extra goes onto overflow pages.
**
** This number is chosen so that at least 4 cells will fit on every page.
*/
#define MX_LOCAL_PAYLOAD ((USABLE_SPACE/4-(sizeof(CellHdr)+sizeof(Pgno)))&~3)

/*
** Data on a database page is stored as a linked list of Cell structures.
** Both the key and the data are stored in aPayload[].  The key always comes
** first.  The aPayload[] field grows as necessary to hold the key and data,
** up to a maximum of MX_LOCAL_PAYLOAD bytes.  If the size of the key and
** data combined exceeds MX_LOCAL_PAYLOAD bytes, then Cell.ovfl is the
** page number of the first overflow page.
**
** Though this structure is fixed in size, the Cell on the database
** page varies in size.  Every cell has a CellHdr and at least 4 bytes
** of payload space.  Additional payload bytes (up to the maximum of
** MX_LOCAL_PAYLOAD) and the Cell.ovfl value are allocated only as
** needed.
*/
struct Cell {
  CellHdr h;                        /* The cell header */
  char aPayload[MX_LOCAL_PAYLOAD];  /* Key and data */
  Pgno ovfl;                        /* The first overflow page */
};

/*
** Free space on a page is remembered using a linked list of the FreeBlk
** structures.  Space on a database page is allocated in increments of
** at least 4 bytes and is always aligned to a 4-byte boundry.  The
** linked list of FreeBlks is always kept in order by address.
*/
struct FreeBlk {
  u16 iSize;      /* Number of bytes in this block of free space */
  u16 iNext;      /* Index in MemPage.u.aDisk[] of the next free block */
};

/*
** The number of bytes of payload that will fit on a single overflow page.
*/
#define OVERFLOW_SIZE (SQLITE_USABLE_SIZE-sizeof(Pgno))

/*
** When the key and data for a single entry in the BTree will not fit in
** the MX_LOCAL_PAYLOAD bytes of space available on the database page,
** then all extra bytes are written to a linked list of overflow pages.
** Each overflow page is an instance of the following structure.
**
** Unused pages in the database are also represented by instances of
** the OverflowPage structure.  The PageOne.freeList field is the
** page number of the first page in a linked list of unused database
** pages.
*/
struct OverflowPage {
  Pgno iNext;
  char aPayload[OVERFLOW_SIZE];
};

/*
** The PageOne.freeList field points to a linked list of overflow pages
** hold information about free pages.  The aPayload section of each
** overflow page contains an instance of the following structure.  The
** aFree[] array holds the page number of nFree unused pages in the disk
** file.
*/
struct FreelistInfo {
  int nFree;
  Pgno aFree[(OVERFLOW_SIZE-sizeof(int))/sizeof(Pgno)];
};

/*
** For every page in the database file, an instance of the following structure
** is stored in memory.  The u.aDisk[] array contains the raw bits read from
** the disk.  The rest is auxiliary information held in memory only. The
** auxiliary info is only valid for regular database pages - it is not
** used for overflow pages and pages on the freelist.
**
** Of particular interest in the auxiliary info is the apCell[] entry.  Each
** apCell[] entry is a pointer to a Cell structure in u.aDisk[].  The cells are
** put in this array so that they can be accessed in constant time, rather
** than in linear time which would be needed if we had to walk the linked 
** list on every access.
**
** Note that apCell[] contains enough space to hold up to two more Cells
** than can possibly fit on one page.  In the steady state, every apCell[]
** points to memory inside u.aDisk[].  But in the middle of an insert
** operation, some apCell[] entries may temporarily point to data space
** outside of u.aDisk[].  This is a transient situation that is quickly
** resolved.  But while it is happening, it is possible for a database
** page to hold as many as two more cells than it might otherwise hold.
** The extra two entries in apCell[] are an allowance for this situation.
**
** The pParent field points back to the parent page.  This allows us to
** walk up the BTree from any leaf to the root.  Care must be taken to
** unref() the parent page pointer when this page is no longer referenced.
** The pageDestructor() routine handles that chore.
*/
struct MemPage {
  union u_page_data {
    char aDisk[SQLITE_PAGE_SIZE];  /* Page data stored on disk */
    PageHdr hdr;                   /* Overlay page header */
  } u;
  u8 isInit;                     /* True if auxiliary data is initialized */
  u8 idxShift;                   /* True if apCell[] indices have changed */
  u8 isOverfull;                 /* Some apCell[] points outside u.aDisk[] */





  MemPage *pParent;              /* The parent of this page.  NULL for root */
  int idxParent;                 /* Index in pParent->apCell[] of this node */
  int nFree;                     /* Number of free bytes in u.aDisk[] */
  int nCell;                     /* Number of entries on this page */
  Cell *apCell[MX_CELL+2];       /* All data entires in sorted order */
};

/*
** The in-memory image of a disk page has the auxiliary information appended
** to the end.  EXTRA_SIZE is the number of bytes of space needed to hold
** that extra information.
*/
#define EXTRA_SIZE (sizeof(MemPage)-sizeof(union u_page_data))

/*
** Everything we need to know about an open database
*/
struct Btree {
  BtOps *pOps;          /* Function table */
  Pager *pPager;        /* The page cache */
  BtCursor *pCursor;    /* A list of all open cursors */
  PageOne *page1;       /* First page of the database */
  u8 inTrans;           /* True if a transaction is in progress */
  u8 inCkpt;            /* True if there is a checkpoint on the transaction */
  u8 readOnly;          /* True if the underlying file is readonly */

  u8 needSwab;          /* Need to byte-swapping */
};
typedef Btree Bt;

/*
** A cursor is a pointer to a particular entry in the BTree.
** The entry is identified by its MemPage and the index in
** MemPage.apCell[] of the entry.

#define SKIP_PREV     2   /* The next sqliteBtreePrevious() is a no-op */
#define SKIP_INVALID  3   /* Calls to Next() and Previous() are invalid */

/* Forward declarations */
static int fileBtreeCloseCursor(BtCursor *pCur);

/*


























** Routines for byte swapping.



*/






u16 swab16(u16 x){
  return ((x & 0xff)<<8) | ((x>>8)&0xff);
}
u32 swab32(u32 x){











  return ((x & 0xff)<<24) | ((x & 0xff00)<<8) |
         ((x>>8) & 0xff00) | ((x>>24)&0xff);
}

/*
** Compute the total number of bytes that a Cell needs on the main
** database page.  The number returned includes the Cell header,
** local payload storage, and the pointer to overflow pages (if
** applicable).  Additional space allocated on overflow pages
** is NOT included in the value returned from this routine.
*/
static int cellSize(Btree *pBt, Cell *pCell){
  int n = NKEY(pBt, pCell->h) + NDATA(pBt, pCell->h);


  if( n>MX_LOCAL_PAYLOAD ){
    n = MX_LOCAL_PAYLOAD + sizeof(Pgno);
  }else{
    n = ROUNDUP(n);
  }









  n += sizeof(CellHdr);






  return n;
}

/*
** Defragment the page given.  All Cells are moved to the
** beginning of the page and all free space is collected 
** into one big FreeBlk at the end of the page.
*/
static void defragmentPage(Btree *pBt, MemPage *pPage){
  int pc, i, n;


  FreeBlk *pFBlk;
  char newPage[SQLITE_USABLE_SIZE];

  assert( sqlitepager_iswriteable(pPage) );
  assert( pPage->isInit );
  pc = sizeof(PageHdr);
  pPage->u.hdr.firstCell = SWAB16(pBt, pc);
  memcpy(newPage, pPage->u.aDisk, pc);


  for(i=0; i<pPage->nCell; i++){
    Cell *pCell = pPage->apCell[i];

    /* This routine should never be called on an overfull page.  The
    ** following asserts verify that constraint. */

    assert( Addr(pCell) > Addr(pPage) );
    assert( Addr(pCell) < Addr(pPage) + SQLITE_USABLE_SIZE );



    n = cellSize(pBt, pCell);
    pCell->h.iNext = SWAB16(pBt, pc + n);

    memcpy(&newPage[pc], pCell, n);
    pPage->apCell[i] = (Cell*)&pPage->u.aDisk[pc];
    pc += n;


  }


  assert( pPage->nFree==SQLITE_USABLE_SIZE-pc );

  memcpy(pPage->u.aDisk, newPage, pc);

  if( pPage->nCell>0 ){
    pPage->apCell[pPage->nCell-1]->h.iNext = 0;
  }


  pFBlk = (FreeBlk*)&pPage->u.aDisk[pc];
  pFBlk->iSize = SWAB16(pBt, SQLITE_USABLE_SIZE - pc);


  pFBlk->iNext = 0;
  pPage->u.hdr.firstFree = SWAB16(pBt, pc);
  memset(&pFBlk[1], 0, SQLITE_USABLE_SIZE - pc - sizeof(FreeBlk));

}

/*
** Allocate nByte bytes of space on a page.  nByte must be a 
** multiple of 4.
**
** Return the index into pPage->u.aDisk[] of the first byte of
** the new allocation. Or return 0 if there is not enough free
** space on the page to satisfy the allocation request.
**
** If the page contains nBytes of free space but does not contain
** nBytes of contiguous free space, then this routine automatically
** calls defragementPage() to consolidate all free space before 
** allocating the new chunk.



*/
static int allocateSpace(Btree *pBt, MemPage *pPage, int nByte){
  FreeBlk *p;
  u16 *pIdx;
  int start;
  int iSize;

#ifndef NDEBUG
  int cnt = 0;
#endif


  assert( sqlitepager_iswriteable(pPage) );
  assert( nByte==ROUNDUP(nByte) );
  assert( pPage->isInit );

  if( pPage->nFree<nByte || pPage->isOverfull ) return 0;
  pIdx = &pPage->u.hdr.firstFree;






  p = (FreeBlk*)&pPage->u.aDisk[SWAB16(pBt, *pIdx)];
  while( (iSize = SWAB16(pBt, p->iSize))<nByte ){


    assert( cnt++ < SQLITE_USABLE_SIZE/4 );

    if( p->iNext==0 ){

      defragmentPage(pBt, pPage);
      pIdx = &pPage->u.hdr.firstFree;
    }else{
      pIdx = &p->iNext;
    }
    p = (FreeBlk*)&pPage->u.aDisk[SWAB16(pBt, *pIdx)];
  }
  if( iSize==nByte ){
    start = SWAB16(pBt, *pIdx);
    *pIdx = p->iNext;





  }else{
    FreeBlk *pNew;
    start = SWAB16(pBt, *pIdx);
    pNew = (FreeBlk*)&pPage->u.aDisk[start + nByte];
    pNew->iNext = p->iNext;
    pNew->iSize = SWAB16(pBt, iSize - nByte);

    *pIdx = SWAB16(pBt, start + nByte);
  }
  pPage->nFree -= nByte;

  return start;
}

/*
** Return a section of the MemPage.u.aDisk[] to the freelist.
** The first byte of the new free block is pPage->u.aDisk[start]
** and the size of the block is "size" bytes.  Size must be
** a multiple of 4.
**
** Most of the effort here is involved in coalesing adjacent
** free blocks into a single big free block.
*/
static void freeSpace(Btree *pBt, MemPage *pPage, int start, int size){
  int end = start + size;





  u16 *pIdx, idx;





  FreeBlk *pFBlk;












  FreeBlk *pNew;
  FreeBlk *pNext;
  int iSize;


  assert( sqlitepager_iswriteable(pPage) );
  assert( size == ROUNDUP(size) );
  assert( start == ROUNDUP(start) );
  assert( pPage->isInit );
  pIdx = &pPage->u.hdr.firstFree;
  idx = SWAB16(pBt, *pIdx);

  while( idx!=0 && idx<start ){
    pFBlk = (FreeBlk*)&pPage->u.aDisk[idx];
    iSize = SWAB16(pBt, pFBlk->iSize);



    if( idx + iSize == start ){
      pFBlk->iSize = SWAB16(pBt, iSize + size);
      if( idx + iSize + size == SWAB16(pBt, pFBlk->iNext) ){


        pNext = (FreeBlk*)&pPage->u.aDisk[idx + iSize + size];

        if( pBt->needSwab ){
          pFBlk->iSize = swab16((u16)swab16(pNext->iSize)+iSize+size);




















        }else{

          pFBlk->iSize += pNext->iSize;
        }
        pFBlk->iNext = pNext->iNext;



      }
      pPage->nFree += size;

      return;
    }






    pIdx = &pFBlk->iNext;
    idx = SWAB16(pBt, *pIdx);
  }
  pNew = (FreeBlk*)&pPage->u.aDisk[start];


  if( idx != end ){





    pNew->iSize = SWAB16(pBt, size);
    pNew->iNext = SWAB16(pBt, idx);



  }else{

    pNext = (FreeBlk*)&pPage->u.aDisk[idx];
    pNew->iSize = SWAB16(pBt, size + SWAB16(pBt, pNext->iSize));
    pNew->iNext = pNext->iNext;




  }





  *pIdx = SWAB16(pBt, start);

  pPage->nFree += size;
}


/*
** Initialize the auxiliary information for a disk block.
**
** The pParent parameter must be a pointer to the MemPage which
** is the parent of the page being initialized.  The root of the
** BTree (usually page 2) has no parent and so for that page, 
** pParent==NULL.
**
** Return SQLITE_OK on success.  If we see that the page does
** not contain a well-formed database page, then return 
** SQLITE_CORRUPT.  Note that a return of SQLITE_OK does not
** guarantee that the page is well-formed.  It only shows that
** we failed to detect any corruption.
*/
static int initPage(Bt *pBt, MemPage *pPage, Pgno pgnoThis, MemPage *pParent){


  int idx;           /* An index into pPage->u.aDisk[] */


  Cell *pCell;       /* A pointer to a Cell in pPage->u.aDisk[] */
  FreeBlk *pFBlk;    /* A pointer to a free block in pPage->u.aDisk[] */
  int sz;            /* The size of a Cell in bytes */
  int freeSpace;     /* Amount of free space on the page */



  if( pPage->pParent ){
    assert( pPage->pParent==pParent );
    return SQLITE_OK;
  }
  if( pParent ){
    pPage->pParent = pParent;
    sqlitepager_ref(pParent);
  }
  if( pPage->isInit ) return SQLITE_OK;
  pPage->isInit = 1;
  pPage->nCell = 0;
  freeSpace = USABLE_SPACE;





  idx = SWAB16(pBt, pPage->u.hdr.firstCell);


  while( idx!=0 ){
    if( idx>SQLITE_USABLE_SIZE-MIN_CELL_SIZE ) goto page_format_error;

    if( idx<sizeof(PageHdr) ) goto page_format_error;



    if( idx!=ROUNDUP(idx) ) goto page_format_error;




    pCell = (Cell*)&pPage->u.aDisk[idx];

    sz = cellSize(pBt, pCell);
    if( idx+sz > SQLITE_USABLE_SIZE ) goto page_format_error;
    freeSpace -= sz;
    pPage->apCell[pPage->nCell++] = pCell;
    idx = SWAB16(pBt, pCell->h.iNext);
  }


  pPage->nFree = 0;
  idx = SWAB16(pBt, pPage->u.hdr.firstFree);

  while( idx!=0 ){
    int iNext;
    if( idx>SQLITE_USABLE_SIZE-sizeof(FreeBlk) ) goto page_format_error;

    if( idx<sizeof(PageHdr) ) goto page_format_error;
    pFBlk = (FreeBlk*)&pPage->u.aDisk[idx];

    pPage->nFree += SWAB16(pBt, pFBlk->iSize);
    iNext = SWAB16(pBt, pFBlk->iNext);
    if( iNext>0 && iNext <= idx ) goto page_format_error;
    idx = iNext;
  }
  if( pPage->nCell==0 && pPage->nFree==0 ){
    /* As a special case, an uninitialized root page appears to be
    ** an empty database */
    return SQLITE_OK;
  }


  if( pPage->nFree!=freeSpace ) goto page_format_error;
  return SQLITE_OK;

page_format_error:
  return SQLITE_CORRUPT;
}

/*
** Set up a raw page so that it looks like a database page holding
** no entries.
*/
static void zeroPage(Btree *pBt, MemPage *pPage){

  PageHdr *pHdr;


  FreeBlk *pFBlk;
  assert( sqlitepager_iswriteable(pPage) );
  memset(pPage, 0, SQLITE_USABLE_SIZE);
  pHdr = &pPage->u.hdr;
  pHdr->firstCell = 0;
  pHdr->firstFree = SWAB16(pBt, sizeof(*pHdr));

  pFBlk = (FreeBlk*)&pHdr[1];
  pFBlk->iNext = 0;
  pPage->nFree = SQLITE_USABLE_SIZE - sizeof(*pHdr);
  pFBlk->iSize = SWAB16(pBt, pPage->nFree);
  pPage->nCell = 0;




  pPage->isOverfull = 0;
}

/*
** This routine is called when the reference count for a page
** reaches zero.  We need to unref the pParent pointer when that
** happens.
*/
static void pageDestructor(void *pData){
  MemPage *pPage = (MemPage*)pData;
  if( pPage->pParent ){
    MemPage *pParent = pPage->pParent;
    pPage->pParent = 0;
    sqlitepager_unref(pParent);
  }


}

/*
** Open a new database.
**
** Actually, this routine just sets up the internal data structures
** for accessing the database.  We do not open the database file 

  int rc;

  /*
  ** The following asserts make sure that structures used by the btree are
  ** the right size.  This is to guard against size changes that result
  ** when compiling on a different architecture.
  */

  assert( sizeof(u32)==4 );
  assert( sizeof(u16)==2 );
  assert( sizeof(Pgno)==4 );
  assert( sizeof(PageHdr)==8 );
  assert( sizeof(CellHdr)==12 );
  assert( sizeof(FreeBlk)==4 );
  assert( sizeof(OverflowPage)==SQLITE_USABLE_SIZE );
  assert( sizeof(FreelistInfo)==OVERFLOW_SIZE );
  assert( sizeof(ptr)==sizeof(char*) );
  assert( sizeof(uptr)==sizeof(ptr) );

  pBt = sqliteMalloc( sizeof(*pBt) );
  if( pBt==0 ){
    *ppBtree = 0;
    return SQLITE_NOMEM;

    return rc;
  }
  sqlitepager_set_destructor(pBt->pPager, pageDestructor);
  pBt->pCursor = 0;
  pBt->page1 = 0;
  pBt->readOnly = sqlitepager_isreadonly(pBt->pPager);
  pBt->pOps = &sqliteBtreeOps;


  *ppBtree = pBt;
  return SQLITE_OK;
}

/*
** Close an open database and invalidate all cursors.
*/

** well-formed database file, then SQLITE_CORRUPT is returned.
** SQLITE_BUSY is returned if the database is locked.  SQLITE_NOMEM
** is returned if we run out of memory.  SQLITE_PROTOCOL is returned
** if there is a locking protocol violation.
*/
static int lockBtree(Btree *pBt){
  int rc;

  if( pBt->page1 ) return SQLITE_OK;
  rc = sqlitepager_get(pBt->pPager, 1, (void**)&pBt->page1);
  if( rc!=SQLITE_OK ) return rc;

  /* Do some checking to help insure the file we opened really is
  ** a valid database file. 
  */
  if( sqlitepager_pagecount(pBt->pPager)>0 ){
    PageOne *pP1 = pBt->page1;
    if( strcmp(pP1->zMagic,zMagicHeader)!=0 ||
          (pP1->iMagic!=MAGIC && swab32(pP1->iMagic)!=MAGIC) ){
      rc = SQLITE_NOTADB;
      goto page1_init_failed;
    }
    pBt->needSwab = pP1->iMagic!=MAGIC;
  }

  return rc;

page1_init_failed:
  sqlitepager_unref(pBt->page1);
  pBt->page1 = 0;
  return rc;
}

/*
** If there are no outstanding cursors and we are not in the middle
** of a transaction but there is a read lock on the database, then
** this routine unrefs the first page of the database file which 
** has the effect of releasing the read lock.
**
** If there are any outstanding cursors, this routine is a no-op.
**
** If there is a transaction in progress, this routine is a no-op.
*/
static void unlockBtreeIfUnused(Btree *pBt){
  if( pBt->inTrans==0 && pBt->pCursor==0 && pBt->page1!=0 ){
    sqlitepager_unref(pBt->page1);
    pBt->page1 = 0;
    pBt->inTrans = 0;
    pBt->inCkpt = 0;
  }
}

/*
** Create a new database by initializing the first two pages of the
** file.
*/
static int newDatabase(Btree *pBt){
  MemPage *pRoot;
  PageOne *pP1;
  int rc;
  if( sqlitepager_pagecount(pBt->pPager)>1 ) return SQLITE_OK;
  pP1 = pBt->page1;


  rc = sqlitepager_write(pBt->page1);
  if( rc ) return rc;
  rc = sqlitepager_get(pBt->pPager, 2, (void**)&pRoot);
  if( rc ) return rc;
  rc = sqlitepager_write(pRoot);
  if( rc ){
    sqlitepager_unref(pRoot);
    return rc;
  }
  strcpy(pP1->zMagic, zMagicHeader);
  if( btree_native_byte_order ){
    pP1->iMagic = MAGIC;

    pBt->needSwab = 0;
  }else{
    pP1->iMagic = swab32(MAGIC);
    pBt->needSwab = 1;
  }

  zeroPage(pBt, pRoot);
  sqlitepager_unref(pRoot);
  return SQLITE_OK;
}

/*
** Attempt to start a new transaction.
**
** A transaction must be started before attempting any changes

  int rc;
  BtCursor *pCur;
  if( pBt->inTrans==0 ) return SQLITE_OK;
  pBt->inTrans = 0;
  pBt->inCkpt = 0;
  rc = pBt->readOnly ? SQLITE_OK : sqlitepager_rollback(pBt->pPager);
  for(pCur=pBt->pCursor; pCur; pCur=pCur->pNext){
    if( pCur->pPage && pCur->pPage->isInit==0 ){
      sqlitepager_unref(pCur->pPage);
      pCur->pPage = 0;
    }
  }
  unlockBtreeIfUnused(pBt);
  return rc;
}