SQLite4  Check-in [8341d438d3]

typedef struct BtDbHdr BtDbHdr;
struct BtDbHdr {
  u32 pgsz;                       /* Page size in bytes */
  u32 nPg;                        /* Size of database file in pages */
  u32 iRoot;                      /* B-tree root page */
  u32 iCookie;                    /* Current value of schema cookie */

  u32 iFreePg;                    /* First page in free-page list trunk */
  u32 iFreeBlk;                   /* First page in free-block list trunk */
};

/*************************************************************************
** Interface to bt_pager.c functionality.
*/

** End of utility interface.
*************************************************************************/

#ifdef NDEBUG
# define sqlite4BtDebugReadPage(a,b,c,d)
# define sqlite4BtDebugKV(a,b,c,d,e,f)
# define sqlite4BtDebugReadlock(a,b,c)
# define sqlite4BtDebugPageAlloc(a,b,c)
# define sqlite4BtDebugPageFree(a,b,c)
# define btErrorBkpt(x) x
#else
void sqlite4BtDebugReadPage(BtLock *pLock, u32 pgno, u8 *aData, int pgsz);
void sqlite4BtDebugKV(BtLock*, const char*,u8 *pK, int nK, u8 *pV, int nV);
void sqlite4BtDebugReadlock(BtLock *pLock, u32 iFirst, u32 iLast);
void sqlite4BtDebugPageAlloc(BtLock *pLock, const char*, u32);
void sqlite4BtDebugPageFree(BtLock *pLock, const char*, u32);
int btErrorBkpt(int rc);
#endif

** VFS method).
*/
struct BtWalHdr {
  u32 iMagic;                     /* Magic number (BT_WAL_MAGIC) */
  u32 iVersion;                   /* File format version */
  u32 iCnt;                       /* 0, 1 or 2 */
  u32 nSector;                    /* Sector size when header written */

  u32 nPgsz;                      /* Database page size in bytes */
  u32 nPg;                        /* Database size in pages at last commit */

  u32 padding;

  u32 iSalt1;                     /* Initial frame cksum-0 value */
  u32 iSalt2;                     /* Initial frame cksum-1 value */

      pLock->iDebugId, zStr, iSide,
      (int)aLog[0], (int)aLog[1], (int)aLog[2], 
      (int)aLog[3], (int)aLog[4], (int)aLog[5]
  );
  fflush(stderr);
#endif
}

static void btDebugDbhdr(BtLock *pLock, const char *zStr, BtDbHdr *pHdr){
#if BT_HDR_DEBUG
  static int nCall = 0;
  fprintf(stderr, "%d:%d: %s db-header "
      "(pgsz=%d nPg=%d iRoot=%d iCookie=%d iFreePg=%d iFreeBlk=%d)\n",
      pLock->iDebugId, nCall++,
      zStr, (int)pHdr->pgsz, (int)pHdr->nPg, (int)pHdr->iRoot, 
      (int)pHdr->iCookie, (int)pHdr->iFreePg, (int)pHdr->iFreeBlk
  );
  fflush(stderr);
#endif
}

#ifndef NDEBUG
void sqlite4BtDebugReadlock(BtLock *pLock, u32 iFirst, u32 iLast){
#if BT_PAGE_DEBUG
  static int nCall = 0;
  fprintf(stderr, "%d:%d: readlock=(%d..%d)\n",
      pLock->iDebugId, nCall++, (int)iFirst, (int)iLast
  );
  fflush(stderr);
#endif
}
#endif

#ifndef NDEBUG
void sqlite4BtDebugPageAlloc(BtLock *pLock, const char *zStr, u32 pgno){
#if BT_PAGE_DEBUG
  static int nCall = 0;
  fprintf(stderr, "%d:%d: allocate page %d from %s\n", 
      pLock->iDebugId, nCall++, pgno, zStr
  );
  fflush(stderr);
#endif
}
#endif

#ifndef NDEBUG
void sqlite4BtDebugPageFree(BtLock *pLock, const char *zStr, u32 pgno){
#if BT_PAGE_DEBUG
  static int nCall = 0;
  fprintf(stderr, "%d:%d: trim page %d (%s)\n", 
      pLock->iDebugId, nCall++, pgno, zStr
  );
  fflush(stderr);
  assert( pgno<1000000 );
#endif
}
#endif


#ifndef NDEBUG
static void btDebugCheckSnapshot(BtShmHdr *pHdr){
  u32 *aLog = pHdr->aLog;
  assert( pHdr->iNextFrame!=1 ||
      (aLog[0]==0 && aLog[1]==0 && aLog[2]==0 && aLog[3]==0)
  );

  return (pLog->snapshot.aLog[4]==0 && pLog->snapshot.iNextFrame==0);
}

typedef struct FrameRecoverCtx FrameRecoverCtx;
struct FrameRecoverCtx {
  u32 iLast;                      /* Frame containing last commit flag in log */
  u32 iNextFrame;                 /* Frame that follows frame iLast */
  u32 iPageOneFrame;              /* Frame containing most recent page 1 */
};

static int btLogRecoverFrame(
  BtLog *pLog,                    /* Log module handle */
  void *pCtx,                     /* woints to type u32 - pgno of last commit*/
  u32 iFrame,                     /* Frame number */
  BtFrameHdr *pHdr                /* Frame header */
){
  FrameRecoverCtx *pFRC = (FrameRecoverCtx*)pCtx;

  if( btLogIsEmpty(pLog) ){
    /* This is the first frame recovered. It is therefore both the first
    ** and last frame of log region (c).  */
    pLog->snapshot.aLog[4] = iFrame;
    pLog->snapshot.aLog[5] = iFrame;
  }else{

      pLog->snapshot.aLog[4] = iFrame;
      pLog->snapshot.aLog[5] = iFrame;
    }
  }

  btLogHashInsert(pLog, pHdr->pgno, iFrame);
  if( pHdr->nPg!=0 ){

    pFRC->iLast = iFrame;
    pFRC->iNextFrame = pHdr->iNext;
    memcpy(pLog->snapshot.aFrameCksum, pHdr->aCksum, sizeof(pHdr->aCksum));
    pLog->snapshot.dbhdr.nPg = pHdr->nPg;
  }

  if( pHdr->pgno==1 ){
    pFRC->iPageOneFrame = iFrame;
  }

#if 0
  fprintf(stderr, "recovered frame=%d pgno=%d\n", iFrame, pHdr->pgno);
  fflush(stderr);
#endif
  return 0;
}

    }
  }

  aLog[5] = iLast;
  return btLogHashRollback(pLog, btLogFrameHash(pLog, iLast), iLast);
}

static void btLogDecodeDbhdr(BtLog *pLog, u8 *aData, BtDbHdr *pHdr){
  BtDbHdrCksum hdr;
  u32 aCksum[2];

  if( aData ){
    memcpy(&hdr, aData, sizeof(BtDbHdrCksum));
    btLogChecksum(1, (u8*)&hdr, offsetof(BtDbHdrCksum, aCksum), 0, aCksum);
  }

  if( aData==0 || aCksum[0]!=hdr.aCksum[0] || aCksum[1]!=hdr.aCksum[1] ){
    memset(&hdr, 0, sizeof(BtDbHdrCksum));
    hdr.hdr.pgsz = BT_DEFAULT_PGSZ;
    hdr.hdr.nPg = 2;
    hdr.hdr.iRoot = 2;
  }

  memcpy(pHdr, &hdr, sizeof(BtDbHdr));
}

static int btLogReadDbhdr(BtLog *pLog, BtDbHdr *pHdr, u32 iFrame){
  BtLock *p = pLog->pLock;        /* BtLock handle */
  int rc;                         /* Return code */
  i64 nByte;                      /* Size of database file in byte */
  u8 *aBuffer[sizeof(BtDbHdrCksum)];
  u8 *aData = 0;

  if( iFrame==0 ){
    rc = p->pVfs->xSize(p->pFd, &nByte);
    if( rc==SQLITE4_OK && nByte>0 ){

      rc = p->pVfs->xRead(p->pFd, 0, aBuffer, sizeof(BtDbHdrCksum));

      aData = aBuffer;

    }



  }else{
    i64 iOff = btLogFrameOffset(pLog, pLog->snapshot.dbhdr.pgsz, iFrame);
    iOff += sizeof(BtFrameHdr);
    rc = p->pVfs->xRead(pLog->pFd, iOff, aBuffer, sizeof(BtDbHdrCksum));
    aData = aBuffer;
  }

  if( rc==SQLITE4_OK ){


    btLogDecodeDbhdr(pLog, aData, pHdr);

  }
  return rc;
}

static int btLogUpdateDbhdr(BtLog *pLog, u8 *aData){
  BtDbHdrCksum hdr;

  memcpy(&hdr.hdr, &pLog->snapshot.dbhdr, sizeof(BtDbHdr));

  btLogChecksum(1, (u8*)&hdr, offsetof(BtDbHdrCksum, aCksum), 0, hdr.aCksum);
  btDebugDbhdr(pLog->pLock, "update", &pLog->snapshot.dbhdr);

  assert( hdr.hdr.iRoot==2 );
  assert( hdr.hdr.pgsz>0 );
  memcpy(aData, &hdr, sizeof(BtDbHdrCksum));

  return SQLITE4_OK;
}

      pShm->ckpt.iFirstRead = pHdr->iFirstFrame;
      pShm->ckpt.iFirstRecover = pHdr->iFirstFrame;
      rc = btLogRollbackRecovery(pLog, &ctx);
      pLog->snapshot.iNextFrame = ctx.iNextFrame;
      pLog->snapshot.dbhdr.pgsz = pHdr->nPgsz;
      assert( pShm->ckpt.iFirstRead>0 );
    }

    /* Based on the wal-header, the page-size and number of pages in the
    ** database are now known and stored in snapshot.dbhdr. But the other
    ** header field values (iCookie, iRoot etc.) are still unknown. Read
    ** them from page 1 of the database file now.  */
    rc = btLogReadDbhdr(pLog, &pLog->snapshot.dbhdr, ctx.iPageOneFrame);

  }else if( rc==SQLITE4_OK ){
    /* There is no data in the log file. Read the database header directly
    ** from offset 0 of the database file.  */
    btLogZeroSnapshot(pLog);



    rc = btLogReadDbhdr(pLog, &pLog->snapshot.dbhdr, 0);
  }

  if( rc==SQLITE4_OK ){
    btDebugTopology(
        pLog->pLock, "recovered", pLog->snapshot.iHashSide, pLog->snapshot.aLog
    );

    btDebugDbhdr(pLog->pLock, "read", &pLog->snapshot.dbhdr);
  }
  return rc;
}

/*
** Open the log file for pager pPager. If successful, return the BtLog* 
** handle via output variable *ppLog. If parameter bRecover is true, then

  BtPager *pPager, 
  u32 pgno, 
  int nDepth
){
  int rc = SQLITE4_OK;

  assert( nDepth<=8 );
  if( nDepth>0 ){


    const int nPgPtr = pgsz / 4;
    BtPage *pPg;
    u8 *aData;
    int i;

    rc = sqlite4BtPageGet(pPager, pgno, &pPg);
    if( rc!=SQLITE4_OK ) return rc;
    aData = sqlite4BtPageData(pPg);

    for(i=0; rc==SQLITE4_OK && i<nPgPtr; i++){
      u32 child = btGetU32(&aData[i*4]);
      if( child==0 ) break;
      rc = btOverflowTrimtree(pgsz, pPager, child, nDepth-1);
    }

    sqlite4BtPageRelease(pPg);
  }
  
  sqlite4BtPageTrimPgno(pPager, pgno);
  return rc;
}

/*
** Cursor pCsr currently points to a leaf page cell. If the leaf page
** cell contains an overflow array, all overflow pages are trimmed here.
**

  int nPg;
  nPg = (nContent + pgsz - 1) / pgsz;
  if( nPg<=BT_MAX_DIRECT_OVERFLOW ){
    return 1 + nPg*4;
  }
  return 1 + (BT_MAX_DIRECT_OVERFLOW+1) * 4;
}

static int btAllocateAndZero(bt_db *db, BtPage **ppPg){
  BtPage *pPg = 0;                /* Allocated page handle */
  int rc;                         /* Return code */

  rc = sqlite4BtPageAllocate(db->pPager, &pPg);
  if( rc==SQLITE4_OK ){
    const int pgsz = sqlite4BtPagerPagesize(db->pPager);
    memset(sqlite4BtPageData(pPg), 0, pgsz);
  }

  *ppPg = pPg;
  return rc;
}

static int btOverflowArrayPopulate(
  bt_db *db, u8 **ppOut,
  u8 *pBuf1, int nBuf1,
  u8 *pBuf2, int nBuf2
){
  const int pgsz = sqlite4BtPagerPagesize(db->pPager);

  while( nOvfl>1 ){
    nDepth++;
    nOvfl = (nOvfl+nPgPtr-1) / nPgPtr;
  }

  for(i=0; rc==SQLITE4_OK && i<nDepth; i++){
    u32 pgno;
    rc = btAllocateAndZero(db, &apHier[i].pPg);
    pgno = sqlite4BtPagePgno(apHier[i].pPg);
    if( i==0 ){
      btPutU32(&aOut[1 + BT_MAX_DIRECT_OVERFLOW*4], pgno);
    }else{
      u8 *a = sqlite4BtPageData(apHier[i-1].pPg);
      btPutU32(a, pgno);
      apHier[i-1].iCell++;

      assert( nDepth>0 );
      for(i=nDepth-1; pgno && i>=0; i--){
        u8 *a = sqlite4BtPageData(apHier[i].pPg);
        if( apHier[i].iCell==nPgPtr ){
          BtPage *pNew = 0;
          rc = sqlite4BtPageRelease(apHier[i].pPg);
          if( rc==SQLITE4_OK ){
            rc = btAllocateAndZero(db, &pNew);
            if( rc==SQLITE4_OK ){
              u8 *a = sqlite4BtPageData(pNew);
              btPutU32(a, pgno);
              pgno = sqlite4BtPagePgno(pNew);
            }
          }

    case BT_CONTROL_LOGSIZECB: {
      bt_logsizecb *p = (bt_logsizecb*)pArg;
      sqlite4BtPagerLogsizeCb(db->pPager, p);
      break;
    }

    case BT_CONTROL_CHECKPOINT: {
      bt_checkpoint *p = (bt_checkpoint*)pArg;
      rc = sqlite4BtPagerCheckpoint(db->pPager, p);
      break;
    }
  }

  return rc;
}

        if( iLevel<2 ) sqlite4BtLogSnapshotEndWrite(p->pLog);
      }else{
        rc = btCloseSavepoints(p, iLevel-1, 1);
        p->nSavepoint++;
      }
    }

    if( p->iTransactionLevel>iLevel ){
      p->iTransactionLevel = iLevel;
      if( iLevel==0 ){
        int rc2 = btCloseReadTransaction(p);
        if( rc==SQLITE4_OK ) rc = rc2;
      }
    }
  }

  return rc;
}

int sqlite4BtPagerRevert(BtPager *p, int iLevel){

      u8 *aData = pTrunk->aData;
      int nFree = (int)sqlite4BtGetU32(aData);

      if( nFree<nMax ){
        sqlite4BtPutU32(&pTrunk->aData[8 + nFree*4], pgno);
        sqlite4BtPutU32(pTrunk->aData, nFree+1);
        bDone = 1;
        sqlite4BtDebugPageFree((BtLock*)p, "free-list", pgno);
      }

      sqlite4BtPageRelease(pTrunk);
    } 
  }

  /* If no error has occurred but the page number has not yet been added 
  ** to the free-list, this page becomes the first trunk in the list.  */
  if( rc==SQLITE4_OK && bDone==0 ){
    BtPage *pRelease = 0;
    if( pPg==0 ){
      rc = sqlite4BtPageGet(p, pgno, &pRelease);
    }
    if( rc==SQLITE4_OK ){
      BtPage *pTrunk = pPg ? pPg : pRelease;
      rc = sqlite4BtPageWrite(pTrunk);
      if( rc==SQLITE4_OK ){
        sqlite4BtPutU32(&pTrunk->aData[0], 0);
        sqlite4BtPutU32(&pTrunk->aData[4], pHdr->iFreePg);
        pHdr->iFreePg = pgno;
        sqlite4BtDebugPageFree((BtLock*)p, "free-list-trunk", pgno);
      }
    }
    sqlite4BtPageRelease(pRelease);
  }

  return rc;
}

    if( rc==SQLITE4_OK ){
      u8 *aData = pTrunk->aData;
      u32 nFree = sqlite4BtGetU32(aData);
      if( nFree==0 ){
        u32 iNext = sqlite4BtGetU32(&aData[4]);
        *pPgno = pHdr->iFreePg;
        pHdr->iFreePg = iNext;
        sqlite4BtDebugPageAlloc((BtLock*)p, "free-list-trunk", *pPgno);
      }else{
        *pPgno = sqlite4BtGetU32(&aData[8 + 4*(nFree-1)]);
        sqlite4BtPutU32(aData, nFree-1);
        sqlite4BtDebugPageAlloc((BtLock*)p, "free-list", *pPgno);
        assert( *pPgno < 1000000 );
      }
    }

    sqlite4BtPageRelease(pTrunk);
  }

  return rc;

  /* Find the page number of the new page. There are two ways a page may
  ** be allocated - from the free-list or by appending it to the end of
  ** the database file. */ 
  rc = btFreelistAlloc(p, &pgno);
  if( rc==SQLITE4_OK && pgno==0 ){
    pgno = p->pHdr->nPg+1;
    sqlite4BtDebugPageAlloc((BtLock*)p, "end-of-file", pgno);
  }

  rc = sqlite4BtPageGet(p, pgno, &pPg);
  if( rc==SQLITE4_OK ){
    rc = sqlite4BtPageWrite(pPg);
    if( rc!=SQLITE4_OK ){
      sqlite4BtPageRelease(pPg);