pager.c

sqlite 嵌入式数据库的源码

#define PAGER_MAX_PGNO 2147483647

/*
** Enable reference count tracking (for debugging) here:
*/
#ifdef SQLITE_DEBUG
  int pager3_refinfo_enable = 0;
  static void pager_refinfo(PgHdr *p){
    static int cnt = 0;
    if( !pager3_refinfo_enable ) return;
    sqlite3DebugPrintf(
       "REFCNT: %4d addr=%p nRef=%d\n",
       p->pgno, PGHDR_TO_DATA(p), p->nRef
    );
    cnt++;   /* Something to set a breakpoint on */
  }
# define REFINFO(X)  pager_refinfo(X)
#else
# define REFINFO(X)
#endif

/*
** Read a 32-bit integer from the given file descriptor.  Store the integer
** that is read in *pRes.  Return SQLITE_OK if everything worked, or an
** error code is something goes wrong.
**
** All values are stored on disk as big-endian.
*/
static int read32bits(OsFile *fd, u32 *pRes){
  u32 res;
  int rc;
  rc = sqlite3OsRead(fd, &res, sizeof(res));
  if( rc==SQLITE_OK ){
    unsigned char ac[4];
    memcpy(ac, &res, 4);
    res = (ac[0]<<24) | (ac[1]<<16) | (ac[2]<<8) | ac[3];
  }
  *pRes = res;
  return rc;
}

/*
** Write a 32-bit integer into the given file descriptor.  Return SQLITE_OK
** on success or an error code is something goes wrong.
*/
static int write32bits(OsFile *fd, u32 val){
  unsigned char ac[4];
  ac[0] = (val>>24) & 0xff;
  ac[1] = (val>>16) & 0xff;
  ac[2] = (val>>8) & 0xff;
  ac[3] = val & 0xff;
  return sqlite3OsWrite(fd, ac, 4);
}

/*
** Write the 32-bit integer 'val' into the page identified by page header
** 'p' at offset 'offset'.
*/
static void store32bits(u32 val, PgHdr *p, int offset){
  unsigned char *ac;
  ac = &((unsigned char*)PGHDR_TO_DATA(p))[offset];
  ac[0] = (val>>24) & 0xff;
  ac[1] = (val>>16) & 0xff;
  ac[2] = (val>>8) & 0xff;
  ac[3] = val & 0xff;
}

/*
** Read a 32-bit integer at offset 'offset' from the page identified by
** page header 'p'.
*/
static u32 retrieve32bits(PgHdr *p, int offset){
  unsigned char *ac;
  ac = &((unsigned char*)PGHDR_TO_DATA(p))[offset];
  return (ac[0]<<24) | (ac[1]<<16) | (ac[2]<<8) | ac[3];
}


/*
** Convert the bits in the pPager->errMask into an approprate
** return code.
*/
static int pager_errcode(Pager *pPager){
  int rc = SQLITE_OK;
  if( pPager->errMask & PAGER_ERR_LOCK )    rc = SQLITE_PROTOCOL;
  if( pPager->errMask & PAGER_ERR_DISK )    rc = SQLITE_IOERR;
  if( pPager->errMask & PAGER_ERR_FULL )    rc = SQLITE_FULL;
  if( pPager->errMask & PAGER_ERR_MEM )     rc = SQLITE_NOMEM;
  if( pPager->errMask & PAGER_ERR_CORRUPT ) rc = SQLITE_CORRUPT;
  return rc;
}

#ifdef SQLITE_CHECK_PAGES
/*
** Return a 32-bit hash of the page data for pPage.
*/
static u32 pager_pagehash(PgHdr *pPage){
  u32 hash = 0;
  int i;
  unsigned char *pData = (unsigned char *)PGHDR_TO_DATA(pPage);
  for(i=0; i<pPage->pPager->pageSize; i++){
    hash = (hash+i)^pData[i];
  }
  return hash;
}

/*
** The CHECK_PAGE macro takes a PgHdr* as an argument. If SQLITE_CHECK_PAGES
** is defined, and NDEBUG is not defined, an assert() statement checks
** that the page is either dirty or still matches the calculated page-hash.
*/
#define CHECK_PAGE(x) checkPage(x)
static void checkPage(PgHdr *pPg){
  Pager *pPager = pPg->pPager;
  assert( !pPg->pageHash || pPager->errMask || MEMDB || pPg->dirty || 
      pPg->pageHash==pager_pagehash(pPg) );
}

#else
#define CHECK_PAGE(x)
#endif

/*
** When this is called the journal file for pager pPager must be open.
** The master journal file name is read from the end of the file and 
** written into memory obtained from sqliteMalloc(). *pzMaster is
** set to point at the memory and SQLITE_OK returned. The caller must
** sqliteFree() *pzMaster.
**
** If no master journal file name is present *pzMaster is set to 0 and
** SQLITE_OK returned.
*/
static int readMasterJournal(OsFile *pJrnl, char **pzMaster){
  int rc;
  u32 len;
  i64 szJ;
  u32 cksum;
  int i;
  unsigned char aMagic[8]; /* A buffer to hold the magic header */

  *pzMaster = 0;

  rc = sqlite3OsFileSize(pJrnl, &szJ);
  if( rc!=SQLITE_OK || szJ<16 ) return rc;

  rc = sqlite3OsSeek(pJrnl, szJ-16);
  if( rc!=SQLITE_OK ) return rc;
 
  rc = read32bits(pJrnl, &len);
  if( rc!=SQLITE_OK ) return rc;

  rc = read32bits(pJrnl, &cksum);
  if( rc!=SQLITE_OK ) return rc;

  rc = sqlite3OsRead(pJrnl, aMagic, 8);
  if( rc!=SQLITE_OK || memcmp(aMagic, aJournalMagic, 8) ) return rc;

  rc = sqlite3OsSeek(pJrnl, szJ-16-len);
  if( rc!=SQLITE_OK ) return rc;

  *pzMaster = (char *)sqliteMalloc(len+1);
  if( !*pzMaster ){
    return SQLITE_NOMEM;
  }
  rc = sqlite3OsRead(pJrnl, *pzMaster, len);
  if( rc!=SQLITE_OK ){
    sqliteFree(*pzMaster);
    *pzMaster = 0;
    return rc;
  }

  /* See if the checksum matches the master journal name */
  for(i=0; i<len; i++){
    cksum -= (*pzMaster)[i];
  }
  if( cksum ){
    /* If the checksum doesn't add up, then one or more of the disk sectors
    ** containing the master journal filename is corrupted. This means
    ** definitely roll back, so just return SQLITE_OK and report a (nul)
    ** master-journal filename.
    */
    sqliteFree(*pzMaster);
    *pzMaster = 0;
  }else{
    (*pzMaster)[len] = '\0';
  }
   
  return SQLITE_OK;
}

/*
** Seek the journal file descriptor to the next sector boundary where a
** journal header may be read or written. Pager.journalOff is updated with
** the new seek offset.
**
** i.e for a sector size of 512:
**
** Input Offset              Output Offset
** ---------------------------------------
** 0                         0
** 512                       512
** 100                       512
** 2000                      2048
** 
*/
static int seekJournalHdr(Pager *pPager){
  i64 offset = 0;
  i64 c = pPager->journalOff;
  if( c ){
    offset = ((c-1)/JOURNAL_HDR_SZ(pPager) + 1) * JOURNAL_HDR_SZ(pPager);
  }
  assert( offset%JOURNAL_HDR_SZ(pPager)==0 );
  assert( offset>=c );
  assert( (offset-c)<JOURNAL_HDR_SZ(pPager) );
  pPager->journalOff = offset;
  return sqlite3OsSeek(&pPager->jfd, pPager->journalOff);
}

/*
** The journal file must be open when this routine is called. A journal
** header (JOURNAL_HDR_SZ bytes) is written into the journal file at the
** current location.
**
** The format for the journal header is as follows:
** - 8 bytes: Magic identifying journal format.
** - 4 bytes: Number of records in journal, or -1 no-sync mode is on.
** - 4 bytes: Random number used for page hash.
** - 4 bytes: Initial database page count.
** - 4 bytes: Sector size used by the process that wrote this journal.
** 
** Followed by (JOURNAL_HDR_SZ - 24) bytes of unused space.
*/
static int writeJournalHdr(Pager *pPager){

  int rc = seekJournalHdr(pPager);
  if( rc ) return rc;

  pPager->journalHdr = pPager->journalOff;
  if( pPager->stmtHdrOff==0 ){
    pPager->stmtHdrOff = pPager->journalHdr;
  }
  pPager->journalOff += JOURNAL_HDR_SZ(pPager);

  /* FIX ME: 
  **
  ** Possibly for a pager not in no-sync mode, the journal magic should not
  ** be written until nRec is filled in as part of next syncJournal(). 
  **
  ** Actually maybe the whole journal header should be delayed until that
  ** point. Think about this.
  */
  rc = sqlite3OsWrite(&pPager->jfd, aJournalMagic, sizeof(aJournalMagic));

  if( rc==SQLITE_OK ){
    /* The nRec Field. 0xFFFFFFFF for no-sync journals. */
    rc = write32bits(&pPager->jfd, pPager->noSync ? 0xffffffff : 0);
  }
  if( rc==SQLITE_OK ){
    /* The random check-hash initialiser */ 
    sqlite3Randomness(sizeof(pPager->cksumInit), &pPager->cksumInit);
    rc = write32bits(&pPager->jfd, pPager->cksumInit);
  }
  if( rc==SQLITE_OK ){
    /* The initial database size */
    rc = write32bits(&pPager->jfd, pPager->dbSize);
  }
  if( rc==SQLITE_OK ){
    /* The assumed sector size for this process */
    rc = write32bits(&pPager->jfd, pPager->sectorSize);
  }

  /* The journal header has been written successfully. Seek the journal
  ** file descriptor to the end of the journal header sector.
  */
  if( rc==SQLITE_OK ){
    sqlite3OsSeek(&pPager->jfd, pPager->journalOff-1);
    rc = sqlite3OsWrite(&pPager->jfd, "\000", 1);
  }
  return rc;
}

/*
** The journal file must be open when this is called. A journal header file
** (JOURNAL_HDR_SZ bytes) is read from the current location in the journal
** file. See comments above function writeJournalHdr() for a description of
** the journal header format.
**
** If the header is read successfully, *nRec is set to the number of
** page records following this header and *dbSize is set to the size of the
** database before the transaction began, in pages. Also, pPager->cksumInit
** is set to the value read from the journal header. SQLITE_OK is returned
** in this case.
**
** If the journal header file appears to be corrupted, SQLITE_DONE is
** returned and *nRec and *dbSize are not set.  If JOURNAL_HDR_SZ bytes
** cannot be read from the journal file an error code is returned.
*/
static int readJournalHdr(
  Pager *pPager, 
  i64 journalSize,
  u32 *pNRec, 
  u32 *pDbSize
){
  int rc;
  unsigned char aMagic[8]; /* A buffer to hold the magic header */

  rc = seekJournalHdr(pPager);
  if( rc ) return rc;

  if( pPager->journalOff+JOURNAL_HDR_SZ(pPager) > journalSize ){
    return SQLITE_DONE;
  }

  rc = sqlite3OsRead(&pPager->jfd, aMagic, sizeof(aMagic));
  if( rc ) return rc;

  if( memcmp(aMagic, aJournalMagic, sizeof(aMagic))!=0 ){
    return SQLITE_DONE;
  }

  rc = read32bits(&pPager->jfd, pNRec);
  if( rc ) return rc;

  rc = read32bits(&pPager->jfd, &pPager->cksumInit);
  if( rc ) return rc;

  rc = read32bits(&pPager->jfd, pDbSize);
  if( rc ) return rc;

  /* Update the assumed sector-size to match the value used by 
  ** the process that created this journal. If this journal was
  ** created by a process other than this one, then this routine
  ** is being called from within pager_playback(). The local value
  ** of Pager.sectorSize is restored at the end of that routine.
  */
  rc = read32bits(&pPager->jfd, (u32 *)&pPager->sectorSize);
  if( rc ) return rc;

  pPager->journalOff += JOURNAL_HDR_SZ(pPager);
  rc = sqlite3OsSeek(&pPager->jfd, pPager->journalOff);
  return rc;
}


/*
** Write the supplied master journal name into the journal file for pager
** pPager at the current location. The master journal name must be the last
** thing written to a journal file. If the pager is in full-sync mode, the
** journal file descriptor is advanced to the next sector boundary before
** anything is written. The format is:
**
** + 4 bytes: PAGER_MJ_PGNO.
** + N bytes: length of master journal name.
** + 4 bytes: N
** + 4 bytes: Master journal name checksum.
** + 8 bytes: aJournalMagic[].
**
** The master journal page checksum is the sum of the bytes in the master
** journal name.
*/
static int writeMasterJournal(Pager *pPager, const char *zMaster){
  int rc;
  int len; 
  int i; 
  u32 cksum = 0; 

  if( !zMaster || pPager->setMaster) return SQLITE_OK;
  pPager->setMaster = 1;

  len = strlen(zMaster);
  for(i=0; i<len; i++){
    cksum += zMaster[i];
  }

  /* If in full-sync mode, advance to the next disk sector before writing
  ** the master journal name. This is in case the previous page written to
  ** the journal has already been synced.
  */
  if( pPager->fullSync ){
    rc = seekJournalHdr(pPager);
    if( rc!=SQLITE_OK ) return rc;
  }
  pPager->journalOff += (len+20);