Many hyperlinks are disabled.
Use anonymous login
to enable hyperlinks.
Overview
Comment: | Add missing calls to xSync(). Fix a problem with recovering wrapped logs. |
---|---|
Downloads: | Tarball | ZIP archive |
Timelines: | family | ancestors | descendants | both | trunk |
Files: | files | file ages | folders |
SHA1: |
93af0d7d05c2f0022b420a37f0a70672 |
User & Date: | dan 2013-11-08 17:50:48.431 |
Context
2013-11-08
| ||
20:18 | Fix a robustness related problem with wrapped logs. check-in: 7dd7b942fd user: dan tags: trunk | |
17:50 | Add missing calls to xSync(). Fix a problem with recovering wrapped logs. check-in: 93af0d7d05 user: dan tags: trunk | |
2013-11-07
| ||
20:22 | Begin adding tests to check that bt database transactions are robust in the face of system failure. check-in: 5d57889261 user: dan tags: trunk | |
Changes
Changes to lsm-test/lsmtest2.c.
︙ | ︙ | |||
432 433 434 435 436 437 438 | /* ** This test verifies that if a system crash occurs while committing a ** transaction to the log file, no earlier transactions are lost or damaged. */ static void crash_test2b(int bCompress, int *pRc){ const char *DBNAME = "testdb.bt"; | | | | 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 | /* ** This test verifies that if a system crash occurs while committing a ** transaction to the log file, no earlier transactions are lost or damaged. */ static void crash_test2b(int bCompress, int *pRc){ const char *DBNAME = "testdb.bt"; const DatasourceDefn defn = {TEST_DATASOURCE_RANDOM, 12, 16, 800, 800}; const int nIter = 200; const int nInsert = 200; int i; int iDot = 0; Datasource *pData; CksumDb *pCksumDb; TestDb *pDb; |
︙ | ︙ | |||
475 476 477 478 479 480 481 | testrc = tdb_write(pDb, pKey, nKey, pVal, nVal); if( testrc ) break; } tdb_close(pDb); /* Check that no data was lost when the system crashed. */ testCompareCksumBtdb(DBNAME, | | | > | 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 | testrc = tdb_write(pDb, pKey, nKey, pVal, nVal); if( testrc ) break; } tdb_close(pDb); /* Check that no data was lost when the system crashed. */ testCompareCksumBtdb(DBNAME, testCksumArrayGet(pCksumDb, 100 + iIns - (testrc==0)), testCksumArrayGet(pCksumDb, 100 + iIns + 1 - (testrc==0)), pRc ); } testDatasourceFree(pData); testCksumArrayFree(pCksumDb); } /* |
︙ | ︙ |
Changes to lsm-test/lsmtest_tdb4.c.
︙ | ︙ | |||
98 99 100 101 102 103 104 | static int btVfsRead(bt_file *pFile, sqlite4_int64 iOff, void *pBuf, int nBuf){ BtFile *p = (BtFile*)pFile; if( p->pBt->bCrash ) return SQLITE4_IOERR; return p->pBt->pVfs->xRead(p->pFile, iOff, pBuf, nBuf); } | | | 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 | static int btVfsRead(bt_file *pFile, sqlite4_int64 iOff, void *pBuf, int nBuf){ BtFile *p = (BtFile*)pFile; if( p->pBt->bCrash ) return SQLITE4_IOERR; return p->pBt->pVfs->xRead(p->pFile, iOff, pBuf, nBuf); } static int btFlushSectors(BtFile *p, int iFile){ sqlite4_int64 iSz; int rc; int i; u8 *aTmp = 0; rc = p->pBt->pVfs->xSize(p->pFile, &iSz); for(i=0; rc==SQLITE4_OK && i<p->nSector; i++){ |
︙ | ︙ | |||
130 131 132 133 134 135 136 | }else if( iOpt==3 ){ if( aTmp==0 ) aTmp = testMalloc(p->nSectorSize); aWrite = aTmp; testPrngArray(i*13, (u32*)aWrite, nWrite/sizeof(u32)); } #if 0 | | > | 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 | }else if( iOpt==3 ){ if( aTmp==0 ) aTmp = testMalloc(p->nSectorSize); aWrite = aTmp; testPrngArray(i*13, (u32*)aWrite, nWrite/sizeof(u32)); } #if 0 fprintf(stderr, "handle sector %d of %s with %s\n", i, iFile==0 ? "db" : "log", iOpt==1 ? "rollback" : iOpt==2 ? "write" : "omit" ); fflush(stderr); #endif if( aWrite ){ rc = p->pBt->pVfs->xWrite(p->pFile, iSOff, aWrite, nWrite); |
︙ | ︙ | |||
160 161 162 163 164 165 166 | int iSector; /* Current sector */ int iLast; /* Last sector affected */ if( p->nSectorSize==0 ){ p->nSectorSize = p->pBt->pVfs->xSectorSize(p->pFile); if( p->nSectorSize<512 ) p->nSectorSize = 512; } | | | 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 | int iSector; /* Current sector */ int iLast; /* Last sector affected */ if( p->nSectorSize==0 ){ p->nSectorSize = p->pBt->pVfs->xSectorSize(p->pFile); if( p->nSectorSize<512 ) p->nSectorSize = 512; } iLast = (iOff+nBuf-1) / p->nSectorSize; iFirst = iOff / p->nSectorSize; rc = p->pBt->pVfs->xSize(p->pFile, &iSz); for(iSector=iFirst; rc==SQLITE4_OK && iSector<=iLast; iSector++){ int nRead; sqlite4_int64 iSOff = iSector * p->nSectorSize; u8 *aBuf = testMalloc(p->nSectorSize); |
︙ | ︙ | |||
213 214 215 216 217 218 219 | BtDb *pBt = p->pBt; if( p->pBt->bCrash ) return SQLITE4_IOERR; if( pBt->nCrashSync ){ pBt->nCrashSync--; pBt->bCrash = (pBt->nCrashSync==0); if( pBt->bCrash ){ | | | | | | 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 | BtDb *pBt = p->pBt; if( p->pBt->bCrash ) return SQLITE4_IOERR; if( pBt->nCrashSync ){ pBt->nCrashSync--; pBt->bCrash = (pBt->nCrashSync==0); if( pBt->bCrash ){ btFlushSectors(pBt->apFile[0], 0); btFlushSectors(pBt->apFile[1], 1); rc = SQLITE4_IOERR; }else{ btFlushSectors(p, 0); } } if( rc==SQLITE4_OK ){ rc = p->pBt->pVfs->xSync(p->pFile); } return rc; } static int btVfsSectorSize(bt_file *pFile){ BtFile *p = (BtFile*)pFile; return p->pBt->pVfs->xSectorSize(p->pFile); } static int btVfsClose(bt_file *pFile){ BtFile *p = (BtFile*)pFile; int rc; assert( p->pBt->apFile[0]==p || p->pBt->apFile[1]==p ); btFlushSectors(p, 0); testFree(p->apSector); rc = p->pBt->pVfs->xClose(p->pFile); testFree(p); return rc; } static int btVfsUnlink(sqlite4_env *pEnv, bt_env *pVfs, const char *zFile){ |
︙ | ︙ |
Changes to src/bt_log.c.
︙ | ︙ | |||
240 241 242 243 244 245 246 | u32 *aOut /* OUT: Final checksum value output */ ){ assert( (nByte&0x00000007)==4 && nByte>=8 ); btLogChecksum(nativeCksum, a, 8, aIn, aOut); btLogChecksum(nativeCksum, &a[4], nByte-4, aOut, aOut); } | | | 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 | u32 *aOut /* OUT: Final checksum value output */ ){ assert( (nByte&0x00000007)==4 && nByte>=8 ); btLogChecksum(nativeCksum, a, 8, aIn, aOut); btLogChecksum(nativeCksum, &a[4], nByte-4, aOut, aOut); } #define BT_PAGE_DEBUG 0 #define BT_VAL_DEBUG 0 static void btDebugTopology(BtLock *pLock, char *zStr, int iSide, u32 *aLog){ #if BT_PAGE_DEBUG fprintf(stderr, "%d:%s: (side=%d) %d..%d %d..%d %d..%d\n", pLock->iDebugId, zStr, iSide, (int)aLog[0], (int)aLog[1], (int)aLog[2], |
︙ | ︙ | |||
363 364 365 366 367 368 369 | #if BT_VAL_DEBUG u8 aKBuf[40]; u8 aVBuf[40]; static int nCall = 0; binToStr(pK, nK, aKBuf, sizeof(aKBuf)); binToStr(pV, nV, aVBuf, sizeof(aVBuf)); | | | | | | | > > > > > > > > > > > > > > > > > > > > | 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 | #if BT_VAL_DEBUG u8 aKBuf[40]; u8 aVBuf[40]; static int nCall = 0; binToStr(pK, nK, aKBuf, sizeof(aKBuf)); binToStr(pV, nV, aVBuf, sizeof(aVBuf)); fprintf(stderr, "%d:%d: %s \"%s\" -> \"%s\" (%d bytes)\n", pLock->iDebugId, nCall++, zStr, aKBuf, aVBuf, nV ); fflush(stderr); #endif } #endif static void btDebugLogSearch(BtLock *pLock, u32 pgno, u32 iSafe, u32 iFrame){ #if BT_PAGE_DEBUG static int nCall = 0; fprintf(stderr, "%d:%d: Search log for page %d (safe=%d) - frame %d\n", pLock->iDebugId, nCall++, (int)pgno, (int)iSafe, (int)iFrame ); fflush(stderr); #endif } static void btDebugSetPgno(BtLock *pLock, int iHash, int iSide, u32 *aPgno, int iFrame, int iZero, u32 pgno ){ #if BT_PAGE_DEBUG static int nCall = 0; fprintf(stderr, "%d:%d: Set iHash=%d/%d aPgno=%p iFrame=%d iZero=%d pgno=%d\n" , pLock->iDebugId, nCall++, iHash, iSide, (void*)aPgno, iFrame, iZero, (int)pgno ); fflush(stderr); #endif } #ifndef NDEBUG static void btDebugLogHeader( BtLock *pLock, const char *z, BtWalHdr *pHdr, int iHdr ){ #if BT_PAGE_DEBUG static int nCall = 0; fprintf(stderr, "%d:%d: %s log-header %d: (iCnt=%d iFirstFrame=%d)\n", pLock->iDebugId, nCall++, z, iHdr, (int)pHdr->iCnt, (int)pHdr->iFirstFrame ); fflush(stderr); #endif } #endif #ifdef NDEBUG # define btDebugLogHeader(a,b,c,d) #endif /* ** Ensure that shared-memory chunk iChunk is mapped and available in ** the BtLog.apShm[] array. If an error occurs, return an SQLite4 error ** code. Otherwise, SQLITE4_OK. */ static int btLogMapShm(BtLog *pLog, int iChunk){ |
︙ | ︙ | |||
484 485 486 487 488 489 490 491 492 493 494 495 496 497 | u32 aCksum[2]; btLogChecksum(1, (u8*)pHdr, offsetof(BtWalHdr, aCksum), 0, aCksum); if( pHdr->iMagic!=BT_WAL_MAGIC || aCksum[0]!=pHdr->aCksum[0] || aCksum[1]!=pHdr->aCksum[1] ){ rc = SQLITE4_NOTFOUND; } } return rc; } /* ** This function is used as part of recovery. It reads the contents of | > > | 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 | u32 aCksum[2]; btLogChecksum(1, (u8*)pHdr, offsetof(BtWalHdr, aCksum), 0, aCksum); if( pHdr->iMagic!=BT_WAL_MAGIC || aCksum[0]!=pHdr->aCksum[0] || aCksum[1]!=pHdr->aCksum[1] ){ rc = SQLITE4_NOTFOUND; }else{ btDebugLogHeader(pLog->pLock, "read", pHdr, iOff!=0); } } return rc; } /* ** This function is used as part of recovery. It reads the contents of |
︙ | ︙ | |||
621 622 623 624 625 626 627 | rc = btLogFindHash(pLog, iSide, iHash, &aHash, &aPgno, &iZero); /* Update the hash table */ if( rc==SQLITE4_OK ){ int iSlot; int nCollide = HASHTABLE_NSLOT*2; aPgno[iFrame-iZero] = pgno; | | | 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 | rc = btLogFindHash(pLog, iSide, iHash, &aHash, &aPgno, &iZero); /* Update the hash table */ if( rc==SQLITE4_OK ){ int iSlot; int nCollide = HASHTABLE_NSLOT*2; aPgno[iFrame-iZero] = pgno; btDebugSetPgno(pLog->pLock, iHash, iSide, aPgno, iFrame, iZero, pgno); if( iFrame==iZero ){ memset(aHash, 0, sizeof(ht_slot) * HASHTABLE_NSLOT); } for(iSlot=btLogHashKey(pLog,pgno); ; iSlot=btLogHashNext(pLog, iSlot)){ if( aHash[iSlot]==0 ){ |
︙ | ︙ | |||
657 658 659 660 661 662 663 | rc = btLogFindHash(pLog, iSide, iHash, &aHash, &aPgno, &iZero); if( rc==SQLITE4_OK ){ int i; ht_slot iMax; iMax = (iFrame - iZero) + 1; for(i=0; i<HASHTABLE_NSLOT; i++){ | | > > | < > | 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 | rc = btLogFindHash(pLog, iSide, iHash, &aHash, &aPgno, &iZero); if( rc==SQLITE4_OK ){ int i; ht_slot iMax; iMax = (iFrame - iZero) + 1; for(i=0; i<HASHTABLE_NSLOT; i++){ if( aHash[i]>iMax ){ aPgno[aHash[i]-1] = 0; aHash[i] = 0; } } } return rc; } /* |
︙ | ︙ | |||
686 687 688 689 690 691 692 693 694 695 696 697 698 699 | 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 */ ){ 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{ u32 iExpect = pLog->snapshot.aLog[5]+1; | > | 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 | 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 */ ){ 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{ u32 iExpect = pLog->snapshot.aLog[5]+1; |
︙ | ︙ | |||
951 952 953 954 955 956 957 958 959 960 961 962 963 964 | } static int btLogWriteHeader(BtLog *pLog, int iHdr, BtWalHdr *pHdr){ int rc; /* Return code */ i64 iOff; /* File offset to write to */ assert( iHdr==0 || iHdr==1 ); /* Calculate a checksum for the header */ btLogChecksum(1, (u8*)pHdr, offsetof(BtWalHdr, aCksum), 0, pHdr->aCksum); /* Write the object to disk */ iOff = iHdr * pLog->snapshot.nSector; rc = btLogWriteData(pLog, iOff, (u8*)pHdr, sizeof(BtWalHdr)); | > > | 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 | } static int btLogWriteHeader(BtLog *pLog, int iHdr, BtWalHdr *pHdr){ int rc; /* Return code */ i64 iOff; /* File offset to write to */ assert( iHdr==0 || iHdr==1 ); btDebugLogHeader(pLog->pLock, "write", pHdr, iHdr); /* Calculate a checksum for the header */ btLogChecksum(1, (u8*)pHdr, offsetof(BtWalHdr, aCksum), 0, pHdr->aCksum); /* Write the object to disk */ iOff = iHdr * pLog->snapshot.nSector; rc = btLogWriteData(pLog, iOff, (u8*)pHdr, sizeof(BtWalHdr)); |
︙ | ︙ | |||
980 981 982 983 984 985 986 | rc = btLogFindHash(pLog, iSide, iHash, &aHash, &aPgno, &iZero); if( rc==SQLITE4_OK ){ int nCollide = HASHTABLE_NSLOT*2; int iSlot; u32 iFrame = 0; | | | 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 | rc = btLogFindHash(pLog, iSide, iHash, &aHash, &aPgno, &iZero); if( rc==SQLITE4_OK ){ int nCollide = HASHTABLE_NSLOT*2; int iSlot; u32 iFrame = 0; iSlot = btLogHashKey(pLog, pgno); for( ; aHash[iSlot]; iSlot=btLogHashNext(pLog, iSlot)){ if( aPgno[aHash[iSlot]-1]==pgno ){ u32 iCandidate = iZero + aHash[iSlot] - 1; if( iCandidate<=iHi ) iFrame = iCandidate; } if( (nCollide--)==0 ) return btErrorBkpt(SQLITE4_CORRUPT); } |
︙ | ︙ | |||
1717 1718 1719 1720 1721 1722 1723 1724 1725 1726 1727 1728 1729 1730 | } btDebugCkptPage(pLog->pLock, pgno, aBuf, pgsz); rc = pVfs->xWrite(pFd, iOff, aBuf, pgsz); }else if( rc==SQLITE4_NOTFOUND ){ rc = SQLITE4_OK; } } /* Update the first field of the checkpoint-header. This tells readers ** that they need not consider anything that in the log before this ** point (since the data has already been copied into the database ** file). */ if( rc==SQLITE4_OK ){ assert( iFirstRead>0 ); | > > > > > | 1744 1745 1746 1747 1748 1749 1750 1751 1752 1753 1754 1755 1756 1757 1758 1759 1760 1761 1762 | } btDebugCkptPage(pLog->pLock, pgno, aBuf, pgsz); rc = pVfs->xWrite(pFd, iOff, aBuf, pgsz); }else if( rc==SQLITE4_NOTFOUND ){ rc = SQLITE4_OK; } } /* Sync the database file to disk. */ if( rc==SQLITE4_OK ){ rc = btLogSyncFile(pLog, pLog->pLock->pFd); } /* Update the first field of the checkpoint-header. This tells readers ** that they need not consider anything that in the log before this ** point (since the data has already been copied into the database ** file). */ if( rc==SQLITE4_OK ){ assert( iFirstRead>0 ); |
︙ | ︙ | |||
1748 1749 1750 1751 1752 1753 1754 1755 1756 1757 1758 1759 1760 1761 | hdr.nSector = pLog->snapshot.nSector; hdr.nPgsz = pgsz; hdr.iFirstFrame = iFirstRead; hdr.iSalt1 = fhdr.aCksum[0]; hdr.iSalt2 = fhdr.aCksum[1]; rc = btLogWriteHeader(pLog, iSlot, &hdr); if( rc==SQLITE4_OK ){ pShm->ckpt.iWalHdr = (iSlot<<2) + hdr.iCnt; } } /* Update the second field of the checkpoint header. This tells future ** writers that it is now safe to recycle pages before this point | > > > | 1780 1781 1782 1783 1784 1785 1786 1787 1788 1789 1790 1791 1792 1793 1794 1795 1796 | hdr.nSector = pLog->snapshot.nSector; hdr.nPgsz = pgsz; hdr.iFirstFrame = iFirstRead; hdr.iSalt1 = fhdr.aCksum[0]; hdr.iSalt2 = fhdr.aCksum[1]; rc = btLogWriteHeader(pLog, iSlot, &hdr); if( rc==SQLITE4_OK ){ rc = btLogSyncFile(pLog, pLog->pFd); } if( rc==SQLITE4_OK ){ pShm->ckpt.iWalHdr = (iSlot<<2) + hdr.iCnt; } } /* Update the second field of the checkpoint header. This tells future ** writers that it is now safe to recycle pages before this point |
︙ | ︙ |