Many hyperlinks are disabled.
Use anonymous login
to enable hyperlinks.
Overview
| Comment: | Fixes to the file locking. 109 tests are now failing. (CVS 1548) |
|---|---|
| Downloads: | Tarball | ZIP archive | SQL archive |
| Timelines: | family | ancestors | descendants | both | trunk |
| Files: | files | file ages | folders |
| SHA1: |
dc0763455bbf54c1d8728e16033709ca |
| User & Date: | drh 2004-06-09 17:37:23 |
Context
|
2004-06-09
| ||
| 19:03 | Remove legacy journal formats. (CVS 1549) (check-in: a12bef4a user: drh tags: trunk) | |
| 17:37 | Fixes to the file locking. 109 tests are now failing. (CVS 1548) (check-in: dc076345 user: drh tags: trunk) | |
| 14:17 | Add the "lock_status" pragma - only available when SQLITE_DEBUG is defined. Used for testing only. (CVS 1547) (check-in: 0ecbba78 user: drh tags: trunk) | |
Changes
Changes to src/btree.c.
5 5 ** a legal notice, here is a blessing: 6 6 ** 7 7 ** May you do good and not evil. 8 8 ** May you find forgiveness for yourself and forgive others. 9 9 ** May you share freely, never taking more than you give. 10 10 ** 11 11 ************************************************************************* 12 -** $Id: btree.c,v 1.161 2004/06/07 16:27:46 drh Exp $ 12 +** $Id: btree.c,v 1.162 2004/06/09 17:37:23 drh Exp $ 13 13 ** 14 14 ** This file implements a external (disk-based) database using BTrees. 15 15 ** For a detailed discussion of BTrees, refer to 16 16 ** 17 17 ** Donald E. Knuth, THE ART OF COMPUTER PROGRAMMING, Volume 3: 18 18 ** "Sorting And Searching", pages 473-480. Addison-Wesley 19 19 ** Publishing Company, Reading, Massachusetts. ................................................................................ 257 257 ** 258 258 ** The pParent field points back to the parent page. This allows us to 259 259 ** walk up the BTree from any leaf to the root. Care must be taken to 260 260 ** unref() the parent page pointer when this page is no longer referenced. 261 261 ** The pageDestructor() routine handles that chore. 262 262 */ 263 263 struct MemPage { 264 - u8 isInit; /* True if previously initialized */ 264 + u8 isInit; /* True if previously initialized. MUST BE FIRST! */ 265 265 u8 idxShift; /* True if Cell indices have changed */ 266 266 u8 nOverflow; /* Number of overflow cell bodies in aCell[] */ 267 267 u8 intKey; /* True if intkey flag is set */ 268 268 u8 leaf; /* True if leaf flag is set */ 269 269 u8 zeroData; /* True if table stores keys only */ 270 270 u8 leafData; /* True if tables stores data on leaves only */ 271 271 u8 hasData; /* True if this page stores data */ ................................................................................ 957 957 if( pPage->pParent ){ 958 958 MemPage *pParent = pPage->pParent; 959 959 pPage->pParent = 0; 960 960 releasePage(pParent); 961 961 } 962 962 pPage->isInit = 0; 963 963 } 964 + 965 +/* 966 +** During a rollback, when the pager reloads information into the cache 967 +** so that the cache is restored to its original state at the start of 968 +** the transaction, for each page restored this routine is called. 969 +** 970 +** This routine needs to reset the extra data section at the end of the 971 +** page to agree with the restored data. 972 +*/ 973 +static void pageReinit(void *pData, int pageSize){ 974 + MemPage *pPage = (MemPage*)&((char*)pData)[pageSize]; 975 + if( pPage->isInit ){ 976 + pPage->isInit = 0; 977 + initPage(pPage, pPage->pParent); 978 + } 979 +} 964 980 965 981 /* 966 982 ** Open a new database. 967 983 ** 968 984 ** Actually, this routine just sets up the internal data structures 969 985 ** for accessing the database. We do not open the database file 970 986 ** until the first page is loaded. ................................................................................ 1007 1023 if( rc!=SQLITE_OK ){ 1008 1024 if( pBt->pPager ) sqlite3pager_close(pBt->pPager); 1009 1025 sqliteFree(pBt); 1010 1026 *ppBtree = 0; 1011 1027 return rc; 1012 1028 } 1013 1029 sqlite3pager_set_destructor(pBt->pPager, pageDestructor); 1030 + sqlite3pager_set_reiniter(pBt->pPager, pageReinit); 1014 1031 pBt->pCursor = 0; 1015 1032 pBt->pPage1 = 0; 1016 1033 pBt->readOnly = sqlite3pager_isreadonly(pBt->pPager); 1017 1034 pBt->pageSize = SQLITE_PAGE_SIZE; /* FIX ME - read from header */ 1018 1035 pBt->usableSize = pBt->pageSize; 1019 1036 pBt->maxEmbedFrac = 64; /* FIX ME - read from header */ 1020 1037 pBt->minEmbedFrac = 32; /* FIX ME - read from header */
Changes to src/os.h.
152 152 int sqlite3OsClose(OsFile*); 153 153 int sqlite3OsRead(OsFile*, void*, int amt); 154 154 int sqlite3OsWrite(OsFile*, const void*, int amt); 155 155 int sqlite3OsSeek(OsFile*, off_t offset); 156 156 int sqlite3OsSync(OsFile*); 157 157 int sqlite3OsTruncate(OsFile*, off_t size); 158 158 int sqlite3OsFileSize(OsFile*, off_t *pSize); 159 -int sqlite3OsReadLock(OsFile*); 160 -int sqlite3OsWriteLock(OsFile*); 161 -int sqlite3OsUnlock(OsFile*); 162 159 int sqlite3OsRandomSeed(char*); 163 160 int sqlite3OsSleep(int ms); 164 161 int sqlite3OsCurrentTime(double*); 165 162 void sqlite3OsEnterMutex(void); 166 163 void sqlite3OsLeaveMutex(void); 167 164 char *sqlite3OsFullPathname(const char*); 168 165 int sqlite3OsLock(OsFile*, int); 169 -int sqlite3OsCheckWriteLock(OsFile *id); 166 +int sqlite3OsUnlock(OsFile*, int); 167 +int sqlite3OsCheckReservedLock(OsFile *id); 170 168 171 169 #endif /* _SQLITE_OS_H_ */
Changes to src/os_common.h.
39 39 #define TIMER_END elapse=hwtime()-g_start 40 40 #define SEEK(X) last_page=(X) 41 41 #define TRACE1(X) if( sqlite3_os_trace ) sqlite3DebugPrintf(X) 42 42 #define TRACE2(X,Y) if( sqlite3_os_trace ) sqlite3DebugPrintf(X,Y) 43 43 #define TRACE3(X,Y,Z) if( sqlite3_os_trace ) sqlite3DebugPrintf(X,Y,Z) 44 44 #define TRACE4(X,Y,Z,A) if( sqlite3_os_trace ) sqlite3DebugPrintf(X,Y,Z,A) 45 45 #define TRACE5(X,Y,Z,A,B) if( sqlite3_os_trace ) sqlite3DebugPrintf(X,Y,Z,A,B) 46 +#define TRACE6(X,Y,Z,A,B,C) if(sqlite3_os_trace) sqlite3DebugPrintf(X,Y,Z,A,B,C) 46 47 #else 47 48 #define TIMER_START 48 49 #define TIMER_END 49 50 #define SEEK(X) 50 51 #define TRACE1(X) 51 52 #define TRACE2(X,Y) 52 53 #define TRACE3(X,Y,Z) 53 54 #define TRACE4(X,Y,Z,A) 54 55 #define TRACE5(X,Y,Z,A,B) 56 +#define TRACE6(X,Y,Z,A,B,C) 55 57 #endif 56 58 57 59 58 60 /* 59 61 ** If we compile with the SQLITE_TEST macro set, then the following block 60 62 ** of code will give us the ability to simulate a disk I/O error. This 61 63 ** is used for testing the I/O recovery logic.
Changes to src/os_unix.c.
338 338 int sqlite3OsOpenReadWrite( 339 339 const char *zFilename, 340 340 OsFile *id, 341 341 int *pReadonly 342 342 ){ 343 343 int rc; 344 344 id->dirfd = -1; 345 - id->fd = open(zFilename, O_RDWR|O_CREAT|O_LARGEFILE|O_BINARY, 0644); 346 - if( id->fd<0 ){ 347 - id->fd = open(zFilename, O_RDONLY|O_LARGEFILE|O_BINARY); 348 - if( id->fd<0 ){ 345 + id->h = open(zFilename, O_RDWR|O_CREAT|O_LARGEFILE|O_BINARY, 0644); 346 + if( id->h<0 ){ 347 + id->h = open(zFilename, O_RDONLY|O_LARGEFILE|O_BINARY); 348 + if( id->h<0 ){ 349 349 return SQLITE_CANTOPEN; 350 350 } 351 351 *pReadonly = 1; 352 352 }else{ 353 353 *pReadonly = 0; 354 354 } 355 355 sqlite3OsEnterMutex(); 356 - rc = findLockInfo(id->fd, &id->pLock, &id->pOpen); 356 + rc = findLockInfo(id->h, &id->pLock, &id->pOpen); 357 357 sqlite3OsLeaveMutex(); 358 358 if( rc ){ 359 - close(id->fd); 359 + close(id->h); 360 360 return SQLITE_NOMEM; 361 361 } 362 362 id->locktype = 0; 363 - TRACE3("OPEN %-3d %s\n", id->fd, zFilename); 363 + TRACE3("OPEN %-3d %s\n", id->h, zFilename); 364 364 OpenCounter(+1); 365 365 return SQLITE_OK; 366 366 } 367 367 368 368 369 369 /* 370 370 ** Attempt to open a new file for exclusive access by this process. ................................................................................ 382 382 */ 383 383 int sqlite3OsOpenExclusive(const char *zFilename, OsFile *id, int delFlag){ 384 384 int rc; 385 385 if( access(zFilename, 0)==0 ){ 386 386 return SQLITE_CANTOPEN; 387 387 } 388 388 id->dirfd = -1; 389 - id->fd = open(zFilename, 389 + id->h = open(zFilename, 390 390 O_RDWR|O_CREAT|O_EXCL|O_NOFOLLOW|O_LARGEFILE|O_BINARY, 0600); 391 - if( id->fd<0 ){ 391 + if( id->h<0 ){ 392 392 return SQLITE_CANTOPEN; 393 393 } 394 394 sqlite3OsEnterMutex(); 395 - rc = findLockInfo(id->fd, &id->pLock, &id->pOpen); 395 + rc = findLockInfo(id->h, &id->pLock, &id->pOpen); 396 396 sqlite3OsLeaveMutex(); 397 397 if( rc ){ 398 - close(id->fd); 398 + close(id->h); 399 399 unlink(zFilename); 400 400 return SQLITE_NOMEM; 401 401 } 402 402 id->locktype = 0; 403 403 if( delFlag ){ 404 404 unlink(zFilename); 405 405 } 406 - TRACE3("OPEN-EX %-3d %s\n", id->fd, zFilename); 406 + TRACE3("OPEN-EX %-3d %s\n", id->h, zFilename); 407 407 OpenCounter(+1); 408 408 return SQLITE_OK; 409 409 } 410 410 411 411 /* 412 412 ** Attempt to open a new file for read-only access. 413 413 ** ................................................................................ 414 414 ** On success, write the file handle into *id and return SQLITE_OK. 415 415 ** 416 416 ** On failure, return SQLITE_CANTOPEN. 417 417 */ 418 418 int sqlite3OsOpenReadOnly(const char *zFilename, OsFile *id){ 419 419 int rc; 420 420 id->dirfd = -1; 421 - id->fd = open(zFilename, O_RDONLY|O_LARGEFILE|O_BINARY); 422 - if( id->fd<0 ){ 421 + id->h = open(zFilename, O_RDONLY|O_LARGEFILE|O_BINARY); 422 + if( id->h<0 ){ 423 423 return SQLITE_CANTOPEN; 424 424 } 425 425 sqlite3OsEnterMutex(); 426 - rc = findLockInfo(id->fd, &id->pLock, &id->pOpen); 426 + rc = findLockInfo(id->h, &id->pLock, &id->pOpen); 427 427 sqlite3OsLeaveMutex(); 428 428 if( rc ){ 429 - close(id->fd); 429 + close(id->h); 430 430 return SQLITE_NOMEM; 431 431 } 432 432 id->locktype = 0; 433 - TRACE3("OPEN-RO %-3d %s\n", id->fd, zFilename); 433 + TRACE3("OPEN-RO %-3d %s\n", id->h, zFilename); 434 434 OpenCounter(+1); 435 435 return SQLITE_OK; 436 436 } 437 437 438 438 /* 439 439 ** Attempt to open a file descriptor for the directory that contains a 440 440 ** file. This file descriptor can be used to fsync() the directory ................................................................................ 451 451 ** On failure, the function returns SQLITE_CANTOPEN and leaves 452 452 ** *id unchanged. 453 453 */ 454 454 int sqlite3OsOpenDirectory( 455 455 const char *zDirname, 456 456 OsFile *id 457 457 ){ 458 - if( id->fd<0 ){ 458 + if( id->h<0 ){ 459 459 /* Do not open the directory if the corresponding file is not already 460 460 ** open. */ 461 461 return SQLITE_CANTOPEN; 462 462 } 463 463 assert( id->dirfd<0 ); 464 464 id->dirfd = open(zDirname, O_RDONLY|O_BINARY, 0644); 465 465 if( id->dirfd<0 ){ ................................................................................ 506 506 return SQLITE_OK; 507 507 } 508 508 509 509 /* 510 510 ** Close a file. 511 511 */ 512 512 int sqlite3OsClose(OsFile *id){ 513 - sqlite3OsUnlock(id); 513 + sqlite3OsUnlock(id, NO_LOCK); 514 514 if( id->dirfd>=0 ) close(id->dirfd); 515 515 id->dirfd = -1; 516 516 sqlite3OsEnterMutex(); 517 517 if( id->pOpen->nLock ){ 518 518 /* If there are outstanding locks, do not actually close the file just 519 519 ** yet because that would clear those locks. Instead, add the file 520 520 ** descriptor to pOpen->aPending. It will be automatically closed when ................................................................................ 524 524 struct openCnt *pOpen = id->pOpen; 525 525 pOpen->nPending++; 526 526 aNew = sqliteRealloc( pOpen->aPending, pOpen->nPending*sizeof(int) ); 527 527 if( aNew==0 ){ 528 528 /* If a malloc fails, just leak the file descriptor */ 529 529 }else{ 530 530 pOpen->aPending = aNew; 531 - pOpen->aPending[pOpen->nPending-1] = id->fd; 531 + pOpen->aPending[pOpen->nPending-1] = id->h; 532 532 } 533 533 }else{ 534 534 /* There are no outstanding locks so we can close the file immediately */ 535 - close(id->fd); 535 + close(id->h); 536 536 } 537 537 releaseLockInfo(id->pLock); 538 538 releaseOpenCnt(id->pOpen); 539 539 sqlite3OsLeaveMutex(); 540 - TRACE2("CLOSE %-3d\n", id->fd); 540 + TRACE2("CLOSE %-3d\n", id->h); 541 541 OpenCounter(-1); 542 542 return SQLITE_OK; 543 543 } 544 544 545 545 /* 546 546 ** Read data from a file into a buffer. Return SQLITE_OK if all 547 547 ** bytes were read successfully and SQLITE_IOERR if anything goes 548 548 ** wrong. 549 549 */ 550 550 int sqlite3OsRead(OsFile *id, void *pBuf, int amt){ 551 551 int got; 552 552 SimulateIOError(SQLITE_IOERR); 553 553 TIMER_START; 554 - got = read(id->fd, pBuf, amt); 554 + got = read(id->h, pBuf, amt); 555 555 TIMER_END; 556 - TRACE4("READ %-3d %7d %d\n", id->fd, last_page, elapse); 556 + TRACE4("READ %-3d %7d %d\n", id->h, last_page, elapse); 557 557 SEEK(0); 558 558 /* if( got<0 ) got = 0; */ 559 559 if( got==amt ){ 560 560 return SQLITE_OK; 561 561 }else{ 562 562 return SQLITE_IOERR; 563 563 } ................................................................................ 567 567 ** Write data from a buffer into a file. Return SQLITE_OK on success 568 568 ** or some other error code on failure. 569 569 */ 570 570 int sqlite3OsWrite(OsFile *id, const void *pBuf, int amt){ 571 571 int wrote = 0; 572 572 SimulateIOError(SQLITE_IOERR); 573 573 TIMER_START; 574 - while( amt>0 && (wrote = write(id->fd, pBuf, amt))>0 ){ 574 + while( amt>0 && (wrote = write(id->h, pBuf, amt))>0 ){ 575 575 amt -= wrote; 576 576 pBuf = &((char*)pBuf)[wrote]; 577 577 } 578 578 TIMER_END; 579 - TRACE4("WRITE %-3d %7d %d\n", id->fd, last_page, elapse); 579 + TRACE4("WRITE %-3d %7d %d\n", id->h, last_page, elapse); 580 580 SEEK(0); 581 581 if( amt>0 ){ 582 582 return SQLITE_FULL; 583 583 } 584 584 return SQLITE_OK; 585 585 } 586 586 587 587 /* 588 588 ** Move the read/write pointer in a file. 589 589 */ 590 590 int sqlite3OsSeek(OsFile *id, off_t offset){ 591 591 SEEK(offset/1024 + 1); 592 - lseek(id->fd, offset, SEEK_SET); 592 + lseek(id->h, offset, SEEK_SET); 593 593 return SQLITE_OK; 594 594 } 595 595 596 596 /* 597 597 ** Make sure all writes to a particular file are committed to disk. 598 598 ** 599 599 ** Under Unix, also make sure that the directory entry for the file ................................................................................ 602 602 ** entry for the journal might not exist after we reboot. The next 603 603 ** SQLite to access the file will not know that the journal exists (because 604 604 ** the directory entry for the journal was never created) and the transaction 605 605 ** will not roll back - possibly leading to database corruption. 606 606 */ 607 607 int sqlite3OsSync(OsFile *id){ 608 608 SimulateIOError(SQLITE_IOERR); 609 - TRACE2("SYNC %-3d\n", id->fd); 610 - if( fsync(id->fd) ){ 609 + TRACE2("SYNC %-3d\n", id->h); 610 + if( fsync(id->h) ){ 611 611 return SQLITE_IOERR; 612 612 }else{ 613 613 if( id->dirfd>=0 ){ 614 614 TRACE2("DIRSYNC %-3d\n", id->dirfd); 615 615 fsync(id->dirfd); 616 616 close(id->dirfd); /* Only need to sync once, so close the directory */ 617 617 id->dirfd = -1; /* when we are done. */ ................................................................................ 621 621 } 622 622 623 623 /* 624 624 ** Truncate an open file to a specified size 625 625 */ 626 626 int sqlite3OsTruncate(OsFile *id, off_t nByte){ 627 627 SimulateIOError(SQLITE_IOERR); 628 - return ftruncate(id->fd, nByte)==0 ? SQLITE_OK : SQLITE_IOERR; 628 + return ftruncate(id->h, nByte)==0 ? SQLITE_OK : SQLITE_IOERR; 629 629 } 630 630 631 631 /* 632 632 ** Determine the current size of a file in bytes 633 633 */ 634 634 int sqlite3OsFileSize(OsFile *id, off_t *pSize){ 635 635 struct stat buf; 636 636 SimulateIOError(SQLITE_IOERR); 637 - if( fstat(id->fd, &buf)!=0 ){ 637 + if( fstat(id->h, &buf)!=0 ){ 638 638 return SQLITE_IOERR; 639 639 } 640 640 *pSize = buf.st_size; 641 641 return SQLITE_OK; 642 642 } 643 643 644 644 /* 645 645 ** This routine checks if there is a RESERVED lock held on the specified 646 646 ** file by this or any other process. If such a lock is held, return 647 647 ** non-zero. If the file is unlocked or holds only SHARED locks, then 648 648 ** return zero. 649 649 */ 650 -int sqlite3OsCheckWriteLock(OsFile *id){ 650 +int sqlite3OsCheckReservedLock(OsFile *id){ 651 651 int r = 0; 652 652 653 653 sqlite3OsEnterMutex(); /* Needed because id->pLock is shared across threads */ 654 654 655 655 /* Check if a thread in this process holds such a lock */ 656 656 if( id->pLock->locktype>SHARED_LOCK ){ 657 657 r = 1; ................................................................................ 661 661 */ 662 662 if( !r ){ 663 663 struct flock lock; 664 664 lock.l_whence = SEEK_SET; 665 665 lock.l_start = RESERVED_BYTE; 666 666 lock.l_len = 1; 667 667 lock.l_type = F_WRLCK; 668 - fcntl(id->fd, F_GETLK, &lock); 668 + fcntl(id->h, F_GETLK, &lock); 669 669 if( lock.l_type!=F_UNLCK ){ 670 670 r = 1; 671 671 } 672 672 } 673 673 674 674 sqlite3OsLeaveMutex(); 675 - TRACE3("TEST WR-LOCK %d %d\n", id->fd, r); 675 + TRACE3("TEST WR-LOCK %d %d\n", id->h, r); 676 676 677 677 return r; 678 678 } 679 679 680 680 /* 681 681 ** Lock the file with the lock specified by parameter locktype - one 682 682 ** of the following: ................................................................................ 694 694 ** 695 695 ** UNLOCKED -> SHARED 696 696 ** SHARED -> RESERVED 697 697 ** SHARED -> (PENDING) -> EXCLUSIVE 698 698 ** RESERVED -> (PENDING) -> EXCLUSIVE 699 699 ** PENDING -> EXCLUSIVE 700 700 ** 701 -** This routine will only increase a lock. The sqlite3OsUnlock() routine 702 -** erases all locks at once and returns us immediately to locking level 0. 703 -** It is not possible to lower the locking level one step at a time. You 704 -** must go straight to locking level 0. 701 +** This routine will only increase a lock. Use the sqlite3OsUnlock() 702 +** routine to lower a locking level. 705 703 */ 706 704 int sqlite3OsLock(OsFile *id, int locktype){ 707 705 int rc = SQLITE_OK; 708 706 struct lockInfo *pLock = id->pLock; 709 707 struct flock lock; 710 708 int s; 711 709 712 - TRACE5("LOCK %d %d was %d(%d)\n", 713 - id->fd, locktype, id->locktype, pLock->locktype); 710 + TRACE6("LOCK %d %d was %d(%d,%d)\n", 711 + id->h, locktype, id->locktype, pLock->locktype, pLock->cnt); 714 712 715 713 /* If there is already a lock of this type or more restrictive on the 716 714 ** OsFile, do nothing. Don't use the end_lock: exit path, as 717 715 ** sqlite3OsEnterMutex() hasn't been called yet. 718 716 */ 719 717 if( id->locktype>=locktype ){ 720 718 return SQLITE_OK; ................................................................................ 767 765 assert( pLock->locktype==0 ); 768 766 769 767 /* Temporarily grab a PENDING lock. This prevents new SHARED locks from 770 768 ** being formed if a PENDING lock is already held. 771 769 */ 772 770 lock.l_type = F_RDLCK; 773 771 lock.l_start = PENDING_BYTE; 774 - s = fcntl(id->fd, F_SETLK, &lock); 772 + s = fcntl(id->h, F_SETLK, &lock); 775 773 if( s ){ 776 774 rc = (errno==EINVAL) ? SQLITE_NOLFS : SQLITE_BUSY; 777 775 goto end_lock; 778 776 } 779 777 780 778 /* Now get the read-lock */ 781 779 lock.l_start = SHARED_FIRST; 782 780 lock.l_len = SHARED_SIZE; 783 - s = fcntl(id->fd, F_SETLK, &lock); 781 + s = fcntl(id->h, F_SETLK, &lock); 784 782 785 783 /* Drop the temporary PENDING lock */ 786 784 lock.l_start = PENDING_BYTE; 787 785 lock.l_len = 1L; 788 786 lock.l_type = F_UNLCK; 789 - fcntl(id->fd, F_SETLK, &lock); 787 + fcntl(id->h, F_SETLK, &lock); 790 788 if( s ){ 791 789 rc = (errno==EINVAL) ? SQLITE_NOLFS : SQLITE_BUSY; 792 790 }else{ 793 791 id->locktype = SHARED_LOCK; 794 792 id->pOpen->nLock++; 795 793 pLock->cnt = 1; 796 794 } ................................................................................ 811 809 case EXCLUSIVE_LOCK: 812 810 lock.l_start = SHARED_FIRST; 813 811 lock.l_len = SHARED_SIZE; 814 812 break; 815 813 default: 816 814 assert(0); 817 815 } 818 - s = fcntl(id->fd, F_SETLK, &lock); 816 + s = fcntl(id->h, F_SETLK, &lock); 819 817 if( s ){ 820 818 rc = (errno==EINVAL) ? SQLITE_NOLFS : SQLITE_BUSY; 821 819 } 822 820 } 823 821 824 822 if( rc==SQLITE_OK ){ 825 823 id->locktype = locktype; 826 824 pLock->locktype = locktype; 827 825 } 828 826 829 827 end_lock: 830 828 sqlite3OsLeaveMutex(); 831 - TRACE4("LOCK %d %d %s\n", id->fd, locktype, rc==SQLITE_OK ? "ok" : "failed"); 829 + TRACE4("LOCK %d %d %s\n", id->h, locktype, rc==SQLITE_OK ? "ok" : "failed"); 832 830 return rc; 833 831 } 834 832 835 833 /* 836 -** Unlock the given file descriptor. If the file descriptor was 837 -** not previously locked, then this routine is a no-op. If this 838 -** library was compiled with large file support (LFS) but LFS is not 839 -** available on the host, then an SQLITE_NOLFS is returned. 834 +** Lower the locking level on file descriptor id to locktype. locktype 835 +** must be either NO_LOCK or SHARED_LOCK. 836 +** 837 +** If the locking level of the file descriptor is already at or below 838 +** the requested locking level, this routine is a no-op. 839 +** 840 +** It is not possible for this routine to fail. 840 841 */ 841 -int sqlite3OsUnlock(OsFile *id){ 842 - int rc; 843 - if( !id->locktype ) return SQLITE_OK; 844 - id->locktype = 0; 842 +int sqlite3OsUnlock(OsFile *id, int locktype){ 843 + struct lockInfo *pLock; 844 + struct flock lock; 845 + 846 + TRACE6("UNLOCK %d %d was %d(%d,%d)\n", 847 + id->h, locktype, id->locktype, id->pLock->locktype, id->pLock->cnt); 848 + 849 + assert( locktype<=SHARED_LOCK ); 850 + if( id->locktype<=locktype ){ 851 + return SQLITE_OK; 852 + } 845 853 sqlite3OsEnterMutex(); 846 - assert( id->pLock->cnt!=0 ); 847 - if( id->pLock->cnt>1 ){ 848 - id->pLock->cnt--; 849 - rc = SQLITE_OK; 850 - }else{ 851 - struct flock lock; 852 - int s; 854 + pLock = id->pLock; 855 + assert( pLock->cnt!=0 ); 856 + if( id->locktype>SHARED_LOCK ){ 857 + assert( pLock->locktype==id->locktype ); 853 858 lock.l_type = F_UNLCK; 854 859 lock.l_whence = SEEK_SET; 855 - lock.l_start = lock.l_len = 0L; 856 - s = fcntl(id->fd, F_SETLK, &lock); 857 - if( s!=0 ){ 858 - rc = (errno==EINVAL) ? SQLITE_NOLFS : SQLITE_BUSY; 859 - }else{ 860 - rc = SQLITE_OK; 861 - id->pLock->cnt = 0; 862 - id->pLock->locktype = 0; 863 - } 860 + lock.l_start = PENDING_BYTE; 861 + lock.l_len = 2L; assert( PENDING_BYTE+1==RESERVED_BYTE ); 862 + fcntl(id->h, F_SETLK, &lock); 863 + pLock->locktype = SHARED_LOCK; 864 864 } 865 + if( locktype==NO_LOCK ){ 866 + struct openCnt *pOpen; 865 867 866 - if( rc==SQLITE_OK ){ 868 + /* Decrement the shared lock counter. Release the lock using an 869 + ** OS call only when all threads in this same process have released 870 + ** the lock. 871 + */ 872 + pLock->cnt--; 873 + if( pLock->cnt==0 ){ 874 + lock.l_type = F_UNLCK; 875 + lock.l_whence = SEEK_SET; 876 + lock.l_start = lock.l_len = 0L; 877 + fcntl(id->h, F_SETLK, &lock); 878 + pLock->locktype = NO_LOCK; 879 + } 880 + 867 881 /* Decrement the count of locks against this same file. When the 868 882 ** count reaches zero, close any other file descriptors whose close 869 883 ** was deferred because of outstanding locks. 870 884 */ 871 - struct openCnt *pOpen = id->pOpen; 885 + pOpen = id->pOpen; 872 886 pOpen->nLock--; 873 887 assert( pOpen->nLock>=0 ); 874 888 if( pOpen->nLock==0 && pOpen->nPending>0 ){ 875 889 int i; 876 890 for(i=0; i<pOpen->nPending; i++){ 877 891 close(pOpen->aPending[i]); 878 892 } 879 893 sqliteFree(pOpen->aPending); 880 894 pOpen->nPending = 0; 881 895 pOpen->aPending = 0; 882 896 } 883 897 } 884 898 sqlite3OsLeaveMutex(); 885 - id->locktype = 0; 886 - return rc; 899 + id->locktype = locktype; 900 + return SQLITE_OK; 887 901 } 888 902 889 903 /* 890 904 ** Get information to seed the random number generator. The seed 891 905 ** is written into the buffer zBuf[256]. The calling function must 892 906 ** supply a sufficiently large buffer. 893 907 */
Changes to src/os_unix.h.
60 60 ** OsFile.locktype takes one of the values SHARED_LOCK, RESERVED_LOCK, 61 61 ** PENDING_LOCK or EXCLUSIVE_LOCK. 62 62 */ 63 63 typedef struct OsFile OsFile; 64 64 struct OsFile { 65 65 struct openCnt *pOpen; /* Info about all open fd's on this inode */ 66 66 struct lockInfo *pLock; /* Info about locks on this inode */ 67 - int fd; /* The file descriptor */ 67 + int h; /* The file descriptor */ 68 68 int locktype; /* The type of lock held on this fd */ 69 69 int dirfd; /* File descriptor for the directory */ 70 70 }; 71 71 72 72 /* 73 73 ** Maximum number of characters in a temporary file name 74 74 */
Changes to src/os_win.c.
504 504 } 505 505 506 506 /* 507 507 ** This routine checks if there is a RESERVED lock held on the specified 508 508 ** file by this or any other process. If such a lock is held, return 509 509 ** non-zero, otherwise zero. 510 510 */ 511 -int sqlite3OsCheckWriteLock(OsFile *id){ 511 +int sqlite3OsCheckReservedLock(OsFile *id){ 512 512 int rc; 513 513 if( id->locktype>=RESERVED_LOCK ){ 514 514 rc = 1; 515 515 TRACE3("TEST WR-LOCK %d %d (local)\n", id->h, rc); 516 516 }else{ 517 517 rc = LockFile(id->h, RESERVED_BYTE, 0, 1, 0); 518 518 if( rc ){ ................................................................................ 521 521 rc = !rc; 522 522 TRACE3("TEST WR-LOCK %d %d (remote)\n", id->h, rc); 523 523 } 524 524 return rc; 525 525 } 526 526 527 527 /* 528 -** Unlock the given file descriptor. If the file descriptor was 529 -** not previously locked, then this routine is a no-op. If this 530 -** library was compiled with large file support (LFS) but LFS is not 531 -** available on the host, then an SQLITE_NOLFS is returned. 528 +** Lower the locking level on file descriptor id to locktype. locktype 529 +** must be either NO_LOCK or SHARED_LOCK. 530 +** 531 +** If the locking level of the file descriptor is already at or below 532 +** the requested locking level, this routine is a no-op. 533 +** 534 +** It is not possible for this routine to fail. 532 535 */ 533 -int sqlite3OsUnlock(OsFile *id){ 536 +int sqlite3OsUnlock(OsFile *id, int locktype){ 534 537 int rc, type; 535 - TRACE4("UNLOCK %d was %d(%d)\n", id->h, id->locktype, id->sharedLockByte); 538 + assert( locktype<=SHARED_LOCK ); 539 + TRACE4("UNLOCK %d to %d was %d(%d)\n", id->h, locktype, 540 + id->locktype, id->sharedLockByte); 536 541 type = id->locktype; 537 542 if( type>=EXCLUSIVE_LOCK ){ 538 543 UnlockFile(id->h, SHARED_FIRST, 0, SHARED_SIZE, 0); 539 544 } 540 545 if( type>=RESERVED_LOCK ){ 541 546 UnlockFile(id->h, RESERVED_BYTE, 0, 1, 0); 542 547 } 543 - if( type>=SHARED_LOCK && type<EXCLUSIVE_LOCK ){ 548 + if( locktype==NO_LOCK && type>=SHARED_LOCK && type<EXCLUSIVE_LOCK ){ 544 549 unlockReadLock(id); 545 550 } 546 551 if( type>=PENDING_LOCK ){ 547 552 UnlockFile(id->h, PENDING_BYTE, 0, 1, 0); 548 553 } 549 - id->locktype = NO_LOCK; 554 + id->locktype = locktype; 550 555 return SQLITE_OK; 551 556 } 552 557 553 558 /* 554 559 ** Get information to seed the random number generator. The seed 555 560 ** is written into the buffer zBuf[256]. The calling function must 556 561 ** supply a sufficiently large buffer.
Changes to src/pager.c.
14 14 ** The pager is used to access a database disk file. It implements 15 15 ** atomic commit and rollback through the use of a journal file that 16 16 ** is separate from the database file. The pager also implements file 17 17 ** locking to prevent two processes from writing the same database 18 18 ** file simultaneously, or one process from reading the database while 19 19 ** another is writing. 20 20 ** 21 -** @(#) $Id: pager.c,v 1.114 2004/06/09 14:17:21 drh Exp $ 21 +** @(#) $Id: pager.c,v 1.115 2004/06/09 17:37:28 drh Exp $ 22 22 */ 23 23 #include "os.h" /* Must be first to enable large file support */ 24 24 #include "sqliteInt.h" 25 25 #include "pager.h" 26 26 #include <assert.h> 27 27 #include <string.h> 28 28 ................................................................................ 45 45 #endif 46 46 47 47 48 48 /* 49 49 ** The page cache as a whole is always in one of the following 50 50 ** states: 51 51 ** 52 -** SQLITE_UNLOCK The page cache is not currently reading or 52 +** PAGER_UNLOCK The page cache is not currently reading or 53 53 ** writing the database file. There is no 54 54 ** data held in memory. This is the initial 55 55 ** state. 56 56 ** 57 -** SQLITE_READLOCK The page cache is reading the database. 57 +** PAGER_SHARED The page cache is reading the database. 58 58 ** Writing is not permitted. There can be 59 59 ** multiple readers accessing the same database 60 60 ** file at the same time. 61 61 ** 62 -** SQLITE_WRITELOCK The page cache is writing the database. 62 +** PAGER_RESERVED Writing is permitted to the page cache only. 63 +** The original database file has not been modified. 64 +** Other processes may still be reading the on-disk 65 +** database file. 66 +** 67 +** PAGER_EXCLUSIVE The page cache is writing the database. 63 68 ** Access is exclusive. No other processes or 64 69 ** threads can be reading or writing while one 65 70 ** process is writing. 66 71 ** 67 -** The page cache comes up in SQLITE_UNLOCK. The first time a 68 -** sqlite_page_get() occurs, the state transitions to SQLITE_READLOCK. 72 +** The page cache comes up in PAGER_UNLOCK. The first time a 73 +** sqlite_page_get() occurs, the state transitions to PAGER_SHARED. 69 74 ** After all pages have been released using sqlite_page_unref(), 70 -** the state transitions back to SQLITE_UNLOCK. The first time 75 +** the state transitions back to PAGER_UNLOCK. The first time 71 76 ** that sqlite_page_write() is called, the state transitions to 72 -** SQLITE_WRITELOCK. (Note that sqlite_page_write() can only be 77 +** PAGER_RESERVED. (Note that sqlite_page_write() can only be 73 78 ** called on an outstanding page which means that the pager must 74 -** be in SQLITE_READLOCK before it transitions to SQLITE_WRITELOCK.) 79 +** be in PAGER_SHARED before it transitions to PAGER_RESERVED.) 75 80 ** The sqlite_page_rollback() and sqlite_page_commit() functions 76 -** transition the state from SQLITE_WRITELOCK back to SQLITE_READLOCK. 81 +** transition the state from PAGER_RESERVED to PAGER_EXCLUSIVE to 82 +** PAGER_SHARED. 77 83 */ 78 -#define SQLITE_UNLOCK 0 79 -#define SQLITE_READLOCK 1 80 -#define SQLITE_WRITELOCK 2 84 +#define PAGER_UNLOCK 0 85 +#define PAGER_SHARED 1 86 +#define PAGER_RESERVED 2 87 +#define PAGER_EXCLUSIVE 3 81 88 82 89 83 90 /* 84 91 ** Each in-memory image of a page begins with the following header. 85 92 ** This header is only visible to this pager module. The client 86 93 ** code that calls pager sees only the data that follows the header. 87 94 ** ................................................................................ 178 185 int stmtSize; /* Size of database (in pages) at stmt_begin() */ 179 186 off_t stmtJSize; /* Size of journal at stmt_begin() */ 180 187 int nRec; /* Number of pages written to the journal */ 181 188 u32 cksumInit; /* Quasi-random value added to every checksum */ 182 189 int stmtNRec; /* Number of records in stmt subjournal */ 183 190 int nExtra; /* Add this many bytes to each in-memory page */ 184 191 void (*xDestructor)(void*,int); /* Call this routine when freeing pages */ 192 + void (*xReiniter)(void*,int); /* Call this routine when reloading pages */ 185 193 int pageSize; /* Number of bytes in a page */ 186 194 int nPage; /* Total number of in-memory pages */ 187 195 int nRef; /* Number of in-memory pages with PgHdr.nRef>0 */ 188 196 int mxPage; /* Maximum number of pages to hold in cache */ 189 197 int nHit, nMiss, nOvfl; /* Cache hits, missing, and LRU overflows */ 190 198 void (*xCodec)(void*,void*,Pgno,int); /* Routine for en/decoding data */ 191 199 void *pCodecArg; /* First argument to xCodec() */ ................................................................................ 193 201 u8 journalStarted; /* True if header of journal is synced */ 194 202 u8 useJournal; /* Use a rollback journal on this file */ 195 203 u8 stmtOpen; /* True if the statement subjournal is open */ 196 204 u8 stmtInUse; /* True we are in a statement subtransaction */ 197 205 u8 stmtAutoopen; /* Open stmt journal when main journal is opened*/ 198 206 u8 noSync; /* Do not sync the journal if true */ 199 207 u8 fullSync; /* Do extra syncs of the journal for robustness */ 200 - u8 state; /* SQLITE_UNLOCK, _READLOCK or _WRITELOCK */ 208 + u8 state; /* PAGER_UNLOCK, _SHARED, _RESERVED, etc. */ 201 209 u8 errMask; /* One of several kinds of errors */ 202 210 u8 tempFile; /* zFilename is a temporary file */ 203 211 u8 readOnly; /* True for a read-only database */ 204 212 u8 needSync; /* True if an fsync() is needed on the journal */ 205 - u8 dirtyFile; /* True if database file has changed in any way */ 213 + u8 dirtyCache; /* True if cached pages have changed */ 206 214 u8 alwaysRollback; /* Disable dont_rollback() for all pages */ 207 215 u8 memDb; /* True to inhibit all file I/O */ 208 216 u8 *aInJournal; /* One bit for each page in the database file */ 209 217 u8 *aInStmt; /* One bit for each page in the database */ 210 218 PgHdr *pFirst, *pLast; /* List of free pages */ 211 219 PgHdr *pFirstSynced; /* First free page with PgHdr.needSync==0 */ 212 220 PgHdr *pAll; /* List of all pages */ ................................................................................ 475 483 } 476 484 pPager->pFirst = 0; 477 485 pPager->pFirstSynced = 0; 478 486 pPager->pLast = 0; 479 487 pPager->pAll = 0; 480 488 memset(pPager->aHash, 0, sizeof(pPager->aHash)); 481 489 pPager->nPage = 0; 482 - if( pPager->state>=SQLITE_WRITELOCK ){ 490 + if( pPager->state>=PAGER_RESERVED ){ 483 491 sqlite3pager_rollback(pPager); 484 492 } 485 - sqlite3OsUnlock(&pPager->fd); 486 - pPager->state = SQLITE_UNLOCK; 493 + sqlite3OsUnlock(&pPager->fd, NO_LOCK); 494 + pPager->state = PAGER_UNLOCK; 487 495 pPager->dbSize = -1; 488 496 pPager->nRef = 0; 489 497 assert( pPager->journalOpen==0 ); 490 498 } 491 499 492 500 /* 493 501 ** When this routine is called, the pager has the journal file open and 494 -** a write lock on the database. This routine releases the database 495 -** write lock and acquires a read lock in its place. The journal file 496 -** is deleted and closed. 502 +** a RESERVED or EXCLUSIVE lock on the database. This routine releases 503 +** the database lock and acquires a SHARED lock in its place. The journal 504 +** file is deleted and closed. 497 505 ** 498 506 ** TODO: Consider keeping the journal file open for temporary databases. 499 507 ** This might give a performance improvement on windows where opening 500 508 ** a file is an expensive operation. 501 509 */ 502 510 static int pager_unwritelock(Pager *pPager){ 503 - int rc; 504 511 PgHdr *pPg; 505 - if( pPager->state<SQLITE_WRITELOCK ) return SQLITE_OK; 512 + if( pPager->state<PAGER_RESERVED ){ 513 + return SQLITE_OK; 514 + } 506 515 sqlite3pager_stmt_commit(pPager); 507 516 if( pPager->stmtOpen ){ 508 517 sqlite3OsClose(&pPager->stfd); 509 518 pPager->stmtOpen = 0; 510 519 } 511 520 if( pPager->journalOpen ){ 512 521 sqlite3OsClose(&pPager->jfd); ................................................................................ 516 525 pPager->aInJournal = 0; 517 526 for(pPg=pPager->pAll; pPg; pPg=pPg->pNextAll){ 518 527 pPg->inJournal = 0; 519 528 pPg->dirty = 0; 520 529 pPg->needSync = 0; 521 530 } 522 531 }else{ 523 - assert( pPager->dirtyFile==0 || pPager->useJournal==0 ); 532 + assert( pPager->dirtyCache==0 || pPager->useJournal==0 ); 524 533 } 525 - rc = sqlite3OsLock(&pPager->fd, SHARED_LOCK); 526 - if( rc==SQLITE_OK ){ 527 - pPager->state = SQLITE_READLOCK; 528 - }else{ 529 - /* This can only happen if a process does a BEGIN, then forks and the 530 - ** child process does the COMMIT. Because of the semantics of unix 531 - ** file locking, the unlock will fail. 532 - */ 533 - pPager->state = SQLITE_UNLOCK; 534 - } 535 - return rc; 534 + sqlite3OsUnlock(&pPager->fd, SHARED_LOCK); 535 + pPager->state = PAGER_SHARED; 536 + return SQLITE_OK; 536 537 } 537 538 538 539 /* 539 540 ** Compute and return a checksum for the page of data. 540 541 ** 541 542 ** This is not a real checksum. It is really just the sum of the 542 543 ** random initial value and the page number. We considered do a checksum ................................................................................ 584 585 } 585 586 } 586 587 587 588 /* Playback the page. Update the in-memory copy of the page 588 589 ** at the same time, if there is one. 589 590 */ 590 591 pPg = pager_lookup(pPager, pgRec.pgno); 591 - TRACE2("PLAYBACK %d\n", pgRec.pgno); 592 + TRACE2("PLAYBACK page %d\n", pgRec.pgno); 592 593 sqlite3OsSeek(&pPager->fd, (pgRec.pgno-1)*(off_t)SQLITE_PAGE_SIZE); 593 594 rc = sqlite3OsWrite(&pPager->fd, pgRec.aData, SQLITE_PAGE_SIZE); 594 595 if( pPg ){ 595 596 /* No page should ever be rolled back that is in use, except for page 596 597 ** 1 which is held in use in order to keep the lock on the database 597 598 ** active. 598 599 */ ................................................................................ 716 717 sqliteFree(zMasterJournal); 717 718 } 718 719 if( master_open ){ 719 720 sqlite3OsClose(&master); 720 721 } 721 722 return rc; 722 723 } 724 + 725 +/* 726 +** Make every page in the cache agree with what is on disk. In other words, 727 +** reread the disk to reset the state of the cache. 728 +** 729 +** This routine is called after a rollback in which some of the dirty cache 730 +** pages had never been written out to disk. We need to roll back the 731 +** cache content and the easiest way to do that is to reread the old content 732 +** back from the disk. 733 +*/ 734 +static int pager_reload_cache(Pager *pPager){ 735 + PgHdr *pPg; 736 + int rc = SQLITE_OK; 737 + for(pPg=pPager->pAll; pPg; pPg=pPg->pNextAll){ 738 + char zBuf[SQLITE_PAGE_SIZE]; 739 + if( !pPg->dirty ) continue; 740 + if( (int)pPg->pgno <= pPager->origDbSize ){ 741 + sqlite3OsSeek(&pPager->fd, SQLITE_PAGE_SIZE*(off_t)(pPg->pgno-1)); 742 + rc = sqlite3OsRead(&pPager->fd, zBuf, SQLITE_PAGE_SIZE); 743 + TRACE2("REFETCH page %d\n", pPg->pgno); 744 + CODEC(pPager, zBuf, pPg->pgno, 2); 745 + if( rc ) break; 746 + }else{ 747 + memset(zBuf, 0, SQLITE_PAGE_SIZE); 748 + } 749 + if( pPg->nRef==0 || memcmp(zBuf, PGHDR_TO_DATA(pPg), SQLITE_PAGE_SIZE) ){ 750 + memcpy(PGHDR_TO_DATA(pPg), zBuf, SQLITE_PAGE_SIZE); 751 + if( pPager->xReiniter ){ 752 + pPager->xReiniter(PGHDR_TO_DATA(pPg), pPager->pageSize); 753 + }else{ 754 + memset(PGHDR_TO_EXTRA(pPg), 0, pPager->nExtra); 755 + } 756 + } 757 + pPg->needSync = 0; 758 + pPg->dirty = 0; 759 + } 760 + return rc; 761 +} 762 + 723 763 724 764 /* 725 765 ** Playback the journal and thus restore the database file to 726 766 ** the state it was in before we started making changes. 727 767 ** 728 768 ** The journal file format is as follows: 729 769 ** ................................................................................ 883 923 } 884 924 885 925 /* Pages that have been written to the journal but never synced 886 926 ** where not restored by the loop above. We have to restore those 887 927 ** pages by reading them back from the original database. 888 928 */ 889 929 if( rc==SQLITE_OK ){ 890 - PgHdr *pPg; 891 - for(pPg=pPager->pAll; pPg; pPg=pPg->pNextAll){ 892 - char zBuf[SQLITE_PAGE_SIZE]; 893 - if( !pPg->dirty ) continue; 894 - if( (int)pPg->pgno <= pPager->origDbSize ){ 895 - sqlite3OsSeek(&pPager->fd, SQLITE_PAGE_SIZE*(off_t)(pPg->pgno-1)); 896 - rc = sqlite3OsRead(&pPager->fd, zBuf, SQLITE_PAGE_SIZE); 897 - TRACE2("REFETCH %d\n", pPg->pgno); 898 - CODEC(pPager, zBuf, pPg->pgno, 2); 899 - if( rc ) break; 900 - }else{ 901 - memset(zBuf, 0, SQLITE_PAGE_SIZE); 902 - } 903 - if( pPg->nRef==0 || memcmp(zBuf, PGHDR_TO_DATA(pPg), SQLITE_PAGE_SIZE) ){ 904 - memcpy(PGHDR_TO_DATA(pPg), zBuf, SQLITE_PAGE_SIZE); 905 - memset(PGHDR_TO_EXTRA(pPg), 0, pPager->nExtra); 906 - } 907 - pPg->needSync = 0; 908 - pPg->dirty = 0; 909 - } 930 + pager_reload_cache(pPager); 910 931 } 911 932 912 933 end_playback: 913 934 if( zMaster ){ 914 935 /* If there was a master journal and this routine will return true, 915 936 ** see if it is possible to delete the master journal. If errors 916 937 ** occur during this process, ignore them. ................................................................................ 1161 1182 pPager->nRef = 0; 1162 1183 pPager->dbSize = memDb-1; 1163 1184 pPager->pageSize = SQLITE_PAGE_SIZE; 1164 1185 pPager->stmtSize = 0; 1165 1186 pPager->stmtJSize = 0; 1166 1187 pPager->nPage = 0; 1167 1188 pPager->mxPage = mxPage>5 ? mxPage : 10; 1168 - pPager->state = SQLITE_UNLOCK; 1189 + pPager->state = PAGER_UNLOCK; 1169 1190 pPager->errMask = 0; 1170 1191 pPager->tempFile = tempFile; 1171 1192 pPager->memDb = memDb; 1172 1193 pPager->readOnly = readOnly; 1173 1194 pPager->needSync = 0; 1174 1195 pPager->noSync = pPager->tempFile || !useJournal; 1175 1196 pPager->pFirst = 0; ................................................................................ 1189 1210 ** 1190 1211 ** The destructor is not called as a result sqlite3pager_close(). 1191 1212 ** Destructors are only called by sqlite3pager_unref(). 1192 1213 */ 1193 1214 void sqlite3pager_set_destructor(Pager *pPager, void (*xDesc)(void*,int)){ 1194 1215 pPager->xDestructor = xDesc; 1195 1216 } 1217 + 1218 +/* 1219 +** Set the reinitializer for this pager. If not NULL, the reinitializer 1220 +** is called when the content of a page in cache is restored to its original 1221 +** value as a result of a rollback. The callback gives higher-level code 1222 +** an opportunity to restore the EXTRA section to agree with the restored 1223 +** page data. 1224 +*/ 1225 +void sqlite3pager_set_reiniter(Pager *pPager, void (*xReinit)(void*,int)){ 1226 + pPager->xReiniter = xReinit; 1227 +} 1196 1228 1197 1229 /* 1198 1230 ** Return the total number of pages in the disk file associated with 1199 1231 ** pPager. 1200 1232 */ 1201 1233 int sqlite3pager_pagecount(Pager *pPager){ 1202 1234 off_t n; ................................................................................ 1205 1237 return pPager->dbSize; 1206 1238 } 1207 1239 if( sqlite3OsFileSize(&pPager->fd, &n)!=SQLITE_OK ){ 1208 1240 pPager->errMask |= PAGER_ERR_DISK; 1209 1241 return 0; 1210 1242 } 1211 1243 n /= SQLITE_PAGE_SIZE; 1212 - if( pPager->state!=SQLITE_UNLOCK ){ 1244 + if( pPager->state!=PAGER_UNLOCK ){ 1213 1245 pPager->dbSize = n; 1214 1246 } 1215 1247 return n; 1216 1248 } 1217 1249 1218 1250 /* 1219 1251 ** Forward declaration ................................................................................ 1326 1358 ** and their memory is freed. Any attempt to use a page associated 1327 1359 ** with this page cache after this function returns will likely 1328 1360 ** result in a coredump. 1329 1361 */ 1330 1362 int sqlite3pager_close(Pager *pPager){ 1331 1363 PgHdr *pPg, *pNext; 1332 1364 switch( pPager->state ){ 1333 - case SQLITE_WRITELOCK: { 1365 + case PAGER_RESERVED: 1366 + case PAGER_EXCLUSIVE: { 1334 1367 sqlite3pager_rollback(pPager); 1335 1368 if( !pPager->memDb ){ 1336 - sqlite3OsUnlock(&pPager->fd); 1369 + sqlite3OsUnlock(&pPager->fd, NO_LOCK); 1337 1370 } 1338 1371 assert( pPager->journalOpen==0 ); 1339 1372 break; 1340 1373 } 1341 - case SQLITE_READLOCK: { 1374 + case PAGER_SHARED: { 1342 1375 if( !pPager->memDb ){ 1343 - sqlite3OsUnlock(&pPager->fd); 1376 + sqlite3OsUnlock(&pPager->fd, NO_LOCK); 1344 1377 } 1345 1378 break; 1346 1379 } 1347 1380 default: { 1348 1381 /* Do nothing */ 1349 1382 break; 1350 1383 } ................................................................................ 1481 1514 assert( pPager->nRec*pgSz+hdrSz==jSz ); 1482 1515 } 1483 1516 #endif 1484 1517 if( journal_format>=3 ){ 1485 1518 /* Write the nRec value into the journal file header */ 1486 1519 off_t szJ; 1487 1520 if( pPager->fullSync ){ 1488 - TRACE1("SYNC\n"); 1521 + TRACE2("SYNC journal of %d\n", pPager->fd.h); 1489 1522 rc = sqlite3OsSync(&pPager->jfd); 1490 1523 if( rc!=0 ) return rc; 1491 1524 } 1492 1525 sqlite3OsSeek(&pPager->jfd, sizeof(aJournalMagic1)); 1493 1526 rc = write32bits(&pPager->jfd, pPager->nRec); 1494 1527 if( rc ) return rc; 1495 1528 ................................................................................ 1502 1535 if( rc ) return rc; 1503 1536 } 1504 1537 1505 1538 szJ = JOURNAL_HDR_SZ(pPager, journal_format) + 1506 1539 pPager->nRec*JOURNAL_PG_SZ(journal_format); 1507 1540 sqlite3OsSeek(&pPager->jfd, szJ); 1508 1541 } 1509 - TRACE1("SYNC\n"); 1542 + TRACE2("SYNC journal of %d\n", pPager->fd.h); 1510 1543 rc = sqlite3OsSync(&pPager->jfd); 1511 1544 if( rc!=0 ) return rc; 1512 1545 pPager->journalStarted = 1; 1513 1546 } 1514 1547 pPager->needSync = 0; 1515 1548 1516 1549 /* Erase the needSync flag from every page. ................................................................................ 1550 1583 if( pList==0 ) return SQLITE_OK; 1551 1584 pPager = pList->pPager; 1552 1585 1553 1586 /* At this point there may be either a RESERVED or EXCLUSIVE lock on the 1554 1587 ** database file. If there is already an EXCLUSIVE lock, the following 1555 1588 ** calls to sqlite3OsLock() are no-ops. 1556 1589 ** 1557 - ** The upgrade from a RESERVED to PENDING might return SQLITE_BUSY on 1558 - ** windows because the windows locking mechanism acquires a transient 1559 - ** PENDING lock during its attempts to get a SHARED lock. So if another 1560 - ** process were trying to get a SHARED lock at the same time this process 1561 - ** is upgrading from RESERVED to PENDING, the two could collide. 1590 + ** Moving the lock from RESERVED to EXCLUSIVE actually involves going 1591 + ** through an intermediate state PENDING. A PENDING lock prevents new 1592 + ** readers from attaching to the database but is unsufficient for us to 1593 + ** write. The idea of a PENDING lock is to prevent new readers from 1594 + ** coming in while we wait for existing readers to clear. 1562 1595 ** 1563 - ** The upgrade from PENDING to EXCLUSIVE can return SQLITE_BUSY if there 1564 - ** are still active readers that were created before the PENDING lock 1565 - ** was acquired. 1596 + ** While the pager is in the RESERVED state, the original database file 1597 + ** is unchanged and we can rollback without having to playback the 1598 + ** journal into the original database file. Once we transition to 1599 + ** EXCLUSIVE, it means the database file has been changed and any rollback 1600 + ** will require a journal playback. 1566 1601 */ 1567 1602 do { 1568 1603 rc = sqlite3OsLock(&pPager->fd, EXCLUSIVE_LOCK); 1569 1604 }while( rc==SQLITE_BUSY && 1570 1605 pPager->pBusyHandler && 1571 1606 pPager->pBusyHandler->xFunc && 1572 1607 pPager->pBusyHandler->xFunc(pPager->pBusyHandler->pArg, "", busy++) 1573 1608 ); 1574 1609 if( rc!=SQLITE_OK ){ 1575 1610 return rc; 1576 1611 } 1612 + pPager->state = PAGER_EXCLUSIVE; 1577 1613 1578 1614 while( pList ){ 1579 1615 assert( pList->dirty ); 1580 1616 sqlite3OsSeek(&pPager->fd, (pList->pgno-1)*(off_t)SQLITE_PAGE_SIZE); 1581 1617 CODEC(pPager, PGHDR_TO_DATA(pList), pList->pgno, 6); 1582 - TRACE2("STORE %d\n", pList->pgno); 1618 + TRACE2("STORE page %d\n", pList->pgno); 1583 1619 rc = sqlite3OsWrite(&pPager->fd, PGHDR_TO_DATA(pList), SQLITE_PAGE_SIZE); 1584 1620 CODEC(pPager, PGHDR_TO_DATA(pList), pList->pgno, 0); 1585 1621 if( rc ) return rc; 1586 1622 pList->dirty = 0; 1587 1623 pList = pList->pDirty; 1588 1624 } 1589 1625 return SQLITE_OK; ................................................................................ 1653 1689 pPager->pBusyHandler && 1654 1690 pPager->pBusyHandler->xFunc && 1655 1691 pPager->pBusyHandler->xFunc(pPager->pBusyHandler->pArg, "", busy++) 1656 1692 ); 1657 1693 if( rc!=SQLITE_OK ){ 1658 1694 return rc; 1659 1695 } 1660 - pPager->state = SQLITE_READLOCK; 1696 + pPager->state = PAGER_SHARED; 1661 1697 1662 1698 /* If a journal file exists, and there is no RESERVED lock on the 1663 1699 ** database file, then it either needs to be played back or deleted. 1664 1700 */ 1665 1701 if( pPager->useJournal && 1666 1702 sqlite3OsFileExists(pPager->zJournal) && 1667 - !sqlite3OsCheckWriteLock(&pPager->fd) 1703 + !sqlite3OsCheckReservedLock(&pPager->fd) 1668 1704 ){ 1669 1705 int rc; 1670 1706 1671 1707 /* Get an EXCLUSIVE lock on the database file. */ 1672 1708 rc = sqlite3OsLock(&pPager->fd, EXCLUSIVE_LOCK); 1673 1709 if( rc!=SQLITE_OK ){ 1674 - if( sqlite3OsUnlock(&pPager->fd)!=SQLITE_OK ){ 1675 - /* This should never happen! */ 1676 - rc = SQLITE_INTERNAL; 1677 - } 1710 + sqlite3OsUnlock(&pPager->fd, NO_LOCK); 1711 + pPager->state = PAGER_UNLOCK; 1678 1712 return rc; 1679 1713 } 1680 - pPager->state = SQLITE_WRITELOCK; 1714 + pPager->state = PAGER_EXCLUSIVE; 1681 1715 1682 1716 /* Open the journal for reading only. Return SQLITE_BUSY if 1683 1717 ** we are unable to open the journal file. 1684 1718 ** 1685 1719 ** The journal file does not need to be locked itself. The 1686 1720 ** journal file is never open unless the main database file holds 1687 1721 ** a write lock, so there is never any chance of two or more 1688 1722 ** processes opening the journal at the same time. 1689 1723 */ 1690 1724 rc = sqlite3OsOpenReadOnly(pPager->zJournal, &pPager->jfd); 1691 1725 if( rc!=SQLITE_OK ){ 1692 - rc = sqlite3OsUnlock(&pPager->fd); 1693 - assert( rc==SQLITE_OK ); 1726 + sqlite3OsUnlock(&pPager->fd, NO_LOCK); 1727 + pPager->state = PAGER_UNLOCK; 1694 1728 return SQLITE_BUSY; 1695 1729 } 1696 1730 pPager->journalOpen = 1; 1697 1731 pPager->journalStarted = 0; 1698 1732 1699 1733 /* Playback and delete the journal. Drop the database write 1700 1734 ** lock and reacquire the read lock. ................................................................................ 1704 1738 return rc; 1705 1739 } 1706 1740 } 1707 1741 pPg = 0; 1708 1742 }else{ 1709 1743 /* Search for page in cache */ 1710 1744 pPg = pager_lookup(pPager, pgno); 1711 - if( pPager->memDb && pPager->state==SQLITE_UNLOCK ){ 1712 - pPager->state = SQLITE_READLOCK; 1745 + if( pPager->memDb && pPager->state==PAGER_UNLOCK ){ 1746 + pPager->state = PAGER_SHARED; 1713 1747 } 1714 1748 } 1715 1749 if( pPg==0 ){ 1716 1750 /* The requested page is not in the page cache. */ 1717 1751 int h; 1718 1752 pPager->nMiss++; 1719 1753 if( pPager->nPage<pPager->mxPage || pPager->pFirst==0 || pPager->memDb ){ ................................................................................ 1824 1858 if( pPager->dbSize<(int)pgno ){ 1825 1859 memset(PGHDR_TO_DATA(pPg), 0, SQLITE_PAGE_SIZE); 1826 1860 }else{ 1827 1861 int rc; 1828 1862 assert( pPager->memDb==0 ); 1829 1863 sqlite3OsSeek(&pPager->fd, (pgno-1)*(off_t)SQLITE_PAGE_SIZE); 1830 1864 rc = sqlite3OsRead(&pPager->fd, PGHDR_TO_DATA(pPg), SQLITE_PAGE_SIZE); 1831 - TRACE2("FETCH %d\n", pPg->pgno); 1865 + TRACE2("FETCH page %d\n", pPg->pgno); 1832 1866 CODEC(pPager, PGHDR_TO_DATA(pPg), pPg->pgno, 3); 1833 1867 if( rc!=SQLITE_OK ){ 1834 1868 off_t fileSize; 1835 1869 if( sqlite3OsFileSize(&pPager->fd,&fileSize)!=SQLITE_OK 1836 1870 || fileSize>=pgno*SQLITE_PAGE_SIZE ){ 1837 1871 sqlite3pager_unref(PGHDR_TO_DATA(pPg)); 1838 1872 return rc; ................................................................................ 1923 1957 pager_reset(pPager); 1924 1958 } 1925 1959 } 1926 1960 return SQLITE_OK; 1927 1961 } 1928 1962 1929 1963 /* 1930 -** Create a journal file for pPager. There should already be a write 1931 -** lock on the database file when this routine is called. 1964 +** Create a journal file for pPager. There should already be a RESERVED 1965 +** or EXCLUSIVE lock on the database file when this routine is called. 1932 1966 ** 1933 1967 ** Return SQLITE_OK if everything. Return an error code and release the 1934 1968 ** write lock if anything goes wrong. 1935 1969 */ 1936 1970 static int pager_open_journal(Pager *pPager){ 1937 1971 int rc; 1938 - assert( pPager->state==SQLITE_WRITELOCK ); 1972 + assert( pPager->state>=PAGER_RESERVED ); 1939 1973 assert( pPager->journalOpen==0 ); 1940 1974 assert( pPager->useJournal ); 1941 1975 sqlite3pager_pagecount(pPager); 1942 1976 pPager->aInJournal = sqliteMalloc( pPager->dbSize/8 + 1 ); 1943 1977 if( pPager->aInJournal==0 ){ 1944 - sqlite3OsLock(&pPager->fd, SHARED_LOCK); 1945 - pPager->state = SQLITE_READLOCK; 1978 + sqlite3OsUnlock(&pPager->fd, SHARED_LOCK); 1979 + pPager->state = PAGER_SHARED; 1946 1980 return SQLITE_NOMEM; 1947 1981 } 1948 1982 rc = sqlite3OsOpenExclusive(pPager->zJournal, &pPager->jfd,pPager->tempFile); 1949 1983 if( rc!=SQLITE_OK ){ 1950 1984 sqliteFree(pPager->aInJournal); 1951 1985 pPager->aInJournal = 0; 1952 - sqlite3OsLock(&pPager->fd, SHARED_LOCK); 1953 - pPager->state = SQLITE_READLOCK; 1986 + sqlite3OsUnlock(&pPager->fd, SHARED_LOCK); 1987 + pPager->state = PAGER_SHARED; 1954 1988 return SQLITE_CANTOPEN; 1955 1989 } 1956 1990 sqlite3OsOpenDirectory(pPager->zDirectory, &pPager->jfd); 1957 1991 pPager->journalOpen = 1; 1958 1992 pPager->journalStarted = 0; 1959 1993 pPager->needSync = 0; 1960 1994 pPager->alwaysRollback = 0; ................................................................................ 2035 2069 ** The second parameter indicates how much space in bytes to reserve for a 2036 2070 ** master journal file-name at the start of the journal when it is created. 2037 2071 ** 2038 2072 ** A journal file is opened if this is not a temporary file. For temporary 2039 2073 ** files, the opening of the journal file is deferred until there is an 2040 2074 ** actual need to write to the journal. 2041 2075 ** 2042 -** If the database is already write-locked, this routine is a no-op. 2076 +** If the database is already reserved for writing, this routine is a no-op. 2043 2077 */ 2044 2078 int sqlite3pager_begin(void *pData, int nMaster){ 2045 2079 PgHdr *pPg = DATA_TO_PGHDR(pData); 2046 2080 Pager *pPager = pPg->pPager; 2047 2081 int rc = SQLITE_OK; 2048 2082 assert( pPg->nRef>0 ); 2049 2083 assert( nMaster>=0 ); 2050 - assert( pPager->state!=SQLITE_UNLOCK ); 2051 - if( pPager->state==SQLITE_READLOCK ){ 2084 + assert( pPager->state!=PAGER_UNLOCK ); 2085 + if( pPager->state==PAGER_SHARED ){ 2052 2086 assert( pPager->aInJournal==0 ); 2053 2087 if( pPager->memDb ){ 2054 - pPager->state = SQLITE_WRITELOCK; 2088 + pPager->state = PAGER_EXCLUSIVE; 2055 2089 pPager->origDbSize = pPager->dbSize; 2056 2090 }else{ 2057 2091 int busy = 1; 2058 2092 do { 2059 - /* If the library grabs an EXCLUSIVE lock here, as in the commented 2060 - ** out line, then it exhibits the old locking behaviour - a writer 2061 - ** excludes all readers, not just other writers. 2062 - */ 2063 - /* rc = sqlite3OsLock(&pPager->fd, EXCLUSIVE_LOCK); */ 2064 2093 rc = sqlite3OsLock(&pPager->fd, RESERVED_LOCK); 2065 2094 }while( rc==SQLITE_BUSY && 2066 2095 pPager->pBusyHandler && 2067 2096 pPager->pBusyHandler->xFunc && 2068 2097 pPager->pBusyHandler->xFunc(pPager->pBusyHandler->pArg, "", busy++) 2069 2098 ); 2070 2099 if( rc!=SQLITE_OK ){ 2071 2100 return rc; 2072 2101 } 2073 2102 pPager->nMaster = nMaster; 2074 - pPager->state = SQLITE_WRITELOCK; 2075 - pPager->dirtyFile = 0; 2076 - TRACE1("TRANSACTION\n"); 2103 + pPager->state = PAGER_RESERVED; 2104 + pPager->dirtyCache = 0; 2105 + TRACE3("TRANSACTION %d nMaster=%d\n", pPager->fd.h, nMaster); 2077 2106 if( pPager->useJournal && !pPager->tempFile ){ 2078 2107 rc = pager_open_journal(pPager); 2079 2108 } 2080 2109 } 2081 2110 } 2082 2111 return rc; 2083 2112 } ................................................................................ 2084 2113 2085 2114 /* 2086 2115 ** Mark a data page as writeable. The page is written into the journal 2087 2116 ** if it is not there already. This routine must be called before making 2088 2117 ** changes to a page. 2089 2118 ** 2090 2119 ** The first time this routine is called, the pager creates a new 2091 -** journal and acquires a write lock on the database. If the write 2120 +** journal and acquires a RESERVED lock on the database. If the RESERVED 2092 2121 ** lock could not be acquired, this routine returns SQLITE_BUSY. The 2093 2122 ** calling routine must check for that return value and be careful not to 2094 2123 ** change any page data until this routine returns SQLITE_OK. 2095 2124 ** 2096 2125 ** If the journal file could not be written because the disk is full, 2097 2126 ** then this routine returns SQLITE_FULL and does an immediate rollback. 2098 2127 ** All subsequent write attempts also return SQLITE_FULL until there ................................................................................ 2114 2143 } 2115 2144 2116 2145 /* Mark the page as dirty. If the page has already been written 2117 2146 ** to the journal then we can return right away. 2118 2147 */ 2119 2148 pPg->dirty = 1; 2120 2149 if( pPg->inJournal && (pPg->inStmt || pPager->stmtInUse==0) ){ 2121 - pPager->dirtyFile = 1; 2150 + pPager->dirtyCache = 1; 2122 2151 return SQLITE_OK; 2123 2152 } 2124 2153 2125 2154 /* If we get this far, it means that the page needs to be 2126 2155 ** written to the transaction journal or the ckeckpoint journal 2127 2156 ** or both. 2128 2157 ** 2129 2158 ** First check to see that the transaction journal exists and 2130 2159 ** create it if it does not. 2131 2160 */ 2132 - assert( pPager->state!=SQLITE_UNLOCK ); 2161 + assert( pPager->state!=PAGER_UNLOCK ); 2133 2162 rc = sqlite3pager_begin(pData, 0); 2134 2163 if( rc!=SQLITE_OK ){ 2135 2164 return rc; 2136 2165 } 2137 - assert( pPager->state==SQLITE_WRITELOCK ); 2166 + assert( pPager->state>=PAGER_RESERVED ); 2138 2167 if( !pPager->journalOpen && pPager->useJournal ){ 2139 2168 rc = pager_open_journal(pPager); 2140 2169 if( rc!=SQLITE_OK ) return rc; 2141 2170 } 2142 2171 assert( pPager->journalOpen || !pPager->useJournal ); 2143 - pPager->dirtyFile = 1; 2172 + pPager->dirtyCache = 1; 2144 2173 2145 - /* The transaction journal now exists and we have a write lock on the 2146 - ** main database file. Write the current page to the transaction 2147 - ** journal if it is not there already. 2174 + /* The transaction journal now exists and we have a RESERVED or an 2175 + ** EXCLUSIVE lock on the main database file. Write the current page to 2176 + ** the transaction journal if it is not there already. 2148 2177 */ 2149 2178 if( !pPg->inJournal && (pPager->useJournal || pPager->memDb) ){ 2150 2179 if( (int)pPg->pgno <= pPager->origDbSize ){ 2151 2180 int szPg; 2152 2181 u32 saved; 2153 2182 if( pPager->memDb ){ 2154 2183 PgHistory *pHist = PGHDR_TO_HIST(pPg, pPager); 2155 - TRACE2("JOURNAL %d\n", pPg->pgno); 2184 + TRACE2("JOURNAL page %d\n", pPg->pgno); 2156 2185 assert( pHist->pOrig==0 ); 2157 2186 pHist->pOrig = sqliteMallocRaw( pPager->pageSize ); 2158 2187 if( pHist->pOrig ){ 2159 2188 memcpy(pHist->pOrig, PGHDR_TO_DATA(pPg), pPager->pageSize); 2160 2189 } 2161 2190 pPg->inJournal = 1; 2162 2191 }else{ ................................................................................ 2167 2196 szPg = SQLITE_PAGE_SIZE+8; 2168 2197 }else{ 2169 2198 szPg = SQLITE_PAGE_SIZE+4; 2170 2199 } 2171 2200 store32bits(pPg->pgno, pPg, -4); 2172 2201 CODEC(pPager, pData, pPg->pgno, 7); 2173 2202 rc = sqlite3OsWrite(&pPager->jfd, &((char*)pData)[-4], szPg); 2174 - TRACE3("JOURNAL %d %d\n", pPg->pgno, pPg->needSync); 2203 + TRACE3("JOURNAL page %d needSync=%d\n", pPg->pgno, pPg->needSync); 2175 2204 CODEC(pPager, pData, pPg->pgno, 0); 2176 2205 if( journal_format>=JOURNAL_FORMAT_3 ){ 2177 2206 *(u32*)PGHDR_TO_EXTRA(pPg) = saved; 2178 2207 } 2179 2208 if( rc!=SQLITE_OK ){ 2180 2209 sqlite3pager_rollback(pPager); 2181 2210 pPager->errMask |= PAGER_ERR_FULL; ................................................................................ 2189 2218 if( pPager->stmtInUse ){ 2190 2219 pPager->aInStmt[pPg->pgno/8] |= 1<<(pPg->pgno&7); 2191 2220 page_add_to_stmt_list(pPg); 2192 2221 } 2193 2222 } 2194 2223 }else{ 2195 2224 pPg->needSync = !pPager->journalStarted && !pPager->noSync; 2196 - TRACE3("APPEND %d %d\n", pPg->pgno, pPg->needSync); 2225 + TRACE3("APPEND page %d needSync=%d\n", pPg->pgno, pPg->needSync); 2197 2226 } 2198 2227 if( pPg->needSync ){ 2199 2228 pPager->needSync = 1; 2200 2229 } 2201 2230 } 2202 2231 2203 2232 /* If the statement journal is open and the page is not in it, ................................................................................ 2210 2239 if( pPager->memDb ){ 2211 2240 PgHistory *pHist = PGHDR_TO_HIST(pPg, pPager); 2212 2241 assert( pHist->pStmt==0 ); 2213 2242 pHist->pStmt = sqliteMallocRaw( pPager->pageSize ); 2214 2243 if( pHist->pStmt ){ 2215 2244 memcpy(pHist->pStmt, PGHDR_TO_DATA(pPg), pPager->pageSize); 2216 2245 } 2217 - TRACE2("STMT-JOURNAL %d\n", pPg->pgno); 2246 + TRACE2("STMT-JOURNAL page %d\n", pPg->pgno); 2218 2247 }else{ 2219 2248 store32bits(pPg->pgno, pPg, -4); 2220 2249 CODEC(pPager, pData, pPg->pgno, 7); 2221 2250 rc = sqlite3OsWrite(&pPager->stfd, ((char*)pData)-4, SQLITE_PAGE_SIZE+4); 2222 - TRACE2("STMT-JOURNAL %d\n", pPg->pgno); 2251 + TRACE2("STMT-JOURNAL page %d\n", pPg->pgno); 2223 2252 CODEC(pPager, pData, pPg->pgno, 0); 2224 2253 if( rc!=SQLITE_OK ){ 2225 2254 sqlite3pager_rollback(pPager); 2226 2255 pPager->errMask |= PAGER_ERR_FULL; 2227 2256 return rc; 2228 2257 } 2229 2258 pPager->stmtNRec++; ................................................................................ 2306 2335 ** When the database file grows, we must make sure that the last page 2307 2336 ** gets written at least once so that the disk file will be the correct 2308 2337 ** size. If you do not write this page and the size of the file 2309 2338 ** on the disk ends up being too small, that can lead to database 2310 2339 ** corruption during the next transaction. 2311 2340 */ 2312 2341 }else{ 2313 - TRACE2("DONT_WRITE %d\n", pgno); 2342 + TRACE3("DONT_WRITE page %d of %d\n", pgno, pPager->fd.h); 2314 2343 pPg->dirty = 0; 2315 2344 } 2316 2345 } 2317 2346 } 2318 2347 2319 2348 /* 2320 2349 ** A call to this routine tells the pager that if a rollback occurs, ................................................................................ 2322 2351 ** means that the pager does not have to record the given page in the 2323 2352 ** rollback journal. 2324 2353 */ 2325 2354 void sqlite3pager_dont_rollback(void *pData){ 2326 2355 PgHdr *pPg = DATA_TO_PGHDR(pData); 2327 2356 Pager *pPager = pPg->pPager; 2328 2357 2329 - if( pPager->state!=SQLITE_WRITELOCK || pPager->journalOpen==0 ) return; 2358 + if( pPager->state!=PAGER_EXCLUSIVE || pPager->journalOpen==0 ) return; 2330 2359 if( pPg->alwaysRollback || pPager->alwaysRollback || pPager->memDb ) return; 2331 2360 if( !pPg->inJournal && (int)pPg->pgno <= pPager->origDbSize ){ 2332 2361 assert( pPager->aInJournal!=0 ); 2333 2362 pPager->aInJournal[pPg->pgno/8] |= 1<<(pPg->pgno&7); 2334 2363 pPg->inJournal = 1; 2335 2364 if( pPager->stmtInUse ){ 2336 2365 pPager->aInStmt[pPg->pgno/8] |= 1<<(pPg->pgno&7); 2337 2366 page_add_to_stmt_list(pPg); 2338 2367 } 2339 - TRACE2("DONT_ROLLBACK %d\n", pPg->pgno); 2368 + TRACE3("DONT_ROLLBACK page %d of %d\n", pPg->pgno, pPager->fd.h); 2340 2369 } 2341 2370 if( pPager->stmtInUse && !pPg->inStmt && (int)pPg->pgno<=pPager->stmtSize ){ 2342 2371 assert( pPg->inJournal || (int)pPg->pgno>pPager->origDbSize ); 2343 2372 assert( pPager->aInStmt!=0 ); 2344 2373 pPager->aInStmt[pPg->pgno/8] |= 1<<(pPg->pgno&7); 2345 2374 page_add_to_stmt_list(pPg); 2346 2375 } ................................................................................ 2375 2404 } 2376 2405 return rc; 2377 2406 } 2378 2407 if( pPager->errMask!=0 ){ 2379 2408 rc = pager_errcode(pPager); 2380 2409 return rc; 2381 2410 } 2382 - if( pPager->state!=SQLITE_WRITELOCK ){ 2411 + if( pPager->state<PAGER_RESERVED ){ 2383 2412 return SQLITE_ERROR; 2384 2413 } 2385 - TRACE1("COMMIT\n"); 2414 + TRACE2("COMMIT %d\n", pPager->fd.h); 2386 2415 if( pPager->memDb ){ 2387 2416 pPg = pager_get_all_dirty_pages(pPager); 2388 2417 while( pPg ){ 2389 2418 clearHistory(PGHDR_TO_HIST(pPg, pPager)); 2390 2419 pPg->dirty = 0; 2391 2420 pPg->inJournal = 0; 2392 2421 pPg->inStmt = 0; 2393 2422 pPg->pPrevStmt = pPg->pNextStmt = 0; 2394 2423 pPg = pPg->pDirty; 2395 2424 } 2396 2425 pPager->pStmt = 0; 2397 - pPager->state = SQLITE_READLOCK; 2426 + pPager->state = PAGER_SHARED; 2398 2427 return SQLITE_OK; 2399 2428 } 2400 -#if 0 2401 - if( pPager->dirtyFile==0 ){ 2429 + if( pPager->dirtyCache==0 ){ 2402 2430 /* Exit early (without doing the time-consuming sqlite3OsSync() calls) 2403 2431 ** if there have been no changes to the database file. */ 2404 2432 assert( pPager->needSync==0 ); 2405 2433 rc = pager_unwritelock(pPager); 2406 2434 pPager->dbSize = -1; 2407 2435 return rc; 2408 2436 } 2409 2437 assert( pPager->journalOpen ); 2438 +#if 0 2410 2439 rc = syncJournal(pPager, 0); 2411 2440 if( rc!=SQLITE_OK ){ 2412 2441 goto commit_abort; 2413 2442 } 2414 2443 pPg = pager_get_all_dirty_pages(pPager); 2415 2444 if( pPg ){ 2416 2445 rc = pager_write_pagelist(pPg); 2417 2446 if( rc || (!pPager->noSync && sqlite3OsSync(&pPager->fd)!=SQLITE_OK) ){ 2418 2447 goto commit_abort; 2419 2448 } 2420 2449 } 2421 2450 #endif 2422 2451 rc = sqlite3pager_sync(pPager, 0); 2423 - if( rc!=SQLITE_OK ) goto commit_abort; 2424 - 2452 + if( rc!=SQLITE_OK ){ 2453 + goto commit_abort; 2454 + } 2425 2455 rc = pager_unwritelock(pPager); 2426 2456 pPager->dbSize = -1; 2427 2457 return rc; 2428 2458 2429 2459 /* Jump here if anything goes wrong during the commit process. 2430 2460 */ 2431 2461 commit_abort: ................................................................................ 2433 2463 if( rc==SQLITE_OK ){ 2434 2464 rc = SQLITE_FULL; 2435 2465 } 2436 2466 return rc; 2437 2467 } 2438 2468 2439 2469 /* 2440 -** Rollback all changes. The database falls back to read-only mode. 2470 +** Rollback all changes. The database falls back to PAGER_SHARED mode. 2441 2471 ** All in-memory cache pages revert to their original data contents. 2442 2472 ** The journal is deleted. 2443 2473 ** 2444 2474 ** This routine cannot fail unless some other process is not following 2445 2475 ** the correct locking protocol (SQLITE_PROTOCOL) or unless some other 2446 2476 ** process is writing trash into the journal file (SQLITE_CORRUPT) or 2447 2477 ** unless a prior malloc() failed (SQLITE_NOMEM). Appropriate error 2448 2478 ** codes are returned for all these occasions. Otherwise, 2449 2479 ** SQLITE_OK is returned. 2450 2480 */ 2451 2481 int sqlite3pager_rollback(Pager *pPager){ 2452 2482 int rc; 2453 - TRACE1("ROLLBACK\n"); 2483 + TRACE2("ROLLBACK %d\n", pPager->fd.h); 2454 2484 if( pPager->memDb ){ 2455 2485 PgHdr *p; 2456 2486 for(p=pPager->pAll; p; p=p->pNextAll){ 2457 2487 PgHistory *pHist; 2458 2488 if( !p->dirty ) continue; 2459 2489 pHist = PGHDR_TO_HIST(p, pPager); 2460 2490 if( pHist->pOrig ){ ................................................................................ 2469 2499 p->inStmt = 0; 2470 2500 p->pPrevStmt = p->pNextStmt = 0; 2471 2501 } 2472 2502 pPager->pStmt = 0; 2473 2503 pPager->dbSize = pPager->origDbSize; 2474 2504 memoryTruncate(pPager); 2475 2505 pPager->stmtInUse = 0; 2476 - pPager->state = SQLITE_READLOCK; 2506 + pPager->state = PAGER_SHARED; 2477 2507 return SQLITE_OK; 2478 2508 } 2479 2509 2480 - if( !pPager->dirtyFile || !pPager->journalOpen ){ 2510 + if( !pPager->dirtyCache || !pPager->journalOpen ){ 2481 2511 rc = pager_unwritelock(pPager); 2482 2512 pPager->dbSize = -1; 2483 2513 return rc; 2484 2514 } 2485 2515 2486 2516 if( pPager->errMask!=0 && pPager->errMask!=PAGER_ERR_FULL ){ 2487 - if( pPager->state>=SQLITE_WRITELOCK ){ 2517 + if( pPager->state>=PAGER_EXCLUSIVE ){ 2488 2518 pager_playback(pPager, 1); 2489 2519 } 2490 2520 return pager_errcode(pPager); 2491 2521 } 2492 - if( pPager->state!=SQLITE_WRITELOCK ){ 2493 - return SQLITE_OK; 2522 + if( pPager->state==PAGER_RESERVED ){ 2523 + int rc2; 2524 + rc = pager_reload_cache(pPager); 2525 + rc2 = pager_unwritelock(pPager); 2526 + if( rc==SQLITE_OK ){ 2527 + rc = rc2; 2528 + } 2529 + }else{ 2530 + rc = pager_playback(pPager, 1); 2494 2531 } 2495 - rc = pager_playback(pPager, 1); 2496 2532 if( rc!=SQLITE_OK ){ 2497 2533 rc = SQLITE_CORRUPT; 2498 2534 pPager->errMask |= PAGER_ERR_CORRUPT; 2499 2535 } 2500 2536 pPager->dbSize = -1; 2501 2537 return rc; 2502 2538 } ................................................................................ 2533 2569 ** open. A new statement journal is created that can be used to rollback 2534 2570 ** changes of a single SQL command within a larger transaction. 2535 2571 */ 2536 2572 int sqlite3pager_stmt_begin(Pager *pPager){ 2537 2573 int rc; 2538 2574 char zTemp[SQLITE_TEMPNAME_SIZE]; 2539 2575 assert( !pPager->stmtInUse ); 2540 - TRACE1("STMT-BEGIN\n"); 2576 + TRACE2("STMT-BEGIN %d\n", pPager->fd.h); 2541 2577 if( pPager->memDb ){ 2542 2578 pPager->stmtInUse = 1; 2543 2579 pPager->stmtSize = pPager->dbSize; 2544 2580 return SQLITE_OK; 2545 2581 } 2546 2582 if( !pPager->journalOpen ){ 2547 2583 pPager->stmtAutoopen = 1; ................................................................................ 2582 2618 2583 2619 /* 2584 2620 ** Commit a statement. 2585 2621 */ 2586 2622 int sqlite3pager_stmt_commit(Pager *pPager){ 2587 2623 if( pPager->stmtInUse ){ 2588 2624 PgHdr *pPg, *pNext; 2589 - TRACE1("STMT-COMMIT\n"); 2625 + TRACE2("STMT-COMMIT %d\n", pPager->fd.h); 2590 2626 if( !pPager->memDb ){ 2591 2627 sqlite3OsSeek(&pPager->stfd, 0); 2592 2628 /* sqlite3OsTruncate(&pPager->stfd, 0); */ 2593 2629 sqliteFree( pPager->aInStmt ); 2594 2630 pPager->aInStmt = 0; 2595 2631 } 2596 2632 for(pPg=pPager->pStmt; pPg; pPg=pNext){ ................................................................................ 2614 2650 2615 2651 /* 2616 2652 ** Rollback a statement. 2617 2653 */ 2618 2654 int sqlite3pager_stmt_rollback(Pager *pPager){ 2619 2655 int rc; 2620 2656 if( pPager->stmtInUse ){ 2621 - TRACE1("STMT-ROLLBACK\n"); 2657 + TRACE2("STMT-ROLLBACK %d\n", pPager->fd.h); 2622 2658 if( pPager->memDb ){ 2623 2659 PgHdr *pPg; 2624 2660 for(pPg=pPager->pStmt; pPg; pPg=pPg->pNextStmt){ 2625 2661 PgHistory *pHist = PGHDR_TO_HIST(pPg, pPager); 2626 2662 if( pHist->pStmt ){ 2627 2663 memcpy(PGHDR_TO_DATA(pPg), pHist->pStmt, pPager->pageSize); 2628 2664 sqliteFree(pHist->pStmt); ................................................................................ 2678 2714 */ 2679 2715 int sqlite3pager_sync(Pager *pPager, const char *zMaster){ 2680 2716 int rc = SQLITE_OK; 2681 2717 2682 2718 /* If this is an in-memory db, or no pages have been written to, this 2683 2719 ** function is a no-op. 2684 2720 */ 2685 - if( !pPager->memDb && pPager->dirtyFile ){ 2721 + if( !pPager->memDb && pPager->dirtyCache ){ 2686 2722 PgHdr *pPg; 2687 2723 assert( pPager->journalOpen ); 2688 2724 2689 2725 /* Sync the journal file */ 2690 2726 rc = syncJournal(pPager, zMaster); 2691 2727 if( rc!=SQLITE_OK ) goto sync_exit; 2692 2728
Changes to src/pager.h.
9 9 ** May you share freely, never taking more than you give. 10 10 ** 11 11 ************************************************************************* 12 12 ** This header file defines the interface that the sqlite page cache 13 13 ** subsystem. The page cache subsystem reads and writes a file a page 14 14 ** at a time and provides a journal for rollback. 15 15 ** 16 -** @(#) $Id: pager.h,v 1.32 2004/06/09 14:17:21 drh Exp $ 16 +** @(#) $Id: pager.h,v 1.33 2004/06/09 17:37:28 drh Exp $ 17 17 */ 18 18 19 19 /* 20 20 ** The size of a page. 21 21 ** 22 22 ** You can change this value to another (reasonable) value you want. 23 23 ** It need not be a power of two, though the interface to the disk ................................................................................ 69 69 ** See source code comments for a detailed description of the following 70 70 ** routines: 71 71 */ 72 72 int sqlite3pager_open(Pager **ppPager, const char *zFilename, 73 73 int nPage, int nExtra, int useJournal, 74 74 void *pBusyHandler); 75 75 void sqlite3pager_set_destructor(Pager*, void(*)(void*,int)); 76 +void sqlite3pager_set_reiniter(Pager*, void(*)(void*,int)); 76 77 void sqlite3pager_set_cachesize(Pager*, int); 77 78 int sqlite3pager_close(Pager *pPager); 78 79 int sqlite3pager_get(Pager *pPager, Pgno pgno, void **ppPage); 79 80 void *sqlite3pager_lookup(Pager *pPager, Pgno pgno); 80 81 int sqlite3pager_ref(void*); 81 82 int sqlite3pager_unref(void*); 82 83 Pgno sqlite3pager_pagenumber(void*);
Changes to src/test1.c.
9 9 ** May you share freely, never taking more than you give. 10 10 ** 11 11 ************************************************************************* 12 12 ** Code for testing the printf() interface to SQLite. This code 13 13 ** is not included in the SQLite library. It is used for automated 14 14 ** testing of the SQLite library. 15 15 ** 16 -** $Id: test1.c,v 1.73 2004/06/09 14:01:51 drh Exp $ 16 +** $Id: test1.c,v 1.74 2004/06/09 17:37:28 drh Exp $ 17 17 */ 18 18 #include "sqliteInt.h" 19 19 #include "tcl.h" 20 20 #include "os.h" 21 21 #include <stdlib.h> 22 22 #include <string.h> 23 23 ................................................................................ 1778 1778 Tcl_GetString(objv[0]), " filehandle", 0); 1779 1779 return TCL_ERROR; 1780 1780 } 1781 1781 1782 1782 if( getFilePointer(interp, Tcl_GetString(objv[1]), &pFile) ){ 1783 1783 return TCL_ERROR; 1784 1784 } 1785 - rc = sqlite3OsUnlock(pFile); 1785 + rc = sqlite3OsUnlock(pFile, NO_LOCK); 1786 1786 if( rc!=SQLITE_OK ){ 1787 1787 Tcl_SetResult(interp, (char *)errorName(rc), TCL_STATIC); 1788 1788 return TCL_ERROR; 1789 1789 } 1790 1790 return TCL_OK; 1791 1791 } 1792 1792
Changes to src/vdbeaux.c.
930 930 } 931 931 } 932 932 933 933 /* The simple case - no more than one database file (not counting the TEMP 934 934 ** database) has a transaction active. There is no need for the 935 935 ** master-journal. 936 936 */ 937 - if( nTrans<=1 ){ 937 + if( nTrans<=100 ){ /**** FIX ME ****/ 938 938 for(i=0; rc==SQLITE_OK && i<db->nDb; i++){ 939 939 Btree *pBt = db->aDb[i].pBt; 940 940 if( pBt ){ 941 941 rc = sqlite3BtreeSync(pBt, 0); 942 942 } 943 943 } 944 944 ................................................................................ 960 960 else{ 961 961 char *zMaster = 0; /* File-name for the master journal */ 962 962 char const *zMainFile = sqlite3BtreeGetFilename(db->aDb[0].pBt); 963 963 OsFile master; 964 964 965 965 /* Select a master journal file name */ 966 966 do { 967 - int random; 968 - if( zMaster ){ 969 - sqliteFree(zMaster); 970 - } 967 + u32 random; 968 + sqliteFree(zMaster); 971 969 sqlite3Randomness(sizeof(random), &random); 972 - zMaster = sqlite3_mprintf("%s%d", zMainFile, random); 970 + zMaster = sqlite3MPrintf("%s-mj%08X", zMainFile, random); 973 971 if( !zMaster ){ 974 972 return SQLITE_NOMEM; 975 973 } 976 974 }while( sqlite3OsFileExists(zMaster) ); 977 975 978 976 /* Open the master journal. */ 979 977 rc = sqlite3OsOpenExclusive(zMaster, &master, 0);
Changes to test/attach2.test.
8 8 # May you share freely, never taking more than you give. 9 9 # 10 10 #*********************************************************************** 11 11 # This file implements regression tests for SQLite library. The 12 12 # focus of this script is testing the ATTACH and DETACH commands 13 13 # and related functionality. 14 14 # 15 -# $Id: attach2.test,v 1.13 2004/06/09 14:01:58 drh Exp $ 15 +# $Id: attach2.test,v 1.14 2004/06/09 17:37:29 drh Exp $ 16 16 # 17 -set sqlite_os_trace 0 18 17 19 18 set testdir [file dirname $argv0] 20 19 source $testdir/tester.tcl 21 20 22 21 23 22 # Ticket #354 24 23 # ................................................................................ 143 142 set rc 144 143 } {0} 145 144 146 145 db close 147 146 for {set i 2} {$i<=15} {incr i} { 148 147 catch {db$i close} 149 148 } 149 + 150 +# A procedure to verify the status of locks on a database. 151 +# 152 +proc lock_status {testnum db expected_result} { 153 + do_test attach2-$testnum [subst { 154 + execsql {PRAGMA lock_status} $db 155 + }] $expected_result 156 +} 150 157 set sqlite_os_trace 0 151 158 152 159 # Tests attach2-4.* test that read-locks work correctly with attached 153 160 # databases. 154 161 do_test attach2-4.1 { 155 162 sqlite db test.db 156 163 sqlite db2 test.db 157 164 execsql {ATTACH 'test2.db' as file2} 158 165 execsql {ATTACH 'test2.db' as file2} db2 159 166 } {} 160 167 168 +lock_status 4.1.1 db {main unlocked temp unlocked file2 unlocked} 169 +lock_status 4.1.2 db2 {main unlocked temp unlocked file2 unlocked} 170 + 161 171 do_test attach2-4.2 { 162 172 # Handle 'db' read-locks test.db 163 173 execsql {BEGIN} 164 174 execsql {SELECT * FROM t1} 165 175 # Lock status: 166 176 # db - shared(main) 167 177 # db2 - 168 178 } {} 179 + 180 +lock_status 4.2.1 db {main shared temp shared file2 unlocked} 181 +lock_status 4.2.2 db2 {main unlocked temp unlocked file2 unlocked} 182 + 169 183 do_test attach2-4.3 { 170 184 # The read lock held by db does not prevent db2 from reading test.db 171 185 execsql {SELECT * FROM t1} db2 172 186 } {} 187 + 188 +lock_status 4.3.1 db {main shared temp shared file2 unlocked} 189 +lock_status 4.3.2 db2 {main unlocked temp unlocked file2 unlocked} 190 + 173 191 do_test attach2-4.4 { 174 192 # db is holding a read lock on test.db, so we should not be able 175 193 # to commit a write to test.db from db2 176 194 catchsql { 177 195 INSERT INTO t1 VALUES(1, 2) 178 196 } db2 179 197 } {1 {database is locked}} 198 + 199 +lock_status 4.4.1 db {main shared temp shared file2 unlocked} 200 +lock_status 4.4.2 db2 {main unlocked temp unlocked file2 unlocked} 201 + 180 202 do_test attach2-4.5 { 181 203 # Handle 'db2' reserves file2. 182 204 execsql {BEGIN} db2 183 205 execsql {INSERT INTO file2.t1 VALUES(1, 2)} db2 184 206 # Lock status: 185 207 # db - shared(main) 186 208 # db2 - reserved(file2) 187 209 } {} 210 + 211 +lock_status 4.5.1 db {main shared temp shared file2 unlocked} 212 +lock_status 4.5.2 db2 {main unlocked temp reserved file2 reserved} 213 + 188 214 do_test attach2-4.6.1 { 189 215 # Reads are allowed against a reserved database. 190 216 catchsql { 191 217 SELECT * FROM file2.t1; 192 218 } 193 219 # Lock status: 194 220 # db - shared(main), shared(file2) 195 221 # db2 - reserved(file2) 196 222 } {0 {}} 223 + 224 +lock_status 4.6.1.1 db {main shared temp shared file2 shared} 225 +lock_status 4.6.1.2 db2 {main unlocked temp reserved file2 reserved} 226 + 197 227 do_test attach2-4.6.2 { 198 228 # Writes against a reserved database are not allowed. 199 229 catchsql { 200 230 UPDATE file2.t1 SET a=0; 201 231 } 202 232 } {1 {database is locked}} 233 + 234 +lock_status 4.6.2.1 db {main shared temp reserved file2 shared} 235 +lock_status 4.6.2.2 db2 {main unlocked temp reserved file2 reserved} 236 + 203 237 do_test attach2-4.7 { 204 238 # Ensure handle 'db' retains the lock on the main file after 205 239 # failing to obtain a write-lock on file2. 206 240 catchsql { 207 241 INSERT INTO t1 VALUES(1, 2) 208 242 } db2 209 -} {1 {database is locked}} 243 +} {0 {}} 244 + 245 +lock_status 4.7.1 db {main shared temp reserved file2 shared} 246 +lock_status 4.7.2 db2 {main reserved temp reserved file2 reserved} 247 + 210 248 do_test attach2-4.8 { 211 - # Read lock the main file with db2. Now both db and db2 have a read lock 212 - # on the main file, db2 has a write-lock on file2. 249 + # We should still be able to read test.db from db2 213 250 execsql {SELECT * FROM t1} db2 214 - # Lock status: 215 - # db - shared(main), shared(file2) 216 - # db2 - shared(main), reserved(file2) 217 -} {} 251 +} {1 2} 252 + 253 +lock_status 4.8.1 db {main shared temp reserved file2 shared} 254 +lock_status 4.8.2 db2 {main reserved temp reserved file2 reserved} 255 + 218 256 do_test attach2-4.9 { 219 257 # Try to upgrade the handle 'db' lock. 220 258 catchsql { 221 259 INSERT INTO t1 VALUES(1, 2) 222 260 } 223 - list $r $msg 224 261 } {1 {database is locked}} 262 + 263 +lock_status 4.9.1 db {main shared temp reserved file2 shared} 264 +lock_status 4.9.2 db2 {main reserved temp reserved file2 reserved} 265 + 225 266 do_test attach2-4.10 { 226 267 # Release the locks held by handle 'db2' 227 268 execsql {COMMIT} db2 228 269 } {} 270 + 271 +lock_status 4.10.1 db {main shared temp reserved file2 shared} 272 +lock_status 4.10.2 db2 {main unlocked temp unlocked file2 unlocked} 273 + 229 274 do_test attach2-4.11 { 230 275 execsql {SELECT * FROM file2.t1} 231 276 } {1 2} 232 277 do_test attach2-4.12 { 233 278 execsql {INSERT INTO t1 VALUES(1, 2)} 234 279 } {} 235 280 do_test attach2-4.13 {