recovery.c revision d920038068bbda80df9060be55da5726a6efe9a4
1/* 2 * linux/fs/jbd/recovery.c 3 * 4 * Written by Stephen C. Tweedie <sct@redhat.com>, 1999 5 * 6 * Copyright 1999-2000 Red Hat Software --- All Rights Reserved 7 * 8 * This file is part of the Linux kernel and is made available under 9 * the terms of the GNU General Public License, version 2, or at your 10 * option, any later version, incorporated herein by reference. 11 * 12 * Journal recovery routines for the generic filesystem journaling code; 13 * part of the ext2fs journaling system. 14 */ 15 16#ifndef __KERNEL__ 17#include "config.h" 18#include "jfs_user.h" 19#else 20#include <linux/time.h> 21#include <linux/fs.h> 22#include <linux/jbd.h> 23#include <linux/errno.h> 24#include <linux/slab.h> 25#endif 26 27/* 28 * Maintain information about the progress of the recovery job, so that 29 * the different passes can carry information between them. 30 */ 31struct recovery_info 32{ 33 tid_t start_transaction; 34 tid_t end_transaction; 35 36 int nr_replays; 37 int nr_revokes; 38 int nr_revoke_hits; 39}; 40 41enum passtype {PASS_SCAN, PASS_REVOKE, PASS_REPLAY}; 42static int do_one_pass(journal_t *journal, 43 struct recovery_info *info, enum passtype pass); 44static int scan_revoke_records(journal_t *, struct buffer_head *, 45 tid_t, struct recovery_info *); 46 47#ifdef __KERNEL__ 48 49/* Release readahead buffers after use */ 50static void journal_brelse_array(struct buffer_head *b[], int n) 51{ 52 while (--n >= 0) 53 brelse (b[n]); 54} 55 56 57/* 58 * When reading from the journal, we are going through the block device 59 * layer directly and so there is no readahead being done for us. We 60 * need to implement any readahead ourselves if we want it to happen at 61 * all. Recovery is basically one long sequential read, so make sure we 62 * do the IO in reasonably large chunks. 63 * 64 * This is not so critical that we need to be enormously clever about 65 * the readahead size, though. 128K is a purely arbitrary, good-enough 66 * fixed value. 67 */ 68 69#define MAXBUF 8 70static int do_readahead(journal_t *journal, unsigned int start) 71{ 72 int err; 73 unsigned int max, nbufs, next; 74 unsigned long long blocknr; 75 struct buffer_head *bh; 76 77 struct buffer_head * bufs[MAXBUF]; 78 79 /* Do up to 128K of readahead */ 80 max = start + (128 * 1024 / journal->j_blocksize); 81 if (max > journal->j_maxlen) 82 max = journal->j_maxlen; 83 84 /* Do the readahead itself. We'll submit MAXBUF buffer_heads at 85 * a time to the block device IO layer. */ 86 87 nbufs = 0; 88 89 for (next = start; next < max; next++) { 90 err = journal_bmap(journal, next, &blocknr); 91 92 if (err) { 93 printk (KERN_ERR "JBD: bad block at offset %u\n", 94 next); 95 goto failed; 96 } 97 98 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize); 99 if (!bh) { 100 err = -ENOMEM; 101 goto failed; 102 } 103 104 if (!buffer_uptodate(bh) && !buffer_locked(bh)) { 105 bufs[nbufs++] = bh; 106 if (nbufs == MAXBUF) { 107 ll_rw_block(READ, nbufs, bufs); 108 journal_brelse_array(bufs, nbufs); 109 nbufs = 0; 110 } 111 } else 112 brelse(bh); 113 } 114 115 if (nbufs) 116 ll_rw_block(READ, nbufs, bufs); 117 err = 0; 118 119failed: 120 if (nbufs) 121 journal_brelse_array(bufs, nbufs); 122 return err; 123} 124 125#endif /* __KERNEL__ */ 126 127 128/* 129 * Read a block from the journal 130 */ 131 132static int jread(struct buffer_head **bhp, journal_t *journal, 133 unsigned int offset) 134{ 135 int err; 136 unsigned long long blocknr; 137 struct buffer_head *bh; 138 139 *bhp = NULL; 140 141 if (offset >= journal->j_maxlen) { 142 printk(KERN_ERR "JBD: corrupted journal superblock\n"); 143 return -EIO; 144 } 145 146 err = journal_bmap(journal, offset, &blocknr); 147 148 if (err) { 149 printk (KERN_ERR "JBD: bad block at offset %u\n", 150 offset); 151 return err; 152 } 153 154 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize); 155 if (!bh) 156 return -ENOMEM; 157 158 if (!buffer_uptodate(bh)) { 159 /* If this is a brand new buffer, start readahead. 160 Otherwise, we assume we are already reading it. */ 161 if (!buffer_req(bh)) 162 do_readahead(journal, offset); 163 wait_on_buffer(bh); 164 } 165 166 if (!buffer_uptodate(bh)) { 167 printk (KERN_ERR "JBD: Failed to read block at offset %u\n", 168 offset); 169 brelse(bh); 170 return -EIO; 171 } 172 173 *bhp = bh; 174 return 0; 175} 176 177static int jbd2_descr_block_csum_verify(journal_t *j, 178 void *buf) 179{ 180 struct journal_block_tail *tail; 181 __u32 provided, calculated; 182 183 if (!JFS_HAS_INCOMPAT_FEATURE(j, JFS_FEATURE_INCOMPAT_CSUM_V2)) 184 return 1; 185 186 tail = (struct journal_block_tail *)(buf + j->j_blocksize - 187 sizeof(struct journal_block_tail)); 188 provided = tail->t_checksum; 189 tail->t_checksum = 0; 190 calculated = ext2fs_crc32c_le(~0, j->j_superblock->s_uuid, 191 sizeof(j->j_superblock->s_uuid)); 192 calculated = ext2fs_crc32c_le(calculated, buf, j->j_blocksize); 193 tail->t_checksum = provided; 194 195 provided = ext2fs_be32_to_cpu(provided); 196 return provided == calculated; 197} 198 199/* 200 * Count the number of in-use tags in a journal descriptor block. 201 */ 202 203static int count_tags(journal_t *journal, struct buffer_head *bh) 204{ 205 char * tagp; 206 journal_block_tag_t * tag; 207 int nr = 0, size = journal->j_blocksize; 208 int tag_bytes = journal_tag_bytes(journal); 209 210 if (JFS_HAS_INCOMPAT_FEATURE(journal, JFS_FEATURE_INCOMPAT_CSUM_V2)) 211 size -= sizeof(struct journal_block_tail); 212 213 tagp = &bh->b_data[sizeof(journal_header_t)]; 214 215 while ((tagp - bh->b_data + tag_bytes) <= size) { 216 tag = (journal_block_tag_t *) tagp; 217 218 nr++; 219 tagp += tag_bytes; 220 if (!(tag->t_flags & cpu_to_be16(JFS_FLAG_SAME_UUID))) 221 tagp += 16; 222 223 if (tag->t_flags & cpu_to_be16(JFS_FLAG_LAST_TAG)) 224 break; 225 } 226 227 return nr; 228} 229 230 231/* Make sure we wrap around the log correctly! */ 232#define wrap(journal, var) \ 233do { \ 234 if (var >= (journal)->j_last) \ 235 var -= ((journal)->j_last - (journal)->j_first); \ 236} while (0) 237 238/** 239 * journal_recover - recovers a on-disk journal 240 * @journal: the journal to recover 241 * 242 * The primary function for recovering the log contents when mounting a 243 * journaled device. 244 * 245 * Recovery is done in three passes. In the first pass, we look for the 246 * end of the log. In the second, we assemble the list of revoke 247 * blocks. In the third and final pass, we replay any un-revoked blocks 248 * in the log. 249 */ 250int journal_recover(journal_t *journal) 251{ 252 int err; 253 journal_superblock_t * sb; 254 255 struct recovery_info info; 256 257 memset(&info, 0, sizeof(info)); 258 sb = journal->j_superblock; 259 260 /* 261 * The journal superblock's s_start field (the current log head) 262 * is always zero if, and only if, the journal was cleanly 263 * unmounted. 264 */ 265 266 if (!sb->s_start) { 267 jbd_debug(1, "No recovery required, last transaction %d\n", 268 be32_to_cpu(sb->s_sequence)); 269 journal->j_transaction_sequence = be32_to_cpu(sb->s_sequence) + 1; 270 return 0; 271 } 272 273 err = do_one_pass(journal, &info, PASS_SCAN); 274 if (!err) 275 err = do_one_pass(journal, &info, PASS_REVOKE); 276 if (!err) 277 err = do_one_pass(journal, &info, PASS_REPLAY); 278 279 jbd_debug(1, "JBD: recovery, exit status %d, " 280 "recovered transactions %u to %u\n", 281 err, info.start_transaction, info.end_transaction); 282 jbd_debug(1, "JBD: Replayed %d and revoked %d/%d blocks\n", 283 info.nr_replays, info.nr_revoke_hits, info.nr_revokes); 284 285 /* Restart the log at the next transaction ID, thus invalidating 286 * any existing commit records in the log. */ 287 journal->j_transaction_sequence = ++info.end_transaction; 288 289 journal_clear_revoke(journal); 290 sync_blockdev(journal->j_fs_dev); 291 return err; 292} 293 294/** 295 * journal_skip_recovery - Start journal and wipe exiting records 296 * @journal: journal to startup 297 * 298 * Locate any valid recovery information from the journal and set up the 299 * journal structures in memory to ignore it (presumably because the 300 * caller has evidence that it is out of date). 301 * This function does'nt appear to be exorted.. 302 * 303 * We perform one pass over the journal to allow us to tell the user how 304 * much recovery information is being erased, and to let us initialise 305 * the journal transaction sequence numbers to the next unused ID. 306 */ 307int journal_skip_recovery(journal_t *journal) 308{ 309 int err; 310 journal_superblock_t * sb; 311 312 struct recovery_info info; 313 314 memset (&info, 0, sizeof(info)); 315 sb = journal->j_superblock; 316 317 err = do_one_pass(journal, &info, PASS_SCAN); 318 319 if (err) { 320 printk(KERN_ERR "JBD: error %d scanning journal\n", err); 321 ++journal->j_transaction_sequence; 322 } else { 323#ifdef CONFIG_JBD_DEBUG 324 int dropped = info.end_transaction - be32_to_cpu(sb->s_sequence); 325#endif 326 jbd_debug(1, 327 "JBD: ignoring %d transaction%s from the journal.\n", 328 dropped, (dropped == 1) ? "" : "s"); 329 journal->j_transaction_sequence = ++info.end_transaction; 330 } 331 332 journal->j_tail = 0; 333 return err; 334} 335 336static inline unsigned long long read_tag_block(int tag_bytes, journal_block_tag_t *tag) 337{ 338 unsigned long long block = be32_to_cpu(tag->t_blocknr); 339 if (tag_bytes > JBD_TAG_SIZE32) 340 block |= (__u64)be32_to_cpu(tag->t_blocknr_high) << 32; 341 return block; 342} 343 344/* 345 * calc_chksums calculates the checksums for the blocks described in the 346 * descriptor block. 347 */ 348static int calc_chksums(journal_t *journal, struct buffer_head *bh, 349 unsigned long long *next_log_block, __u32 *crc32_sum) 350{ 351 int i, num_blks, err; 352 unsigned long long io_block; 353 struct buffer_head *obh; 354 355 num_blks = count_tags(journal, bh); 356 /* Calculate checksum of the descriptor block. */ 357 *crc32_sum = crc32_be(*crc32_sum, (void *)bh->b_data, bh->b_size); 358 359 for (i = 0; i < num_blks; i++) { 360 io_block = (*next_log_block)++; 361 wrap(journal, *next_log_block); 362 err = jread(&obh, journal, io_block); 363 if (err) { 364 printk(KERN_ERR "JBD: IO error %d recovering block " 365 "%llu in log\n", err, io_block); 366 return 1; 367 } else { 368 *crc32_sum = crc32_be(*crc32_sum, (void *)obh->b_data, 369 obh->b_size); 370 } 371 brelse(obh); 372 } 373 return 0; 374} 375 376static int jbd2_commit_block_csum_verify(journal_t *j, void *buf) 377{ 378 struct commit_header *h; 379 __u32 provided, calculated; 380 381 if (!JFS_HAS_INCOMPAT_FEATURE(j, JFS_FEATURE_INCOMPAT_CSUM_V2)) 382 return 1; 383 384 h = buf; 385 provided = h->h_chksum[0]; 386 h->h_chksum[0] = 0; 387 calculated = ext2fs_crc32c_le(~0, j->j_superblock->s_uuid, 388 sizeof(j->j_superblock->s_uuid)); 389 calculated = ext2fs_crc32c_le(calculated, buf, j->j_blocksize); 390 h->h_chksum[0] = provided; 391 392 provided = ext2fs_be32_to_cpu(provided); 393 return provided == calculated; 394} 395 396static int jbd2_block_tag_csum_verify(journal_t *j, journal_block_tag_t *tag, 397 void *buf, __u32 sequence) 398{ 399 __u32 provided, calculated; 400 401 if (!JFS_HAS_INCOMPAT_FEATURE(j, JFS_FEATURE_INCOMPAT_CSUM_V2)) 402 return 1; 403 404 sequence = ext2fs_cpu_to_be32(sequence); 405 calculated = ext2fs_crc32c_le(~0, j->j_superblock->s_uuid, 406 sizeof(j->j_superblock->s_uuid)); 407 calculated = ext2fs_crc32c_le(calculated, (__u8 *)&sequence, 408 sizeof(sequence)); 409 calculated = ext2fs_crc32c_le(calculated, buf, j->j_blocksize) & 0xffff; 410 provided = ext2fs_be16_to_cpu(tag->t_checksum); 411 412 return provided == ext2fs_cpu_to_be32(calculated); 413} 414 415static int do_one_pass(journal_t *journal, 416 struct recovery_info *info, enum passtype pass) 417{ 418 unsigned int first_commit_ID, next_commit_ID; 419 unsigned long long next_log_block; 420 int err, success = 0; 421 journal_superblock_t * sb; 422 journal_header_t * tmp; 423 struct buffer_head * bh; 424 unsigned int sequence; 425 int blocktype; 426 int tag_bytes = journal_tag_bytes(journal); 427 __u32 crc32_sum = ~0; /* Transactional Checksums */ 428 int descr_csum_size = 0; 429 430 /* Precompute the maximum metadata descriptors in a descriptor block */ 431 int MAX_BLOCKS_PER_DESC; 432 MAX_BLOCKS_PER_DESC = ((journal->j_blocksize-sizeof(journal_header_t)) 433 / tag_bytes); 434 435 /* 436 * First thing is to establish what we expect to find in the log 437 * (in terms of transaction IDs), and where (in terms of log 438 * block offsets): query the superblock. 439 */ 440 441 sb = journal->j_superblock; 442 next_commit_ID = be32_to_cpu(sb->s_sequence); 443 next_log_block = be32_to_cpu(sb->s_start); 444 445 first_commit_ID = next_commit_ID; 446 if (pass == PASS_SCAN) 447 info->start_transaction = first_commit_ID; 448 449 jbd_debug(1, "Starting recovery pass %d\n", pass); 450 451 /* 452 * Now we walk through the log, transaction by transaction, 453 * making sure that each transaction has a commit block in the 454 * expected place. Each complete transaction gets replayed back 455 * into the main filesystem. 456 */ 457 458 while (1) { 459 int flags; 460 char * tagp; 461 journal_block_tag_t * tag; 462 struct buffer_head * obh; 463 struct buffer_head * nbh; 464 465 cond_resched(); 466 467 /* If we already know where to stop the log traversal, 468 * check right now that we haven't gone past the end of 469 * the log. */ 470 471 if (pass != PASS_SCAN) 472 if (tid_geq(next_commit_ID, info->end_transaction)) 473 break; 474 475 jbd_debug(2, "Scanning for sequence ID %u at %lu/%lu\n", 476 next_commit_ID, next_log_block, journal->j_last); 477 478 /* Skip over each chunk of the transaction looking 479 * either the next descriptor block or the final commit 480 * record. */ 481 482 jbd_debug(3, "JBD: checking block %ld\n", next_log_block); 483 err = jread(&bh, journal, next_log_block); 484 if (err) 485 goto failed; 486 487 next_log_block++; 488 wrap(journal, next_log_block); 489 490 /* What kind of buffer is it? 491 * 492 * If it is a descriptor block, check that it has the 493 * expected sequence number. Otherwise, we're all done 494 * here. */ 495 496 tmp = (journal_header_t *)bh->b_data; 497 498 if (tmp->h_magic != cpu_to_be32(JFS_MAGIC_NUMBER)) { 499 brelse(bh); 500 break; 501 } 502 503 blocktype = be32_to_cpu(tmp->h_blocktype); 504 sequence = be32_to_cpu(tmp->h_sequence); 505 jbd_debug(3, "Found magic %d, sequence %d\n", 506 blocktype, sequence); 507 508 if (sequence != next_commit_ID) { 509 brelse(bh); 510 break; 511 } 512 513 /* OK, we have a valid descriptor block which matches 514 * all of the sequence number checks. What are we going 515 * to do with it? That depends on the pass... */ 516 517 switch(blocktype) { 518 case JFS_DESCRIPTOR_BLOCK: 519 /* Verify checksum first */ 520 if (JFS_HAS_INCOMPAT_FEATURE(journal, 521 JFS_FEATURE_INCOMPAT_CSUM_V2)) 522 descr_csum_size = 523 sizeof(struct journal_block_tail); 524 if (descr_csum_size > 0 && 525 !jbd2_descr_block_csum_verify(journal, 526 bh->b_data)) { 527 err = -EIO; 528 goto failed; 529 } 530 531 /* If it is a valid descriptor block, replay it 532 * in pass REPLAY; if journal_checksums enabled, then 533 * calculate checksums in PASS_SCAN, otherwise, 534 * just skip over the blocks it describes. */ 535 if (pass != PASS_REPLAY) { 536 if (pass == PASS_SCAN && 537 JFS_HAS_COMPAT_FEATURE(journal, 538 JFS_FEATURE_COMPAT_CHECKSUM) && 539 !info->end_transaction) { 540 if (calc_chksums(journal, bh, 541 &next_log_block, 542 &crc32_sum)) { 543 brelse(bh); 544 break; 545 } 546 brelse(bh); 547 continue; 548 } 549 next_log_block += count_tags(journal, bh); 550 wrap(journal, next_log_block); 551 brelse(bh); 552 continue; 553 } 554 555 /* A descriptor block: we can now write all of 556 * the data blocks. Yay, useful work is finally 557 * getting done here! */ 558 559 tagp = &bh->b_data[sizeof(journal_header_t)]; 560 while ((tagp - bh->b_data + tag_bytes) 561 <= journal->j_blocksize - descr_csum_size) { 562 unsigned long long io_block; 563 564 tag = (journal_block_tag_t *) tagp; 565 flags = be16_to_cpu(tag->t_flags); 566 567 io_block = next_log_block++; 568 wrap(journal, next_log_block); 569 err = jread(&obh, journal, io_block); 570 if (err) { 571 /* Recover what we can, but 572 * report failure at the end. */ 573 success = err; 574 printk (KERN_ERR 575 "JBD: IO error %d recovering " 576 "block %llu in log\n", 577 err, io_block); 578 } else { 579 unsigned long long blocknr; 580 581 J_ASSERT(obh != NULL); 582 blocknr = read_tag_block(tag_bytes, 583 tag); 584 585 /* If the block has been 586 * revoked, then we're all done 587 * here. */ 588 if (journal_test_revoke 589 (journal, blocknr, 590 next_commit_ID)) { 591 brelse(obh); 592 ++info->nr_revoke_hits; 593 goto skip_write; 594 } 595 596 /* Look for block corruption */ 597 if (!jbd2_block_tag_csum_verify( 598 journal, tag, obh->b_data, 599 be32_to_cpu(tmp->h_sequence))) { 600 brelse(obh); 601 success = -EIO; 602 printk(KERN_ERR "JBD: Invalid " 603 "checksum recovering " 604 "block %ld in log\n", 605 blocknr); 606 continue; 607 } 608 609 /* Find a buffer for the new 610 * data being restored */ 611 nbh = __getblk(journal->j_fs_dev, 612 blocknr, 613 journal->j_blocksize); 614 if (nbh == NULL) { 615 printk(KERN_ERR 616 "JBD: Out of memory " 617 "during recovery.\n"); 618 err = -ENOMEM; 619 brelse(bh); 620 brelse(obh); 621 goto failed; 622 } 623 624 lock_buffer(nbh); 625 memcpy(nbh->b_data, obh->b_data, 626 journal->j_blocksize); 627 if (flags & JFS_FLAG_ESCAPE) { 628 journal_header_t *header; 629 630 header = (journal_header_t *) &nbh->b_data[0]; 631 header->h_magic = cpu_to_be32(JFS_MAGIC_NUMBER); 632 } 633 634 BUFFER_TRACE(nbh, "marking dirty"); 635 set_buffer_uptodate(nbh); 636 mark_buffer_dirty(nbh); 637 BUFFER_TRACE(nbh, "marking uptodate"); 638 ++info->nr_replays; 639 /* ll_rw_block(WRITE, 1, &nbh); */ 640 unlock_buffer(nbh); 641 brelse(obh); 642 brelse(nbh); 643 } 644 645 skip_write: 646 tagp += tag_bytes; 647 if (!(flags & JFS_FLAG_SAME_UUID)) 648 tagp += 16; 649 650 if (flags & JFS_FLAG_LAST_TAG) 651 break; 652 } 653 654 brelse(bh); 655 continue; 656 657 case JFS_COMMIT_BLOCK: 658 jbd_debug(3, "Commit block for #%u found\n", 659 next_commit_ID); 660 /* How to differentiate between interrupted commit 661 * and journal corruption ? 662 * 663 * {nth transaction} 664 * Checksum Verification Failed 665 * | 666 * ____________________ 667 * | | 668 * async_commit sync_commit 669 * | | 670 * | GO TO NEXT "Journal Corruption" 671 * | TRANSACTION 672 * | 673 * {(n+1)th transanction} 674 * | 675 * _______|______________ 676 * | | 677 * Commit block found Commit block not found 678 * | | 679 * "Journal Corruption" | 680 * _____________|_________ 681 * | | 682 * nth trans corrupt OR nth trans 683 * and (n+1)th interrupted interrupted 684 * before commit block 685 * could reach the disk. 686 * (Cannot find the difference in above 687 * mentioned conditions. Hence assume 688 * "Interrupted Commit".) 689 */ 690 691 /* Found an expected commit block: if checksums 692 * are present verify them in PASS_SCAN; else not 693 * much to do other than move on to the next sequence 694 * number. */ 695 if (pass == PASS_SCAN && 696 JFS_HAS_COMPAT_FEATURE(journal, 697 JFS_FEATURE_COMPAT_CHECKSUM)) { 698 int chksum_err, chksum_seen; 699 struct commit_header *cbh = 700 (struct commit_header *)bh->b_data; 701 unsigned found_chksum = 702 be32_to_cpu(cbh->h_chksum[0]); 703 704 chksum_err = chksum_seen = 0; 705 706 jbd_debug(3, "Checksums %x %x\n", 707 crc32_sum, found_chksum); 708 if (info->end_transaction) { 709 journal->j_failed_commit = 710 info->end_transaction; 711 brelse(bh); 712 break; 713 } 714 715 if (crc32_sum == found_chksum && 716 cbh->h_chksum_type == JBD2_CRC32_CHKSUM && 717 cbh->h_chksum_size == 718 JBD2_CRC32_CHKSUM_SIZE) 719 chksum_seen = 1; 720 else if (!(cbh->h_chksum_type == 0 && 721 cbh->h_chksum_size == 0 && 722 found_chksum == 0 && 723 !chksum_seen)) 724 /* 725 * If fs is mounted using an old kernel and then 726 * kernel with journal_chksum is used then we 727 * get a situation where the journal flag has 728 * checksum flag set but checksums are not 729 * present i.e chksum = 0, in the individual 730 * commit blocks. 731 * Hence to avoid checksum failures, in this 732 * situation, this extra check is added. 733 */ 734 chksum_err = 1; 735 736 if (chksum_err) { 737 info->end_transaction = next_commit_ID; 738 jbd_debug(1, "Checksum_err %x %x\n", 739 crc32_sum, found_chksum); 740 if (!JFS_HAS_INCOMPAT_FEATURE(journal, 741 JFS_FEATURE_INCOMPAT_ASYNC_COMMIT)){ 742 journal->j_failed_commit = 743 next_commit_ID; 744 brelse(bh); 745 break; 746 } 747 } 748 crc32_sum = ~0; 749 } 750 if (pass == PASS_SCAN && 751 !jbd2_commit_block_csum_verify(journal, 752 bh->b_data)) { 753 info->end_transaction = next_commit_ID; 754 755 if (!JFS_HAS_INCOMPAT_FEATURE(journal, 756 JFS_FEATURE_INCOMPAT_ASYNC_COMMIT)) { 757 journal->j_failed_commit = 758 next_commit_ID; 759 brelse(bh); 760 break; 761 } 762 } 763 brelse(bh); 764 next_commit_ID++; 765 continue; 766 767 case JFS_REVOKE_BLOCK: 768 /* If we aren't in the REVOKE pass, then we can 769 * just skip over this block. */ 770 if (pass != PASS_REVOKE) { 771 brelse(bh); 772 continue; 773 } 774 775 err = scan_revoke_records(journal, bh, 776 next_commit_ID, info); 777 brelse(bh); 778 if (err) 779 goto failed; 780 continue; 781 782 default: 783 jbd_debug(3, "Unrecognised magic %d, end of scan.\n", 784 blocktype); 785 brelse(bh); 786 goto done; 787 } 788 } 789 790 done: 791 /* 792 * We broke out of the log scan loop: either we came to the 793 * known end of the log or we found an unexpected block in the 794 * log. If the latter happened, then we know that the "current" 795 * transaction marks the end of the valid log. 796 */ 797 798 if (pass == PASS_SCAN) { 799 if (!info->end_transaction) 800 info->end_transaction = next_commit_ID; 801 } else { 802 /* It's really bad news if different passes end up at 803 * different places (but possible due to IO errors). */ 804 if (info->end_transaction != next_commit_ID) { 805 printk (KERN_ERR "JBD: recovery pass %d ended at " 806 "transaction %u, expected %u\n", 807 pass, next_commit_ID, info->end_transaction); 808 if (!success) 809 success = -EIO; 810 } 811 } 812 813 return success; 814 815 failed: 816 return err; 817} 818 819static int jbd2_revoke_block_csum_verify(journal_t *j, 820 void *buf) 821{ 822 struct journal_revoke_tail *tail; 823 __u32 provided, calculated; 824 825 if (!JFS_HAS_INCOMPAT_FEATURE(j, JFS_FEATURE_INCOMPAT_CSUM_V2)) 826 return 1; 827 828 tail = (struct journal_revoke_tail *)(buf + j->j_blocksize - 829 sizeof(struct journal_revoke_tail)); 830 provided = tail->r_checksum; 831 tail->r_checksum = 0; 832 calculated = ext2fs_crc32c_le(~0, j->j_superblock->s_uuid, 833 sizeof(j->j_superblock->s_uuid)); 834 calculated = ext2fs_crc32c_le(calculated, buf, j->j_blocksize); 835 tail->r_checksum = provided; 836 837 provided = ext2fs_be32_to_cpu(provided); 838 return provided == calculated; 839} 840 841/* Scan a revoke record, marking all blocks mentioned as revoked. */ 842 843static int scan_revoke_records(journal_t *journal, struct buffer_head *bh, 844 tid_t sequence, struct recovery_info *info) 845{ 846 journal_revoke_header_t *header; 847 int offset, max; 848 int record_len = 4; 849 850 header = (journal_revoke_header_t *) bh->b_data; 851 offset = sizeof(journal_revoke_header_t); 852 max = be32_to_cpu(header->r_count); 853 854 if (!jbd2_revoke_block_csum_verify(journal, header)) 855 return -EINVAL; 856 857 if (JFS_HAS_INCOMPAT_FEATURE(journal, JFS_FEATURE_INCOMPAT_64BIT)) 858 record_len = 8; 859 860 while (offset < max) { 861 unsigned long long blocknr; 862 int err; 863 864 if (record_len == 4) 865 blocknr = ext2fs_be32_to_cpu(*((__be32 *)(bh->b_data + 866 offset))); 867 else 868 blocknr = ext2fs_be64_to_cpu(*((__be64 *)(bh->b_data + 869 offset))); 870 offset += record_len; 871 err = journal_set_revoke(journal, blocknr, sequence); 872 if (err) 873 return err; 874 ++info->nr_revokes; 875 } 876 return 0; 877} 878