e2fsck/recovery.c

   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  */
  31 struct 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
  41 enum passtype {PASS_SCAN, PASS_REVOKE, PASS_REPLAY};
  42 static int do_one_pass(journal_t *journal,
  43                                 struct recovery_info *info, enum passtype pass);
  44 static 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 */
  50 static 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
  70 static 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
 119 failed:
 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
 132 static 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
 177 static 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
 203 static 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)                                              \
 233 do {                                                                    \
 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  */
 250 int 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  */
 307 int 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
 336 static 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  */
 348 static 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
 376 static int do_one_pass(journal_t *journal,
 377                         struct recovery_info *info, enum passtype pass)
 378 {
 379         unsigned int            first_commit_ID, next_commit_ID;
 380         unsigned long long      next_log_block;
 381         int                     err, success = 0;
 382         journal_superblock_t *  sb;
 383         journal_header_t *      tmp;
 384         struct buffer_head *    bh;
 385         unsigned int            sequence;
 386         int                     blocktype;
 387         int                     tag_bytes = journal_tag_bytes(journal);
 388         __u32                   crc32_sum = ~0; /* Transactional Checksums */
 389         int                     descr_csum_size = 0;
 390
 391         /* Precompute the maximum metadata descriptors in a descriptor block */
 392         int                     MAX_BLOCKS_PER_DESC;
 393         MAX_BLOCKS_PER_DESC = ((journal->j_blocksize-sizeof(journal_header_t))
 394                                / tag_bytes);
 395
 396         /*
 397          * First thing is to establish what we expect to find in the log
 398          * (in terms of transaction IDs), and where (in terms of log
 399          * block offsets): query the superblock.
 400          */
 401
 402         sb = journal->j_superblock;
 403         next_commit_ID = be32_to_cpu(sb->s_sequence);
 404         next_log_block = be32_to_cpu(sb->s_start);
 405
 406         first_commit_ID = next_commit_ID;
 407         if (pass == PASS_SCAN)
 408                 info->start_transaction = first_commit_ID;
 409
 410         jbd_debug(1, "Starting recovery pass %d\n", pass);
 411
 412         /*
 413          * Now we walk through the log, transaction by transaction,
 414          * making sure that each transaction has a commit block in the
 415          * expected place.  Each complete transaction gets replayed back
 416          * into the main filesystem.
 417          */
 418
 419         while (1) {
 420                 int                     flags;
 421                 char *                  tagp;
 422                 journal_block_tag_t *   tag;
 423                 struct buffer_head *    obh;
 424                 struct buffer_head *    nbh;
 425
 426                 cond_resched();
 427
 428                 /* If we already know where to stop the log traversal,
 429                  * check right now that we haven't gone past the end of
 430                  * the log. */
 431
 432                 if (pass != PASS_SCAN)
 433                         if (tid_geq(next_commit_ID, info->end_transaction))
 434                                 break;
 435
 436                 jbd_debug(2, "Scanning for sequence ID %u at %lu/%lu\n",
 437                           next_commit_ID, next_log_block, journal->j_last);
 438
 439                 /* Skip over each chunk of the transaction looking
 440                  * either the next descriptor block or the final commit
 441                  * record. */
 442
 443                 jbd_debug(3, "JBD: checking block %ld\n", next_log_block);
 444                 err = jread(&bh, journal, next_log_block);
 445                 if (err)
 446                         goto failed;
 447
 448                 next_log_block++;
 449                 wrap(journal, next_log_block);
 450
 451                 /* What kind of buffer is it?
 452                  *
 453                  * If it is a descriptor block, check that it has the
 454                  * expected sequence number.  Otherwise, we're all done
 455                  * here. */
 456
 457                 tmp = (journal_header_t *)bh->b_data;
 458
 459                 if (tmp->h_magic != cpu_to_be32(JFS_MAGIC_NUMBER)) {
 460                         brelse(bh);
 461                         break;
 462                 }
 463
 464                 blocktype = be32_to_cpu(tmp->h_blocktype);
 465                 sequence = be32_to_cpu(tmp->h_sequence);
 466                 jbd_debug(3, "Found magic %d, sequence %d\n",
 467                           blocktype, sequence);
 468
 469                 if (sequence != next_commit_ID) {
 470                         brelse(bh);
 471                         break;
 472                 }
 473
 474                 /* OK, we have a valid descriptor block which matches
 475                  * all of the sequence number checks.  What are we going
 476                  * to do with it?  That depends on the pass... */
 477
 478                 switch(blocktype) {
 479                 case JFS_DESCRIPTOR_BLOCK:
 480                         /* Verify checksum first */
 481                         if (JFS_HAS_INCOMPAT_FEATURE(journal,
 482                                         JFS_FEATURE_INCOMPAT_CSUM_V2))
 483                                 descr_csum_size =
 484                                         sizeof(struct journal_block_tail);
 485                         if (descr_csum_size > 0 &&
 486                             !jbd2_descr_block_csum_verify(journal,
 487                                                           bh->b_data)) {
 488                                 err = -EIO;
 489                                 goto failed;
 490                         }
 491
 492                         /* If it is a valid descriptor block, replay it
 493                          * in pass REPLAY; if journal_checksums enabled, then
 494                          * calculate checksums in PASS_SCAN, otherwise,
 495                          * just skip over the blocks it describes. */
 496                         if (pass != PASS_REPLAY) {
 497                                 if (pass == PASS_SCAN &&
 498                                     JFS_HAS_COMPAT_FEATURE(journal,
 499                                             JFS_FEATURE_COMPAT_CHECKSUM) &&
 500                                     !info->end_transaction) {
 501                                         if (calc_chksums(journal, bh,
 502                                                         &next_log_block,
 503                                                         &crc32_sum)) {
 504                                                 brelse(bh);
 505                                                 break;
 506                                         }
 507                                         brelse(bh);
 508                                         continue;
 509                                 }
 510                                 next_log_block += count_tags(journal, bh);
 511                                 wrap(journal, next_log_block);
 512                                 brelse(bh);
 513                                 continue;
 514                         }
 515
 516                         /* A descriptor block: we can now write all of
 517                          * the data blocks.  Yay, useful work is finally
 518                          * getting done here! */
 519
 520                         tagp = &bh->b_data[sizeof(journal_header_t)];
 521                         while ((tagp - bh->b_data + tag_bytes)
 522                                <= journal->j_blocksize - descr_csum_size) {
 523                                 unsigned long long io_block;
 524
 525                                 tag = (journal_block_tag_t *) tagp;
 526                                 flags = be16_to_cpu(tag->t_flags);
 527
 528                                 io_block = next_log_block++;
 529                                 wrap(journal, next_log_block);
 530                                 err = jread(&obh, journal, io_block);
 531                                 if (err) {
 532                                         /* Recover what we can, but
 533                                          * report failure at the end. */
 534                                         success = err;
 535                                         printk (KERN_ERR
 536                                                 "JBD: IO error %d recovering "
 537                                                 "block %llu in log\n",
 538                                                 err, io_block);
 539                                 } else {
 540                                         unsigned long long blocknr;
 541
 542                                         J_ASSERT(obh != NULL);
 543                                         blocknr = read_tag_block(tag_bytes,
 544                                                                  tag);
 545
 546                                         /* If the block has been
 547                                          * revoked, then we're all done
 548                                          * here. */
 549                                         if (journal_test_revoke
 550                                             (journal, blocknr,
 551                                              next_commit_ID)) {
 552                                                 brelse(obh);
 553                                                 ++info->nr_revoke_hits;
 554                                                 goto skip_write;
 555                                         }
 556
 557                                         /* Find a buffer for the new
 558                                          * data being restored */
 559                                         nbh = __getblk(journal->j_fs_dev,
 560                                                         blocknr,
 561                                                         journal->j_blocksize);
 562                                         if (nbh == NULL) {
 563                                                 printk(KERN_ERR
 564                                                        "JBD: Out of memory "
 565                                                        "during recovery.\n");
 566                                                 err = -ENOMEM;
 567                                                 brelse(bh);
 568                                                 brelse(obh);
 569                                                 goto failed;
 570                                         }
 571
 572                                         lock_buffer(nbh);
 573                                         memcpy(nbh->b_data, obh->b_data,
 574                                                         journal->j_blocksize);
 575                                         if (flags & JFS_FLAG_ESCAPE) {
 576                                                 journal_header_t *header;
 577
 578                                                 header = (journal_header_t *) &nbh->b_data[0];
 579                                                 header->h_magic = cpu_to_be32(JFS_MAGIC_NUMBER);
 580                                         }
 581
 582                                         BUFFER_TRACE(nbh, "marking dirty");
 583                                         set_buffer_uptodate(nbh);
 584                                         mark_buffer_dirty(nbh);
 585                                         BUFFER_TRACE(nbh, "marking uptodate");
 586                                         ++info->nr_replays;
 587                                         /* ll_rw_block(WRITE, 1, &nbh); */
 588                                         unlock_buffer(nbh);
 589                                         brelse(obh);
 590                                         brelse(nbh);
 591                                 }
 592
 593                         skip_write:
 594                                 tagp += tag_bytes;
 595                                 if (!(flags & JFS_FLAG_SAME_UUID))
 596                                         tagp += 16;
 597
 598                                 if (flags & JFS_FLAG_LAST_TAG)
 599                                         break;
 600                         }
 601
 602                         brelse(bh);
 603                         continue;
 604
 605                 case JFS_COMMIT_BLOCK:
 606                         jbd_debug(3, "Commit block for #%u found\n",
 607                                   next_commit_ID);
 608                         /*     How to differentiate between interrupted commit
 609                          *               and journal corruption ?
 610                          *
 611                          * {nth transaction}
 612                          *        Checksum Verification Failed
 613                          *                       |
 614                          *               ____________________
 615                          *              |                    |
 616                          *      async_commit             sync_commit
 617                          *              |                    |
 618                          *              | GO TO NEXT    "Journal Corruption"
 619                          *              | TRANSACTION
 620                          *              |
 621                          * {(n+1)th transanction}
 622                          *              |
 623                          *       _______|______________
 624                          *      |                     |
 625                          * Commit block found   Commit block not found
 626                          *      |                     |
 627                          * "Journal Corruption"       |
 628                          *               _____________|_________
 629                          *              |                       |
 630                          *      nth trans corrupt       OR   nth trans
 631                          *      and (n+1)th interrupted     interrupted
 632                          *      before commit block
 633                          *      could reach the disk.
 634                          *      (Cannot find the difference in above
 635                          *       mentioned conditions. Hence assume
 636                          *       "Interrupted Commit".)
 637                          */
 638
 639                         /* Found an expected commit block: if checksums
 640                          * are present verify them in PASS_SCAN; else not
 641                          * much to do other than move on to the next sequence
 642                          * number. */
 643                         if (pass == PASS_SCAN &&
 644                             JFS_HAS_COMPAT_FEATURE(journal,
 645                                     JFS_FEATURE_COMPAT_CHECKSUM)) {
 646                                 int chksum_err, chksum_seen;
 647                                 struct commit_header *cbh =
 648                                         (struct commit_header *)bh->b_data;
 649                                 unsigned found_chksum =
 650                                         be32_to_cpu(cbh->h_chksum[0]);
 651
 652                                 chksum_err = chksum_seen = 0;
 653
 654                                 jbd_debug(3, "Checksums %x %x\n",
 655                                           crc32_sum, found_chksum);
 656                                 if (info->end_transaction) {
 657                                         journal->j_failed_commit =
 658                                                 info->end_transaction;
 659                                         brelse(bh);
 660                                         break;
 661                                 }
 662
 663                                 if (crc32_sum == found_chksum &&
 664                                     cbh->h_chksum_type == JBD2_CRC32_CHKSUM &&
 665                                     cbh->h_chksum_size ==
 666                                                 JBD2_CRC32_CHKSUM_SIZE)
 667                                        chksum_seen = 1;
 668                                 else if (!(cbh->h_chksum_type == 0 &&
 669                                              cbh->h_chksum_size == 0 &&
 670                                              found_chksum == 0 &&
 671                                              !chksum_seen))
 672                                 /*
 673                                  * If fs is mounted using an old kernel and then
 674                                  * kernel with journal_chksum is used then we
 675                                  * get a situation where the journal flag has
 676                                  * checksum flag set but checksums are not
 677                                  * present i.e chksum = 0, in the individual
 678                                  * commit blocks.
 679                                  * Hence to avoid checksum failures, in this
 680                                  * situation, this extra check is added.
 681                                  */
 682                                                 chksum_err = 1;
 683
 684                                 if (chksum_err) {
 685                                         info->end_transaction = next_commit_ID;
 686                                         jbd_debug(1, "Checksum_err %x %x\n",
 687                                                   crc32_sum, found_chksum);
 688                                         if (!JFS_HAS_INCOMPAT_FEATURE(journal,
 689                                            JFS_FEATURE_INCOMPAT_ASYNC_COMMIT)){
 690                                                 journal->j_failed_commit =
 691                                                         next_commit_ID;
 692                                                 brelse(bh);
 693                                                 break;
 694                                         }
 695                                 }
 696                                 crc32_sum = ~0;
 697                         }
 698                         brelse(bh);
 699                         next_commit_ID++;
 700                         continue;
 701
 702                 case JFS_REVOKE_BLOCK:
 703                         /* If we aren't in the REVOKE pass, then we can
 704                          * just skip over this block. */
 705                         if (pass != PASS_REVOKE) {
 706                                 brelse(bh);
 707                                 continue;
 708                         }
 709
 710                         err = scan_revoke_records(journal, bh,
 711                                                   next_commit_ID, info);
 712                         brelse(bh);
 713                         if (err)
 714                                 goto failed;
 715                         continue;
 716
 717                 default:
 718                         jbd_debug(3, "Unrecognised magic %d, end of scan.\n",
 719                                   blocktype);
 720                         brelse(bh);
 721                         goto done;
 722                 }
 723         }
 724
 725  done:
 726         /*
 727          * We broke out of the log scan loop: either we came to the
 728          * known end of the log or we found an unexpected block in the
 729          * log.  If the latter happened, then we know that the "current"
 730          * transaction marks the end of the valid log.
 731          */
 732
 733         if (pass == PASS_SCAN) {
 734                 if (!info->end_transaction)
 735                         info->end_transaction = next_commit_ID;
 736         } else {
 737                 /* It's really bad news if different passes end up at
 738                  * different places (but possible due to IO errors). */
 739                 if (info->end_transaction != next_commit_ID) {
 740                         printk (KERN_ERR "JBD: recovery pass %d ended at "
 741                                 "transaction %u, expected %u\n",
 742                                 pass, next_commit_ID, info->end_transaction);
 743                         if (!success)
 744                                 success = -EIO;
 745                 }
 746         }
 747
 748         return success;
 749
 750  failed:
 751         return err;
 752 }
 753
 754 static int jbd2_revoke_block_csum_verify(journal_t *j,
 755                                          void *buf)
 756 {
 757         struct journal_revoke_tail *tail;
 758         __u32 provided, calculated;
 759
 760         if (!JFS_HAS_INCOMPAT_FEATURE(j, JFS_FEATURE_INCOMPAT_CSUM_V2))
 761                 return 1;
 762
 763         tail = (struct journal_revoke_tail *)(buf + j->j_blocksize -
 764                         sizeof(struct journal_revoke_tail));
 765         provided = tail->r_checksum;
 766         tail->r_checksum = 0;
 767         calculated = ext2fs_crc32c_le(~0, j->j_superblock->s_uuid,
 768                                       sizeof(j->j_superblock->s_uuid));
 769         calculated = ext2fs_crc32c_le(calculated, buf, j->j_blocksize);
 770         tail->r_checksum = provided;
 771
 772         provided = ext2fs_be32_to_cpu(provided);
 773         return provided == calculated;
 774 }
 775
 776 /* Scan a revoke record, marking all blocks mentioned as revoked. */
 777
 778 static int scan_revoke_records(journal_t *journal, struct buffer_head *bh,
 779                                tid_t sequence, struct recovery_info *info)
 780 {
 781         journal_revoke_header_t *header;
 782         int offset, max;
 783         int record_len = 4;
 784
 785         header = (journal_revoke_header_t *) bh->b_data;
 786         offset = sizeof(journal_revoke_header_t);
 787         max = be32_to_cpu(header->r_count);
 788
 789         if (!jbd2_revoke_block_csum_verify(journal, header))
 790                 return -EINVAL;
 791
 792         if (JFS_HAS_INCOMPAT_FEATURE(journal, JFS_FEATURE_INCOMPAT_64BIT))
 793                 record_len = 8;
 794
 795         while (offset < max) {
 796                 unsigned long long blocknr;
 797                 int err;
 798
 799                 if (record_len == 4)
 800                         blocknr = ext2fs_be32_to_cpu(*((__be32 *)(bh->b_data +
 801                                                                   offset)));
 802                 else
 803                         blocknr = ext2fs_be64_to_cpu(*((__be64 *)(bh->b_data +
 804                                                                   offset)));
 805                 offset += record_len;
 806                 err = journal_set_revoke(journal, blocknr, sequence);
 807                 if (err)
 808                         return err;
 809                 ++info->nr_revokes;
 810         }
 811         return 0;
 812 }