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         struct recovery_info    info;
 311
 312         memset (&info, 0, sizeof(info));
 313
 314         err = do_one_pass(journal, &info, PASS_SCAN);
 315
 316         if (err) {
 317                 printk(KERN_ERR "JBD: error %d scanning journal\n", err);
 318                 ++journal->j_transaction_sequence;
 319         } else {
 320 #ifdef CONFIG_JBD_DEBUG
 321                 journal_superblock_t *sb = journal->j_superblock;
 322
 323                 int dropped = info.end_transaction - be32_to_cpu(sb->s_sequence);
 324 #endif
 325                 jbd_debug(1,
 326                           "JBD: ignoring %d transaction%s from the journal.\n",
 327                           dropped, (dropped == 1) ? "" : "s");
 328                 journal->j_transaction_sequence = ++info.end_transaction;
 329         }
 330
 331         journal->j_tail = 0;
 332         return err;
 333 }
 334
 335 static inline unsigned long long read_tag_block(int tag_bytes, journal_block_tag_t *tag)
 336 {
 337         unsigned long long block = be32_to_cpu(tag->t_blocknr);
 338         if (tag_bytes > JBD_TAG_SIZE32)
 339                 block |= (__u64)be32_to_cpu(tag->t_blocknr_high) << 32;
 340         return block;
 341 }
 342
 343 /*
 344  * calc_chksums calculates the checksums for the blocks described in the
 345  * descriptor block.
 346  */
 347 static int calc_chksums(journal_t *journal, struct buffer_head *bh,
 348                         unsigned long long *next_log_block, __u32 *crc32_sum)
 349 {
 350         int i, num_blks, err;
 351         unsigned long long io_block;
 352         struct buffer_head *obh;
 353
 354         num_blks = count_tags(journal, bh);
 355         /* Calculate checksum of the descriptor block. */
 356         *crc32_sum = ext2fs_crc32_be(*crc32_sum, (void *)bh->b_data,
 357                                      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 = ext2fs_crc32_be(*crc32_sum,
 369                                                      (void *)obh->b_data,
 370                                                      obh->b_size);
 371                 }
 372                 brelse(obh);
 373         }
 374         return 0;
 375 }
 376
 377 static int jbd2_commit_block_csum_verify(journal_t *j, void *buf)
 378 {
 379         struct commit_header *h;
 380         __u32 provided, calculated;
 381
 382         if (!JFS_HAS_INCOMPAT_FEATURE(j, JFS_FEATURE_INCOMPAT_CSUM_V2))
 383                 return 1;
 384
 385         h = buf;
 386         provided = h->h_chksum[0];
 387         h->h_chksum[0] = 0;
 388         calculated = ext2fs_crc32c_le(~0, j->j_superblock->s_uuid,
 389                                       sizeof(j->j_superblock->s_uuid));
 390         calculated = ext2fs_crc32c_le(calculated, buf, j->j_blocksize);
 391         h->h_chksum[0] = provided;
 392
 393         provided = ext2fs_be32_to_cpu(provided);
 394         return provided == calculated;
 395 }
 396
 397 static int jbd2_block_tag_csum_verify(journal_t *j, journal_block_tag_t *tag,
 398                                       void *buf, __u32 sequence)
 399 {
 400         __u32 provided, calculated;
 401
 402         if (!JFS_HAS_INCOMPAT_FEATURE(j, JFS_FEATURE_INCOMPAT_CSUM_V2))
 403                 return 1;
 404
 405         sequence = ext2fs_cpu_to_be32(sequence);
 406         calculated = ext2fs_crc32c_le(~0, j->j_superblock->s_uuid,
 407                                       sizeof(j->j_superblock->s_uuid));
 408         calculated = ext2fs_crc32c_le(calculated, (__u8 *)&sequence,
 409                                       sizeof(sequence));
 410         calculated = ext2fs_crc32c_le(calculated, buf, j->j_blocksize) & 0xffff;
 411         provided = ext2fs_be16_to_cpu(tag->t_checksum);
 412
 413         return provided == ext2fs_cpu_to_be32(calculated);
 414 }
 415
 416 static int do_one_pass(journal_t *journal,
 417                         struct recovery_info *info, enum passtype pass)
 418 {
 419         unsigned int            first_commit_ID, next_commit_ID;
 420         unsigned long long      next_log_block;
 421         int                     err, success = 0;
 422         journal_superblock_t *  sb;
 423         journal_header_t *      tmp;
 424         struct buffer_head *    bh;
 425         unsigned int            sequence;
 426         int                     blocktype;
 427         int                     tag_bytes = journal_tag_bytes(journal);
 428         __u32                   crc32_sum = ~0; /* Transactional Checksums */
 429         int                     descr_csum_size = 0;
 430
 431         /*
 432          * First thing is to establish what we expect to find in the log
 433          * (in terms of transaction IDs), and where (in terms of log
 434          * block offsets): query the superblock.
 435          */
 436
 437         sb = journal->j_superblock;
 438         next_commit_ID = be32_to_cpu(sb->s_sequence);
 439         next_log_block = be32_to_cpu(sb->s_start);
 440
 441         first_commit_ID = next_commit_ID;
 442         if (pass == PASS_SCAN)
 443                 info->start_transaction = first_commit_ID;
 444
 445         jbd_debug(1, "Starting recovery pass %d\n", pass);
 446
 447         /*
 448          * Now we walk through the log, transaction by transaction,
 449          * making sure that each transaction has a commit block in the
 450          * expected place.  Each complete transaction gets replayed back
 451          * into the main filesystem.
 452          */
 453
 454         while (1) {
 455                 int                     flags;
 456                 char *                  tagp;
 457                 journal_block_tag_t *   tag;
 458                 struct buffer_head *    obh;
 459                 struct buffer_head *    nbh;
 460
 461                 cond_resched();
 462
 463                 /* If we already know where to stop the log traversal,
 464                  * check right now that we haven't gone past the end of
 465                  * the log. */
 466
 467                 if (pass != PASS_SCAN)
 468                         if (tid_geq(next_commit_ID, info->end_transaction))
 469                                 break;
 470
 471                 jbd_debug(2, "Scanning for sequence ID %u at %llu/%lu\n",
 472                           next_commit_ID, next_log_block, journal->j_last);
 473
 474                 /* Skip over each chunk of the transaction looking
 475                  * either the next descriptor block or the final commit
 476                  * record. */
 477
 478                 jbd_debug(3, "JBD: checking block %llu\n", next_log_block);
 479                 err = jread(&bh, journal, next_log_block);
 480                 if (err)
 481                         goto failed;
 482
 483                 next_log_block++;
 484                 wrap(journal, next_log_block);
 485
 486                 /* What kind of buffer is it?
 487                  *
 488                  * If it is a descriptor block, check that it has the
 489                  * expected sequence number.  Otherwise, we're all done
 490                  * here. */
 491
 492                 tmp = (journal_header_t *)bh->b_data;
 493
 494                 if (tmp->h_magic != cpu_to_be32(JFS_MAGIC_NUMBER)) {
 495                         brelse(bh);
 496                         break;
 497                 }
 498
 499                 blocktype = be32_to_cpu(tmp->h_blocktype);
 500                 sequence = be32_to_cpu(tmp->h_sequence);
 501                 jbd_debug(3, "Found magic %d, sequence %d\n",
 502                           blocktype, sequence);
 503
 504                 if (sequence != next_commit_ID) {
 505                         brelse(bh);
 506                         break;
 507                 }
 508
 509                 /* OK, we have a valid descriptor block which matches
 510                  * all of the sequence number checks.  What are we going
 511                  * to do with it?  That depends on the pass... */
 512
 513                 switch(blocktype) {
 514                 case JFS_DESCRIPTOR_BLOCK:
 515                         /* Verify checksum first */
 516                         if (JFS_HAS_INCOMPAT_FEATURE(journal,
 517                                         JFS_FEATURE_INCOMPAT_CSUM_V2))
 518                                 descr_csum_size =
 519                                         sizeof(struct journal_block_tail);
 520                         if (descr_csum_size > 0 &&
 521                             !jbd2_descr_block_csum_verify(journal,
 522                                                           bh->b_data)) {
 523                                 err = -EIO;
 524                                 goto failed;
 525                         }
 526
 527                         /* If it is a valid descriptor block, replay it
 528                          * in pass REPLAY; if journal_checksums enabled, then
 529                          * calculate checksums in PASS_SCAN, otherwise,
 530                          * just skip over the blocks it describes. */
 531                         if (pass != PASS_REPLAY) {
 532                                 if (pass == PASS_SCAN &&
 533                                     JFS_HAS_COMPAT_FEATURE(journal,
 534                                             JFS_FEATURE_COMPAT_CHECKSUM) &&
 535                                     !info->end_transaction) {
 536                                         if (calc_chksums(journal, bh,
 537                                                         &next_log_block,
 538                                                         &crc32_sum)) {
 539                                                 brelse(bh);
 540                                                 break;
 541                                         }
 542                                         brelse(bh);
 543                                         continue;
 544                                 }
 545                                 next_log_block += count_tags(journal, bh);
 546                                 wrap(journal, next_log_block);
 547                                 brelse(bh);
 548                                 continue;
 549                         }
 550
 551                         /* A descriptor block: we can now write all of
 552                          * the data blocks.  Yay, useful work is finally
 553                          * getting done here! */
 554
 555                         tagp = &bh->b_data[sizeof(journal_header_t)];
 556                         while ((tagp - bh->b_data + tag_bytes)
 557                                <= journal->j_blocksize - descr_csum_size) {
 558                                 unsigned long long io_block;
 559
 560                                 tag = (journal_block_tag_t *) tagp;
 561                                 flags = be16_to_cpu(tag->t_flags);
 562
 563                                 io_block = next_log_block++;
 564                                 wrap(journal, next_log_block);
 565                                 err = jread(&obh, journal, io_block);
 566                                 if (err) {
 567                                         /* Recover what we can, but
 568                                          * report failure at the end. */
 569                                         success = err;
 570                                         printk (KERN_ERR
 571                                                 "JBD: IO error %d recovering "
 572                                                 "block %llu in log\n",
 573                                                 err, io_block);
 574                                 } else {
 575                                         unsigned long long blocknr;
 576
 577                                         J_ASSERT(obh != NULL);
 578                                         blocknr = read_tag_block(tag_bytes,
 579                                                                  tag);
 580
 581                                         /* If the block has been
 582                                          * revoked, then we're all done
 583                                          * here. */
 584                                         if (journal_test_revoke
 585                                             (journal, blocknr,
 586                                              next_commit_ID)) {
 587                                                 brelse(obh);
 588                                                 ++info->nr_revoke_hits;
 589                                                 goto skip_write;
 590                                         }
 591
 592                                         /* Look for block corruption */
 593                                         if (!jbd2_block_tag_csum_verify(
 594                                                 journal, tag, obh->b_data,
 595                                                 be32_to_cpu(tmp->h_sequence))) {
 596                                                 brelse(obh);
 597                                                 success = -EIO;
 598                                                 printk(KERN_ERR "JBD: Invalid "
 599                                                        "checksum recovering "
 600                                                        "block %lld in log\n",
 601                                                        blocknr);
 602                                                 continue;
 603                                         }
 604
 605                                         /* Find a buffer for the new
 606                                          * data being restored */
 607                                         nbh = __getblk(journal->j_fs_dev,
 608                                                         blocknr,
 609                                                         journal->j_blocksize);
 610                                         if (nbh == NULL) {
 611                                                 printk(KERN_ERR
 612                                                        "JBD: Out of memory "
 613                                                        "during recovery.\n");
 614                                                 err = -ENOMEM;
 615                                                 brelse(bh);
 616                                                 brelse(obh);
 617                                                 goto failed;
 618                                         }
 619
 620                                         lock_buffer(nbh);
 621                                         memcpy(nbh->b_data, obh->b_data,
 622                                                         journal->j_blocksize);
 623                                         if (flags & JFS_FLAG_ESCAPE) {
 624                                                 journal_header_t *header;
 625
 626                                                 header = (journal_header_t *) &nbh->b_data[0];
 627                                                 header->h_magic = cpu_to_be32(JFS_MAGIC_NUMBER);
 628                                         }
 629
 630                                         BUFFER_TRACE(nbh, "marking dirty");
 631                                         set_buffer_uptodate(nbh);
 632                                         mark_buffer_dirty(nbh);
 633                                         BUFFER_TRACE(nbh, "marking uptodate");
 634                                         ++info->nr_replays;
 635                                         /* ll_rw_block(WRITE, 1, &nbh); */
 636                                         unlock_buffer(nbh);
 637                                         brelse(obh);
 638                                         brelse(nbh);
 639                                 }
 640
 641                         skip_write:
 642                                 tagp += tag_bytes;
 643                                 if (!(flags & JFS_FLAG_SAME_UUID))
 644                                         tagp += 16;
 645
 646                                 if (flags & JFS_FLAG_LAST_TAG)
 647                                         break;
 648                         }
 649
 650                         brelse(bh);
 651                         continue;
 652
 653                 case JFS_COMMIT_BLOCK:
 654                         jbd_debug(3, "Commit block for #%u found\n",
 655                                   next_commit_ID);
 656                         /*     How to differentiate between interrupted commit
 657                          *               and journal corruption ?
 658                          *
 659                          * {nth transaction}
 660                          *        Checksum Verification Failed
 661                          *                       |
 662                          *               ____________________
 663                          *              |                    |
 664                          *      async_commit             sync_commit
 665                          *              |                    |
 666                          *              | GO TO NEXT    "Journal Corruption"
 667                          *              | TRANSACTION
 668                          *              |
 669                          * {(n+1)th transanction}
 670                          *              |
 671                          *       _______|______________
 672                          *      |                     |
 673                          * Commit block found   Commit block not found
 674                          *      |                     |
 675                          * "Journal Corruption"       |
 676                          *               _____________|_________
 677                          *              |                       |
 678                          *      nth trans corrupt       OR   nth trans
 679                          *      and (n+1)th interrupted     interrupted
 680                          *      before commit block
 681                          *      could reach the disk.
 682                          *      (Cannot find the difference in above
 683                          *       mentioned conditions. Hence assume
 684                          *       "Interrupted Commit".)
 685                          */
 686
 687                         /* Found an expected commit block: if checksums
 688                          * are present verify them in PASS_SCAN; else not
 689                          * much to do other than move on to the next sequence
 690                          * number. */
 691                         if (pass == PASS_SCAN &&
 692                             JFS_HAS_COMPAT_FEATURE(journal,
 693                                     JFS_FEATURE_COMPAT_CHECKSUM)) {
 694                                 int chksum_err, chksum_seen;
 695                                 struct commit_header *cbh =
 696                                         (struct commit_header *)bh->b_data;
 697                                 unsigned found_chksum =
 698                                         be32_to_cpu(cbh->h_chksum[0]);
 699
 700                                 chksum_err = chksum_seen = 0;
 701
 702                                 jbd_debug(3, "Checksums %x %x\n",
 703                                           crc32_sum, found_chksum);
 704                                 if (info->end_transaction) {
 705                                         journal->j_failed_commit =
 706                                                 info->end_transaction;
 707                                         brelse(bh);
 708                                         break;
 709                                 }
 710
 711                                 if (crc32_sum == found_chksum &&
 712                                     cbh->h_chksum_type == JBD2_CRC32_CHKSUM &&
 713                                     cbh->h_chksum_size ==
 714                                                 JBD2_CRC32_CHKSUM_SIZE)
 715                                        chksum_seen = 1;
 716                                 else if (!(cbh->h_chksum_type == 0 &&
 717                                              cbh->h_chksum_size == 0 &&
 718                                              found_chksum == 0 &&
 719                                              !chksum_seen))
 720                                 /*
 721                                  * If fs is mounted using an old kernel and then
 722                                  * kernel with journal_chksum is used then we
 723                                  * get a situation where the journal flag has
 724                                  * checksum flag set but checksums are not
 725                                  * present i.e chksum = 0, in the individual
 726                                  * commit blocks.
 727                                  * Hence to avoid checksum failures, in this
 728                                  * situation, this extra check is added.
 729                                  */
 730                                                 chksum_err = 1;
 731
 732                                 if (chksum_err) {
 733                                         info->end_transaction = next_commit_ID;
 734                                         jbd_debug(1, "Checksum_err %x %x\n",
 735                                                   crc32_sum, found_chksum);
 736                                         if (!JFS_HAS_INCOMPAT_FEATURE(journal,
 737                                            JFS_FEATURE_INCOMPAT_ASYNC_COMMIT)){
 738                                                 journal->j_failed_commit =
 739                                                         next_commit_ID;
 740                                                 brelse(bh);
 741                                                 break;
 742                                         }
 743                                 }
 744                                 crc32_sum = ~0;
 745                         }
 746                         if (pass == PASS_SCAN &&
 747                             !jbd2_commit_block_csum_verify(journal,
 748                                                            bh->b_data)) {
 749                                 info->end_transaction = next_commit_ID;
 750
 751                                 if (!JFS_HAS_INCOMPAT_FEATURE(journal,
 752                                      JFS_FEATURE_INCOMPAT_ASYNC_COMMIT)) {
 753                                         journal->j_failed_commit =
 754                                                 next_commit_ID;
 755                                         brelse(bh);
 756                                         break;
 757                                 }
 758                         }
 759                         brelse(bh);
 760                         next_commit_ID++;
 761                         continue;
 762
 763                 case JFS_REVOKE_BLOCK:
 764                         /* If we aren't in the REVOKE pass, then we can
 765                          * just skip over this block. */
 766                         if (pass != PASS_REVOKE) {
 767                                 brelse(bh);
 768                                 continue;
 769                         }
 770
 771                         err = scan_revoke_records(journal, bh,
 772                                                   next_commit_ID, info);
 773                         brelse(bh);
 774                         if (err)
 775                                 goto failed;
 776                         continue;
 777
 778                 default:
 779                         jbd_debug(3, "Unrecognised magic %d, end of scan.\n",
 780                                   blocktype);
 781                         brelse(bh);
 782                         goto done;
 783                 }
 784         }
 785
 786  done:
 787         /*
 788          * We broke out of the log scan loop: either we came to the
 789          * known end of the log or we found an unexpected block in the
 790          * log.  If the latter happened, then we know that the "current"
 791          * transaction marks the end of the valid log.
 792          */
 793
 794         if (pass == PASS_SCAN) {
 795                 if (!info->end_transaction)
 796                         info->end_transaction = next_commit_ID;
 797         } else {
 798                 /* It's really bad news if different passes end up at
 799                  * different places (but possible due to IO errors). */
 800                 if (info->end_transaction != next_commit_ID) {
 801                         printk (KERN_ERR "JBD: recovery pass %d ended at "
 802                                 "transaction %u, expected %u\n",
 803                                 pass, next_commit_ID, info->end_transaction);
 804                         if (!success)
 805                                 success = -EIO;
 806                 }
 807         }
 808
 809         return success;
 810
 811  failed:
 812         return err;
 813 }
 814
 815 static int jbd2_revoke_block_csum_verify(journal_t *j,
 816                                          void *buf)
 817 {
 818         struct journal_revoke_tail *tail;
 819         __u32 provided, calculated;
 820
 821         if (!JFS_HAS_INCOMPAT_FEATURE(j, JFS_FEATURE_INCOMPAT_CSUM_V2))
 822                 return 1;
 823
 824         tail = (struct journal_revoke_tail *)(buf + j->j_blocksize -
 825                         sizeof(struct journal_revoke_tail));
 826         provided = tail->r_checksum;
 827         tail->r_checksum = 0;
 828         calculated = ext2fs_crc32c_le(~0, j->j_superblock->s_uuid,
 829                                       sizeof(j->j_superblock->s_uuid));
 830         calculated = ext2fs_crc32c_le(calculated, buf, j->j_blocksize);
 831         tail->r_checksum = provided;
 832
 833         provided = ext2fs_be32_to_cpu(provided);
 834         return provided == calculated;
 835 }
 836
 837 /* Scan a revoke record, marking all blocks mentioned as revoked. */
 838
 839 static int scan_revoke_records(journal_t *journal, struct buffer_head *bh,
 840                                tid_t sequence, struct recovery_info *info)
 841 {
 842         journal_revoke_header_t *header;
 843         int offset, max;
 844         int record_len = 4;
 845
 846         header = (journal_revoke_header_t *) bh->b_data;
 847         offset = sizeof(journal_revoke_header_t);
 848         max = be32_to_cpu(header->r_count);
 849
 850         if (!jbd2_revoke_block_csum_verify(journal, header))
 851                 return -EINVAL;
 852
 853         if (JFS_HAS_INCOMPAT_FEATURE(journal, JFS_FEATURE_INCOMPAT_64BIT))
 854                 record_len = 8;
 855
 856         while (offset < max) {
 857                 unsigned long long blocknr;
 858                 int err;
 859
 860                 if (record_len == 4) {
 861                         __be32 b;
 862                         memcpy(&b, bh->b_data + offset, sizeof(__be32));
 863                         blocknr = ext2fs_be32_to_cpu(b);
 864                 } else {
 865                         __be64 b;
 866                         memcpy(&b, bh->b_data + offset, sizeof(__be64));
 867                         blocknr = ext2fs_be64_to_cpu(b);
 868                 }
 869
 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 }