fs/jbd2/recovery.c

   1 /*
   2  * linux/fs/jbd2/recovery.c
   3  *
   4  * Written by Stephen C. Tweedie <[email protected]>, 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 "jfs_user.h"
  18 #else
  19 #include <linux/time.h>
  20 #include <linux/fs.h>
  21 #include <linux/jbd2.h>
  22 #include <linux/errno.h>
  23 #include <linux/crc32.h>
  24 #include <linux/blkdev.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 = jbd2_journal_bmap(journal, next, &blocknr);
  91
  92                 if (err) {
  93                         printk(KERN_ERR "JBD2: 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(REQ_OP_READ, 0, 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(REQ_OP_READ, 0, 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 "JBD2: corrupted journal superblock\n");
 143                 return -EFSCORRUPTED;
 144         }
 145
 146         err = jbd2_journal_bmap(journal, offset, &blocknr);
 147
 148         if (err) {
 149                 printk(KERN_ERR "JBD2: 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 "JBD2: 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_descriptor_block_csum_verify(journal_t *j, void *buf)
 178 {
 179         struct jbd2_journal_block_tail *tail;
 180         __be32 provided;
 181         __u32 calculated;
 182
 183         if (!jbd2_journal_has_csum_v2or3(j))
 184                 return 1;
 185
 186         tail = (struct jbd2_journal_block_tail *)(buf + j->j_blocksize -
 187                         sizeof(struct jbd2_journal_block_tail));
 188         provided = tail->t_checksum;
 189         tail->t_checksum = 0;
 190         calculated = jbd2_chksum(j, j->j_csum_seed, buf, j->j_blocksize);
 191         tail->t_checksum = provided;
 192
 193         return provided == cpu_to_be32(calculated);
 194 }
 195
 196 /*
 197  * Count the number of in-use tags in a journal descriptor block.
 198  */
 199
 200 static int count_tags(journal_t *journal, struct buffer_head *bh)
 201 {
 202         char *                  tagp;
 203         journal_block_tag_t *   tag;
 204         int                     nr = 0, size = journal->j_blocksize;
 205         int                     tag_bytes = journal_tag_bytes(journal);
 206
 207         if (jbd2_journal_has_csum_v2or3(journal))
 208                 size -= sizeof(struct jbd2_journal_block_tail);
 209
 210         tagp = &bh->b_data[sizeof(journal_header_t)];
 211
 212         while ((tagp - bh->b_data + tag_bytes) <= size) {
 213                 tag = (journal_block_tag_t *) tagp;
 214
 215                 nr++;
 216                 tagp += tag_bytes;
 217                 if (!(tag->t_flags & cpu_to_be16(JBD2_FLAG_SAME_UUID)))
 218                         tagp += 16;
 219
 220                 if (tag->t_flags & cpu_to_be16(JBD2_FLAG_LAST_TAG))
 221                         break;
 222         }
 223
 224         return nr;
 225 }
 226
 227
 228 /* Make sure we wrap around the log correctly! */
 229 #define wrap(journal, var)                                              \
 230 do {                                                                    \
 231         if (var >= (journal)->j_last)                                   \
 232                 var -= ((journal)->j_last - (journal)->j_first);        \
 233 } while (0)
 234
 235 /**
 236  * jbd2_journal_recover - recovers a on-disk journal
 237  * @journal: the journal to recover
 238  *
 239  * The primary function for recovering the log contents when mounting a
 240  * journaled device.
 241  *
 242  * Recovery is done in three passes.  In the first pass, we look for the
 243  * end of the log.  In the second, we assemble the list of revoke
 244  * blocks.  In the third and final pass, we replay any un-revoked blocks
 245  * in the log.
 246  */
 247 int jbd2_journal_recover(journal_t *journal)
 248 {
 249         int                     err, err2;
 250         journal_superblock_t *  sb;
 251
 252         struct recovery_info    info;
 253
 254         memset(&info, 0, sizeof(info));
 255         sb = journal->j_superblock;
 256
 257         /*
 258          * The journal superblock's s_start field (the current log head)
 259          * is always zero if, and only if, the journal was cleanly
 260          * unmounted.
 261          */
 262
 263         if (!sb->s_start) {
 264                 jbd_debug(1, "No recovery required, last transaction %d\n",
 265                           be32_to_cpu(sb->s_sequence));
 266                 journal->j_transaction_sequence = be32_to_cpu(sb->s_sequence) + 1;
 267                 return 0;
 268         }
 269
 270         err = do_one_pass(journal, &info, PASS_SCAN);
 271         if (!err)
 272                 err = do_one_pass(journal, &info, PASS_REVOKE);
 273         if (!err)
 274                 err = do_one_pass(journal, &info, PASS_REPLAY);
 275
 276         jbd_debug(1, "JBD2: recovery, exit status %d, "
 277                   "recovered transactions %u to %u\n",
 278                   err, info.start_transaction, info.end_transaction);
 279         jbd_debug(1, "JBD2: Replayed %d and revoked %d/%d blocks\n",
 280                   info.nr_replays, info.nr_revoke_hits, info.nr_revokes);
 281
 282         /* Restart the log at the next transaction ID, thus invalidating
 283          * any existing commit records in the log. */
 284         journal->j_transaction_sequence = ++info.end_transaction;
 285
 286         jbd2_journal_clear_revoke(journal);
 287         err2 = sync_blockdev(journal->j_fs_dev);
 288         if (!err)
 289                 err = err2;
 290         /* Make sure all replayed data is on permanent storage */
 291         if (journal->j_flags & JBD2_BARRIER) {
 292                 err2 = blkdev_issue_flush(journal->j_fs_dev, GFP_KERNEL, NULL);
 293                 if (!err)
 294                         err = err2;
 295         }
 296         return err;
 297 }
 298
 299 /**
 300  * jbd2_journal_skip_recovery - Start journal and wipe exiting records
 301  * @journal: journal to startup
 302  *
 303  * Locate any valid recovery information from the journal and set up the
 304  * journal structures in memory to ignore it (presumably because the
 305  * caller has evidence that it is out of date).
 306  * This function doesn't appear to be exported..
 307  *
 308  * We perform one pass over the journal to allow us to tell the user how
 309  * much recovery information is being erased, and to let us initialise
 310  * the journal transaction sequence numbers to the next unused ID.
 311  */
 312 int jbd2_journal_skip_recovery(journal_t *journal)
 313 {
 314         int                     err;
 315
 316         struct recovery_info    info;
 317
 318         memset (&info, 0, sizeof(info));
 319
 320         err = do_one_pass(journal, &info, PASS_SCAN);
 321
 322         if (err) {
 323                 printk(KERN_ERR "JBD2: error %d scanning journal\n", err);
 324                 ++journal->j_transaction_sequence;
 325         } else {
 326 #ifdef CONFIG_JBD2_DEBUG
 327                 int dropped = info.end_transaction -
 328                         be32_to_cpu(journal->j_superblock->s_sequence);
 329                 jbd_debug(1,
 330                           "JBD2: ignoring %d transaction%s from the journal.\n",
 331                           dropped, (dropped == 1) ? "" : "s");
 332 #endif
 333                 journal->j_transaction_sequence = ++info.end_transaction;
 334         }
 335
 336         journal->j_tail = 0;
 337         return err;
 338 }
 339
 340 static inline unsigned long long read_tag_block(journal_t *journal,
 341                                                 journal_block_tag_t *tag)
 342 {
 343         unsigned long long block = be32_to_cpu(tag->t_blocknr);
 344         if (jbd2_has_feature_64bit(journal))
 345                 block |= (u64)be32_to_cpu(tag->t_blocknr_high) << 32;
 346         return block;
 347 }
 348
 349 /*
 350  * calc_chksums calculates the checksums for the blocks described in the
 351  * descriptor block.
 352  */
 353 static int calc_chksums(journal_t *journal, struct buffer_head *bh,
 354                         unsigned long *next_log_block, __u32 *crc32_sum)
 355 {
 356         int i, num_blks, err;
 357         unsigned long io_block;
 358         struct buffer_head *obh;
 359
 360         num_blks = count_tags(journal, bh);
 361         /* Calculate checksum of the descriptor block. */
 362         *crc32_sum = crc32_be(*crc32_sum, (void *)bh->b_data, bh->b_size);
 363
 364         for (i = 0; i < num_blks; i++) {
 365                 io_block = (*next_log_block)++;
 366                 wrap(journal, *next_log_block);
 367                 err = jread(&obh, journal, io_block);
 368                 if (err) {
 369                         printk(KERN_ERR "JBD2: IO error %d recovering block "
 370                                 "%lu in log\n", err, io_block);
 371                         return 1;
 372                 } else {
 373                         *crc32_sum = crc32_be(*crc32_sum, (void *)obh->b_data,
 374                                      obh->b_size);
 375                 }
 376                 put_bh(obh);
 377         }
 378         return 0;
 379 }
 380
 381 static int jbd2_commit_block_csum_verify(journal_t *j, void *buf)
 382 {
 383         struct commit_header *h;
 384         __be32 provided;
 385         __u32 calculated;
 386
 387         if (!jbd2_journal_has_csum_v2or3(j))
 388                 return 1;
 389
 390         h = buf;
 391         provided = h->h_chksum[0];
 392         h->h_chksum[0] = 0;
 393         calculated = jbd2_chksum(j, j->j_csum_seed, buf, j->j_blocksize);
 394         h->h_chksum[0] = provided;
 395
 396         return provided == cpu_to_be32(calculated);
 397 }
 398
 399 static int jbd2_block_tag_csum_verify(journal_t *j, journal_block_tag_t *tag,
 400                                       void *buf, __u32 sequence)
 401 {
 402         journal_block_tag3_t *tag3 = (journal_block_tag3_t *)tag;
 403         __u32 csum32;
 404         __be32 seq;
 405
 406         if (!jbd2_journal_has_csum_v2or3(j))
 407                 return 1;
 408
 409         seq = cpu_to_be32(sequence);
 410         csum32 = jbd2_chksum(j, j->j_csum_seed, (__u8 *)&seq, sizeof(seq));
 411         csum32 = jbd2_chksum(j, csum32, buf, j->j_blocksize);
 412
 413         if (jbd2_has_feature_csum3(j))
 414                 return tag3->t_checksum == cpu_to_be32(csum32);
 415         else
 416                 return tag->t_checksum == cpu_to_be16(csum32);
 417 }
 418
 419 static int do_one_pass(journal_t *journal,
 420                         struct recovery_info *info, enum passtype pass)
 421 {
 422         unsigned int            first_commit_ID, next_commit_ID;
 423         unsigned long           next_log_block;
 424         int                     err, success = 0;
 425         journal_superblock_t *  sb;
 426         journal_header_t *      tmp;
 427         struct buffer_head *    bh;
 428         unsigned int            sequence;
 429         int                     blocktype;
 430         int                     tag_bytes = journal_tag_bytes(journal);
 431         __u32                   crc32_sum = ~0; /* Transactional Checksums */
 432         int                     descr_csum_size = 0;
 433         int                     block_error = 0;
 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, "JBD2: 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(JBD2_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 JBD2_DESCRIPTOR_BLOCK:
 519                         /* Verify checksum first */
 520                         if (jbd2_journal_has_csum_v2or3(journal))
 521                                 descr_csum_size =
 522                                         sizeof(struct jbd2_journal_block_tail);
 523                         if (descr_csum_size > 0 &&
 524                             !jbd2_descriptor_block_csum_verify(journal,
 525                                                                bh->b_data)) {
 526                                 printk(KERN_ERR "JBD2: Invalid checksum "
 527                                        "recovering block %lu in log\n",
 528                                        next_log_block);
 529                                 err = -EFSBADCRC;
 530                                 brelse(bh);
 531                                 goto failed;
 532                         }
 533
 534                         /* If it is a valid descriptor block, replay it
 535                          * in pass REPLAY; if journal_checksums enabled, then
 536                          * calculate checksums in PASS_SCAN, otherwise,
 537                          * just skip over the blocks it describes. */
 538                         if (pass != PASS_REPLAY) {
 539                                 if (pass == PASS_SCAN &&
 540                                     jbd2_has_feature_checksum(journal) &&
 541                                     !info->end_transaction) {
 542                                         if (calc_chksums(journal, bh,
 543                                                         &next_log_block,
 544                                                         &crc32_sum)) {
 545                                                 put_bh(bh);
 546                                                 break;
 547                                         }
 548                                         put_bh(bh);
 549                                         continue;
 550                                 }
 551                                 next_log_block += count_tags(journal, bh);
 552                                 wrap(journal, next_log_block);
 553                                 put_bh(bh);
 554                                 continue;
 555                         }
 556
 557                         /* A descriptor block: we can now write all of
 558                          * the data blocks.  Yay, useful work is finally
 559                          * getting done here! */
 560
 561                         tagp = &bh->b_data[sizeof(journal_header_t)];
 562                         while ((tagp - bh->b_data + tag_bytes)
 563                                <= journal->j_blocksize - descr_csum_size) {
 564                                 unsigned long io_block;
 565
 566                                 tag = (journal_block_tag_t *) tagp;
 567                                 flags = be16_to_cpu(tag->t_flags);
 568
 569                                 io_block = next_log_block++;
 570                                 wrap(journal, next_log_block);
 571                                 err = jread(&obh, journal, io_block);
 572                                 if (err) {
 573                                         /* Recover what we can, but
 574                                          * report failure at the end. */
 575                                         success = err;
 576                                         printk(KERN_ERR
 577                                                 "JBD2: IO error %d recovering "
 578                                                 "block %ld in log\n",
 579                                                 err, io_block);
 580                                 } else {
 581                                         unsigned long long blocknr;
 582
 583                                         J_ASSERT(obh != NULL);
 584                                         blocknr = read_tag_block(journal,
 585                                                                  tag);
 586
 587                                         /* If the block has been
 588                                          * revoked, then we're all done
 589                                          * here. */
 590                                         if (jbd2_journal_test_revoke
 591                                             (journal, blocknr,
 592                                              next_commit_ID)) {
 593                                                 brelse(obh);
 594                                                 ++info->nr_revoke_hits;
 595                                                 goto skip_write;
 596                                         }
 597
 598                                         /* Look for block corruption */
 599                                         if (!jbd2_block_tag_csum_verify(
 600                                                 journal, tag, obh->b_data,
 601                                                 be32_to_cpu(tmp->h_sequence))) {
 602                                                 brelse(obh);
 603                                                 success = -EFSBADCRC;
 604                                                 printk(KERN_ERR "JBD2: Invalid "
 605                                                        "checksum recovering "
 606                                                        "block %llu in log\n",
 607                                                        blocknr);
 608                                                 block_error = 1;
 609                                                 goto skip_write;
 610                                         }
 611
 612                                         /* Find a buffer for the new
 613                                          * data being restored */
 614                                         nbh = __getblk(journal->j_fs_dev,
 615                                                         blocknr,
 616                                                         journal->j_blocksize);
 617                                         if (nbh == NULL) {
 618                                                 printk(KERN_ERR
 619                                                        "JBD2: Out of memory "
 620                                                        "during recovery.\n");
 621                                                 err = -ENOMEM;
 622                                                 brelse(bh);
 623                                                 brelse(obh);
 624                                                 goto failed;
 625                                         }
 626
 627                                         lock_buffer(nbh);
 628                                         memcpy(nbh->b_data, obh->b_data,
 629                                                         journal->j_blocksize);
 630                                         if (flags & JBD2_FLAG_ESCAPE) {
 631                                                 *((__be32 *)nbh->b_data) =
 632                                                 cpu_to_be32(JBD2_MAGIC_NUMBER);
 633                                         }
 634
 635                                         BUFFER_TRACE(nbh, "marking dirty");
 636                                         set_buffer_uptodate(nbh);
 637                                         mark_buffer_dirty(nbh);
 638                                         BUFFER_TRACE(nbh, "marking uptodate");
 639                                         ++info->nr_replays;
 640                                         /* ll_rw_block(WRITE, 1, &nbh); */
 641                                         unlock_buffer(nbh);
 642                                         brelse(obh);
 643                                         brelse(nbh);
 644                                 }
 645
 646                         skip_write:
 647                                 tagp += tag_bytes;
 648                                 if (!(flags & JBD2_FLAG_SAME_UUID))
 649                                         tagp += 16;
 650
 651                                 if (flags & JBD2_FLAG_LAST_TAG)
 652                                         break;
 653                         }
 654
 655                         brelse(bh);
 656                         continue;
 657
 658                 case JBD2_COMMIT_BLOCK:
 659                         /*     How to differentiate between interrupted commit
 660                          *               and journal corruption ?
 661                          *
 662                          * {nth transaction}
 663                          *        Checksum Verification Failed
 664                          *                       |
 665                          *               ____________________
 666                          *              |                    |
 667                          *      async_commit             sync_commit
 668                          *              |                    |
 669                          *              | GO TO NEXT    "Journal Corruption"
 670                          *              | TRANSACTION
 671                          *              |
 672                          * {(n+1)th transanction}
 673                          *              |
 674                          *       _______|______________
 675                          *      |                     |
 676                          * Commit block found   Commit block not found
 677                          *      |                     |
 678                          * "Journal Corruption"       |
 679                          *               _____________|_________
 680                          *              |                       |
 681                          *      nth trans corrupt       OR   nth trans
 682                          *      and (n+1)th interrupted     interrupted
 683                          *      before commit block
 684                          *      could reach the disk.
 685                          *      (Cannot find the difference in above
 686                          *       mentioned conditions. Hence assume
 687                          *       "Interrupted Commit".)
 688                          */
 689
 690                         /* Found an expected commit block: if checksums
 691                          * are present verify them in PASS_SCAN; else not
 692                          * much to do other than move on to the next sequence
 693                          * number. */
 694                         if (pass == PASS_SCAN &&
 695                             jbd2_has_feature_checksum(journal)) {
 696                                 int chksum_err, chksum_seen;
 697                                 struct commit_header *cbh =
 698                                         (struct commit_header *)bh->b_data;
 699                                 unsigned found_chksum =
 700                                         be32_to_cpu(cbh->h_chksum[0]);
 701
 702                                 chksum_err = chksum_seen = 0;
 703
 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
 735                                         if (!jbd2_has_feature_async_commit(journal)) {
 736                                                 journal->j_failed_commit =
 737                                                         next_commit_ID;
 738                                                 brelse(bh);
 739                                                 break;
 740                                         }
 741                                 }
 742                                 crc32_sum = ~0;
 743                         }
 744                         if (pass == PASS_SCAN &&
 745                             !jbd2_commit_block_csum_verify(journal,
 746                                                            bh->b_data)) {
 747                                 info->end_transaction = next_commit_ID;
 748
 749                                 if (!jbd2_has_feature_async_commit(journal)) {
 750                                         journal->j_failed_commit =
 751                                                 next_commit_ID;
 752                                         brelse(bh);
 753                                         break;
 754                                 }
 755                         }
 756                         brelse(bh);
 757                         next_commit_ID++;
 758                         continue;
 759
 760                 case JBD2_REVOKE_BLOCK:
 761                         /* If we aren't in the REVOKE pass, then we can
 762                          * just skip over this block. */
 763                         if (pass != PASS_REVOKE) {
 764                                 brelse(bh);
 765                                 continue;
 766                         }
 767
 768                         err = scan_revoke_records(journal, bh,
 769                                                   next_commit_ID, info);
 770                         brelse(bh);
 771                         if (err)
 772                                 goto failed;
 773                         continue;
 774
 775                 default:
 776                         jbd_debug(3, "Unrecognised magic %d, end of scan.\n",
 777                                   blocktype);
 778                         brelse(bh);
 779                         goto done;
 780                 }
 781         }
 782
 783  done:
 784         /*
 785          * We broke out of the log scan loop: either we came to the
 786          * known end of the log or we found an unexpected block in the
 787          * log.  If the latter happened, then we know that the "current"
 788          * transaction marks the end of the valid log.
 789          */
 790
 791         if (pass == PASS_SCAN) {
 792                 if (!info->end_transaction)
 793                         info->end_transaction = next_commit_ID;
 794         } else {
 795                 /* It's really bad news if different passes end up at
 796                  * different places (but possible due to IO errors). */
 797                 if (info->end_transaction != next_commit_ID) {
 798                         printk(KERN_ERR "JBD2: recovery pass %d ended at "
 799                                 "transaction %u, expected %u\n",
 800                                 pass, next_commit_ID, info->end_transaction);
 801                         if (!success)
 802                                 success = -EIO;
 803                 }
 804         }
 805         if (block_error && success == 0)
 806                 success = -EIO;
 807         return success;
 808
 809  failed:
 810         return err;
 811 }
 812
 813 /* Scan a revoke record, marking all blocks mentioned as revoked. */
 814
 815 static int scan_revoke_records(journal_t *journal, struct buffer_head *bh,
 816                                tid_t sequence, struct recovery_info *info)
 817 {
 818         jbd2_journal_revoke_header_t *header;
 819         int offset, max;
 820         int csum_size = 0;
 821         __u32 rcount;
 822         int record_len = 4;
 823
 824         header = (jbd2_journal_revoke_header_t *) bh->b_data;
 825         offset = sizeof(jbd2_journal_revoke_header_t);
 826         rcount = be32_to_cpu(header->r_count);
 827
 828         if (!jbd2_descriptor_block_csum_verify(journal, header))
 829                 return -EFSBADCRC;
 830
 831         if (jbd2_journal_has_csum_v2or3(journal))
 832                 csum_size = sizeof(struct jbd2_journal_block_tail);
 833         if (rcount > journal->j_blocksize - csum_size)
 834                 return -EINVAL;
 835         max = rcount;
 836
 837         if (jbd2_has_feature_64bit(journal))
 838                 record_len = 8;
 839
 840         while (offset + record_len <= max) {
 841                 unsigned long long blocknr;
 842                 int err;
 843
 844                 if (record_len == 4)
 845                         blocknr = be32_to_cpu(* ((__be32 *) (bh->b_data+offset)));
 846                 else
 847                         blocknr = be64_to_cpu(* ((__be64 *) (bh->b_data+offset)));
 848                 offset += record_len;
 849                 err = jbd2_journal_set_revoke(journal, blocknr, sequence);
 850                 if (err)
 851                         return err;
 852                 ++info->nr_revokes;
 853         }
 854         return 0;
 855 }