fs/jbd2/recovery.c

   1 // SPDX-License-Identifier: GPL-2.0+
   2 /*
   3  * linux/fs/jbd2/recovery.c
   4  *
   5  * Written by Stephen C. Tweedie <sct@redhat.com>, 1999
   6  *
   7  * Copyright 1999-2000 Red Hat Software --- All Rights Reserved
   8  *
   9  * Journal recovery routines for the generic filesystem journaling code;
  10  * part of the ext2fs journaling system.
  11  */
  12
  13 #ifndef __KERNEL__
  14 #include "jfs_user.h"
  15 #else
  16 #include <linux/time.h>
  17 #include <linux/fs.h>
  18 #include <linux/jbd2.h>
  19 #include <linux/errno.h>
  20 #include <linux/crc32.h>
  21 #include <linux/blkdev.h>
  22 #endif
  23
  24 /*
  25  * Maintain information about the progress of the recovery job, so that
  26  * the different passes can carry information between them.
  27  */
  28 struct recovery_info
  29 {
  30         tid_t           start_transaction;
  31         tid_t           end_transaction;
  32         unsigned long   head_block;
  33
  34         int             nr_replays;
  35         int             nr_revokes;
  36         int             nr_revoke_hits;
  37 };
  38
  39 static int do_one_pass(journal_t *journal,
  40                                 struct recovery_info *info, enum passtype pass);
  41 static int scan_revoke_records(journal_t *, struct buffer_head *,
  42                                 tid_t, struct recovery_info *);
  43
  44 #ifdef __KERNEL__
  45
  46 /* Release readahead buffers after use */
  47 static void journal_brelse_array(struct buffer_head *b[], int n)
  48 {
  49         while (--n >= 0)
  50                 brelse (b[n]);
  51 }
  52
  53
  54 /*
  55  * When reading from the journal, we are going through the block device
  56  * layer directly and so there is no readahead being done for us.  We
  57  * need to implement any readahead ourselves if we want it to happen at
  58  * all.  Recovery is basically one long sequential read, so make sure we
  59  * do the IO in reasonably large chunks.
  60  *
  61  * This is not so critical that we need to be enormously clever about
  62  * the readahead size, though.  128K is a purely arbitrary, good-enough
  63  * fixed value.
  64  */
  65
  66 #define MAXBUF 8
  67 static int do_readahead(journal_t *journal, unsigned int start)
  68 {
  69         int err;
  70         unsigned int max, nbufs, next;
  71         unsigned long long blocknr;
  72         struct buffer_head *bh;
  73
  74         struct buffer_head * bufs[MAXBUF];
  75
  76         /* Do up to 128K of readahead */
  77         max = start + (128 * 1024 / journal->j_blocksize);
  78         if (max > journal->j_total_len)
  79                 max = journal->j_total_len;
  80
  81         /* Do the readahead itself.  We'll submit MAXBUF buffer_heads at
  82          * a time to the block device IO layer. */
  83
  84         nbufs = 0;
  85
  86         for (next = start; next < max; next++) {
  87                 err = jbd2_journal_bmap(journal, next, &blocknr);
  88
  89                 if (err) {
  90                         printk(KERN_ERR "JBD2: bad block at offset %u\n",
  91                                 next);
  92                         goto failed;
  93                 }
  94
  95                 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
  96                 if (!bh) {
  97                         err = -ENOMEM;
  98                         goto failed;
  99                 }
 100
 101                 if (!buffer_uptodate(bh) && !buffer_locked(bh)) {
 102                         bufs[nbufs++] = bh;
 103                         if (nbufs == MAXBUF) {
 104                                 bh_readahead_batch(nbufs, bufs, 0);
 105                                 journal_brelse_array(bufs, nbufs);
 106                                 nbufs = 0;
 107                         }
 108                 } else
 109                         brelse(bh);
 110         }
 111
 112         if (nbufs)
 113                 bh_readahead_batch(nbufs, bufs, 0);
 114         err = 0;
 115
 116 failed:
 117         if (nbufs)
 118                 journal_brelse_array(bufs, nbufs);
 119         return err;
 120 }
 121
 122 #endif /* __KERNEL__ */
 123
 124
 125 /*
 126  * Read a block from the journal
 127  */
 128
 129 static int jread(struct buffer_head **bhp, journal_t *journal,
 130                  unsigned int offset)
 131 {
 132         int err;
 133         unsigned long long blocknr;
 134         struct buffer_head *bh;
 135
 136         *bhp = NULL;
 137
 138         if (offset >= journal->j_total_len) {
 139                 printk(KERN_ERR "JBD2: corrupted journal superblock\n");
 140                 return -EFSCORRUPTED;
 141         }
 142
 143         err = jbd2_journal_bmap(journal, offset, &blocknr);
 144
 145         if (err) {
 146                 printk(KERN_ERR "JBD2: bad block at offset %u\n",
 147                         offset);
 148                 return err;
 149         }
 150
 151         bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
 152         if (!bh)
 153                 return -ENOMEM;
 154
 155         if (!buffer_uptodate(bh)) {
 156                 /*
 157                  * If this is a brand new buffer, start readahead.
 158                  * Otherwise, we assume we are already reading it.
 159                  */
 160                 bool need_readahead = !buffer_req(bh);
 161
 162                 bh_read_nowait(bh, 0);
 163                 if (need_readahead)
 164                         do_readahead(journal, offset);
 165                 wait_on_buffer(bh);
 166         }
 167
 168         if (!buffer_uptodate(bh)) {
 169                 printk(KERN_ERR "JBD2: Failed to read block at offset %u\n",
 170                         offset);
 171                 brelse(bh);
 172                 return -EIO;
 173         }
 174
 175         *bhp = bh;
 176         return 0;
 177 }
 178
 179 static int jbd2_descriptor_block_csum_verify(journal_t *j, void *buf)
 180 {
 181         struct jbd2_journal_block_tail *tail;
 182         __be32 provided;
 183         __u32 calculated;
 184
 185         if (!jbd2_journal_has_csum_v2or3(j))
 186                 return 1;
 187
 188         tail = (struct jbd2_journal_block_tail *)((char *)buf +
 189                 j->j_blocksize - sizeof(struct jbd2_journal_block_tail));
 190         provided = tail->t_checksum;
 191         tail->t_checksum = 0;
 192         calculated = jbd2_chksum(j, j->j_csum_seed, buf, j->j_blocksize);
 193         tail->t_checksum = provided;
 194
 195         return provided == cpu_to_be32(calculated);
 196 }
 197
 198 /*
 199  * Count the number of in-use tags in a journal descriptor block.
 200  */
 201
 202 static int count_tags(journal_t *journal, struct buffer_head *bh)
 203 {
 204         char *                  tagp;
 205         journal_block_tag_t     tag;
 206         int                     nr = 0, size = journal->j_blocksize;
 207         int                     tag_bytes = journal_tag_bytes(journal);
 208
 209         if (jbd2_journal_has_csum_v2or3(journal))
 210                 size -= sizeof(struct jbd2_journal_block_tail);
 211
 212         tagp = &bh->b_data[sizeof(journal_header_t)];
 213
 214         while ((tagp - bh->b_data + tag_bytes) <= size) {
 215                 memcpy(&tag, tagp, sizeof(tag));
 216
 217                 nr++;
 218                 tagp += tag_bytes;
 219                 if (!(tag.t_flags & cpu_to_be16(JBD2_FLAG_SAME_UUID)))
 220                         tagp += 16;
 221
 222                 if (tag.t_flags & cpu_to_be16(JBD2_FLAG_LAST_TAG))
 223                         break;
 224         }
 225
 226         return nr;
 227 }
 228
 229
 230 /* Make sure we wrap around the log correctly! */
 231 #define wrap(journal, var)                                              \
 232 do {                                                                    \
 233         unsigned long _wrap_last =                                      \
 234                 jbd2_has_feature_fast_commit(journal) ?                 \
 235                         (journal)->j_fc_last : (journal)->j_last;       \
 236                                                                         \
 237         if (var >= _wrap_last)                                          \
 238                 var -= (_wrap_last - (journal)->j_first);               \
 239 } while (0)
 240
 241 static int fc_do_one_pass(journal_t *journal,
 242                           struct recovery_info *info, enum passtype pass)
 243 {
 244         unsigned int expected_commit_id = info->end_transaction;
 245         unsigned long next_fc_block;
 246         struct buffer_head *bh;
 247         int err = 0;
 248
 249         next_fc_block = journal->j_fc_first;
 250         if (!journal->j_fc_replay_callback)
 251                 return 0;
 252
 253         while (next_fc_block <= journal->j_fc_last) {
 254                 jbd2_debug(3, "Fast commit replay: next block %ld\n",
 255                           next_fc_block);
 256                 err = jread(&bh, journal, next_fc_block);
 257                 if (err) {
 258                         jbd2_debug(3, "Fast commit replay: read error\n");
 259                         break;
 260                 }
 261
 262                 err = journal->j_fc_replay_callback(journal, bh, pass,
 263                                         next_fc_block - journal->j_fc_first,
 264                                         expected_commit_id);
 265                 brelse(bh);
 266                 next_fc_block++;
 267                 if (err < 0 || err == JBD2_FC_REPLAY_STOP)
 268                         break;
 269                 err = 0;
 270         }
 271
 272         if (err)
 273                 jbd2_debug(3, "Fast commit replay failed, err = %d\n", err);
 274
 275         return err;
 276 }
 277
 278 /**
 279  * jbd2_journal_recover - recovers a on-disk journal
 280  * @journal: the journal to recover
 281  *
 282  * The primary function for recovering the log contents when mounting a
 283  * journaled device.
 284  *
 285  * Recovery is done in three passes.  In the first pass, we look for the
 286  * end of the log.  In the second, we assemble the list of revoke
 287  * blocks.  In the third and final pass, we replay any un-revoked blocks
 288  * in the log.
 289  */
 290 int jbd2_journal_recover(journal_t *journal)
 291 {
 292         int                     err, err2;
 293         journal_superblock_t *  sb;
 294
 295         struct recovery_info    info;
 296
 297         memset(&info, 0, sizeof(info));
 298         sb = journal->j_superblock;
 299
 300         /*
 301          * The journal superblock's s_start field (the current log head)
 302          * is always zero if, and only if, the journal was cleanly
 303          * unmounted.
 304          */
 305         if (!sb->s_start) {
 306                 jbd2_debug(1, "No recovery required, last transaction %d, head block %u\n",
 307                           be32_to_cpu(sb->s_sequence), be32_to_cpu(sb->s_head));
 308                 journal->j_transaction_sequence = be32_to_cpu(sb->s_sequence) + 1;
 309                 journal->j_head = be32_to_cpu(sb->s_head);
 310                 return 0;
 311         }
 312
 313         err = do_one_pass(journal, &info, PASS_SCAN);
 314         if (!err)
 315                 err = do_one_pass(journal, &info, PASS_REVOKE);
 316         if (!err)
 317                 err = do_one_pass(journal, &info, PASS_REPLAY);
 318
 319         jbd2_debug(1, "JBD2: recovery, exit status %d, "
 320                   "recovered transactions %u to %u\n",
 321                   err, info.start_transaction, info.end_transaction);
 322         jbd2_debug(1, "JBD2: Replayed %d and revoked %d/%d blocks\n",
 323                   info.nr_replays, info.nr_revoke_hits, info.nr_revokes);
 324
 325         /* Restart the log at the next transaction ID, thus invalidating
 326          * any existing commit records in the log. */
 327         journal->j_transaction_sequence = ++info.end_transaction;
 328         journal->j_head = info.head_block;
 329         jbd2_debug(1, "JBD2: last transaction %d, head block %lu\n",
 330                   journal->j_transaction_sequence, journal->j_head);
 331
 332         jbd2_journal_clear_revoke(journal);
 333         err2 = sync_blockdev(journal->j_fs_dev);
 334         if (!err)
 335                 err = err2;
 336         /* Make sure all replayed data is on permanent storage */
 337         if (journal->j_flags & JBD2_BARRIER) {
 338                 err2 = blkdev_issue_flush(journal->j_fs_dev);
 339                 if (!err)
 340                         err = err2;
 341         }
 342         return err;
 343 }
 344
 345 /**
 346  * jbd2_journal_skip_recovery - Start journal and wipe exiting records
 347  * @journal: journal to startup
 348  *
 349  * Locate any valid recovery information from the journal and set up the
 350  * journal structures in memory to ignore it (presumably because the
 351  * caller has evidence that it is out of date).
 352  * This function doesn't appear to be exported..
 353  *
 354  * We perform one pass over the journal to allow us to tell the user how
 355  * much recovery information is being erased, and to let us initialise
 356  * the journal transaction sequence numbers to the next unused ID.
 357  */
 358 int jbd2_journal_skip_recovery(journal_t *journal)
 359 {
 360         int                     err;
 361
 362         struct recovery_info    info;
 363
 364         memset (&info, 0, sizeof(info));
 365
 366         err = do_one_pass(journal, &info, PASS_SCAN);
 367
 368         if (err) {
 369                 printk(KERN_ERR "JBD2: error %d scanning journal\n", err);
 370                 ++journal->j_transaction_sequence;
 371                 journal->j_head = journal->j_first;
 372         } else {
 373 #ifdef CONFIG_JBD2_DEBUG
 374                 int dropped = info.end_transaction -
 375                         be32_to_cpu(journal->j_superblock->s_sequence);
 376                 jbd2_debug(1,
 377                           "JBD2: ignoring %d transaction%s from the journal.\n",
 378                           dropped, (dropped == 1) ? "" : "s");
 379 #endif
 380                 journal->j_transaction_sequence = ++info.end_transaction;
 381                 journal->j_head = info.head_block;
 382         }
 383
 384         journal->j_tail = 0;
 385         return err;
 386 }
 387
 388 static inline unsigned long long read_tag_block(journal_t *journal,
 389                                                 journal_block_tag_t *tag)
 390 {
 391         unsigned long long block = be32_to_cpu(tag->t_blocknr);
 392         if (jbd2_has_feature_64bit(journal))
 393                 block |= (u64)be32_to_cpu(tag->t_blocknr_high) << 32;
 394         return block;
 395 }
 396
 397 /*
 398  * calc_chksums calculates the checksums for the blocks described in the
 399  * descriptor block.
 400  */
 401 static int calc_chksums(journal_t *journal, struct buffer_head *bh,
 402                         unsigned long *next_log_block, __u32 *crc32_sum)
 403 {
 404         int i, num_blks, err;
 405         unsigned long io_block;
 406         struct buffer_head *obh;
 407
 408         num_blks = count_tags(journal, bh);
 409         /* Calculate checksum of the descriptor block. */
 410         *crc32_sum = crc32_be(*crc32_sum, (void *)bh->b_data, bh->b_size);
 411
 412         for (i = 0; i < num_blks; i++) {
 413                 io_block = (*next_log_block)++;
 414                 wrap(journal, *next_log_block);
 415                 err = jread(&obh, journal, io_block);
 416                 if (err) {
 417                         printk(KERN_ERR "JBD2: IO error %d recovering block "
 418                                 "%lu in log\n", err, io_block);
 419                         return 1;
 420                 } else {
 421                         *crc32_sum = crc32_be(*crc32_sum, (void *)obh->b_data,
 422                                      obh->b_size);
 423                 }
 424                 put_bh(obh);
 425         }
 426         return 0;
 427 }
 428
 429 static int jbd2_commit_block_csum_verify(journal_t *j, void *buf)
 430 {
 431         struct commit_header *h;
 432         __be32 provided;
 433         __u32 calculated;
 434
 435         if (!jbd2_journal_has_csum_v2or3(j))
 436                 return 1;
 437
 438         h = buf;
 439         provided = h->h_chksum[0];
 440         h->h_chksum[0] = 0;
 441         calculated = jbd2_chksum(j, j->j_csum_seed, buf, j->j_blocksize);
 442         h->h_chksum[0] = provided;
 443
 444         return provided == cpu_to_be32(calculated);
 445 }
 446
 447 static int jbd2_block_tag_csum_verify(journal_t *j, journal_block_tag_t *tag,
 448                                       journal_block_tag3_t *tag3,
 449                                       void *buf, __u32 sequence)
 450 {
 451         __u32 csum32;
 452         __be32 seq;
 453
 454         if (!jbd2_journal_has_csum_v2or3(j))
 455                 return 1;
 456
 457         seq = cpu_to_be32(sequence);
 458         csum32 = jbd2_chksum(j, j->j_csum_seed, (__u8 *)&seq, sizeof(seq));
 459         csum32 = jbd2_chksum(j, csum32, buf, j->j_blocksize);
 460
 461         if (jbd2_has_feature_csum3(j))
 462                 return tag3->t_checksum == cpu_to_be32(csum32);
 463         else
 464                 return tag->t_checksum == cpu_to_be16(csum32);
 465 }
 466
 467 static int do_one_pass(journal_t *journal,
 468                         struct recovery_info *info, enum passtype pass)
 469 {
 470         unsigned int            first_commit_ID, next_commit_ID;
 471         unsigned long           next_log_block, head_block;
 472         int                     err, success = 0;
 473         journal_superblock_t *  sb;
 474         journal_header_t *      tmp;
 475         struct buffer_head *    bh;
 476         unsigned int            sequence;
 477         int                     blocktype;
 478         int                     tag_bytes = journal_tag_bytes(journal);
 479         __u32                   crc32_sum = ~0; /* Transactional Checksums */
 480         int                     descr_csum_size = 0;
 481         int                     block_error = 0;
 482         bool                    need_check_commit_time = false;
 483         __u64                   last_trans_commit_time = 0, commit_time;
 484
 485         /*
 486          * First thing is to establish what we expect to find in the log
 487          * (in terms of transaction IDs), and where (in terms of log
 488          * block offsets): query the superblock.
 489          */
 490
 491         sb = journal->j_superblock;
 492         next_commit_ID = be32_to_cpu(sb->s_sequence);
 493         next_log_block = be32_to_cpu(sb->s_start);
 494         head_block = next_log_block;
 495
 496         first_commit_ID = next_commit_ID;
 497         if (pass == PASS_SCAN)
 498                 info->start_transaction = first_commit_ID;
 499
 500         jbd2_debug(1, "Starting recovery pass %d\n", pass);
 501
 502         /*
 503          * Now we walk through the log, transaction by transaction,
 504          * making sure that each transaction has a commit block in the
 505          * expected place.  Each complete transaction gets replayed back
 506          * into the main filesystem.
 507          */
 508
 509         while (1) {
 510                 int                     flags;
 511                 char *                  tagp;
 512                 journal_block_tag_t     tag;
 513                 struct buffer_head *    obh;
 514                 struct buffer_head *    nbh;
 515
 516                 cond_resched();
 517
 518                 /* If we already know where to stop the log traversal,
 519                  * check right now that we haven't gone past the end of
 520                  * the log. */
 521
 522                 if (pass != PASS_SCAN)
 523                         if (tid_geq(next_commit_ID, info->end_transaction))
 524                                 break;
 525
 526                 jbd2_debug(2, "Scanning for sequence ID %u at %lu/%lu\n",
 527                           next_commit_ID, next_log_block,
 528                           jbd2_has_feature_fast_commit(journal) ?
 529                           journal->j_fc_last : journal->j_last);
 530
 531                 /* Skip over each chunk of the transaction looking
 532                  * either the next descriptor block or the final commit
 533                  * record. */
 534
 535                 jbd2_debug(3, "JBD2: checking block %ld\n", next_log_block);
 536                 err = jread(&bh, journal, next_log_block);
 537                 if (err)
 538                         goto failed;
 539
 540                 next_log_block++;
 541                 wrap(journal, next_log_block);
 542
 543                 /* What kind of buffer is it?
 544                  *
 545                  * If it is a descriptor block, check that it has the
 546                  * expected sequence number.  Otherwise, we're all done
 547                  * here. */
 548
 549                 tmp = (journal_header_t *)bh->b_data;
 550
 551                 if (tmp->h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER)) {
 552                         brelse(bh);
 553                         break;
 554                 }
 555
 556                 blocktype = be32_to_cpu(tmp->h_blocktype);
 557                 sequence = be32_to_cpu(tmp->h_sequence);
 558                 jbd2_debug(3, "Found magic %d, sequence %d\n",
 559                           blocktype, sequence);
 560
 561                 if (sequence != next_commit_ID) {
 562                         brelse(bh);
 563                         break;
 564                 }
 565
 566                 /* OK, we have a valid descriptor block which matches
 567                  * all of the sequence number checks.  What are we going
 568                  * to do with it?  That depends on the pass... */
 569
 570                 switch(blocktype) {
 571                 case JBD2_DESCRIPTOR_BLOCK:
 572                         /* Verify checksum first */
 573                         if (jbd2_journal_has_csum_v2or3(journal))
 574                                 descr_csum_size =
 575                                         sizeof(struct jbd2_journal_block_tail);
 576                         if (descr_csum_size > 0 &&
 577                             !jbd2_descriptor_block_csum_verify(journal,
 578                                                                bh->b_data)) {
 579                                 /*
 580                                  * PASS_SCAN can see stale blocks due to lazy
 581                                  * journal init. Don't error out on those yet.
 582                                  */
 583                                 if (pass != PASS_SCAN) {
 584                                         pr_err("JBD2: Invalid checksum recovering block %lu in log\n",
 585                                                next_log_block);
 586                                         err = -EFSBADCRC;
 587                                         brelse(bh);
 588                                         goto failed;
 589                                 }
 590                                 need_check_commit_time = true;
 591                                 jbd2_debug(1,
 592                                         "invalid descriptor block found in %lu\n",
 593                                         next_log_block);
 594                         }
 595
 596                         /* If it is a valid descriptor block, replay it
 597                          * in pass REPLAY; if journal_checksums enabled, then
 598                          * calculate checksums in PASS_SCAN, otherwise,
 599                          * just skip over the blocks it describes. */
 600                         if (pass != PASS_REPLAY) {
 601                                 if (pass == PASS_SCAN &&
 602                                     jbd2_has_feature_checksum(journal) &&
 603                                     !need_check_commit_time &&
 604                                     !info->end_transaction) {
 605                                         if (calc_chksums(journal, bh,
 606                                                         &next_log_block,
 607                                                         &crc32_sum)) {
 608                                                 put_bh(bh);
 609                                                 break;
 610                                         }
 611                                         put_bh(bh);
 612                                         continue;
 613                                 }
 614                                 next_log_block += count_tags(journal, bh);
 615                                 wrap(journal, next_log_block);
 616                                 put_bh(bh);
 617                                 continue;
 618                         }
 619
 620                         /* A descriptor block: we can now write all of
 621                          * the data blocks.  Yay, useful work is finally
 622                          * getting done here! */
 623
 624                         tagp = &bh->b_data[sizeof(journal_header_t)];
 625                         while ((tagp - bh->b_data + tag_bytes)
 626                                <= journal->j_blocksize - descr_csum_size) {
 627                                 unsigned long io_block;
 628
 629                                 memcpy(&tag, tagp, sizeof(tag));
 630                                 flags = be16_to_cpu(tag.t_flags);
 631
 632                                 io_block = next_log_block++;
 633                                 wrap(journal, next_log_block);
 634                                 err = jread(&obh, journal, io_block);
 635                                 if (err) {
 636                                         /* Recover what we can, but
 637                                          * report failure at the end. */
 638                                         success = err;
 639                                         printk(KERN_ERR
 640                                                 "JBD2: IO error %d recovering "
 641                                                 "block %ld in log\n",
 642                                                 err, io_block);
 643                                 } else {
 644                                         unsigned long long blocknr;
 645
 646                                         J_ASSERT(obh != NULL);
 647                                         blocknr = read_tag_block(journal,
 648                                                                  &tag);
 649
 650                                         /* If the block has been
 651                                          * revoked, then we're all done
 652                                          * here. */
 653                                         if (jbd2_journal_test_revoke
 654                                             (journal, blocknr,
 655                                              next_commit_ID)) {
 656                                                 brelse(obh);
 657                                                 ++info->nr_revoke_hits;
 658                                                 goto skip_write;
 659                                         }
 660
 661                                         /* Look for block corruption */
 662                                         if (!jbd2_block_tag_csum_verify(
 663                         journal, &tag, (journal_block_tag3_t *)tagp,
 664                         obh->b_data, be32_to_cpu(tmp->h_sequence))) {
 665                                                 brelse(obh);
 666                                                 success = -EFSBADCRC;
 667                                                 printk(KERN_ERR "JBD2: Invalid "
 668                                                        "checksum recovering "
 669                                                        "data block %llu in "
 670                                                        "log\n", blocknr);
 671                                                 block_error = 1;
 672                                                 goto skip_write;
 673                                         }
 674
 675                                         /* Find a buffer for the new
 676                                          * data being restored */
 677                                         nbh = __getblk(journal->j_fs_dev,
 678                                                         blocknr,
 679                                                         journal->j_blocksize);
 680                                         if (nbh == NULL) {
 681                                                 printk(KERN_ERR
 682                                                        "JBD2: Out of memory "
 683                                                        "during recovery.\n");
 684                                                 err = -ENOMEM;
 685                                                 brelse(bh);
 686                                                 brelse(obh);
 687                                                 goto failed;
 688                                         }
 689
 690                                         lock_buffer(nbh);
 691                                         memcpy(nbh->b_data, obh->b_data,
 692                                                         journal->j_blocksize);
 693                                         if (flags & JBD2_FLAG_ESCAPE) {
 694                                                 *((__be32 *)nbh->b_data) =
 695                                                 cpu_to_be32(JBD2_MAGIC_NUMBER);
 696                                         }
 697
 698                                         BUFFER_TRACE(nbh, "marking dirty");
 699                                         set_buffer_uptodate(nbh);
 700                                         mark_buffer_dirty(nbh);
 701                                         BUFFER_TRACE(nbh, "marking uptodate");
 702                                         ++info->nr_replays;
 703                                         unlock_buffer(nbh);
 704                                         brelse(obh);
 705                                         brelse(nbh);
 706                                 }
 707
 708                         skip_write:
 709                                 tagp += tag_bytes;
 710                                 if (!(flags & JBD2_FLAG_SAME_UUID))
 711                                         tagp += 16;
 712
 713                                 if (flags & JBD2_FLAG_LAST_TAG)
 714                                         break;
 715                         }
 716
 717                         brelse(bh);
 718                         continue;
 719
 720                 case JBD2_COMMIT_BLOCK:
 721                         /*     How to differentiate between interrupted commit
 722                          *               and journal corruption ?
 723                          *
 724                          * {nth transaction}
 725                          *        Checksum Verification Failed
 726                          *                       |
 727                          *               ____________________
 728                          *              |                    |
 729                          *      async_commit             sync_commit
 730                          *              |                    |
 731                          *              | GO TO NEXT    "Journal Corruption"
 732                          *              | TRANSACTION
 733                          *              |
 734                          * {(n+1)th transanction}
 735                          *              |
 736                          *       _______|______________
 737                          *      |                     |
 738                          * Commit block found   Commit block not found
 739                          *      |                     |
 740                          * "Journal Corruption"       |
 741                          *               _____________|_________
 742                          *              |                       |
 743                          *      nth trans corrupt       OR   nth trans
 744                          *      and (n+1)th interrupted     interrupted
 745                          *      before commit block
 746                          *      could reach the disk.
 747                          *      (Cannot find the difference in above
 748                          *       mentioned conditions. Hence assume
 749                          *       "Interrupted Commit".)
 750                          */
 751                         commit_time = be64_to_cpu(
 752                                 ((struct commit_header *)bh->b_data)->h_commit_sec);
 753                         /*
 754                          * If need_check_commit_time is set, it means we are in
 755                          * PASS_SCAN and csum verify failed before. If
 756                          * commit_time is increasing, it's the same journal,
 757                          * otherwise it is stale journal block, just end this
 758                          * recovery.
 759                          */
 760                         if (need_check_commit_time) {
 761                                 if (commit_time >= last_trans_commit_time) {
 762                                         pr_err("JBD2: Invalid checksum found in transaction %u\n",
 763                                                next_commit_ID);
 764                                         err = -EFSBADCRC;
 765                                         brelse(bh);
 766                                         goto failed;
 767                                 }
 768                         ignore_crc_mismatch:
 769                                 /*
 770                                  * It likely does not belong to same journal,
 771                                  * just end this recovery with success.
 772                                  */
 773                                 jbd2_debug(1, "JBD2: Invalid checksum ignored in transaction %u, likely stale data\n",
 774                                           next_commit_ID);
 775                                 brelse(bh);
 776                                 goto done;
 777                         }
 778
 779                         /*
 780                          * Found an expected commit block: if checksums
 781                          * are present, verify them in PASS_SCAN; else not
 782                          * much to do other than move on to the next sequence
 783                          * number.
 784                          */
 785                         if (pass == PASS_SCAN &&
 786                             jbd2_has_feature_checksum(journal)) {
 787                                 struct commit_header *cbh =
 788                                         (struct commit_header *)bh->b_data;
 789                                 unsigned found_chksum =
 790                                         be32_to_cpu(cbh->h_chksum[0]);
 791
 792                                 if (info->end_transaction) {
 793                                         journal->j_failed_commit =
 794                                                 info->end_transaction;
 795                                         brelse(bh);
 796                                         break;
 797                                 }
 798
 799                                 /* Neither checksum match nor unused? */
 800                                 if (!((crc32_sum == found_chksum &&
 801                                        cbh->h_chksum_type ==
 802                                                 JBD2_CRC32_CHKSUM &&
 803                                        cbh->h_chksum_size ==
 804                                                 JBD2_CRC32_CHKSUM_SIZE) ||
 805                                       (cbh->h_chksum_type == 0 &&
 806                                        cbh->h_chksum_size == 0 &&
 807                                        found_chksum == 0)))
 808                                         goto chksum_error;
 809
 810                                 crc32_sum = ~0;
 811                         }
 812                         if (pass == PASS_SCAN &&
 813                             !jbd2_commit_block_csum_verify(journal,
 814                                                            bh->b_data)) {
 815                         chksum_error:
 816                                 if (commit_time < last_trans_commit_time)
 817                                         goto ignore_crc_mismatch;
 818                                 info->end_transaction = next_commit_ID;
 819                                 info->head_block = head_block;
 820
 821                                 if (!jbd2_has_feature_async_commit(journal)) {
 822                                         journal->j_failed_commit =
 823                                                 next_commit_ID;
 824                                         brelse(bh);
 825                                         break;
 826                                 }
 827                         }
 828                         if (pass == PASS_SCAN) {
 829                                 last_trans_commit_time = commit_time;
 830                                 head_block = next_log_block;
 831                         }
 832                         brelse(bh);
 833                         next_commit_ID++;
 834                         continue;
 835
 836                 case JBD2_REVOKE_BLOCK:
 837                         /*
 838                          * Check revoke block crc in pass_scan, if csum verify
 839                          * failed, check commit block time later.
 840                          */
 841                         if (pass == PASS_SCAN &&
 842                             !jbd2_descriptor_block_csum_verify(journal,
 843                                                                bh->b_data)) {
 844                                 jbd2_debug(1, "JBD2: invalid revoke block found in %lu\n",
 845                                           next_log_block);
 846                                 need_check_commit_time = true;
 847                         }
 848                         /* If we aren't in the REVOKE pass, then we can
 849                          * just skip over this block. */
 850                         if (pass != PASS_REVOKE) {
 851                                 brelse(bh);
 852                                 continue;
 853                         }
 854
 855                         err = scan_revoke_records(journal, bh,
 856                                                   next_commit_ID, info);
 857                         brelse(bh);
 858                         if (err)
 859                                 goto failed;
 860                         continue;
 861
 862                 default:
 863                         jbd2_debug(3, "Unrecognised magic %d, end of scan.\n",
 864                                   blocktype);
 865                         brelse(bh);
 866                         goto done;
 867                 }
 868         }
 869
 870  done:
 871         /*
 872          * We broke out of the log scan loop: either we came to the
 873          * known end of the log or we found an unexpected block in the
 874          * log.  If the latter happened, then we know that the "current"
 875          * transaction marks the end of the valid log.
 876          */
 877
 878         if (pass == PASS_SCAN) {
 879                 if (!info->end_transaction)
 880                         info->end_transaction = next_commit_ID;
 881                 if (!info->head_block)
 882                         info->head_block = head_block;
 883         } else {
 884                 /* It's really bad news if different passes end up at
 885                  * different places (but possible due to IO errors). */
 886                 if (info->end_transaction != next_commit_ID) {
 887                         printk(KERN_ERR "JBD2: recovery pass %d ended at "
 888                                 "transaction %u, expected %u\n",
 889                                 pass, next_commit_ID, info->end_transaction);
 890                         if (!success)
 891                                 success = -EIO;
 892                 }
 893         }
 894
 895         if (jbd2_has_feature_fast_commit(journal) &&  pass != PASS_REVOKE) {
 896                 err = fc_do_one_pass(journal, info, pass);
 897                 if (err)
 898                         success = err;
 899         }
 900
 901         if (block_error && success == 0)
 902                 success = -EIO;
 903         return success;
 904
 905  failed:
 906         return err;
 907 }
 908
 909 /* Scan a revoke record, marking all blocks mentioned as revoked. */
 910
 911 static int scan_revoke_records(journal_t *journal, struct buffer_head *bh,
 912                                tid_t sequence, struct recovery_info *info)
 913 {
 914         jbd2_journal_revoke_header_t *header;
 915         int offset, max;
 916         unsigned csum_size = 0;
 917         __u32 rcount;
 918         int record_len = 4;
 919
 920         header = (jbd2_journal_revoke_header_t *) bh->b_data;
 921         offset = sizeof(jbd2_journal_revoke_header_t);
 922         rcount = be32_to_cpu(header->r_count);
 923
 924         if (jbd2_journal_has_csum_v2or3(journal))
 925                 csum_size = sizeof(struct jbd2_journal_block_tail);
 926         if (rcount > journal->j_blocksize - csum_size)
 927                 return -EINVAL;
 928         max = rcount;
 929
 930         if (jbd2_has_feature_64bit(journal))
 931                 record_len = 8;
 932
 933         while (offset + record_len <= max) {
 934                 unsigned long long blocknr;
 935                 int err;
 936
 937                 if (record_len == 4)
 938                         blocknr = be32_to_cpu(* ((__be32 *) (bh->b_data+offset)));
 939                 else
 940                         blocknr = be64_to_cpu(* ((__be64 *) (bh->b_data+offset)));
 941                 offset += record_len;
 942                 err = jbd2_journal_set_revoke(journal, blocknr, sequence);
 943                 if (err)
 944                         return err;
 945                 ++info->nr_revokes;
 946         }
 947         return 0;
 948 }