e2fsck/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
  33         int             nr_replays;
  34         int             nr_revokes;
  35         int             nr_revoke_hits;
  36 };
  37
  38 enum passtype {PASS_SCAN, PASS_REVOKE, PASS_REPLAY};
  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_maxlen)
  79                 max = journal->j_maxlen;
  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                                 ll_rw_block(REQ_OP_READ, 0, nbufs, bufs);
 105                                 journal_brelse_array(bufs, nbufs);
 106                                 nbufs = 0;
 107                         }
 108                 } else
 109                         brelse(bh);
 110         }
 111
 112         if (nbufs)
 113                 ll_rw_block(REQ_OP_READ, 0, nbufs, bufs);
 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 static inline __u32 get_be32(__be32 *p)
 125 {
 126         unsigned char *cp = (unsigned char *) p;
 127         __u32 ret;
 128
 129         ret = *cp++;
 130         ret = (ret << 8) + *cp++;
 131         ret = (ret << 8) + *cp++;
 132         ret = (ret << 8) + *cp++;
 133         return ret;
 134 }
 135
 136 static inline __u16 get_be16(__be16 *p)
 137 {
 138         unsigned char *cp = (unsigned char *) p;
 139         __u16 ret;
 140
 141         ret = *cp++;
 142         ret = (ret << 8) + *cp++;
 143         return ret;
 144 }
 145
 146 /*
 147  * Read a block from the journal
 148  */
 149
 150 static int jread(struct buffer_head **bhp, journal_t *journal,
 151                  unsigned int offset)
 152 {
 153         int err;
 154         unsigned long long blocknr;
 155         struct buffer_head *bh;
 156
 157         *bhp = NULL;
 158
 159         if (offset >= journal->j_maxlen) {
 160                 printk(KERN_ERR "JBD2: corrupted journal superblock\n");
 161                 return -EFSCORRUPTED;
 162         }
 163
 164         err = jbd2_journal_bmap(journal, offset, &blocknr);
 165
 166         if (err) {
 167                 printk(KERN_ERR "JBD2: bad block at offset %u\n",
 168                         offset);
 169                 return err;
 170         }
 171
 172         bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
 173         if (!bh)
 174                 return -ENOMEM;
 175
 176         if (!buffer_uptodate(bh)) {
 177                 /* If this is a brand new buffer, start readahead.
 178                    Otherwise, we assume we are already reading it.  */
 179                 if (!buffer_req(bh))
 180                         do_readahead(journal, offset);
 181                 wait_on_buffer(bh);
 182         }
 183
 184         if (!buffer_uptodate(bh)) {
 185                 printk(KERN_ERR "JBD2: Failed to read block at offset %u\n",
 186                         offset);
 187                 brelse(bh);
 188                 return -EIO;
 189         }
 190
 191         *bhp = bh;
 192         return 0;
 193 }
 194
 195 static int jbd2_descriptor_block_csum_verify(journal_t *j, void *buf)
 196 {
 197         struct jbd2_journal_block_tail *tail;
 198         __be32 provided;
 199         __u32 calculated;
 200
 201         if (!jbd2_journal_has_csum_v2or3(j))
 202                 return 1;
 203
 204         tail = (struct jbd2_journal_block_tail *)((char *)buf + j->j_blocksize -
 205                         sizeof(struct jbd2_journal_block_tail));
 206         provided = tail->t_checksum;
 207         tail->t_checksum = 0;
 208         calculated = jbd2_chksum(j, j->j_csum_seed, buf, j->j_blocksize);
 209         tail->t_checksum = provided;
 210
 211         return provided == cpu_to_be32(calculated);
 212 }
 213
 214 /*
 215  * Count the number of in-use tags in a journal descriptor block.
 216  */
 217
 218 static int count_tags(journal_t *journal, struct buffer_head *bh)
 219 {
 220         char *                  tagp;
 221         journal_block_tag_t *   tag;
 222         int                     nr = 0, size = journal->j_blocksize;
 223         int                     tag_bytes = journal_tag_bytes(journal);
 224
 225         if (jbd2_journal_has_csum_v2or3(journal))
 226                 size -= sizeof(struct jbd2_journal_block_tail);
 227
 228         tagp = &bh->b_data[sizeof(journal_header_t)];
 229
 230         while ((tagp - bh->b_data + tag_bytes) <= size) {
 231                 tag = (journal_block_tag_t *) tagp;
 232
 233                 nr++;
 234                 tagp += tag_bytes;
 235                 if (!(get_be16(&tag->t_flags) & JBD2_FLAG_SAME_UUID))
 236                         tagp += 16;
 237
 238                 if (get_be16(&tag->t_flags) & JBD2_FLAG_LAST_TAG)
 239                         break;
 240         }
 241
 242         return nr;
 243 }
 244
 245
 246 /* Make sure we wrap around the log correctly! */
 247 #define wrap(journal, var)                                              \
 248 do {                                                                    \
 249         if (var >= (journal)->j_last)                                   \
 250                 var -= ((journal)->j_last - (journal)->j_first);        \
 251 } while (0)
 252
 253 /**
 254  * jbd2_journal_recover - recovers a on-disk journal
 255  * @journal: the journal to recover
 256  *
 257  * The primary function for recovering the log contents when mounting a
 258  * journaled device.
 259  *
 260  * Recovery is done in three passes.  In the first pass, we look for the
 261  * end of the log.  In the second, we assemble the list of revoke
 262  * blocks.  In the third and final pass, we replay any un-revoked blocks
 263  * in the log.
 264  */
 265 int jbd2_journal_recover(journal_t *journal)
 266 {
 267         int                     err, err2;
 268         journal_superblock_t *  sb;
 269
 270         struct recovery_info    info;
 271
 272         memset(&info, 0, sizeof(info));
 273         sb = journal->j_superblock;
 274
 275         /*
 276          * The journal superblock's s_start field (the current log head)
 277          * is always zero if, and only if, the journal was cleanly
 278          * unmounted.
 279          */
 280
 281         if (!sb->s_start) {
 282                 jbd_debug(1, "No recovery required, last transaction %d\n",
 283                           be32_to_cpu(sb->s_sequence));
 284                 journal->j_transaction_sequence = be32_to_cpu(sb->s_sequence) + 1;
 285                 return 0;
 286         }
 287
 288         err = do_one_pass(journal, &info, PASS_SCAN);
 289         if (!err)
 290                 err = do_one_pass(journal, &info, PASS_REVOKE);
 291         if (!err)
 292                 err = do_one_pass(journal, &info, PASS_REPLAY);
 293
 294         jbd_debug(1, "JBD2: recovery, exit status %d, "
 295                   "recovered transactions %u to %u\n",
 296                   err, info.start_transaction, info.end_transaction);
 297         jbd_debug(1, "JBD2: Replayed %d and revoked %d/%d blocks\n",
 298                   info.nr_replays, info.nr_revoke_hits, info.nr_revokes);
 299
 300         /* Restart the log at the next transaction ID, thus invalidating
 301          * any existing commit records in the log. */
 302         journal->j_transaction_sequence = ++info.end_transaction;
 303
 304         jbd2_journal_clear_revoke(journal);
 305         err2 = sync_blockdev(journal->j_fs_dev);
 306         if (!err)
 307                 err = err2;
 308         /* Make sure all replayed data is on permanent storage */
 309         if (journal->j_flags & JBD2_BARRIER) {
 310                 err2 = blkdev_issue_flush(journal->j_fs_dev, GFP_KERNEL, NULL);
 311                 if (!err)
 312                         err = err2;
 313         }
 314         return err;
 315 }
 316
 317 /**
 318  * jbd2_journal_skip_recovery - Start journal and wipe exiting records
 319  * @journal: journal to startup
 320  *
 321  * Locate any valid recovery information from the journal and set up the
 322  * journal structures in memory to ignore it (presumably because the
 323  * caller has evidence that it is out of date).
 324  * This function doesn't appear to be exported..
 325  *
 326  * We perform one pass over the journal to allow us to tell the user how
 327  * much recovery information is being erased, and to let us initialise
 328  * the journal transaction sequence numbers to the next unused ID.
 329  */
 330 int jbd2_journal_skip_recovery(journal_t *journal)
 331 {
 332         int                     err;
 333
 334         struct recovery_info    info;
 335
 336         memset (&info, 0, sizeof(info));
 337
 338         err = do_one_pass(journal, &info, PASS_SCAN);
 339
 340         if (err) {
 341                 printk(KERN_ERR "JBD2: error %d scanning journal\n", err);
 342                 ++journal->j_transaction_sequence;
 343         } else {
 344 #ifdef CONFIG_JBD2_DEBUG
 345                 int dropped = info.end_transaction -
 346                         be32_to_cpu(journal->j_superblock->s_sequence);
 347                 jbd_debug(1,
 348                           "JBD2: ignoring %d transaction%s from the journal.\n",
 349                           dropped, (dropped == 1) ? "" : "s");
 350 #endif
 351                 journal->j_transaction_sequence = ++info.end_transaction;
 352         }
 353
 354         journal->j_tail = 0;
 355         return err;
 356 }
 357
 358 static inline unsigned long long read_tag_block(journal_t *journal,
 359                                                 journal_block_tag_t *tag)
 360 {
 361         unsigned long long block = get_be32(&tag->t_blocknr);
 362         if (jbd2_has_feature_64bit(journal))
 363                 block |= (u64)get_be32(&tag->t_blocknr_high) << 32;
 364         return block;
 365 }
 366
 367 /*
 368  * calc_chksums calculates the checksums for the blocks described in the
 369  * descriptor block.
 370  */
 371 static int calc_chksums(journal_t *journal, struct buffer_head *bh,
 372                         unsigned long *next_log_block, __u32 *crc32_sum)
 373 {
 374         int i, num_blks, err;
 375         unsigned long io_block;
 376         struct buffer_head *obh;
 377
 378         num_blks = count_tags(journal, bh);
 379         /* Calculate checksum of the descriptor block. */
 380         *crc32_sum = crc32_be(*crc32_sum, (void *)bh->b_data, bh->b_size);
 381
 382         for (i = 0; i < num_blks; i++) {
 383                 io_block = (*next_log_block)++;
 384                 wrap(journal, *next_log_block);
 385                 err = jread(&obh, journal, io_block);
 386                 if (err) {
 387                         printk(KERN_ERR "JBD2: IO error %d recovering block "
 388                                 "%lu in log\n", err, io_block);
 389                         return 1;
 390                 } else {
 391                         *crc32_sum = crc32_be(*crc32_sum, (void *)obh->b_data,
 392                                      obh->b_size);
 393                 }
 394                 put_bh(obh);
 395         }
 396         return 0;
 397 }
 398
 399 static int jbd2_commit_block_csum_verify(journal_t *j, void *buf)
 400 {
 401         struct commit_header *h;
 402         __be32 provided;
 403         __u32 calculated;
 404
 405         if (!jbd2_journal_has_csum_v2or3(j))
 406                 return 1;
 407
 408         h = buf;
 409         provided = h->h_chksum[0];
 410         h->h_chksum[0] = 0;
 411         calculated = jbd2_chksum(j, j->j_csum_seed, buf, j->j_blocksize);
 412         h->h_chksum[0] = provided;
 413
 414         return provided == cpu_to_be32(calculated);
 415 }
 416
 417 static int jbd2_block_tag_csum_verify(journal_t *j, journal_block_tag_t *tag,
 418                                       void *buf, __u32 sequence)
 419 {
 420         journal_block_tag3_t *tag3 = (journal_block_tag3_t *)tag;
 421         __u32 csum32;
 422         __be32 seq;
 423
 424         if (!jbd2_journal_has_csum_v2or3(j))
 425                 return 1;
 426
 427         seq = cpu_to_be32(sequence);
 428         csum32 = jbd2_chksum(j, j->j_csum_seed, (__u8 *)&seq, sizeof(seq));
 429         csum32 = jbd2_chksum(j, csum32, buf, j->j_blocksize);
 430
 431         if (jbd2_has_feature_csum3(j))
 432                 return get_be32(&tag3->t_checksum) == csum32;
 433
 434         return get_be16(&tag->t_checksum) == (csum32 & 0xFFFF);
 435 }
 436
 437 static int do_one_pass(journal_t *journal,
 438                         struct recovery_info *info, enum passtype pass)
 439 {
 440         unsigned int            first_commit_ID, next_commit_ID;
 441         unsigned long           next_log_block;
 442         int                     err, success = 0;
 443         journal_superblock_t *  sb;
 444         journal_header_t *      tmp;
 445         struct buffer_head *    bh;
 446         unsigned int            sequence;
 447         int                     blocktype;
 448         int                     tag_bytes = journal_tag_bytes(journal);
 449         __u32                   crc32_sum = ~0; /* Transactional Checksums */
 450         int                     descr_csum_size = 0;
 451         int                     block_error = 0;
 452
 453         /*
 454          * First thing is to establish what we expect to find in the log
 455          * (in terms of transaction IDs), and where (in terms of log
 456          * block offsets): query the superblock.
 457          */
 458
 459         sb = journal->j_superblock;
 460         next_commit_ID = be32_to_cpu(sb->s_sequence);
 461         next_log_block = be32_to_cpu(sb->s_start);
 462
 463         first_commit_ID = next_commit_ID;
 464         if (pass == PASS_SCAN)
 465                 info->start_transaction = first_commit_ID;
 466
 467         jbd_debug(1, "Starting recovery pass %d\n", pass);
 468
 469         /*
 470          * Now we walk through the log, transaction by transaction,
 471          * making sure that each transaction has a commit block in the
 472          * expected place.  Each complete transaction gets replayed back
 473          * into the main filesystem.
 474          */
 475
 476         while (1) {
 477                 int                     flags;
 478                 char *                  tagp;
 479                 journal_block_tag_t *   tag;
 480                 struct buffer_head *    obh;
 481                 struct buffer_head *    nbh;
 482
 483                 cond_resched();
 484
 485                 /* If we already know where to stop the log traversal,
 486                  * check right now that we haven't gone past the end of
 487                  * the log. */
 488
 489                 if (pass != PASS_SCAN)
 490                         if (tid_geq(next_commit_ID, info->end_transaction))
 491                                 break;
 492
 493                 jbd_debug(2, "Scanning for sequence ID %u at %lu/%lu\n",
 494                           next_commit_ID, next_log_block, journal->j_last);
 495
 496                 /* Skip over each chunk of the transaction looking
 497                  * either the next descriptor block or the final commit
 498                  * record. */
 499
 500                 jbd_debug(3, "JBD2: checking block %ld\n", next_log_block);
 501                 err = jread(&bh, journal, next_log_block);
 502                 if (err)
 503                         goto failed;
 504
 505                 next_log_block++;
 506                 wrap(journal, next_log_block);
 507
 508                 /* What kind of buffer is it?
 509                  *
 510                  * If it is a descriptor block, check that it has the
 511                  * expected sequence number.  Otherwise, we're all done
 512                  * here. */
 513
 514                 tmp = (journal_header_t *)bh->b_data;
 515
 516                 if (tmp->h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER)) {
 517                         brelse(bh);
 518                         break;
 519                 }
 520
 521                 blocktype = be32_to_cpu(tmp->h_blocktype);
 522                 sequence = be32_to_cpu(tmp->h_sequence);
 523                 jbd_debug(3, "Found magic %d, sequence %d\n",
 524                           blocktype, sequence);
 525
 526                 if (sequence != next_commit_ID) {
 527                         brelse(bh);
 528                         break;
 529                 }
 530
 531                 /* OK, we have a valid descriptor block which matches
 532                  * all of the sequence number checks.  What are we going
 533                  * to do with it?  That depends on the pass... */
 534
 535                 switch(blocktype) {
 536                 case JBD2_DESCRIPTOR_BLOCK:
 537                         /* Verify checksum first */
 538                         if (jbd2_journal_has_csum_v2or3(journal))
 539                                 descr_csum_size =
 540                                         sizeof(struct jbd2_journal_block_tail);
 541                         if (descr_csum_size > 0 &&
 542                             !jbd2_descriptor_block_csum_verify(journal,
 543                                                                bh->b_data)) {
 544                                 err = -EFSBADCRC;
 545                                 brelse(bh);
 546                                 goto failed;
 547                         }
 548
 549                         /* If it is a valid descriptor block, replay it
 550                          * in pass REPLAY; if journal_checksums enabled, then
 551                          * calculate checksums in PASS_SCAN, otherwise,
 552                          * just skip over the blocks it describes. */
 553                         if (pass != PASS_REPLAY) {
 554                                 if (pass == PASS_SCAN &&
 555                                     jbd2_has_feature_checksum(journal) &&
 556                                     !info->end_transaction) {
 557                                         if (calc_chksums(journal, bh,
 558                                                         &next_log_block,
 559                                                         &crc32_sum)) {
 560                                                 put_bh(bh);
 561                                                 break;
 562                                         }
 563                                         put_bh(bh);
 564                                         continue;
 565                                 }
 566                                 next_log_block += count_tags(journal, bh);
 567                                 wrap(journal, next_log_block);
 568                                 put_bh(bh);
 569                                 continue;
 570                         }
 571
 572                         /* A descriptor block: we can now write all of
 573                          * the data blocks.  Yay, useful work is finally
 574                          * getting done here! */
 575
 576                         tagp = &bh->b_data[sizeof(journal_header_t)];
 577                         while ((tagp - bh->b_data + tag_bytes)
 578                                <= journal->j_blocksize - descr_csum_size) {
 579                                 unsigned long io_block;
 580
 581                                 tag = (journal_block_tag_t *) tagp;
 582                                 flags = get_be16(&tag->t_flags);
 583
 584                                 io_block = next_log_block++;
 585                                 wrap(journal, next_log_block);
 586                                 err = jread(&obh, journal, io_block);
 587                                 if (err) {
 588                                         /* Recover what we can, but
 589                                          * report failure at the end. */
 590                                         success = err;
 591                                         printk(KERN_ERR
 592                                                 "JBD2: IO error %d recovering "
 593                                                 "block %ld in log\n",
 594                                                 err, io_block);
 595                                 } else {
 596                                         unsigned long long blocknr;
 597
 598                                         J_ASSERT(obh != NULL);
 599                                         blocknr = read_tag_block(journal,
 600                                                                  tag);
 601
 602                                         /* If the block has been
 603                                          * revoked, then we're all done
 604                                          * here. */
 605                                         if (jbd2_journal_test_revoke
 606                                             (journal, blocknr,
 607                                              next_commit_ID)) {
 608                                                 brelse(obh);
 609                                                 ++info->nr_revoke_hits;
 610                                                 goto skip_write;
 611                                         }
 612
 613                                         /* Look for block corruption */
 614                                         if (!jbd2_block_tag_csum_verify(
 615                                                 journal, tag, obh->b_data,
 616                                                 be32_to_cpu(tmp->h_sequence))) {
 617                                                 brelse(obh);
 618                                                 success = -EFSBADCRC;
 619                                                 printk(KERN_ERR "JBD2: Invalid "
 620                                                        "checksum recovering "
 621                                                        "data block %llu in "
 622                                                        "log\n", blocknr);
 623                                                 block_error = 1;
 624                                                 goto skip_write;
 625                                         }
 626
 627                                         /* Find a buffer for the new
 628                                          * data being restored */
 629                                         nbh = __getblk(journal->j_fs_dev,
 630                                                         blocknr,
 631                                                         journal->j_blocksize);
 632                                         if (nbh == NULL) {
 633                                                 printk(KERN_ERR
 634                                                        "JBD2: Out of memory "
 635                                                        "during recovery.\n");
 636                                                 err = -ENOMEM;
 637                                                 brelse(bh);
 638                                                 brelse(obh);
 639                                                 goto failed;
 640                                         }
 641
 642                                         lock_buffer(nbh);
 643                                         memcpy(nbh->b_data, obh->b_data,
 644                                                         journal->j_blocksize);
 645                                         if (flags & JBD2_FLAG_ESCAPE) {
 646                                                 __be32 magic = cpu_to_be32(JBD2_MAGIC_NUMBER);
 647                                                 memcpy(nbh->b_data, &magic,
 648                                                        sizeof(magic));
 649                                         }
 650
 651                                         BUFFER_TRACE(nbh, "marking dirty");
 652                                         set_buffer_uptodate(nbh);
 653                                         mark_buffer_dirty(nbh);
 654                                         BUFFER_TRACE(nbh, "marking uptodate");
 655                                         ++info->nr_replays;
 656                                         /* ll_rw_block(WRITE, 1, &nbh); */
 657                                         unlock_buffer(nbh);
 658                                         brelse(obh);
 659                                         brelse(nbh);
 660                                 }
 661
 662                         skip_write:
 663                                 tagp += tag_bytes;
 664                                 if (!(flags & JBD2_FLAG_SAME_UUID))
 665                                         tagp += 16;
 666
 667                                 if (flags & JBD2_FLAG_LAST_TAG)
 668                                         break;
 669                         }
 670
 671                         brelse(bh);
 672                         continue;
 673
 674                 case JBD2_COMMIT_BLOCK:
 675                         /*     How to differentiate between interrupted commit
 676                          *               and journal corruption ?
 677                          *
 678                          * {nth transaction}
 679                          *        Checksum Verification Failed
 680                          *                       |
 681                          *               ____________________
 682                          *              |                    |
 683                          *      async_commit             sync_commit
 684                          *              |                    |
 685                          *              | GO TO NEXT    "Journal Corruption"
 686                          *              | TRANSACTION
 687                          *              |
 688                          * {(n+1)th transaction}
 689                          *              |
 690                          *       _______|______________
 691                          *      |                     |
 692                          * Commit block found   Commit block not found
 693                          *      |                     |
 694                          * "Journal Corruption"       |
 695                          *               _____________|_________
 696                          *              |                       |
 697                          *      nth trans corrupt       OR   nth trans
 698                          *      and (n+1)th interrupted     interrupted
 699                          *      before commit block
 700                          *      could reach the disk.
 701                          *      (Cannot find the difference in above
 702                          *       mentioned conditions. Hence assume
 703                          *       "Interrupted Commit".)
 704                          */
 705
 706                         /* Found an expected commit block: if checksums
 707                          * are present verify them in PASS_SCAN; else not
 708                          * much to do other than move on to the next sequence
 709                          * number. */
 710                         if (pass == PASS_SCAN &&
 711                             jbd2_has_feature_checksum(journal)) {
 712                                 int chksum_err, chksum_seen;
 713                                 struct commit_header *cbh =
 714                                         (struct commit_header *)bh->b_data;
 715                                 unsigned found_chksum =
 716                                         be32_to_cpu(cbh->h_chksum[0]);
 717
 718                                 chksum_err = chksum_seen = 0;
 719
 720                                 if (info->end_transaction) {
 721                                         journal->j_failed_commit =
 722                                                 info->end_transaction;
 723                                         brelse(bh);
 724                                         break;
 725                                 }
 726
 727                                 if (crc32_sum == found_chksum &&
 728                                     cbh->h_chksum_type == JBD2_CRC32_CHKSUM &&
 729                                     cbh->h_chksum_size ==
 730                                                 JBD2_CRC32_CHKSUM_SIZE)
 731                                        chksum_seen = 1;
 732                                 else if (!(cbh->h_chksum_type == 0 &&
 733                                              cbh->h_chksum_size == 0 &&
 734                                              found_chksum == 0 &&
 735                                              !chksum_seen))
 736                                 /*
 737                                  * If fs is mounted using an old kernel and then
 738                                  * kernel with journal_chksum is used then we
 739                                  * get a situation where the journal flag has
 740                                  * checksum flag set but checksums are not
 741                                  * present i.e chksum = 0, in the individual
 742                                  * commit blocks.
 743                                  * Hence to avoid checksum failures, in this
 744                                  * situation, this extra check is added.
 745                                  */
 746                                                 chksum_err = 1;
 747
 748                                 if (chksum_err) {
 749                                         info->end_transaction = next_commit_ID;
 750
 751                                         if (!jbd2_has_feature_async_commit(journal)) {
 752                                                 journal->j_failed_commit =
 753                                                         next_commit_ID;
 754                                                 brelse(bh);
 755                                                 break;
 756                                         }
 757                                 }
 758                                 crc32_sum = ~0;
 759                         }
 760                         if (pass == PASS_SCAN &&
 761                             !jbd2_commit_block_csum_verify(journal,
 762                                                            bh->b_data)) {
 763                                 info->end_transaction = next_commit_ID;
 764
 765                                 if (!jbd2_has_feature_async_commit(journal)) {
 766                                         journal->j_failed_commit =
 767                                                 next_commit_ID;
 768                                         brelse(bh);
 769                                         break;
 770                                 }
 771                         }
 772                         brelse(bh);
 773                         next_commit_ID++;
 774                         continue;
 775
 776                 case JBD2_REVOKE_BLOCK:
 777                         /* If we aren't in the REVOKE pass, then we can
 778                          * just skip over this block. */
 779                         if (pass != PASS_REVOKE) {
 780                                 brelse(bh);
 781                                 continue;
 782                         }
 783
 784                         err = scan_revoke_records(journal, bh,
 785                                                   next_commit_ID, info);
 786                         brelse(bh);
 787                         if (err)
 788                                 goto failed;
 789                         continue;
 790
 791                 default:
 792                         jbd_debug(3, "Unrecognised magic %d, end of scan.\n",
 793                                   blocktype);
 794                         brelse(bh);
 795                         goto done;
 796                 }
 797         }
 798
 799  done:
 800         /*
 801          * We broke out of the log scan loop: either we came to the
 802          * known end of the log or we found an unexpected block in the
 803          * log.  If the latter happened, then we know that the "current"
 804          * transaction marks the end of the valid log.
 805          */
 806
 807         if (pass == PASS_SCAN) {
 808                 if (!info->end_transaction)
 809                         info->end_transaction = next_commit_ID;
 810         } else {
 811                 /* It's really bad news if different passes end up at
 812                  * different places (but possible due to IO errors). */
 813                 if (info->end_transaction != next_commit_ID) {
 814                         printk(KERN_ERR "JBD2: recovery pass %d ended at "
 815                                 "transaction %u, expected %u\n",
 816                                 pass, next_commit_ID, info->end_transaction);
 817                         if (!success)
 818                                 success = -EIO;
 819                 }
 820         }
 821         if (block_error && success == 0)
 822                 success = -EIO;
 823         return success;
 824
 825  failed:
 826         return err;
 827 }
 828
 829 /* Scan a revoke record, marking all blocks mentioned as revoked. */
 830
 831 static int scan_revoke_records(journal_t *journal, struct buffer_head *bh,
 832                                tid_t sequence, struct recovery_info *info)
 833 {
 834         jbd2_journal_revoke_header_t *header;
 835         int offset, max;
 836         unsigned csum_size = 0;
 837         __u32 rcount;
 838         int record_len = 4;
 839
 840         header = (jbd2_journal_revoke_header_t *) bh->b_data;
 841         offset = sizeof(jbd2_journal_revoke_header_t);
 842         rcount = be32_to_cpu(header->r_count);
 843
 844         if (!jbd2_descriptor_block_csum_verify(journal, header))
 845                 return -EFSBADCRC;
 846
 847         if (jbd2_journal_has_csum_v2or3(journal))
 848                 csum_size = sizeof(struct jbd2_journal_block_tail);
 849         if (rcount > journal->j_blocksize - csum_size)
 850                 return -EINVAL;
 851         max = rcount;
 852
 853         if (jbd2_has_feature_64bit(journal))
 854                 record_len = 8;
 855
 856         while (offset + record_len <= max) {
 857                 unsigned long long blocknr;
 858                 int err;
 859
 860                 if (record_len == 4)
 861                         blocknr = be32_to_cpu(* ((__be32 *) (bh->b_data+offset)));
 862                 else
 863                         blocknr = be64_to_cpu(* ((__be64 *) (bh->b_data+offset)));
 864                 offset += record_len;
 865                 err = jbd2_journal_set_revoke(journal, blocknr, sequence);
 866                 if (err)
 867                         return err;
 868                 ++info->nr_revokes;
 869         }
 870         return 0;
 871 }