e2fsck/recovery.c

   1 /*
   2  * linux/fs/jbd2/recovery.c
   3  *
   4  * Written by Stephen C. Tweedie <sct@redhat.com>, 1999
   5  *
   6  * Copyright 1999-2000 Red Hat Software --- All Rights Reserved
   7  *
   8  * This file is part of the Linux kernel and is made available under
   9  * the terms of the GNU General Public License, version 2, or at your
  10  * option, any later version, incorporated herein by reference.
  11  *
  12  * Journal recovery routines for the generic filesystem journaling code;
  13  * part of the ext2fs journaling system.
  14  */
  15
  16 #ifndef __KERNEL__
  17 #include "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 = 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(READ, nbufs, bufs);
 108                                 journal_brelse_array(bufs, nbufs);
 109                                 nbufs = 0;
 110                         }
 111                 } else
 112                         brelse(bh);
 113         }
 114
 115         if (nbufs)
 116                 ll_rw_block(READ, nbufs, bufs);
 117         err = 0;
 118
 119 failed:
 120         if (nbufs)
 121                 journal_brelse_array(bufs, nbufs);
 122         return err;
 123 }
 124
 125 #endif /* __KERNEL__ */
 126
 127 static inline __u32 get_be32(__be32 *p)
 128 {
 129         unsigned char *cp = (unsigned char *) p;
 130         __u32 ret;
 131
 132         ret = *cp++;
 133         ret = (ret << 8) + *cp++;
 134         ret = (ret << 8) + *cp++;
 135         ret = (ret << 8) + *cp++;
 136         return ret;
 137 }
 138
 139 static inline __u16 get_be16(__be16 *p)
 140 {
 141         unsigned char *cp = (unsigned char *) p;
 142         __u16 ret;
 143
 144         ret = *cp++;
 145         ret = (ret << 8) + *cp++;
 146         return ret;
 147 }
 148
 149 /*
 150  * Read a block from the journal
 151  */
 152
 153 static int jread(struct buffer_head **bhp, journal_t *journal,
 154                  unsigned int offset)
 155 {
 156         int err;
 157         unsigned long long blocknr;
 158         struct buffer_head *bh;
 159
 160         *bhp = NULL;
 161
 162         if (offset >= journal->j_maxlen) {
 163                 printk(KERN_ERR "JBD2: corrupted journal superblock\n");
 164                 return -EFSCORRUPTED;
 165         }
 166
 167         err = journal_bmap(journal, offset, &blocknr);
 168
 169         if (err) {
 170                 printk(KERN_ERR "JBD2: bad block at offset %u\n",
 171                         offset);
 172                 return err;
 173         }
 174
 175         bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
 176         if (!bh)
 177                 return -ENOMEM;
 178
 179         if (!buffer_uptodate(bh)) {
 180                 /* If this is a brand new buffer, start readahead.
 181                    Otherwise, we assume we are already reading it.  */
 182                 if (!buffer_req(bh))
 183                         do_readahead(journal, offset);
 184                 wait_on_buffer(bh);
 185         }
 186
 187         if (!buffer_uptodate(bh)) {
 188                 printk(KERN_ERR "JBD2: Failed to read block at offset %u\n",
 189                         offset);
 190                 brelse(bh);
 191                 return -EIO;
 192         }
 193
 194         *bhp = bh;
 195         return 0;
 196 }
 197
 198 static int jbd2_descr_block_csum_verify(journal_t *j,
 199                                         void *buf)
 200 {
 201         struct journal_block_tail *tail;
 202         __u32 provided;
 203         __u32 calculated;
 204
 205         if (!journal_has_csum_v2or3(j))
 206                 return 1;
 207
 208         tail = (struct journal_block_tail *)((char *)buf + j->j_blocksize -
 209                         sizeof(struct journal_block_tail));
 210         provided = tail->t_checksum;
 211         tail->t_checksum = 0;
 212         calculated = jbd2_chksum(j, j->j_csum_seed, buf, j->j_blocksize);
 213         tail->t_checksum = provided;
 214
 215         return provided == ext2fs_cpu_to_be32(calculated);
 216 }
 217
 218 /*
 219  * Count the number of in-use tags in a journal descriptor block.
 220  */
 221
 222 static int count_tags(journal_t *journal, struct buffer_head *bh)
 223 {
 224         char *                  tagp;
 225         journal_block_tag_t *   tag;
 226         int                     nr = 0, size = journal->j_blocksize;
 227         int                     tag_bytes = journal_tag_bytes(journal);
 228
 229         if (journal_has_csum_v2or3(journal))
 230                 size -= sizeof(struct journal_block_tail);
 231
 232         tagp = &bh->b_data[sizeof(journal_header_t)];
 233
 234         while ((tagp - bh->b_data + tag_bytes) <= size) {
 235                 tag = (journal_block_tag_t *) tagp;
 236
 237                 nr++;
 238                 tagp += tag_bytes;
 239                 if (!(get_be16(&tag->t_flags) & JFS_FLAG_SAME_UUID))
 240                         tagp += 16;
 241
 242                 if (get_be16(&tag->t_flags) & JFS_FLAG_LAST_TAG)
 243                         break;
 244         }
 245
 246         return nr;
 247 }
 248
 249
 250 /* Make sure we wrap around the log correctly! */
 251 #define wrap(journal, var)                                              \
 252 do {                                                                    \
 253         if (var >= (journal)->j_last)                                   \
 254                 var -= ((journal)->j_last - (journal)->j_first);        \
 255 } while (0)
 256
 257 /**
 258  * journal_recover - recovers a on-disk journal
 259  * @journal: the journal to recover
 260  *
 261  * The primary function for recovering the log contents when mounting a
 262  * journaled device.
 263  *
 264  * Recovery is done in three passes.  In the first pass, we look for the
 265  * end of the log.  In the second, we assemble the list of revoke
 266  * blocks.  In the third and final pass, we replay any un-revoked blocks
 267  * in the log.
 268  */
 269 int journal_recover(journal_t *journal)
 270 {
 271         int                     err, err2;
 272         journal_superblock_t *  sb;
 273
 274         struct recovery_info    info;
 275
 276         memset(&info, 0, sizeof(info));
 277         sb = journal->j_superblock;
 278
 279         /*
 280          * The journal superblock's s_start field (the current log head)
 281          * is always zero if, and only if, the journal was cleanly
 282          * unmounted.
 283          */
 284
 285         if (!sb->s_start) {
 286                 jbd_debug(1, "No recovery required, last transaction %d\n",
 287                           ext2fs_be32_to_cpu(sb->s_sequence));
 288                 journal->j_transaction_sequence = ext2fs_be32_to_cpu(sb->s_sequence) + 1;
 289                 return 0;
 290         }
 291
 292         err = do_one_pass(journal, &info, PASS_SCAN);
 293         if (!err)
 294                 err = do_one_pass(journal, &info, PASS_REVOKE);
 295         if (!err)
 296                 err = do_one_pass(journal, &info, PASS_REPLAY);
 297
 298         jbd_debug(1, "JBD2: recovery, exit status %d, "
 299                   "recovered transactions %u to %u\n",
 300                   err, info.start_transaction, info.end_transaction);
 301         jbd_debug(1, "JBD2: Replayed %d and revoked %d/%d blocks\n",
 302                   info.nr_replays, info.nr_revoke_hits, info.nr_revokes);
 303
 304         /* Restart the log at the next transaction ID, thus invalidating
 305          * any existing commit records in the log. */
 306         journal->j_transaction_sequence = ++info.end_transaction;
 307
 308         journal_clear_revoke(journal);
 309         err2 = sync_blockdev(journal->j_fs_dev);
 310         if (!err)
 311                 err = err2;
 312         /* Make sure all replayed data is on permanent storage */
 313         if (journal->j_flags & JFS_BARRIER) {
 314                 err2 = blkdev_issue_flush(journal->j_fs_dev, GFP_KERNEL, NULL);
 315                 if (!err)
 316                         err = err2;
 317         }
 318         return err;
 319 }
 320
 321 /**
 322  * journal_skip_recovery - Start journal and wipe exiting records
 323  * @journal: journal to startup
 324  *
 325  * Locate any valid recovery information from the journal and set up the
 326  * journal structures in memory to ignore it (presumably because the
 327  * caller has evidence that it is out of date).
 328  * This function doesn't appear to be exported..
 329  *
 330  * We perform one pass over the journal to allow us to tell the user how
 331  * much recovery information is being erased, and to let us initialise
 332  * the journal transaction sequence numbers to the next unused ID.
 333  */
 334 int journal_skip_recovery(journal_t *journal)
 335 {
 336         int                     err;
 337
 338         struct recovery_info    info;
 339
 340         memset (&info, 0, sizeof(info));
 341
 342         err = do_one_pass(journal, &info, PASS_SCAN);
 343
 344         if (err) {
 345                 printk(KERN_ERR "JBD2: error %d scanning journal\n", err);
 346                 ++journal->j_transaction_sequence;
 347         } else {
 348 #ifdef CONFIG_JFS_DEBUG
 349                 int dropped = info.end_transaction -
 350                         ext2fs_be32_to_cpu(journal->j_superblock->s_sequence);
 351                 jbd_debug(1,
 352                           "JBD2: ignoring %d transaction%s from the journal.\n",
 353                           dropped, (dropped == 1) ? "" : "s");
 354 #endif
 355                 journal->j_transaction_sequence = ++info.end_transaction;
 356         }
 357
 358         journal->j_tail = 0;
 359         return err;
 360 }
 361
 362 static inline unsigned long long read_tag_block(journal_t *journal,
 363                                                 journal_block_tag_t *tag)
 364 {
 365         unsigned long long block = get_be32(&tag->t_blocknr);
 366         if (jfs_has_feature_64bit(journal))
 367                 block |= (u64)get_be32(&tag->t_blocknr_high) << 32;
 368         return block;
 369 }
 370
 371 /*
 372  * calc_chksums calculates the checksums for the blocks described in the
 373  * descriptor block.
 374  */
 375 static int calc_chksums(journal_t *journal, struct buffer_head *bh,
 376                         unsigned long *next_log_block, __u32 *crc32_sum)
 377 {
 378         int i, num_blks, err;
 379         unsigned long io_block;
 380         struct buffer_head *obh;
 381
 382         num_blks = count_tags(journal, bh);
 383         /* Calculate checksum of the descriptor block. */
 384         *crc32_sum = crc32_be(*crc32_sum, (void *)bh->b_data, bh->b_size);
 385
 386         for (i = 0; i < num_blks; i++) {
 387                 io_block = (*next_log_block)++;
 388                 wrap(journal, *next_log_block);
 389                 err = jread(&obh, journal, io_block);
 390                 if (err) {
 391                         printk(KERN_ERR "JBD2: IO error %d recovering block "
 392                                 "%lu in log\n", err, io_block);
 393                         return 1;
 394                 } else {
 395                         *crc32_sum = crc32_be(*crc32_sum, (void *)obh->b_data,
 396                                      obh->b_size);
 397                 }
 398                 put_bh(obh);
 399         }
 400         return 0;
 401 }
 402
 403 static int jbd2_commit_block_csum_verify(journal_t *j, void *buf)
 404 {
 405         struct commit_header *h;
 406         __u32 provided;
 407         __u32 calculated;
 408
 409         if (!journal_has_csum_v2or3(j))
 410                 return 1;
 411
 412         h = buf;
 413         provided = h->h_chksum[0];
 414         h->h_chksum[0] = 0;
 415         calculated = jbd2_chksum(j, j->j_csum_seed, buf, j->j_blocksize);
 416         h->h_chksum[0] = provided;
 417
 418         return provided == ext2fs_cpu_to_be32(calculated);
 419 }
 420
 421 static int jbd2_block_tag_csum_verify(journal_t *j, journal_block_tag_t *tag,
 422                                       void *buf, __u32 sequence)
 423 {
 424         journal_block_tag3_t *tag3 = (journal_block_tag3_t *)tag;
 425         __u32 csum32;
 426         __u32 seq;
 427
 428         if (!journal_has_csum_v2or3(j))
 429                 return 1;
 430
 431         seq = ext2fs_cpu_to_be32(sequence);
 432         csum32 = jbd2_chksum(j, j->j_csum_seed, (__u8 *)&seq, sizeof(seq));
 433         csum32 = jbd2_chksum(j, csum32, buf, j->j_blocksize);
 434
 435         if (jfs_has_feature_csum3(j))
 436                 return get_be32(&tag3->t_checksum) == csum32;
 437
 438         return get_be16(&tag->t_checksum) == (csum32 & 0xFFFF);
 439 }
 440
 441 static int do_one_pass(journal_t *journal,
 442                         struct recovery_info *info, enum passtype pass)
 443 {
 444         unsigned int            first_commit_ID, next_commit_ID;
 445         unsigned long           next_log_block;
 446         int                     err, success = 0;
 447         journal_superblock_t *  sb;
 448         journal_header_t *      tmp;
 449         struct buffer_head *    bh;
 450         unsigned int            sequence;
 451         int                     blocktype;
 452         int                     tag_bytes = journal_tag_bytes(journal);
 453         __u32                   crc32_sum = ~0; /* Transactional Checksums */
 454         int                     descr_csum_size = 0;
 455         int                     block_error = 0;
 456
 457         /*
 458          * First thing is to establish what we expect to find in the log
 459          * (in terms of transaction IDs), and where (in terms of log
 460          * block offsets): query the superblock.
 461          */
 462
 463         sb = journal->j_superblock;
 464         next_commit_ID = ext2fs_be32_to_cpu(sb->s_sequence);
 465         next_log_block = ext2fs_be32_to_cpu(sb->s_start);
 466
 467         first_commit_ID = next_commit_ID;
 468         if (pass == PASS_SCAN)
 469                 info->start_transaction = first_commit_ID;
 470
 471         jbd_debug(1, "Starting recovery pass %d\n", pass);
 472
 473         /*
 474          * Now we walk through the log, transaction by transaction,
 475          * making sure that each transaction has a commit block in the
 476          * expected place.  Each complete transaction gets replayed back
 477          * into the main filesystem.
 478          */
 479
 480         while (1) {
 481                 int                     flags;
 482                 char *                  tagp;
 483                 journal_block_tag_t *   tag;
 484                 struct buffer_head *    obh;
 485                 struct buffer_head *    nbh;
 486
 487                 cond_resched();
 488
 489                 /* If we already know where to stop the log traversal,
 490                  * check right now that we haven't gone past the end of
 491                  * the log. */
 492
 493                 if (pass != PASS_SCAN)
 494                         if (tid_geq(next_commit_ID, info->end_transaction))
 495                                 break;
 496
 497                 jbd_debug(2, "Scanning for sequence ID %u at %lu/%lu\n",
 498                           next_commit_ID, next_log_block, journal->j_last);
 499
 500                 /* Skip over each chunk of the transaction looking
 501                  * either the next descriptor block or the final commit
 502                  * record. */
 503
 504                 jbd_debug(3, "JBD2: checking block %ld\n", next_log_block);
 505                 err = jread(&bh, journal, next_log_block);
 506                 if (err)
 507                         goto failed;
 508
 509                 next_log_block++;
 510                 wrap(journal, next_log_block);
 511
 512                 /* What kind of buffer is it?
 513                  *
 514                  * If it is a descriptor block, check that it has the
 515                  * expected sequence number.  Otherwise, we're all done
 516                  * here. */
 517
 518                 tmp = (journal_header_t *)bh->b_data;
 519
 520                 if (tmp->h_magic != ext2fs_cpu_to_be32(JFS_MAGIC_NUMBER)) {
 521                         brelse(bh);
 522                         break;
 523                 }
 524
 525                 blocktype = ext2fs_be32_to_cpu(tmp->h_blocktype);
 526                 sequence = ext2fs_be32_to_cpu(tmp->h_sequence);
 527                 jbd_debug(3, "Found magic %d, sequence %d\n",
 528                           blocktype, sequence);
 529
 530                 if (sequence != next_commit_ID) {
 531                         brelse(bh);
 532                         break;
 533                 }
 534
 535                 /* OK, we have a valid descriptor block which matches
 536                  * all of the sequence number checks.  What are we going
 537                  * to do with it?  That depends on the pass... */
 538
 539                 switch(blocktype) {
 540                 case JFS_DESCRIPTOR_BLOCK:
 541                         /* Verify checksum first */
 542                         if (journal_has_csum_v2or3(journal))
 543                                 descr_csum_size =
 544                                         sizeof(struct journal_block_tail);
 545                         if (descr_csum_size > 0 &&
 546                             !jbd2_descr_block_csum_verify(journal,
 547                                                           bh->b_data)) {
 548                                 err = -EFSBADCRC;
 549                                 brelse(bh);
 550                                 goto failed;
 551                         }
 552
 553                         /* If it is a valid descriptor block, replay it
 554                          * in pass REPLAY; if journal_checksums enabled, then
 555                          * calculate checksums in PASS_SCAN, otherwise,
 556                          * just skip over the blocks it describes. */
 557                         if (pass != PASS_REPLAY) {
 558                                 if (pass == PASS_SCAN &&
 559                                     jfs_has_feature_checksum(journal) &&
 560                                     !info->end_transaction) {
 561                                         if (calc_chksums(journal, bh,
 562                                                         &next_log_block,
 563                                                         &crc32_sum)) {
 564                                                 put_bh(bh);
 565                                                 break;
 566                                         }
 567                                         put_bh(bh);
 568                                         continue;
 569                                 }
 570                                 next_log_block += count_tags(journal, bh);
 571                                 wrap(journal, next_log_block);
 572                                 put_bh(bh);
 573                                 continue;
 574                         }
 575
 576                         /* A descriptor block: we can now write all of
 577                          * the data blocks.  Yay, useful work is finally
 578                          * getting done here! */
 579
 580                         tagp = &bh->b_data[sizeof(journal_header_t)];
 581                         while ((tagp - bh->b_data + tag_bytes)
 582                                <= journal->j_blocksize - descr_csum_size) {
 583                                 unsigned long io_block;
 584
 585                                 tag = (journal_block_tag_t *) tagp;
 586                                 flags = get_be16(&tag->t_flags);
 587
 588                                 io_block = next_log_block++;
 589                                 wrap(journal, next_log_block);
 590                                 err = jread(&obh, journal, io_block);
 591                                 if (err) {
 592                                         /* Recover what we can, but
 593                                          * report failure at the end. */
 594                                         success = err;
 595                                         printk(KERN_ERR
 596                                                 "JBD2: IO error %d recovering "
 597                                                 "block %ld in log\n",
 598                                                 err, io_block);
 599                                 } else {
 600                                         unsigned long long blocknr;
 601
 602                                         J_ASSERT(obh != NULL);
 603                                         blocknr = read_tag_block(journal,
 604                                                                  tag);
 605
 606                                         /* If the block has been
 607                                          * revoked, then we're all done
 608                                          * here. */
 609                                         if (journal_test_revoke
 610                                             (journal, blocknr,
 611                                              next_commit_ID)) {
 612                                                 brelse(obh);
 613                                                 ++info->nr_revoke_hits;
 614                                                 goto skip_write;
 615                                         }
 616
 617                                         /* Look for block corruption */
 618                                         if (!jbd2_block_tag_csum_verify(
 619                                                 journal, tag, obh->b_data,
 620                                                 ext2fs_be32_to_cpu(tmp->h_sequence))) {
 621                                                 brelse(obh);
 622                                                 success = -EFSBADCRC;
 623                                                 printk(KERN_ERR "JBD2: Invalid "
 624                                                        "checksum recovering "
 625                                                        "block %llu in log\n",
 626                                                        blocknr);
 627                                                 block_error = 1;
 628                                                 goto skip_write;
 629                                         }
 630
 631                                         /* Find a buffer for the new
 632                                          * data being restored */
 633                                         nbh = __getblk(journal->j_fs_dev,
 634                                                         blocknr,
 635                                                         journal->j_blocksize);
 636                                         if (nbh == NULL) {
 637                                                 printk(KERN_ERR
 638                                                        "JBD2: Out of memory "
 639                                                        "during recovery.\n");
 640                                                 err = -ENOMEM;
 641                                                 brelse(bh);
 642                                                 brelse(obh);
 643                                                 goto failed;
 644                                         }
 645
 646                                         lock_buffer(nbh);
 647                                         memcpy(nbh->b_data, obh->b_data,
 648                                                         journal->j_blocksize);
 649                                         if (flags & JFS_FLAG_ESCAPE) {
 650                                                 __u32 magic = ext2fs_cpu_to_be32(JFS_MAGIC_NUMBER);
 651                                                 memcpy(nbh->b_data, &magic,
 652                                                        sizeof(magic));
 653                                         }
 654
 655                                         BUFFER_TRACE(nbh, "marking dirty");
 656                                         set_buffer_uptodate(nbh);
 657                                         mark_buffer_dirty(nbh);
 658                                         BUFFER_TRACE(nbh, "marking uptodate");
 659                                         ++info->nr_replays;
 660                                         /* ll_rw_block(WRITE, 1, &nbh); */
 661                                         unlock_buffer(nbh);
 662                                         brelse(obh);
 663                                         brelse(nbh);
 664                                 }
 665
 666                         skip_write:
 667                                 tagp += tag_bytes;
 668                                 if (!(flags & JFS_FLAG_SAME_UUID))
 669                                         tagp += 16;
 670
 671                                 if (flags & JFS_FLAG_LAST_TAG)
 672                                         break;
 673                         }
 674
 675                         brelse(bh);
 676                         continue;
 677
 678                 case JFS_COMMIT_BLOCK:
 679                         /*     How to differentiate between interrupted commit
 680                          *               and journal corruption ?
 681                          *
 682                          * {nth transaction}
 683                          *        Checksum Verification Failed
 684                          *                       |
 685                          *               ____________________
 686                          *              |                    |
 687                          *      async_commit             sync_commit
 688                          *              |                    |
 689                          *              | GO TO NEXT    "Journal Corruption"
 690                          *              | TRANSACTION
 691                          *              |
 692                          * {(n+1)th transaction}
 693                          *              |
 694                          *       _______|______________
 695                          *      |                     |
 696                          * Commit block found   Commit block not found
 697                          *      |                     |
 698                          * "Journal Corruption"       |
 699                          *               _____________|_________
 700                          *              |                       |
 701                          *      nth trans corrupt       OR   nth trans
 702                          *      and (n+1)th interrupted     interrupted
 703                          *      before commit block
 704                          *      could reach the disk.
 705                          *      (Cannot find the difference in above
 706                          *       mentioned conditions. Hence assume
 707                          *       "Interrupted Commit".)
 708                          */
 709
 710                         /* Found an expected commit block: if checksums
 711                          * are present verify them in PASS_SCAN; else not
 712                          * much to do other than move on to the next sequence
 713                          * number. */
 714                         if (pass == PASS_SCAN &&
 715                             jfs_has_feature_checksum(journal)) {
 716                                 int chksum_err, chksum_seen;
 717                                 struct commit_header *cbh =
 718                                         (struct commit_header *)bh->b_data;
 719                                 unsigned found_chksum =
 720                                         ext2fs_be32_to_cpu(cbh->h_chksum[0]);
 721
 722                                 chksum_err = chksum_seen = 0;
 723
 724                                 if (info->end_transaction) {
 725                                         journal->j_failed_commit =
 726                                                 info->end_transaction;
 727                                         brelse(bh);
 728                                         break;
 729                                 }
 730
 731                                 if (crc32_sum == found_chksum &&
 732                                     cbh->h_chksum_type == JFS_CRC32_CHKSUM &&
 733                                     cbh->h_chksum_size ==
 734                                                 JFS_CRC32_CHKSUM_SIZE)
 735                                        chksum_seen = 1;
 736                                 else if (!(cbh->h_chksum_type == 0 &&
 737                                              cbh->h_chksum_size == 0 &&
 738                                              found_chksum == 0 &&
 739                                              !chksum_seen))
 740                                 /*
 741                                  * If fs is mounted using an old kernel and then
 742                                  * kernel with journal_chksum is used then we
 743                                  * get a situation where the journal flag has
 744                                  * checksum flag set but checksums are not
 745                                  * present i.e chksum = 0, in the individual
 746                                  * commit blocks.
 747                                  * Hence to avoid checksum failures, in this
 748                                  * situation, this extra check is added.
 749                                  */
 750                                                 chksum_err = 1;
 751
 752                                 if (chksum_err) {
 753                                         info->end_transaction = next_commit_ID;
 754
 755                                         if (!jfs_has_feature_async_commit(journal)){
 756                                                 journal->j_failed_commit =
 757                                                         next_commit_ID;
 758                                                 brelse(bh);
 759                                                 break;
 760                                         }
 761                                 }
 762                                 crc32_sum = ~0;
 763                         }
 764                         if (pass == PASS_SCAN &&
 765                             !jbd2_commit_block_csum_verify(journal,
 766                                                            bh->b_data)) {
 767                                 info->end_transaction = next_commit_ID;
 768
 769                                 if (!jfs_has_feature_async_commit(journal)) {
 770                                         journal->j_failed_commit =
 771                                                 next_commit_ID;
 772                                         brelse(bh);
 773                                         break;
 774                                 }
 775                         }
 776                         brelse(bh);
 777                         next_commit_ID++;
 778                         continue;
 779
 780                 case JFS_REVOKE_BLOCK:
 781                         /* If we aren't in the REVOKE pass, then we can
 782                          * just skip over this block. */
 783                         if (pass != PASS_REVOKE) {
 784                                 brelse(bh);
 785                                 continue;
 786                         }
 787
 788                         err = scan_revoke_records(journal, bh,
 789                                                   next_commit_ID, info);
 790                         brelse(bh);
 791                         if (err)
 792                                 goto failed;
 793                         continue;
 794
 795                 default:
 796                         jbd_debug(3, "Unrecognised magic %d, end of scan.\n",
 797                                   blocktype);
 798                         brelse(bh);
 799                         goto done;
 800                 }
 801         }
 802
 803  done:
 804         /*
 805          * We broke out of the log scan loop: either we came to the
 806          * known end of the log or we found an unexpected block in the
 807          * log.  If the latter happened, then we know that the "current"
 808          * transaction marks the end of the valid log.
 809          */
 810
 811         if (pass == PASS_SCAN) {
 812                 if (!info->end_transaction)
 813                         info->end_transaction = next_commit_ID;
 814         } else {
 815                 /* It's really bad news if different passes end up at
 816                  * different places (but possible due to IO errors). */
 817                 if (info->end_transaction != next_commit_ID) {
 818                         printk(KERN_ERR "JBD2: recovery pass %d ended at "
 819                                 "transaction %u, expected %u\n",
 820                                 pass, next_commit_ID, info->end_transaction);
 821                         if (!success)
 822                                 success = -EIO;
 823                 }
 824         }
 825         if (block_error && success == 0)
 826                 success = -EIO;
 827         return success;
 828
 829  failed:
 830         return err;
 831 }
 832
 833 static int jbd2_revoke_block_csum_verify(journal_t *j,
 834                                          void *buf)
 835 {
 836         struct journal_revoke_tail *tail;
 837         __u32 provided;
 838         __u32 calculated;
 839
 840         if (!journal_has_csum_v2or3(j))
 841                 return 1;
 842
 843         tail = (struct journal_revoke_tail *)((char *)buf + j->j_blocksize -
 844                         sizeof(struct journal_revoke_tail));
 845         provided = tail->r_checksum;
 846         tail->r_checksum = 0;
 847         calculated = jbd2_chksum(j, j->j_csum_seed, buf, j->j_blocksize);
 848         tail->r_checksum = provided;
 849
 850         return provided == ext2fs_cpu_to_be32(calculated);
 851 }
 852
 853 /* Scan a revoke record, marking all blocks mentioned as revoked. */
 854
 855 static int scan_revoke_records(journal_t *journal, struct buffer_head *bh,
 856                                tid_t sequence, struct recovery_info *info)
 857 {
 858         journal_revoke_header_t *header;
 859         int offset, max;
 860         unsigned csum_size = 0;
 861         __u32 rcount;
 862         int record_len = 4;
 863
 864         header = (journal_revoke_header_t *) bh->b_data;
 865         offset = sizeof(journal_revoke_header_t);
 866         rcount = ext2fs_be32_to_cpu(header->r_count);
 867
 868         if (!jbd2_revoke_block_csum_verify(journal, header))
 869                 return -EFSBADCRC;
 870
 871         if (journal_has_csum_v2or3(journal))
 872                 csum_size = sizeof(struct journal_revoke_tail);
 873         if (rcount > journal->j_blocksize - csum_size)
 874                 return -EINVAL;
 875         max = rcount;
 876
 877         if (jfs_has_feature_64bit(journal))
 878                 record_len = 8;
 879
 880         while (offset + record_len <= max) {
 881                 unsigned long long blocknr;
 882                 int err;
 883
 884                 if (record_len == 4)
 885                         blocknr = ext2fs_be32_to_cpu(* ((__u32 *) (bh->b_data+offset)));
 886                 else
 887                         blocknr = ext2fs_be64_to_cpu(* ((__u64 *) (bh->b_data+offset)));
 888                 offset += record_len;
 889                 err = journal_set_revoke(journal, blocknr, sequence);
 890                 if (err)
 891                         return err;
 892                 ++info->nr_revokes;
 893         }
 894         return 0;
 895 }