lustre/kernel_patches/patches/jbd-journal-chksum-2.6-sles10.patch

   1 Index: linux-2.6.16.60-0.37/fs/jbd/commit.c
   2 ===================================================================
   3 --- linux-2.6.16.60-0.37.orig/fs/jbd/commit.c   2009-06-02 23:33:33.000000000 -0600
   4 +++ linux-2.6.16.60-0.37/fs/jbd/commit.c        2009-06-02 23:33:54.000000000 -0600
   5 @@ -22,6 +22,7 @@
   6  #include <linux/pagemap.h>
   7  #include <linux/smp_lock.h>
   8  #include <linux/jiffies.h>
   9 +#include <linux/crc32.h>
  10
  11  /*
  12   * Default IO end handler for temporary BJ_IO buffer_heads.
  13 @@ -94,19 +95,23 @@
  14         return 1;
  15  }
  16
  17 -/* Done it all: now write the commit record.  We should have
  18 +/*
  19 + * Done it all: now submit the commit record.  We should have
  20   * cleaned up our previous buffers by now, so if we are in abort
  21   * mode we can now just skip the rest of the journal write
  22   * entirely.
  23   *
  24   * Returns 1 if the journal needs to be aborted or 0 on success
  25   */
  26 -static int journal_write_commit_record(journal_t *journal,
  27 -                                       transaction_t *commit_transaction)
  28 +static int journal_submit_commit_record(journal_t *journal,
  29 +                                       transaction_t *commit_transaction,
  30 +                                       struct buffer_head **cbh,
  31 +                                       __u32 crc32_sum)
  32  {
  33         struct journal_head *descriptor;
  34 +       struct commit_header *tmp;
  35         struct buffer_head *bh;
  36 -       int i, ret;
  37 +       int ret;
  38         int barrier_done = 0;
  39
  40         if (is_journal_aborted(journal))
  41 @@ -118,21 +123,34 @@
  42
  43         bh = jh2bh(descriptor);
  44
  45 -       /* AKPM: buglet - add `i' to tmp! */
  46 -       for (i = 0; i < bh->b_size; i += 512) {
  47 -               journal_header_t *tmp = (journal_header_t*)bh->b_data;
  48 -               tmp->h_magic = cpu_to_be32(JFS_MAGIC_NUMBER);
  49 -               tmp->h_blocktype = cpu_to_be32(JFS_COMMIT_BLOCK);
  50 -               tmp->h_sequence = cpu_to_be32(commit_transaction->t_tid);
  51 +       tmp = (struct commit_header *)bh->b_data;
  52 +       tmp->h_magic = cpu_to_be32(JFS_MAGIC_NUMBER);
  53 +       tmp->h_blocktype = cpu_to_be32(JFS_COMMIT_BLOCK);
  54 +       tmp->h_sequence = cpu_to_be32(commit_transaction->t_tid);
  55 +
  56 +       if (JFS_HAS_COMPAT_FEATURE(journal,
  57 +                               JFS_FEATURE_COMPAT_CHECKSUM)) {
  58 +               tmp->h_chksum_type      = JFS_CRC32_CHKSUM;
  59 +               tmp->h_chksum_size      = JFS_CRC32_CHKSUM_SIZE;
  60 +               tmp->h_chksum[0]        = cpu_to_be32(crc32_sum);
  61         }
  62
  63 -       JBUFFER_TRACE(descriptor, "write commit block");
  64 +       JBUFFER_TRACE(descriptor, "submit commit block");
  65 +       lock_buffer(bh);
  66 +
  67         set_buffer_dirty(bh);
  68 -       if (journal->j_flags & JFS_BARRIER) {
  69 +       set_buffer_uptodate(bh);
  70 +       bh->b_end_io = journal_end_buffer_io_sync;
  71 +
  72 +       if (journal->j_flags & JFS_BARRIER &&
  73 +               !JFS_HAS_INCOMPAT_FEATURE(journal,
  74 +                                        JFS_FEATURE_INCOMPAT_ASYNC_COMMIT)) {
  75 +
  76                 set_buffer_ordered(bh);
  77                 barrier_done = 1;
  78         }
  79 -       ret = sync_dirty_buffer(bh);
  80 +       ret = submit_bh(WRITE, bh);
  81 +
  82         if (barrier_done)
  83                 clear_buffer_ordered(bh);
  84         /* is it possible for another commit to fail at roughly
  85 @@ -154,12 +172,84 @@
  86                 /* And try again, without the barrier */
  87                 set_buffer_uptodate(bh);
  88                 set_buffer_dirty(bh);
  89 -               ret = sync_dirty_buffer(bh);
  90 +               ret = submit_bh(WRITE, bh);
  91         }
  92 -       put_bh(bh);             /* One for getblk() */
  93 -       journal_put_journal_head(descriptor);
  94 +       *cbh = bh;
  95 +       return ret;
  96 +}
  97
  98 -       return (ret == -EIO);
  99 +/*
 100 + * This function along with journal_submit_commit_record
 101 + * allows to write the commit record asynchronously.
 102 + */
 103 +static int journal_wait_on_commit_record(struct buffer_head *bh)
 104 +{
 105 +       int ret = 0;
 106 +
 107 +       clear_buffer_dirty(bh);
 108 +       wait_on_buffer(bh);
 109 +
 110 +       if (unlikely(!buffer_uptodate(bh)))
 111 +               ret = -EIO;
 112 +       put_bh(bh);            /* One for getblk() */
 113 +       journal_put_journal_head(bh2jh(bh));
 114 +
 115 +       return ret;
 116 +}
 117 +
 118 +/*
 119 + * Wait for all submitted IO to complete.
 120 + */
 121 +static int journal_wait_on_locked_list(journal_t *journal,
 122 +                                      transaction_t *commit_transaction)
 123 +{
 124 +       int ret = 0;
 125 +       struct journal_head *jh;
 126 +
 127 +       while (commit_transaction->t_locked_list) {
 128 +               struct buffer_head *bh;
 129 +
 130 +               jh = commit_transaction->t_locked_list->b_tprev;
 131 +               bh = jh2bh(jh);
 132 +               get_bh(bh);
 133 +               if (buffer_locked(bh)) {
 134 +                       spin_unlock(&journal->j_list_lock);
 135 +                       wait_on_buffer(bh);
 136 +                       if (unlikely(!buffer_uptodate(bh)))
 137 +                               ret = -EIO;
 138 +                       spin_lock(&journal->j_list_lock);
 139 +               }
 140 +               if (!inverted_lock(journal, bh)) {
 141 +                       put_bh(bh);
 142 +                       spin_lock(&journal->j_list_lock);
 143 +                       continue;
 144 +               }
 145 +               if (buffer_jbd(bh) && jh->b_jlist == BJ_Locked) {
 146 +                       __journal_unfile_buffer(jh);
 147 +                       jbd_unlock_bh_state(bh);
 148 +                       journal_remove_journal_head(bh);
 149 +                       put_bh(bh);
 150 +               } else {
 151 +                       jbd_unlock_bh_state(bh);
 152 +               }
 153 +               put_bh(bh);
 154 +               cond_resched_lock(&journal->j_list_lock);
 155 +       }
 156 +       return ret;
 157 +}
 158 +
 159 +static inline __u32 jbd_checksum_data(__u32 crc32_sum, struct buffer_head *bh)
 160 +{
 161 +       struct page *page = bh->b_page;
 162 +       char *addr;
 163 +       __u32 checksum;
 164 +
 165 +       addr = kmap_atomic(page, KM_USER0);
 166 +       checksum = crc32_be(crc32_sum,
 167 +                           (void *)(addr + offset_in_page(bh->b_data)),
 168 +                           bh->b_size);
 169 +       kunmap_atomic(addr, KM_USER0);
 170 +       return checksum;
 171  }
 172
 173  void journal_do_submit_data(struct buffer_head **wbuf, int bufs)
 174 @@ -296,6 +386,8 @@
 175         int first_tag = 0;
 176         int tag_flag;
 177         int i;
 178 +       struct buffer_head *cbh = NULL; /* For transactional checksums */
 179 +       __u32 crc32_sum = ~0;
 180
 181         /*
 182          * First job: lock down the current transaction and wait for
 183 @@ -439,38 +531,15 @@
 184         journal_submit_data_buffers(journal, commit_transaction);
 185
 186         /*
 187 -        * Wait for all previously submitted IO to complete.
 188 +        * Wait for all previously submitted IO to complete if commit
 189 +        * record is to be written synchronously.
 190          */
 191         spin_lock(&journal->j_list_lock);
 192 -       while (commit_transaction->t_locked_list) {
 193 -               struct buffer_head *bh;
 194 +       if (!JFS_HAS_INCOMPAT_FEATURE(journal,
 195 +               JFS_FEATURE_INCOMPAT_ASYNC_COMMIT))
 196 +               err = journal_wait_on_locked_list(journal,
 197 +                                                 commit_transaction);
 198
 199 -               jh = commit_transaction->t_locked_list->b_tprev;
 200 -               bh = jh2bh(jh);
 201 -               get_bh(bh);
 202 -               if (buffer_locked(bh)) {
 203 -                       spin_unlock(&journal->j_list_lock);
 204 -                       wait_on_buffer(bh);
 205 -                       if (unlikely(!buffer_uptodate(bh)))
 206 -                               err = -EIO;
 207 -                       spin_lock(&journal->j_list_lock);
 208 -               }
 209 -               if (!inverted_lock(journal, bh)) {
 210 -                       put_bh(bh);
 211 -                       spin_lock(&journal->j_list_lock);
 212 -                       continue;
 213 -               }
 214 -               if (buffer_jbd(bh) && jh->b_jlist == BJ_Locked) {
 215 -                       __journal_unfile_buffer(jh);
 216 -                       jbd_unlock_bh_state(bh);
 217 -                       journal_remove_journal_head(bh);
 218 -                       put_bh(bh);
 219 -               } else {
 220 -                       jbd_unlock_bh_state(bh);
 221 -               }
 222 -               put_bh(bh);
 223 -               cond_resched_lock(&journal->j_list_lock);
 224 -       }
 225         spin_unlock(&journal->j_list_lock);
 226
 227         if (err)
 228 @@ -643,6 +712,16 @@
 229  start_journal_io:
 230                         for (i = 0; i < bufs; i++) {
 231                                 struct buffer_head *bh = wbuf[i];
 232 +                               /*
 233 +                                * Compute checksum.
 234 +                                */
 235 +                               if (JFS_HAS_COMPAT_FEATURE(journal,
 236 +                                       JFS_FEATURE_COMPAT_CHECKSUM)) {
 237 +                                       crc32_sum =
 238 +                                               jbd_checksum_data(crc32_sum,
 239 +                                                                  bh);
 240 +                               }
 241 +
 242                                 lock_buffer(bh);
 243                                 clear_buffer_dirty(bh);
 244                                 set_buffer_uptodate(bh);
 245 @@ -659,6 +738,23 @@
 246                 }
 247         }
 248
 249 +       /* Done it all: now write the commit record asynchronously. */
 250 +
 251 +       if (JFS_HAS_INCOMPAT_FEATURE(journal,
 252 +               JFS_FEATURE_INCOMPAT_ASYNC_COMMIT)) {
 253 +               err = journal_submit_commit_record(journal, commit_transaction,
 254 +                                                &cbh, crc32_sum);
 255 +               if (err)
 256 +                       __journal_abort_hard(journal);
 257 +
 258 +               spin_lock(&journal->j_list_lock);
 259 +               err = journal_wait_on_locked_list(journal,
 260 +                                               commit_transaction);
 261 +               spin_unlock(&journal->j_list_lock);
 262 +               if (err)
 263 +                       __journal_abort_hard(journal);
 264 +       }
 265 +
 266         /* Lo and behold: we have just managed to send a transaction to
 267             the log.  Before we can commit it, wait for the IO so far to
 268             complete.  Control buffers being written are on the
 269 @@ -757,9 +853,15 @@
 270         }
 271
 272         jbd_debug(3, "JBD: commit phase 6\n");
 273 -
 274 -       if (journal_write_commit_record(journal, commit_transaction))
 275 -               err = -EIO;
 276 +
 277 +       if (!JFS_HAS_INCOMPAT_FEATURE(journal,
 278 +               JFS_FEATURE_INCOMPAT_ASYNC_COMMIT)) {
 279 +               err = journal_submit_commit_record(journal, commit_transaction,
 280 +                                               &cbh, crc32_sum);
 281 +               if (err)
 282 +                       __journal_abort_hard(journal);
 283 +       }
 284 +       err = journal_wait_on_commit_record(cbh);
 285
 286         if (err)
 287                 __journal_abort_hard(journal);
 288 Index: linux-2.6.16.60-0.37/fs/jbd/recovery.c
 289 ===================================================================
 290 --- linux-2.6.16.60-0.37.orig/fs/jbd/recovery.c 2006-03-19 22:53:29.000000000 -0700
 291 +++ linux-2.6.16.60-0.37/fs/jbd/recovery.c      2009-06-02 23:33:54.000000000 -0600
 292 @@ -21,6 +21,7 @@
 293  #include <linux/jbd.h>
 294  #include <linux/errno.h>
 295  #include <linux/slab.h>
 296 +#include <linux/crc32.h>
 297  #endif
 298
 299  /*
 300 @@ -307,6 +308,38 @@
 301         return err;
 302  }
 303
 304 +/*
 305 + * calc_chksums calculates the checksums for the blocks described in the
 306 + * descriptor block.
 307 + */
 308 +static int calc_chksums(journal_t *journal, struct buffer_head *bh,
 309 +                      unsigned long *next_log_block, __u32 *crc32_sum)
 310 +{
 311 +       int i, num_blks, err;
 312 +       unsigned long io_block;
 313 +       struct buffer_head *obh;
 314 +
 315 +       num_blks = count_tags(bh, journal->j_blocksize);
 316 +       /* Calculate checksum of the descriptor block. */
 317 +       *crc32_sum = crc32_be(*crc32_sum, (void *)bh->b_data, bh->b_size);
 318 +
 319 +       for (i = 0; i < num_blks; i++) {
 320 +               io_block = (*next_log_block)++;
 321 +               wrap(journal, *next_log_block);
 322 +               err = jread(&obh, journal, io_block);
 323 +               if (err) {
 324 +                       printk(KERN_ERR "JBD: IO error %d recovering block "
 325 +                               "%lu in log\n", err, io_block);
 326 +                       return 1;
 327 +               } else {
 328 +                       *crc32_sum = crc32_be(*crc32_sum, (void *)obh->b_data,
 329 +                                    obh->b_size);
 330 +               }
 331 +               put_bh(obh);
 332 +       }
 333 +       return 0;
 334 +}
 335 +
 336  static int do_one_pass(journal_t *journal,
 337                         struct recovery_info *info, enum passtype pass)
 338  {
 339 @@ -318,6 +351,7 @@
 340         struct buffer_head *    bh;
 341         unsigned int            sequence;
 342         int                     blocktype;
 343 +       __u32                   crc32_sum = ~0; /* Transactional Checksums */
 344
 345         /* Precompute the maximum metadata descriptors in a descriptor block */
 346         int                     MAX_BLOCKS_PER_DESC;
 347 @@ -409,9 +443,24 @@
 348                 switch(blocktype) {
 349                 case JFS_DESCRIPTOR_BLOCK:
 350                         /* If it is a valid descriptor block, replay it
 351 -                        * in pass REPLAY; otherwise, just skip over the
 352 -                        * blocks it describes. */
 353 +                        * in pass REPLAY; if journal_checksums enabled, then
 354 +                        * calculate checksums in PASS_SCAN, otherwise,
 355 +                        * just skip over the blocks it describes. */
 356                         if (pass != PASS_REPLAY) {
 357 +                               if (pass == PASS_SCAN &&
 358 +                                   JFS_HAS_COMPAT_FEATURE(journal,
 359 +                                           JFS_FEATURE_COMPAT_CHECKSUM) &&
 360 +                                   !info->end_transaction) {
 361 +                                       if (calc_chksums(journal, bh,
 362 +                                                       &next_log_block,
 363 +                                                       &crc32_sum)) {
 364 +                                               put_bh(bh);
 365 +                                               break;
 366 +                                       }
 367 +                                       put_bh(bh);
 368 +                                       continue;
 369 +                               }
 370 +
 371                                 next_log_block +=
 372                                         count_tags(bh, journal->j_blocksize);
 373                                 wrap(journal, next_log_block);
 374 @@ -506,9 +555,97 @@
 375                         continue;
 376
 377                 case JFS_COMMIT_BLOCK:
 378 -                       /* Found an expected commit block: not much to
 379 -                        * do other than move on to the next sequence
 380 +                       /*     How to differentiate between interrupted commit
 381 +                        *               and journal corruption ?
 382 +                        *
 383 +                        * {nth transaction}
 384 +                        *        Checksum Verification Failed
 385 +                        *                       |
 386 +                        *               ____________________
 387 +                        *              |                    |
 388 +                        *      async_commit             sync_commit
 389 +                        *              |                    |
 390 +                        *              | GO TO NEXT    "Journal Corruption"
 391 +                        *              | TRANSACTION
 392 +                        *              |
 393 +                        * {(n+1)th transanction}
 394 +                        *              |
 395 +                        *       _______|______________
 396 +                        *      |                     |
 397 +                        * Commit block found   Commit block not found
 398 +                        *      |                     |
 399 +                        * "Journal Corruption"       |
 400 +                        *               _____________|__________
 401 +                        *              |                       |
 402 +                        *      nth trans corrupt       OR   nth trans
 403 +                        *      and (n+1)th interrupted     interrupted
 404 +                        *      before commit block
 405 +                        *      could reach the disk.
 406 +                        *      (Cannot find the difference in above
 407 +                        *       mentioned conditions. Hence assume
 408 +                        *       "Interrupted Commit".)
 409 +                        */
 410 +
 411 +                       /* Found an expected commit block: if checksums
 412 +                        * are present verify them in PASS_SCAN; else not
 413 +                        * much to do other than move on to the next sequence
 414                          * number. */
 415 +                       if (pass == PASS_SCAN &&
 416 +                           JFS_HAS_COMPAT_FEATURE(journal,
 417 +                                   JFS_FEATURE_COMPAT_CHECKSUM)) {
 418 +                               int chksum_err, chksum_seen;
 419 +                               struct commit_header *cbh =
 420 +                                       (struct commit_header *)bh->b_data;
 421 +                               unsigned found_chksum =
 422 +                                               be32_to_cpu(cbh->h_chksum[0]);
 423 +
 424 +                               chksum_err = chksum_seen = 0;
 425 +
 426 +                               if (info->end_transaction) {
 427 +                                       printk(KERN_ERR "JBD: Transaction %u "
 428 +                                               "found to be corrupt.\n",
 429 +                                               next_commit_ID - 1);
 430 +                                       brelse(bh);
 431 +                                       break;
 432 +                               }
 433 +
 434 +                               if (crc32_sum == found_chksum &&
 435 +                                   cbh->h_chksum_type == JFS_CRC32_CHKSUM &&
 436 +                                   cbh->h_chksum_size ==
 437 +                                               JFS_CRC32_CHKSUM_SIZE) {
 438 +                                      chksum_seen = 1;
 439 +                               } else if (!(cbh->h_chksum_type == 0 &&
 440 +                                            cbh->h_chksum_size == 0 &&
 441 +                                            found_chksum == 0 &&
 442 +                                            !chksum_seen)) {
 443 +                               /*
 444 +                                * If fs is mounted using an old kernel and then
 445 +                                * kernel with journal_chksum is used then we
 446 +                                * get a situation where the journal flag has
 447 +                                * checksum flag set but checksums are not
 448 +                                * present i.e chksum = 0, in the individual
 449 +                                * commit blocks.
 450 +                                * Hence to avoid checksum failures, in this
 451 +                                * situation, this extra check is added.
 452 +                                */
 453 +                                               chksum_err = 1;
 454 +                               }
 455 +
 456 +                               if (chksum_err) {
 457 +                                       info->end_transaction = next_commit_ID;
 458 +
 459 +                                       if (!JFS_HAS_INCOMPAT_FEATURE(journal,
 460 +                                           JFS_FEATURE_INCOMPAT_ASYNC_COMMIT)){
 461 +                                               printk(KERN_ERR
 462 +                                                      "JBD: Transaction %u "
 463 +                                                      "found to be corrupt.\n",
 464 +                                                      next_commit_ID);
 465 +                                               brelse(bh);
 466 +                                               break;
 467 +                                       }
 468 +                               }
 469 +                               crc32_sum = ~0;
 470 +                       }
 471                         brelse(bh);
 472                         next_commit_ID++;
 473                         continue;
 474 @@ -543,9 +680,10 @@
 475          * transaction marks the end of the valid log.
 476          */
 477
 478 -       if (pass == PASS_SCAN)
 479 -               info->end_transaction = next_commit_ID;
 480 -       else {
 481 +       if (pass == PASS_SCAN) {
 482 +               if (!info->end_transaction)
 483 +                       info->end_transaction = next_commit_ID;
 484 +       } else {
 485                 /* It's really bad news if different passes end up at
 486                  * different places (but possible due to IO errors). */
 487                 if (info->end_transaction != next_commit_ID) {
 488 Index: linux-2.6.16.60-0.37/fs/jbd/journal.c
 489 ===================================================================
 490 --- linux-2.6.16.60-0.37.orig/fs/jbd/journal.c  2009-06-02 23:33:33.000000000 -0600
 491 +++ linux-2.6.16.60-0.37/fs/jbd/journal.c       2009-06-02 23:33:54.000000000 -0600
 492 @@ -64,6 +64,7 @@
 493  EXPORT_SYMBOL(journal_check_used_features);
 494  EXPORT_SYMBOL(journal_check_available_features);
 495  EXPORT_SYMBOL(journal_set_features);
 496 +EXPORT_SYMBOL(journal_clear_features);
 497  EXPORT_SYMBOL(journal_create);
 498  EXPORT_SYMBOL(journal_load);
 499  EXPORT_SYMBOL(journal_destroy);
 500 @@ -1565,6 +1566,33 @@
 501         return 1;
 502  }
 503
 504 +/**
 505 + * int journal_clear_features () - Clear a given journal feature in the superblock
 506 + * @journal: Journal to act on.
 507 + * @compat: bitmask of compatible features
 508 + * @ro: bitmask of features that force read-only mount
 509 + * @incompat: bitmask of incompatible features
 510 + *
 511 + * Clear a given journal feature as present on the
 512 + * superblock.  Returns true if the requested features could be reset.
 513 + *
 514 + */
 515 +int journal_clear_features (journal_t *journal, unsigned long compat,
 516 +                         unsigned long ro, unsigned long incompat)
 517 +{
 518 +       journal_superblock_t *sb;
 519 +
 520 +       jbd_debug(1, "Clear features 0x%lx/0x%lx/0x%lx\n",
 521 +                 compat, ro, incompat);
 522 +
 523 +       sb = journal->j_superblock;
 524 +
 525 +       sb->s_feature_compat    &= ~cpu_to_be32(compat);
 526 +       sb->s_feature_ro_compat &= ~cpu_to_be32(ro);
 527 +       sb->s_feature_incompat  &= ~cpu_to_be32(incompat);
 528 +
 529 +       return 1;
 530 +}
 531
 532  /**
 533   * int journal_update_format () - Update on-disk journal structure.
 534 Index: linux-2.6.16.60-0.37/fs/Kconfig
 535 ===================================================================
 536 --- linux-2.6.16.60-0.37.orig/fs/Kconfig        2009-03-24 05:46:35.000000000 -0700
 537 +++ linux-2.6.16.60-0.37/fs/Kconfig     2009-06-02 23:33:54.000000000 -0600
 538 @@ -140,6 +140,7 @@
 539
 540  config JBD
 541         tristate
 542 +       select CRC32
 543         help
 544           This is a generic journaling layer for block devices.  It is
 545           currently used by the ext3 and OCFS2 file systems, but it could
 546 Index: linux-2.6.16.60-0.37/include/linux/jbd.h
 547 ===================================================================
 548 --- linux-2.6.16.60-0.37.orig/include/linux/jbd.h       2009-06-02 23:33:33.000000000 -0600
 549 +++ linux-2.6.16.60-0.37/include/linux/jbd.h    2009-06-02 23:33:54.000000000 -0600
 550 @@ -142,6 +142,29 @@
 551         __be32          h_sequence;
 552  } journal_header_t;
 553
 554 +/*
 555 + * Checksum types.
 556 + */
 557 +#define JFS_CRC32_CHKSUM   1
 558 +#define JFS_MD5_CHKSUM     2
 559 +#define JFS_SHA1_CHKSUM    3
 560 +
 561 +#define JFS_CRC32_CHKSUM_SIZE 4
 562 +
 563 +#define JFS_CHECKSUM_BYTES (32 / sizeof(u32))
 564 +/*
 565 + * Commit block header for storing transactional checksums:
 566 + */
 567 +struct commit_header
 568 +{
 569 +       __be32          h_magic;
 570 +       __be32          h_blocktype;
 571 +       __be32          h_sequence;
 572 +       unsigned char   h_chksum_type;
 573 +       unsigned char   h_chksum_size;
 574 +       unsigned char   h_padding[2];
 575 +       __be32          h_chksum[JFS_CHECKSUM_BYTES];
 576 +};
 577
 578  /*
 579   * The block tag: used to describe a single buffer in the journal
 580 @@ -228,12 +251,16 @@
 581         ((j)->j_format_version >= 2 &&                                  \
 582          ((j)->j_superblock->s_feature_incompat & cpu_to_be32((mask))))
 583
 584 -#define JFS_FEATURE_INCOMPAT_REVOKE    0x00000001
 585 +#define JFS_FEATURE_COMPAT_CHECKSUM    0x00000001
 586 +
 587 +#define JFS_FEATURE_INCOMPAT_REVOKE            0x00000001
 588 +#define JFS_FEATURE_INCOMPAT_ASYNC_COMMIT      0x00000004
 589
 590  /* Features known to this kernel version: */
 591 -#define JFS_KNOWN_COMPAT_FEATURES      0
 592 +#define JFS_KNOWN_COMPAT_FEATURES      JFS_FEATURE_COMPAT_CHECKSUM
 593  #define JFS_KNOWN_ROCOMPAT_FEATURES    0
 594 -#define JFS_KNOWN_INCOMPAT_FEATURES    JFS_FEATURE_INCOMPAT_REVOKE
 595 +#define JFS_KNOWN_INCOMPAT_FEATURES    (JFS_FEATURE_INCOMPAT_REVOKE | \
 596 +                                       JFS_FEATURE_INCOMPAT_ASYNC_COMMIT)
 597
 598  #ifdef __KERNEL__
 599
 600 @@ -1041,6 +1068,8 @@
 601                    (journal_t *, unsigned long, unsigned long, unsigned long);
 602  extern int        journal_set_features
 603                    (journal_t *, unsigned long, unsigned long, unsigned long);
 604 +extern int        journal_clear_features
 605 +                  (journal_t *, unsigned long, unsigned long, unsigned long);
 606  extern int        journal_create     (journal_t *);
 607  extern int        journal_load       (journal_t *journal);
 608  extern void       journal_destroy    (journal_t *);
 609 Index: linux-2.6.16.60-0.37/Documentation/filesystems/ext3.txt
 610 ===================================================================
 611 --- linux-2.6.16.60-0.37.orig/Documentation/filesystems/ext3.txt        2006-03-19 22:53:29.000000000 -0700
 612 +++ linux-2.6.16.60-0.37/Documentation/filesystems/ext3.txt     2009-06-02 23:33:54.000000000 -0600
 613 @@ -14,6 +14,16 @@
 614  When mounting an ext3 filesystem, the following option are accepted:
 615  (*) == default
 616
 617 +journal_checksum       Enable checksumming of the journal transactions.
 618 +                       This will allow the recovery code in e2fsck and the
 619 +                       kernel to detect corruption in the kernel.  It is a
 620 +                       compatible change and will be ignored by older kernels.
 621 +
 622 +journal_async_commit   Commit block can be written to disk without waiting
 623 +                       for descriptor blocks. If enabled older kernels cannot
 624 +                       mount the device. This will enable 'journal_checksum'
 625 +                       internally.
 626 +
 627  journal=update         Update the ext3 file system's journal to the current
 628                         format.
 629