e2fsck/journal.c

   1 /*
   2  * journal.c --- code for handling the "ext3" journal
   3  *
   4  * Copyright (C) 2000 Andreas Dilger
   5  * Copyright (C) 2000 Theodore Ts'o
   6  *
   7  * Parts of the code are based on fs/jfs/journal.c by Stephen C. Tweedie
   8  * Copyright (C) 1999 Red Hat Software
   9  *
  10  * This file may be redistributed under the terms of the
  11  * GNU General Public License version 2 or at your discretion
  12  * any later version.
  13  */
  14
  15 #ifdef HAVE_SYS_MOUNT_H
  16 #include <sys/mount.h>
  17 #define MNT_FL (MS_MGC_VAL | MS_RDONLY)
  18 #endif
  19 #ifdef HAVE_SYS_STAT_H
  20 #include <sys/stat.h>
  21 #endif
  22
  23 #define E2FSCK_INCLUDE_INLINE_FUNCS
  24 #include "jfs_user.h"
  25 #include "problem.h"
  26 #include "uuid/uuid.h"
  27
  28 #ifdef JFS_DEBUG                /* Enabled by configure --enable-jfs-debug */
  29 static int bh_count = 0;
  30 int journal_enable_debug = 2;
  31 #endif
  32
  33 /* Kernel compatibility functions for handling the journal.  These allow us
  34  * to use the recovery.c file virtually unchanged from the kernel, so we
  35  * don't have to do much to keep kernel and user recovery in sync.
  36  */
  37 int bmap(struct inode *inode, int block)
  38 {
  39         int retval;
  40         blk_t phys;
  41
  42         retval = ext2fs_bmap(inode->i_ctx->fs, inode->i_ino, &inode->i_ext2,
  43                              NULL, 0, block, &phys);
  44
  45         if (retval)
  46                 com_err(inode->i_ctx->device_name, retval,
  47                         _("bmap journal inode %ld, block %d\n"),
  48                         inode->i_ino, block);
  49
  50         return phys;
  51 }
  52
  53 struct buffer_head *getblk(e2fsck_t ctx, blk_t blocknr, int blocksize)
  54 {
  55         struct buffer_head *bh;
  56
  57         bh = e2fsck_allocate_memory(ctx, sizeof(*bh), "block buffer");
  58         if (!bh)
  59                 return NULL;
  60
  61         jfs_debug(4, "getblk for block %lu (%d bytes)(total %d)\n",
  62                   (unsigned long) blocknr, blocksize, ++bh_count);
  63
  64         bh->b_ctx = ctx;
  65         bh->b_size = blocksize;
  66         bh->b_blocknr = blocknr;
  67
  68         return bh;
  69 }
  70
  71 void ll_rw_block(int rw, int nr, struct buffer_head *bhp[])
  72 {
  73         int retval;
  74         struct buffer_head *bh;
  75
  76         for (; nr > 0; --nr) {
  77                 bh = *bhp++;
  78                 if (rw == READ && !bh->b_uptodate) {
  79                         jfs_debug(3, "reading block %lu/%p\n",
  80                                   (unsigned long) bh->b_blocknr, (void *) bh);
  81                         retval = io_channel_read_blk(bh->b_ctx->fs->io,
  82                                                      bh->b_blocknr,
  83                                                      1, bh->b_data);
  84                         if (retval) {
  85                                 com_err(bh->b_ctx->device_name, retval,
  86                                         "while reading block %ld\n",
  87                                         bh->b_blocknr);
  88                                 bh->b_err = retval;
  89                                 continue;
  90                         }
  91                         bh->b_uptodate = 1;
  92                 } else if (rw == WRITE && bh->b_dirty) {
  93                         jfs_debug(3, "writing block %lu/%p\n",
  94                                   (unsigned long) bh->b_blocknr, (void *) bh);
  95                         retval = io_channel_write_blk(bh->b_ctx->fs->io,
  96                                                       bh->b_blocknr,
  97                                                       1, bh->b_data);
  98                         if (retval) {
  99                                 com_err(bh->b_ctx->device_name, retval,
 100                                         "while writing block %ld\n",
 101                                         bh->b_blocknr);
 102                                 bh->b_err = retval;
 103                                 continue;
 104                         }
 105                         bh->b_dirty = 0;
 106                         bh->b_uptodate = 1;
 107                 } else
 108                         jfs_debug(3, "no-op %s for block %lu\n",
 109                                   rw == READ ? "read" : "write",
 110                                   (unsigned long) bh->b_blocknr);
 111         }
 112 }
 113
 114 void mark_buffer_dirty(struct buffer_head *bh, int dummy)
 115 {
 116         bh->b_dirty = dummy | 1; /* use dummy to avoid unused variable */
 117 }
 118
 119 static void mark_buffer_clean(struct buffer_head * bh)
 120 {
 121         bh->b_dirty = 0;
 122 }
 123
 124 void brelse(struct buffer_head *bh)
 125 {
 126         if (bh->b_dirty)
 127                 ll_rw_block(WRITE, 1, &bh);
 128         jfs_debug(3, "freeing block %lu/%p (total %d)\n",
 129                   (unsigned long) bh->b_blocknr, (void *) bh, --bh_count);
 130         ext2fs_free_mem((void **) &bh);
 131 }
 132
 133 int buffer_uptodate(struct buffer_head *bh)
 134 {
 135         return bh->b_uptodate;
 136 }
 137
 138 void mark_buffer_uptodate(struct buffer_head *bh, int val)
 139 {
 140         bh->b_uptodate = val;
 141 }
 142
 143 void wait_on_buffer(struct buffer_head *bh)
 144 {
 145         if (!bh->b_uptodate)
 146                 ll_rw_block(READ, 1, &bh);
 147 }
 148
 149
 150 static void e2fsck_clear_recover(e2fsck_t ctx, int error)
 151 {
 152         ctx->fs->super->s_feature_incompat &= ~EXT3_FEATURE_INCOMPAT_RECOVER;
 153
 154         /* if we had an error doing journal recovery, we need a full fsck */
 155         if (error)
 156                 ctx->fs->super->s_state &= ~EXT2_VALID_FS;
 157         ext2fs_mark_super_dirty(ctx->fs);
 158 }
 159
 160 static errcode_t e2fsck_journal_init_inode(e2fsck_t ctx,
 161                                            struct ext2_super_block *s,
 162                                            ext2_ino_t journal_inum,
 163                                            journal_t **journal)
 164 {
 165         struct inode *inode;
 166         struct buffer_head *bh;
 167         blk_t start;
 168         int retval;
 169
 170         jfs_debug(1, "Using journal inode %u\n", journal_inum);
 171         *journal = e2fsck_allocate_memory(ctx, sizeof(journal_t), "journal");
 172         if (!*journal) {
 173                 return EXT2_ET_NO_MEMORY;
 174         }
 175
 176         inode = e2fsck_allocate_memory(ctx, sizeof(*inode), "journal inode");
 177         if (!inode) {
 178                 retval = EXT2_ET_NO_MEMORY;
 179                 goto exit_journal;
 180         }
 181
 182         inode->i_ctx = ctx;
 183         inode->i_ino = journal_inum;
 184         retval = ext2fs_read_inode(ctx->fs, journal_inum, &inode->i_ext2);
 185         if (retval)
 186                 goto exit_inode;
 187
 188         (*journal)->j_dev = ctx;
 189         (*journal)->j_inode = inode;
 190         (*journal)->j_blocksize = ctx->fs->blocksize;
 191         (*journal)->j_maxlen = inode->i_ext2.i_size / (*journal)->j_blocksize;
 192
 193         if (!inode->i_ext2.i_links_count ||
 194             !LINUX_S_ISREG(inode->i_ext2.i_mode) ||
 195             (*journal)->j_maxlen < JFS_MIN_JOURNAL_BLOCKS ||
 196             (start = bmap(inode, 0)) == 0) {
 197                 retval = EXT2_ET_BAD_INODE_NUM;
 198                 goto exit_inode;
 199         }
 200
 201         bh = getblk(ctx, start, (*journal)->j_blocksize);
 202         if (!bh) {
 203                 retval = EXT2_ET_NO_MEMORY;
 204                 goto exit_inode;
 205         }
 206         (*journal)->j_sb_buffer = bh;
 207         (*journal)->j_superblock = (journal_superblock_t *)bh->b_data;
 208
 209         return 0;
 210
 211 exit_inode:
 212         ext2fs_free_mem((void **)&inode);
 213 exit_journal:
 214         ext2fs_free_mem((void **)journal);
 215
 216         return retval;
 217 }
 218
 219 static errcode_t e2fsck_get_journal(e2fsck_t ctx, journal_t **journal)
 220 {
 221         char uuid_str[40];
 222         struct problem_context pctx;
 223         struct ext2_super_block *sb = ctx->fs->super;
 224
 225         clear_problem_context(&pctx);
 226
 227         if (sb->s_feature_compat & EXT3_FEATURE_COMPAT_HAS_JOURNAL) {
 228                 /* FIXME: check if dev is valid block dev, has a journal */
 229                 if (sb->s_journal_dev) {
 230                         pctx.num = sb->s_journal_dev;
 231                         /* this problem aborts on -y, -p, unsupported on -n */
 232                         if (!fix_problem(ctx, PR_0_JOURNAL_UNSUPP_DEV, &pctx))
 233                                 return EXT2_ET_UNSUPP_FEATURE;
 234                         sb->s_journal_dev = 0;
 235                         sb->s_state &= ~EXT2_VALID_FS;
 236                         ext2fs_mark_super_dirty(ctx->fs);
 237                 }
 238                 /* FIXME: check if UUID is valid block dev, has a journal */
 239                 if (!uuid_is_null(sb->s_journal_uuid)) {
 240                         uuid_unparse(sb->s_journal_uuid, uuid_str);
 241                         pctx.str = uuid_str;
 242                         /* this problem aborts on -y, -p, unsupported on -n */
 243                         if (!fix_problem(ctx, PR_0_JOURNAL_UNSUPP_UUID, &pctx))
 244                                 return EXT2_ET_UNSUPP_FEATURE;
 245                         uuid_clear(sb->s_journal_uuid);
 246                         sb->s_state &= ~EXT2_VALID_FS;
 247                         ext2fs_mark_super_dirty(ctx->fs);
 248                 }
 249                 if (!sb->s_journal_inum)
 250                         return EXT2_ET_BAD_INODE_NUM;
 251         }
 252
 253         if (sb->s_journal_dev) {
 254                 pctx.num = sb->s_journal_dev;
 255                 if (!fix_problem(ctx, PR_0_JOURNAL_BAD_DEV, &pctx))
 256                         return EXT2_ET_UNSUPP_FEATURE;
 257                 sb->s_journal_dev = 0;
 258                 sb->s_state &= ~EXT2_VALID_FS;
 259                 ext2fs_mark_super_dirty(ctx->fs);
 260         }
 261         if (!uuid_is_null(sb->s_journal_uuid)) {
 262                 uuid_unparse(sb->s_journal_uuid, uuid_str);
 263                 pctx.str = uuid_str;
 264                 if (!fix_problem(ctx, PR_0_JOURNAL_BAD_UUID, &pctx))
 265                         return EXT2_ET_UNSUPP_FEATURE;
 266                 uuid_clear(sb->s_journal_uuid);
 267                 sb->s_state &= ~EXT2_VALID_FS;
 268                 ext2fs_mark_super_dirty(ctx->fs);
 269         }
 270
 271         return e2fsck_journal_init_inode(ctx, sb, sb->s_journal_inum, journal);
 272 }
 273
 274 static errcode_t e2fsck_journal_fix_bad_inode(e2fsck_t ctx,
 275                                               struct problem_context *pctx)
 276 {
 277         struct ext2_super_block *sb = ctx->fs->super;
 278         int recover = ctx->fs->super->s_feature_incompat &
 279                 EXT3_FEATURE_INCOMPAT_RECOVER;
 280         int has_journal = ctx->fs->super->s_feature_compat &
 281                 EXT3_FEATURE_COMPAT_HAS_JOURNAL;
 282
 283         if (has_journal || sb->s_journal_inum) {
 284                 /* The journal inode is bogus, remove and force full fsck */
 285                 pctx->ino = sb->s_journal_inum;
 286                 if (fix_problem(ctx, PR_0_JOURNAL_BAD_INODE, pctx)) {
 287                         if (has_journal && sb->s_journal_inum)
 288                                 printf("*** ext3 journal has been deleted - "
 289                                        "filesystem is now ext2 only ***\n\n");
 290                         sb->s_feature_compat &= ~EXT3_FEATURE_COMPAT_HAS_JOURNAL;
 291                         sb->s_journal_inum = 0;
 292                         ctx->flags |= E2F_FLAG_JOURNAL_INODE; /* FIXME: todo */
 293                         e2fsck_clear_recover(ctx, 1);
 294                         return 0;
 295                 }
 296                 return EXT2_ET_BAD_INODE_NUM;
 297         } else if (recover) {
 298                 if (fix_problem(ctx, PR_0_JOURNAL_RECOVER_SET, pctx)) {
 299                         e2fsck_clear_recover(ctx, 1);
 300                         return 0;
 301                 }
 302                 return EXT2_ET_UNSUPP_FEATURE;
 303         }
 304         return 0;
 305 }
 306
 307 static errcode_t e2fsck_journal_load(journal_t *journal)
 308 {
 309         e2fsck_t ctx = journal->j_dev;
 310         journal_superblock_t *jsb;
 311         struct buffer_head *jbh = journal->j_sb_buffer;
 312         struct problem_context pctx;
 313
 314         clear_problem_context(&pctx);
 315
 316         ll_rw_block(READ, 1, &jbh);
 317         if (jbh->b_err) {
 318                 com_err(ctx->device_name, jbh->b_err,
 319                         _("reading journal superblock\n"));
 320                 return jbh->b_err;
 321         }
 322
 323         jsb = journal->j_superblock;
 324         /* If we don't even have JFS_MAGIC, we probably have a wrong inode */
 325         if (jsb->s_header.h_magic != htonl(JFS_MAGIC_NUMBER))
 326                 return e2fsck_journal_fix_bad_inode(ctx, &pctx);
 327
 328         switch (ntohl(jsb->s_header.h_blocktype)) {
 329         case JFS_SUPERBLOCK_V1:
 330                 journal->j_format_version = 1;
 331                 break;
 332
 333         case JFS_SUPERBLOCK_V2:
 334                 journal->j_format_version = 2;
 335                 break;
 336
 337         /*
 338          * These should never appear in a journal super block, so if
 339          * they do, the journal is badly corrupted.
 340          */
 341         case JFS_DESCRIPTOR_BLOCK:
 342         case JFS_COMMIT_BLOCK:
 343         case JFS_REVOKE_BLOCK:
 344                 return EXT2_ET_CORRUPT_SUPERBLOCK;
 345
 346         /* If we don't understand the superblock major type, but there
 347          * is a magic number, then it is likely to be a new format we
 348          * just don't understand, so leave it alone. */
 349         default:
 350                 return EXT2_ET_JOURNAL_UNSUPP_VERSION;
 351         }
 352
 353         if (JFS_HAS_INCOMPAT_FEATURE(journal, ~JFS_KNOWN_INCOMPAT_FEATURES))
 354                 return EXT2_ET_UNSUPP_FEATURE;
 355
 356         if (JFS_HAS_RO_COMPAT_FEATURE(journal, ~JFS_KNOWN_ROCOMPAT_FEATURES))
 357                 return EXT2_ET_RO_UNSUPP_FEATURE;
 358
 359         /* We have now checked whether we know enough about the journal
 360          * format to be able to proceed safely, so any other checks that
 361          * fail we should attempt to recover from. */
 362         if (jsb->s_blocksize != htonl(journal->j_blocksize)) {
 363                 com_err(ctx->program_name, EXT2_ET_CORRUPT_SUPERBLOCK,
 364                         _("%s: no valid journal superblock found\n"),
 365                         ctx->device_name);
 366                 return EXT2_ET_CORRUPT_SUPERBLOCK;
 367         }
 368
 369         if (ntohl(jsb->s_maxlen) < journal->j_maxlen)
 370                 journal->j_maxlen = ntohl(jsb->s_maxlen);
 371         else if (ntohl(jsb->s_maxlen) > journal->j_maxlen) {
 372                 com_err(ctx->program_name, EXT2_ET_CORRUPT_SUPERBLOCK,
 373                         _("%s: journal too short\n"),
 374                         ctx->device_name);
 375                 return EXT2_ET_CORRUPT_SUPERBLOCK;
 376         }
 377
 378         journal->j_tail_sequence = ntohl(jsb->s_sequence);
 379         journal->j_transaction_sequence = journal->j_tail_sequence;
 380         journal->j_tail = ntohl(jsb->s_start);
 381         journal->j_first = ntohl(jsb->s_first);
 382         journal->j_last = ntohl(jsb->s_maxlen);
 383
 384         return 0;
 385 }
 386
 387 static void e2fsck_journal_reset_super(e2fsck_t ctx, journal_superblock_t *jsb,
 388                                        journal_t *journal)
 389 {
 390         char *p;
 391         union {
 392                 uuid_t uuid;
 393                 __u32 val[4];
 394         } u;
 395         __u32 new_seq = 0;
 396         int i;
 397
 398         /* Leave a valid existing V1 superblock signature alone.
 399          * Anything unrecognisable we overwrite with a new V2
 400          * signature. */
 401
 402         if (jsb->s_header.h_magic != htonl(JFS_MAGIC_NUMBER) ||
 403             jsb->s_header.h_blocktype != htonl(JFS_SUPERBLOCK_V1)) {
 404                 jsb->s_header.h_magic = htonl(JFS_MAGIC_NUMBER);
 405                 jsb->s_header.h_blocktype = htonl(JFS_SUPERBLOCK_V2);
 406         }
 407
 408         /* Zero out everything else beyond the superblock header */
 409
 410         p = ((char *) jsb) + sizeof(journal_header_t);
 411         memset (p, 0, ctx->fs->blocksize-sizeof(journal_header_t));
 412
 413         jsb->s_blocksize = htonl(ctx->fs->blocksize);
 414         jsb->s_maxlen = htonl(journal->j_maxlen);
 415         jsb->s_first = htonl(1);
 416
 417         /* Initialize the journal sequence number so that there is "no"
 418          * chance we will find old "valid" transactions in the journal.
 419          * This avoids the need to zero the whole journal (slow to do,
 420          * and risky when we are just recovering the filesystem).
 421          */
 422         uuid_generate(u.uuid);
 423         for (i = 0; i < 4; i ++)
 424                 new_seq ^= u.val[i];
 425         jsb->s_sequence = htonl(new_seq);
 426
 427         mark_buffer_dirty(journal->j_sb_buffer, 1);
 428         ll_rw_block(WRITE, 1, &journal->j_sb_buffer);
 429 }
 430
 431 static errcode_t e2fsck_journal_fix_corrupt_super(e2fsck_t ctx,
 432                                                   journal_t *journal,
 433                                                   struct problem_context *pctx)
 434 {
 435         struct ext2_super_block *sb = ctx->fs->super;
 436         int recover = ctx->fs->super->s_feature_incompat &
 437                 EXT3_FEATURE_INCOMPAT_RECOVER;
 438
 439         pctx->num = journal->j_inode->i_ino;
 440
 441         if (sb->s_feature_compat & EXT3_FEATURE_COMPAT_HAS_JOURNAL) {
 442                 if (fix_problem(ctx, PR_0_JOURNAL_BAD_SUPER, pctx)) {
 443                         e2fsck_journal_reset_super(ctx, journal->j_superblock,
 444                                                    journal);
 445                         journal->j_transaction_sequence = 1;
 446                         e2fsck_clear_recover(ctx, recover);
 447                         return 0;
 448                 }
 449                 return EXT2_ET_CORRUPT_SUPERBLOCK;
 450         } else if (e2fsck_journal_fix_bad_inode(ctx, pctx))
 451                 return EXT2_ET_CORRUPT_SUPERBLOCK;
 452
 453         return 0;
 454 }
 455
 456 static void e2fsck_journal_release(e2fsck_t ctx, journal_t *journal,
 457                                    int reset, int drop)
 458 {
 459         journal_superblock_t *jsb;
 460
 461         if (drop)
 462                 mark_buffer_clean(journal->j_sb_buffer);
 463         else if (!(ctx->options & E2F_OPT_READONLY)) {
 464                 jsb = journal->j_superblock;
 465                 jsb->s_sequence = htonl(journal->j_transaction_sequence);
 466                 if (reset)
 467                         jsb->s_start = 0; /* this marks the journal as empty */
 468                 mark_buffer_dirty(journal->j_sb_buffer, 1);
 469         }
 470         brelse(journal->j_sb_buffer);
 471
 472         if (journal->j_inode)
 473                 ext2fs_free_mem((void **)&journal->j_inode);
 474         ext2fs_free_mem((void **)&journal);
 475 }
 476
 477 /*
 478  * This function makes sure that the superblock fields regarding the
 479  * journal are consistent.
 480  */
 481 int e2fsck_check_ext3_journal(e2fsck_t ctx)
 482 {
 483         struct ext2_super_block *sb = ctx->fs->super;
 484         journal_t *journal;
 485         int recover = ctx->fs->super->s_feature_incompat &
 486                 EXT3_FEATURE_INCOMPAT_RECOVER;
 487         struct problem_context pctx;
 488         int reset = 0, force_fsck = 0;
 489         int retval;
 490
 491         /* If we don't have any journal features, don't do anything more */
 492         if (!(sb->s_feature_compat & EXT3_FEATURE_COMPAT_HAS_JOURNAL) &&
 493             !recover && sb->s_journal_inum == 0 && sb->s_journal_dev == 0 &&
 494             uuid_is_null(sb->s_journal_uuid))
 495                 return 0;
 496
 497         clear_problem_context(&pctx);
 498         pctx.num = sb->s_journal_inum;
 499
 500         retval = e2fsck_get_journal(ctx, &journal);
 501         if (retval) {
 502                 if (retval == EXT2_ET_BAD_INODE_NUM)
 503                         return e2fsck_journal_fix_bad_inode(ctx, &pctx);
 504                 return retval;
 505         }
 506
 507         retval = e2fsck_journal_load(journal);
 508         if (retval) {
 509                 if ((retval == EXT2_ET_CORRUPT_SUPERBLOCK) ||
 510                     ((retval == EXT2_ET_UNSUPP_FEATURE) &&
 511                     (!fix_problem(ctx, PR_0_JOURNAL_UNSUPP_INCOMPAT,
 512                                   &pctx))) ||
 513                     ((retval == EXT2_ET_RO_UNSUPP_FEATURE) &&
 514                     (!fix_problem(ctx, PR_0_JOURNAL_UNSUPP_ROCOMPAT,
 515                                   &pctx))) ||
 516                     ((retval == EXT2_ET_JOURNAL_UNSUPP_VERSION) &&
 517                     (!fix_problem(ctx, PR_0_JOURNAL_UNSUPP_VERSION, &pctx))))
 518                         retval = e2fsck_journal_fix_corrupt_super(ctx, journal,
 519                                                                   &pctx);
 520                 e2fsck_journal_release(ctx, journal, 0, 1);
 521                 return retval;
 522         }
 523
 524         /*
 525          * We want to make the flags consistent here.  We will not leave with
 526          * needs_recovery set but has_journal clear.  We can't get in a loop
 527          * with -y, -n, or -p, only if a user isn't making up their mind.
 528          */
 529 no_has_journal:
 530         if (!(sb->s_feature_compat & EXT3_FEATURE_COMPAT_HAS_JOURNAL)) {
 531                 recover = sb->s_feature_incompat & EXT3_FEATURE_INCOMPAT_RECOVER;
 532                 pctx.str = "inode";
 533                 if (fix_problem(ctx, PR_0_JOURNAL_HAS_JOURNAL, &pctx)) {
 534                         if (recover &&
 535                             !fix_problem(ctx, PR_0_JOURNAL_RECOVER_SET, &pctx))
 536                                 goto no_has_journal;
 537                         /*
 538                          * Need a full fsck if we are releasing a
 539                          * journal stored on a reserved inode.
 540                          */
 541                         force_fsck = recover ||
 542                                 (sb->s_journal_inum < EXT2_FIRST_INODE(sb));
 543                         /* Clear all of the journal fields */
 544                         sb->s_journal_inum = 0;
 545                         sb->s_journal_dev = 0;
 546                         memset(sb->s_journal_uuid, 0,
 547                                sizeof(sb->s_journal_uuid));
 548                         e2fsck_clear_recover(ctx, force_fsck);
 549                 } else if (!(ctx->options & E2F_OPT_READONLY)) {
 550                         sb->s_feature_compat |= EXT3_FEATURE_COMPAT_HAS_JOURNAL;
 551                         ext2fs_mark_super_dirty(ctx->fs);
 552                 }
 553         }
 554
 555         if (sb->s_feature_compat & EXT3_FEATURE_COMPAT_HAS_JOURNAL &&
 556             !(sb->s_feature_incompat & EXT3_FEATURE_INCOMPAT_RECOVER) &&
 557             journal->j_superblock->s_start != 0) {
 558                 if (fix_problem(ctx, PR_0_JOURNAL_RESET_JOURNAL, &pctx)) {
 559                         reset = 1;
 560                         sb->s_state &= ~EXT2_VALID_FS;
 561                         ext2fs_mark_super_dirty(ctx->fs);
 562                 }
 563                 /*
 564                  * If the user answers no to the above question, we
 565                  * ignore the fact that journal apparently has data;
 566                  * accidentally replaying over valid data would be far
 567                  * worse than skipping a questionable recovery.
 568                  *
 569                  * XXX should we abort with a fatal error here?  What
 570                  * will the ext3 kernel code do if a filesystem with
 571                  * !NEEDS_RECOVERY but with a non-zero
 572                  * journal->j_superblock->s_start is mounted?
 573                  */
 574         }
 575
 576         e2fsck_journal_release(ctx, journal, reset, 0);
 577         return retval;
 578 }
 579
 580 static errcode_t recover_ext3_journal(e2fsck_t ctx)
 581 {
 582         journal_t *journal;
 583         int retval;
 584
 585         retval = e2fsck_get_journal(ctx, &journal);
 586         if (retval)
 587                 return retval;
 588
 589         retval = e2fsck_journal_load(journal);
 590         if (retval)
 591                 goto errout;
 592
 593         retval = journal_init_revoke(journal, 1024);
 594         if (retval)
 595                 goto errout;
 596
 597         retval = -journal_recover(journal);
 598 errout:
 599         e2fsck_journal_release(ctx, journal, 1, 0);
 600         return retval;
 601 }
 602
 603
 604 #if 0
 605 #define TEMPLATE "/tmp/ext3.XXXXXX"
 606
 607 /*
 608  * This function attempts to mount and unmount an ext3 filesystem,
 609  * which is a cheap way to force the kernel to run the journal and
 610  * handle the recovery for us.
 611  */
 612 static errcode_t recover_ext3_journal_via_mount(e2fsck_t ctx)
 613 {
 614         ext2_filsys fs = ctx->fs;
 615         char    *dirlist[] = {"/mnt","/lost+found","/tmp","/root","/boot",0};
 616         errcode_t        retval, retval2;
 617         int      count = 0;
 618         char     template[] = TEMPLATE;
 619         struct stat buf;
 620         char    *tmpdir;
 621
 622         if (ctx->options & E2F_OPT_READONLY) {
 623                 printf("%s: won't do journal recovery while read-only\n",
 624                        ctx->device_name);
 625                 return EXT2_ET_FILE_RO;
 626         }
 627
 628         printf(_("%s: trying for ext3 kernel journal recovery\n"),
 629                ctx->device_name);
 630         /*
 631          * First try to make a temporary directory.  This may fail if
 632          * the root partition is still mounted read-only.
 633          */
 634 newtemp:
 635         tmpdir = mktemp(template);
 636         if (tmpdir) {
 637                 jfs_debug(2, "trying %s as ext3 temp mount point\n", tmpdir);
 638                 if (mkdir(template, 0700)) {
 639                         if (errno == EROFS) {
 640                                 tmpdir = NULL;
 641                                 template[0] = '\0';
 642                         } else if (errno == EEXIST && count++ < 10) {
 643                                 strcpy(template, TEMPLATE);
 644                                 goto newtemp;
 645                         }
 646                         return errno;
 647                 }
 648         }
 649
 650         /*
 651          * OK, creating a temporary directory didn't work.
 652          * Let's try a list of possible temporary mountpoints.
 653          */
 654         if (!tmpdir) {
 655                 dev_t   rootdev;
 656                 char    **cpp, *dir;
 657
 658                 if (stat("/", &buf))
 659                         return errno;
 660
 661                 rootdev = buf.st_dev;
 662
 663                 /*
 664                  * Check that dir is on the same device as root (no other
 665                  * filesystem is mounted there), and it's a directory.
 666                  */
 667                 for (cpp = dirlist; (dir = *cpp); cpp++)
 668                         if (stat(dir, &buf) == 0 && buf.st_dev == rootdev &&
 669                             S_ISDIR(buf.st_mode)) {
 670                                 tmpdir = dir;
 671                                 break;
 672                         }
 673         }
 674
 675         if (tmpdir) {
 676                 io_manager      io_ptr = fs->io->manager;
 677                 int             blocksize = fs->blocksize;
 678
 679                 jfs_debug(2, "using %s for ext3 mount\n", tmpdir);
 680                 /* FIXME - need to handle loop devices here */
 681                 if (mount(ctx->device_name, tmpdir, "ext3", MNT_FL, NULL)) {
 682                         retval = errno;
 683                         com_err(ctx->program_name, errno,
 684                                 "when mounting %s", ctx->device_name);
 685                         if (template[0])
 686                                 rmdir(tmpdir);
 687                         return retval;
 688                 }
 689                 /*
 690                  * Now that it mounted cleanly, the filesystem will have been
 691                  * recovered, so we can now unmount it.
 692                  */
 693                 if (umount(tmpdir))
 694                         return errno;
 695
 696                 /*
 697                  * Remove the temporary directory, if it was created.
 698                  */
 699                 if (template[0])
 700                         rmdir(tmpdir);
 701                 return 0;
 702         }
 703 }
 704 #endif
 705
 706 int e2fsck_run_ext3_journal(e2fsck_t ctx)
 707 {
 708         io_manager io_ptr = ctx->fs->io->manager;
 709         int blocksize = ctx->fs->blocksize;
 710         errcode_t       retval, recover_retval;
 711
 712         printf(_("%s: recovering journal\n"), ctx->device_name);
 713         if (ctx->options & E2F_OPT_READONLY) {
 714                 printf(_("%s: won't do journal recovery while read-only\n"),
 715                        ctx->device_name);
 716                 return EXT2_ET_FILE_RO;
 717         }
 718
 719         if (ctx->fs->flags & EXT2_FLAG_DIRTY)
 720                 ext2fs_flush(ctx->fs);  /* Force out any modifications */
 721
 722         recover_retval = recover_ext3_journal(ctx);
 723
 724         /*
 725          * Reload the filesystem context to get up-to-date data from disk
 726          * because journal recovery will change the filesystem under us.
 727          */
 728         ext2fs_close(ctx->fs);
 729         retval = ext2fs_open(ctx->filesystem_name, EXT2_FLAG_RW,
 730                              ctx->superblock, blocksize, io_ptr,
 731                              &ctx->fs);
 732
 733         if (retval) {
 734                 com_err(ctx->program_name, retval,
 735                         _("while trying to re-open %s"),
 736                         ctx->device_name);
 737                 fatal_error(ctx, 0);
 738         }
 739         ctx->fs->priv_data = ctx;
 740
 741         /* Set the superblock flags */
 742         e2fsck_clear_recover(ctx, recover_retval);
 743         return recover_retval;
 744 }