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->journal_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->journal_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                                            journal_t **journal)
 163 {
 164         struct inode *inode;
 165         struct buffer_head *bh;
 166         blk_t start;
 167         int retval;
 168
 169         jfs_debug(1, "Using journal inode %u\n", s->s_journal_inum);
 170         *journal = e2fsck_allocate_memory(ctx, sizeof(journal_t), "journal");
 171         if (!*journal) {
 172                 return EXT2_ET_NO_MEMORY;
 173         }
 174
 175         inode = e2fsck_allocate_memory(ctx, sizeof(*inode), "journal inode");
 176         if (!inode) {
 177                 retval = EXT2_ET_NO_MEMORY;
 178                 goto exit_journal;
 179         }
 180
 181         inode->i_ctx = ctx;
 182         inode->i_ino = s->s_journal_inum;
 183         retval = ext2fs_read_inode(ctx->fs, s->s_journal_inum, &inode->i_ext2);
 184         if (retval)
 185                 goto exit_inode;
 186
 187         (*journal)->j_dev = ctx;
 188         (*journal)->j_inode = inode;
 189         (*journal)->j_blocksize = ctx->fs->blocksize;
 190         (*journal)->j_maxlen = inode->i_ext2.i_size / (*journal)->j_blocksize;
 191         ctx->journal_io = ctx->fs->io;
 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_journal_init_dev(e2fsck_t ctx,
 220                                          struct ext2_super_block *s,
 221                                          journal_t **journal)
 222 {
 223         struct buffer_head *bh;
 224         io_manager      io_ptr;
 225         blk_t           start;
 226         int             retval;
 227         int             blocksize = ctx->fs->blocksize;
 228         struct ext2_super_block jsuper;
 229         struct problem_context pctx;
 230         const char      *journal_name;
 231
 232         clear_problem_context(&pctx);
 233         journal_name = ctx->journal_name;
 234         if (!journal_name)
 235                 journal_name = ext2fs_find_block_device(s->s_journal_dev);
 236
 237         if (!journal_name) {
 238                 fix_problem(ctx, PR_0_CANT_FIND_JOURNAL, &pctx);
 239                 return EXT2_ET_LOAD_EXT_JOURNAL;
 240         }
 241
 242         jfs_debug(1, "Using journal file %s\n", journal_name);
 243
 244 #if 1
 245         io_ptr = unix_io_manager;
 246 #else
 247         io_ptr = test_io_manager;
 248         test_io_backing_manager = unix_io_manager;
 249 #endif
 250         retval = io_ptr->open(journal_name, IO_FLAG_RW, &ctx->journal_io);
 251         if (!ctx->journal_name)
 252                 free((void *) journal_name);
 253         if (retval)
 254                 return retval;
 255
 256         io_channel_set_blksize(ctx->journal_io, blocksize);
 257         start = (blocksize == 1024) ? 1 : 0;
 258         bh = getblk(ctx, start, blocksize);
 259         if (!bh)
 260                 return EXT2_ET_NO_MEMORY;
 261         ll_rw_block(READ, 1, &bh);
 262         if (bh->b_err)
 263                 return bh->b_err;
 264         memcpy(&jsuper, start ? bh->b_data :  bh->b_data + 1024,
 265                sizeof(jsuper));
 266         brelse(bh);
 267 #ifdef EXT2FS_ENABLE_SWAPFS
 268         if (jsuper.s_magic == ext2fs_swab16(EXT2_SUPER_MAGIC))
 269                 ext2fs_swap_super(&jsuper);
 270 #endif
 271         if (jsuper.s_magic != EXT2_SUPER_MAGIC ||
 272             !(jsuper.s_feature_incompat & EXT3_FEATURE_INCOMPAT_JOURNAL_DEV)) {
 273                 fix_problem(ctx, PR_0_EXT_JOURNAL_BAD_SUPER, &pctx);
 274                 return EXT2_ET_LOAD_EXT_JOURNAL;
 275         }
 276         /* Make sure the journal UUID is correct */
 277         if (memcmp(jsuper.s_uuid, ctx->fs->super->s_journal_uuid,
 278                    sizeof(jsuper.s_uuid))) {
 279                 fix_problem(ctx, PR_0_JOURNAL_BAD_UUID, &pctx);
 280                 return EXT2_ET_LOAD_EXT_JOURNAL;
 281         }
 282
 283         *journal = e2fsck_allocate_memory(ctx, sizeof(journal_t), "journal");
 284         if (!*journal) {
 285                 return EXT2_ET_NO_MEMORY;
 286         }
 287
 288         (*journal)->j_dev = ctx;
 289         (*journal)->j_inode = NULL;
 290         (*journal)->j_blocksize = ctx->fs->blocksize;
 291         (*journal)->j_maxlen = jsuper.s_blocks_count;
 292
 293         bh = getblk(ctx, start+1, (*journal)->j_blocksize);
 294         if (!bh) {
 295                 retval = EXT2_ET_NO_MEMORY;
 296                 goto errout;
 297         }
 298         (*journal)->j_sb_buffer = bh;
 299         (*journal)->j_superblock = (journal_superblock_t *)bh->b_data;
 300
 301         return 0;
 302
 303 errout:
 304         ext2fs_free_mem((void **)journal);
 305         return retval;
 306 }
 307
 308 static errcode_t e2fsck_get_journal(e2fsck_t ctx, journal_t **journal)
 309 {
 310         struct ext2_super_block *sb = ctx->fs->super;
 311
 312         if (uuid_is_null(sb->s_journal_uuid)) {
 313                 if (!sb->s_journal_inum)
 314                         return EXT2_ET_BAD_INODE_NUM;
 315                 return e2fsck_journal_init_inode(ctx, sb, journal);
 316         } else {
 317                 return e2fsck_journal_init_dev(ctx, sb, journal);
 318         }
 319 }
 320
 321 static errcode_t e2fsck_journal_fix_bad_inode(e2fsck_t ctx,
 322                                               struct problem_context *pctx)
 323 {
 324         struct ext2_super_block *sb = ctx->fs->super;
 325         int recover = ctx->fs->super->s_feature_incompat &
 326                 EXT3_FEATURE_INCOMPAT_RECOVER;
 327         int has_journal = ctx->fs->super->s_feature_compat &
 328                 EXT3_FEATURE_COMPAT_HAS_JOURNAL;
 329
 330         if (has_journal || sb->s_journal_inum) {
 331                 /* The journal inode is bogus, remove and force full fsck */
 332                 pctx->ino = sb->s_journal_inum;
 333                 if (fix_problem(ctx, PR_0_JOURNAL_BAD_INODE, pctx)) {
 334                         if (has_journal && sb->s_journal_inum)
 335                                 printf("*** ext3 journal has been deleted - "
 336                                        "filesystem is now ext2 only ***\n\n");
 337                         sb->s_feature_compat &= ~EXT3_FEATURE_COMPAT_HAS_JOURNAL;
 338                         sb->s_journal_inum = 0;
 339                         ctx->flags |= E2F_FLAG_JOURNAL_INODE; /* FIXME: todo */
 340                         e2fsck_clear_recover(ctx, 1);
 341                         return 0;
 342                 }
 343                 return EXT2_ET_BAD_INODE_NUM;
 344         } else if (recover) {
 345                 if (fix_problem(ctx, PR_0_JOURNAL_RECOVER_SET, pctx)) {
 346                         e2fsck_clear_recover(ctx, 1);
 347                         return 0;
 348                 }
 349                 return EXT2_ET_UNSUPP_FEATURE;
 350         }
 351         return 0;
 352 }
 353
 354 #define V1_SB_SIZE      0x0024
 355 static void clear_v2_journal_fields(journal_t *journal)
 356 {
 357         e2fsck_t ctx = journal->j_dev;
 358         struct buffer_head *jbh = journal->j_sb_buffer;
 359         struct problem_context pctx;
 360
 361         clear_problem_context(&pctx);
 362
 363         if (!fix_problem(ctx, PR_0_CLEAR_V2_JOURNAL, &pctx))
 364                 return;
 365
 366         memset(((char *) journal->j_superblock) + V1_SB_SIZE, 0,
 367                ctx->fs->blocksize-V1_SB_SIZE);
 368         mark_buffer_dirty(journal->j_sb_buffer, 1);
 369 }
 370
 371
 372 static errcode_t e2fsck_journal_load(journal_t *journal)
 373 {
 374         e2fsck_t ctx = journal->j_dev;
 375         journal_superblock_t *jsb;
 376         struct buffer_head *jbh = journal->j_sb_buffer;
 377         struct problem_context pctx;
 378
 379         clear_problem_context(&pctx);
 380
 381         ll_rw_block(READ, 1, &jbh);
 382         if (jbh->b_err) {
 383                 com_err(ctx->device_name, jbh->b_err,
 384                         _("reading journal superblock\n"));
 385                 return jbh->b_err;
 386         }
 387
 388         jsb = journal->j_superblock;
 389         /* If we don't even have JFS_MAGIC, we probably have a wrong inode */
 390         if (jsb->s_header.h_magic != htonl(JFS_MAGIC_NUMBER))
 391                 return e2fsck_journal_fix_bad_inode(ctx, &pctx);
 392
 393         switch (ntohl(jsb->s_header.h_blocktype)) {
 394         case JFS_SUPERBLOCK_V1:
 395                 journal->j_format_version = 1;
 396                 if (jsb->s_feature_compat ||
 397                     jsb->s_feature_incompat ||
 398                     jsb->s_feature_ro_compat ||
 399                     jsb->s_nr_users)
 400                         clear_v2_journal_fields(journal);
 401                 break;
 402
 403         case JFS_SUPERBLOCK_V2:
 404                 journal->j_format_version = 2;
 405                 if (jsb->s_nr_users > 1 &&
 406                     (ctx->fs->io == ctx->journal_io))
 407                         clear_v2_journal_fields(journal);
 408                 if (ntohl(jsb->s_nr_users) > 1) {
 409                         fix_problem(ctx, PR_0_JOURNAL_UNSUPP_MULTIFS, &pctx);
 410                         return EXT2_ET_JOURNAL_UNSUPP_VERSION;
 411                 }
 412                 break;
 413
 414         /*
 415          * These should never appear in a journal super block, so if
 416          * they do, the journal is badly corrupted.
 417          */
 418         case JFS_DESCRIPTOR_BLOCK:
 419         case JFS_COMMIT_BLOCK:
 420         case JFS_REVOKE_BLOCK:
 421                 return EXT2_ET_CORRUPT_SUPERBLOCK;
 422
 423         /* If we don't understand the superblock major type, but there
 424          * is a magic number, then it is likely to be a new format we
 425          * just don't understand, so leave it alone. */
 426         default:
 427                 return EXT2_ET_JOURNAL_UNSUPP_VERSION;
 428         }
 429
 430         if (JFS_HAS_INCOMPAT_FEATURE(journal, ~JFS_KNOWN_INCOMPAT_FEATURES))
 431                 return EXT2_ET_UNSUPP_FEATURE;
 432
 433         if (JFS_HAS_RO_COMPAT_FEATURE(journal, ~JFS_KNOWN_ROCOMPAT_FEATURES))
 434                 return EXT2_ET_RO_UNSUPP_FEATURE;
 435
 436         /* We have now checked whether we know enough about the journal
 437          * format to be able to proceed safely, so any other checks that
 438          * fail we should attempt to recover from. */
 439         if (jsb->s_blocksize != htonl(journal->j_blocksize)) {
 440                 com_err(ctx->program_name, EXT2_ET_CORRUPT_SUPERBLOCK,
 441                         _("%s: no valid journal superblock found\n"),
 442                         ctx->device_name);
 443                 return EXT2_ET_CORRUPT_SUPERBLOCK;
 444         }
 445
 446         if (ntohl(jsb->s_maxlen) < journal->j_maxlen)
 447                 journal->j_maxlen = ntohl(jsb->s_maxlen);
 448         else if (ntohl(jsb->s_maxlen) > journal->j_maxlen) {
 449                 com_err(ctx->program_name, EXT2_ET_CORRUPT_SUPERBLOCK,
 450                         _("%s: journal too short\n"),
 451                         ctx->device_name);
 452                 return EXT2_ET_CORRUPT_SUPERBLOCK;
 453         }
 454
 455         journal->j_tail_sequence = ntohl(jsb->s_sequence);
 456         journal->j_transaction_sequence = journal->j_tail_sequence;
 457         journal->j_tail = ntohl(jsb->s_start);
 458         journal->j_first = ntohl(jsb->s_first);
 459         journal->j_last = ntohl(jsb->s_maxlen);
 460
 461         return 0;
 462 }
 463
 464 static void e2fsck_journal_reset_super(e2fsck_t ctx, journal_superblock_t *jsb,
 465                                        journal_t *journal)
 466 {
 467         char *p;
 468         union {
 469                 uuid_t uuid;
 470                 __u32 val[4];
 471         } u;
 472         __u32 new_seq = 0;
 473         int i;
 474
 475         /* Leave a valid existing V1 superblock signature alone.
 476          * Anything unrecognisable we overwrite with a new V2
 477          * signature. */
 478
 479         if (jsb->s_header.h_magic != htonl(JFS_MAGIC_NUMBER) ||
 480             jsb->s_header.h_blocktype != htonl(JFS_SUPERBLOCK_V1)) {
 481                 jsb->s_header.h_magic = htonl(JFS_MAGIC_NUMBER);
 482                 jsb->s_header.h_blocktype = htonl(JFS_SUPERBLOCK_V2);
 483         }
 484
 485         /* Zero out everything else beyond the superblock header */
 486
 487         p = ((char *) jsb) + sizeof(journal_header_t);
 488         memset (p, 0, ctx->fs->blocksize-sizeof(journal_header_t));
 489
 490         jsb->s_blocksize = htonl(ctx->fs->blocksize);
 491         jsb->s_maxlen = htonl(journal->j_maxlen);
 492         jsb->s_first = htonl(1);
 493
 494         /* Initialize the journal sequence number so that there is "no"
 495          * chance we will find old "valid" transactions in the journal.
 496          * This avoids the need to zero the whole journal (slow to do,
 497          * and risky when we are just recovering the filesystem).
 498          */
 499         uuid_generate(u.uuid);
 500         for (i = 0; i < 4; i ++)
 501                 new_seq ^= u.val[i];
 502         jsb->s_sequence = htonl(new_seq);
 503
 504         mark_buffer_dirty(journal->j_sb_buffer, 1);
 505         ll_rw_block(WRITE, 1, &journal->j_sb_buffer);
 506 }
 507
 508 static errcode_t e2fsck_journal_fix_corrupt_super(e2fsck_t ctx,
 509                                                   journal_t *journal,
 510                                                   struct problem_context *pctx)
 511 {
 512         struct ext2_super_block *sb = ctx->fs->super;
 513         int recover = ctx->fs->super->s_feature_incompat &
 514                 EXT3_FEATURE_INCOMPAT_RECOVER;
 515
 516         pctx->num = journal->j_inode->i_ino;
 517
 518         if (sb->s_feature_compat & EXT3_FEATURE_COMPAT_HAS_JOURNAL) {
 519                 if (fix_problem(ctx, PR_0_JOURNAL_BAD_SUPER, pctx)) {
 520                         e2fsck_journal_reset_super(ctx, journal->j_superblock,
 521                                                    journal);
 522                         journal->j_transaction_sequence = 1;
 523                         e2fsck_clear_recover(ctx, recover);
 524                         return 0;
 525                 }
 526                 return EXT2_ET_CORRUPT_SUPERBLOCK;
 527         } else if (e2fsck_journal_fix_bad_inode(ctx, pctx))
 528                 return EXT2_ET_CORRUPT_SUPERBLOCK;
 529
 530         return 0;
 531 }
 532
 533 static void e2fsck_journal_release(e2fsck_t ctx, journal_t *journal,
 534                                    int reset, int drop)
 535 {
 536         journal_superblock_t *jsb;
 537
 538         if (drop)
 539                 mark_buffer_clean(journal->j_sb_buffer);
 540         else if (!(ctx->options & E2F_OPT_READONLY)) {
 541                 jsb = journal->j_superblock;
 542                 jsb->s_sequence = htonl(journal->j_transaction_sequence);
 543                 if (reset)
 544                         jsb->s_start = 0; /* this marks the journal as empty */
 545                 mark_buffer_dirty(journal->j_sb_buffer, 1);
 546         }
 547         brelse(journal->j_sb_buffer);
 548
 549         if (ctx->journal_io) {
 550                 if (ctx->fs && ctx->fs->io != ctx->journal_io)
 551                         io_channel_close(ctx->journal_io);
 552                 ctx->journal_io = 0;
 553         }
 554
 555         if (journal->j_inode)
 556                 ext2fs_free_mem((void **)&journal->j_inode);
 557         ext2fs_free_mem((void **)&journal);
 558 }
 559
 560 /*
 561  * This function makes sure that the superblock fields regarding the
 562  * journal are consistent.
 563  */
 564 int e2fsck_check_ext3_journal(e2fsck_t ctx)
 565 {
 566         struct ext2_super_block *sb = ctx->fs->super;
 567         journal_t *journal;
 568         int recover = ctx->fs->super->s_feature_incompat &
 569                 EXT3_FEATURE_INCOMPAT_RECOVER;
 570         struct problem_context pctx;
 571         int reset = 0, force_fsck = 0;
 572         int retval;
 573
 574         /* If we don't have any journal features, don't do anything more */
 575         if (!(sb->s_feature_compat & EXT3_FEATURE_COMPAT_HAS_JOURNAL) &&
 576             !recover && sb->s_journal_inum == 0 && sb->s_journal_dev == 0 &&
 577             uuid_is_null(sb->s_journal_uuid))
 578                 return 0;
 579
 580         clear_problem_context(&pctx);
 581         pctx.num = sb->s_journal_inum;
 582
 583         retval = e2fsck_get_journal(ctx, &journal);
 584         if (retval) {
 585                 if (retval == EXT2_ET_BAD_INODE_NUM)
 586                         return e2fsck_journal_fix_bad_inode(ctx, &pctx);
 587                 return retval;
 588         }
 589
 590         retval = e2fsck_journal_load(journal);
 591         if (retval) {
 592                 if ((retval == EXT2_ET_CORRUPT_SUPERBLOCK) ||
 593                     ((retval == EXT2_ET_UNSUPP_FEATURE) &&
 594                     (!fix_problem(ctx, PR_0_JOURNAL_UNSUPP_INCOMPAT,
 595                                   &pctx))) ||
 596                     ((retval == EXT2_ET_RO_UNSUPP_FEATURE) &&
 597                     (!fix_problem(ctx, PR_0_JOURNAL_UNSUPP_ROCOMPAT,
 598                                   &pctx))) ||
 599                     ((retval == EXT2_ET_JOURNAL_UNSUPP_VERSION) &&
 600                     (!fix_problem(ctx, PR_0_JOURNAL_UNSUPP_VERSION, &pctx))))
 601                         retval = e2fsck_journal_fix_corrupt_super(ctx, journal,
 602                                                                   &pctx);
 603                 e2fsck_journal_release(ctx, journal, 0, 1);
 604                 return retval;
 605         }
 606
 607         /*
 608          * We want to make the flags consistent here.  We will not leave with
 609          * needs_recovery set but has_journal clear.  We can't get in a loop
 610          * with -y, -n, or -p, only if a user isn't making up their mind.
 611          */
 612 no_has_journal:
 613         if (!(sb->s_feature_compat & EXT3_FEATURE_COMPAT_HAS_JOURNAL)) {
 614                 recover = sb->s_feature_incompat & EXT3_FEATURE_INCOMPAT_RECOVER;
 615                 pctx.str = "inode";
 616                 if (fix_problem(ctx, PR_0_JOURNAL_HAS_JOURNAL, &pctx)) {
 617                         if (recover &&
 618                             !fix_problem(ctx, PR_0_JOURNAL_RECOVER_SET, &pctx))
 619                                 goto no_has_journal;
 620                         /*
 621                          * Need a full fsck if we are releasing a
 622                          * journal stored on a reserved inode.
 623                          */
 624                         force_fsck = recover ||
 625                                 (sb->s_journal_inum < EXT2_FIRST_INODE(sb));
 626                         /* Clear all of the journal fields */
 627                         sb->s_journal_inum = 0;
 628                         sb->s_journal_dev = 0;
 629                         memset(sb->s_journal_uuid, 0,
 630                                sizeof(sb->s_journal_uuid));
 631                         e2fsck_clear_recover(ctx, force_fsck);
 632                 } else if (!(ctx->options & E2F_OPT_READONLY)) {
 633                         sb->s_feature_compat |= EXT3_FEATURE_COMPAT_HAS_JOURNAL;
 634                         ext2fs_mark_super_dirty(ctx->fs);
 635                 }
 636         }
 637
 638         if (sb->s_feature_compat & EXT3_FEATURE_COMPAT_HAS_JOURNAL &&
 639             !(sb->s_feature_incompat & EXT3_FEATURE_INCOMPAT_RECOVER) &&
 640             journal->j_superblock->s_start != 0) {
 641                 if (fix_problem(ctx, PR_0_JOURNAL_RESET_JOURNAL, &pctx)) {
 642                         reset = 1;
 643                         sb->s_state &= ~EXT2_VALID_FS;
 644                         ext2fs_mark_super_dirty(ctx->fs);
 645                 }
 646                 /*
 647                  * If the user answers no to the above question, we
 648                  * ignore the fact that journal apparently has data;
 649                  * accidentally replaying over valid data would be far
 650                  * worse than skipping a questionable recovery.
 651                  *
 652                  * XXX should we abort with a fatal error here?  What
 653                  * will the ext3 kernel code do if a filesystem with
 654                  * !NEEDS_RECOVERY but with a non-zero
 655                  * journal->j_superblock->s_start is mounted?
 656                  */
 657         }
 658
 659         e2fsck_journal_release(ctx, journal, reset, 0);
 660         return retval;
 661 }
 662
 663 static errcode_t recover_ext3_journal(e2fsck_t ctx)
 664 {
 665         journal_t *journal;
 666         int retval;
 667
 668         retval = e2fsck_get_journal(ctx, &journal);
 669         if (retval)
 670                 return retval;
 671
 672         retval = e2fsck_journal_load(journal);
 673         if (retval)
 674                 goto errout;
 675
 676         retval = journal_init_revoke(journal, 1024);
 677         if (retval)
 678                 goto errout;
 679
 680         retval = -journal_recover(journal);
 681         if (retval)
 682                 goto errout;
 683
 684         if (journal->j_superblock->s_errno) {
 685                 ctx->fs->super->s_state |= EXT2_ERROR_FS;
 686                 ext2fs_mark_super_dirty(ctx->fs);
 687                 journal->j_superblock->s_errno = 0;
 688                 mark_buffer_dirty(journal->j_sb_buffer, 1);
 689         }
 690
 691 errout:
 692         e2fsck_journal_release(ctx, journal, 1, 0);
 693         return retval;
 694 }
 695
 696 int e2fsck_run_ext3_journal(e2fsck_t ctx)
 697 {
 698         io_manager io_ptr = ctx->fs->io->manager;
 699         int blocksize = ctx->fs->blocksize;
 700         errcode_t       retval, recover_retval;
 701
 702         printf(_("%s: recovering journal\n"), ctx->device_name);
 703         if (ctx->options & E2F_OPT_READONLY) {
 704                 printf(_("%s: won't do journal recovery while read-only\n"),
 705                        ctx->device_name);
 706                 return EXT2_ET_FILE_RO;
 707         }
 708
 709         if (ctx->fs->flags & EXT2_FLAG_DIRTY)
 710                 ext2fs_flush(ctx->fs);  /* Force out any modifications */
 711
 712         recover_retval = recover_ext3_journal(ctx);
 713
 714         /*
 715          * Reload the filesystem context to get up-to-date data from disk
 716          * because journal recovery will change the filesystem under us.
 717          */
 718         ext2fs_close(ctx->fs);
 719         retval = ext2fs_open(ctx->filesystem_name, EXT2_FLAG_RW,
 720                              ctx->superblock, blocksize, io_ptr,
 721                              &ctx->fs);
 722
 723         if (retval) {
 724                 com_err(ctx->program_name, retval,
 725                         _("while trying to re-open %s"),
 726                         ctx->device_name);
 727                 fatal_error(ctx, 0);
 728         }
 729         ctx->fs->priv_data = ctx;
 730
 731         /* Set the superblock flags */
 732         e2fsck_clear_recover(ctx, recover_retval);
 733         return recover_retval;
 734 }
 735
 736 /*
 737  * This function will move the journal inode from a visible file in
 738  * the filesystem directory hierarchy to the reserved inode if necessary.
 739  */
 740 const static char * const journal_names[] = {
 741         ".journal", "journal", ".journal.dat", "journal.dat", 0 };
 742
 743 void e2fsck_move_ext3_journal(e2fsck_t ctx)
 744 {
 745         struct ext2_super_block *sb = ctx->fs->super;
 746         struct problem_context  pctx;
 747         struct ext2_inode       inode;
 748         ext2_filsys             fs = ctx->fs;
 749         ext2_ino_t              ino;
 750         errcode_t               retval;
 751         const char * const *    cpp;
 752         int                     group, mount_flags;
 753
 754         /*
 755          * If the filesystem is opened read-only, or there is no
 756          * journal, or the journal is already in the hidden inode,
 757          * then do nothing.
 758          */
 759         if ((ctx->options & E2F_OPT_READONLY) ||
 760             (sb->s_journal_inum == 0) ||
 761             (sb->s_journal_inum == EXT2_JOURNAL_INO) ||
 762             !(sb->s_feature_compat & EXT3_FEATURE_COMPAT_HAS_JOURNAL))
 763                 return;
 764
 765         /*
 766          * If the filesystem is mounted, or we can't tell whether
 767          * or not it's mounted, do nothing.
 768          */
 769         retval = ext2fs_check_if_mounted(ctx->filesystem_name, &mount_flags);
 770         if (retval || (mount_flags & EXT2_MF_MOUNTED))
 771                 return;
 772
 773         /*
 774          * If we can't find the name of the journal inode, then do
 775          * nothing.
 776          */
 777         for (cpp = journal_names; *cpp; cpp++) {
 778                 retval = ext2fs_lookup(fs, EXT2_ROOT_INO, *cpp,
 779                                        strlen(*cpp), 0, &ino);
 780                 if ((retval == 0) && (ino == sb->s_journal_inum))
 781                         break;
 782         }
 783         if (*cpp == 0)
 784                 return;
 785
 786         /*
 787          * The inode had better have only one link and not be readable.
 788          */
 789         if (ext2fs_read_inode(fs, ino, &inode) != 0)
 790                 return;
 791         if (inode.i_links_count != 1)
 792                 return;
 793
 794         /* We need the inode bitmap to be loaded */
 795         retval = ext2fs_read_bitmaps(fs);
 796         if (retval)
 797                 return;
 798
 799         clear_problem_context(&pctx);
 800         pctx.str = *cpp;
 801         if (!fix_problem(ctx, PR_0_MOVE_JOURNAL, &pctx))
 802                 return;
 803
 804         /*
 805          * OK, we've done all the checks, let's actually move the
 806          * journal inode.  Errors at this point mean we need to force
 807          * an ext2 filesystem check.
 808          */
 809         if ((retval = ext2fs_unlink(fs, EXT2_ROOT_INO, *cpp, ino, 0)) != 0)
 810                 goto err_out;
 811         if ((retval = ext2fs_write_inode(fs, EXT2_JOURNAL_INO, &inode)) != 0)
 812                 goto err_out;
 813         sb->s_journal_inum = EXT2_JOURNAL_INO;
 814         ext2fs_mark_super_dirty(fs);
 815         inode.i_links_count = 0;
 816         inode.i_dtime = time(0);
 817         if ((retval = ext2fs_write_inode(fs, ino, &inode)) != 0)
 818                 goto err_out;
 819
 820         group = ext2fs_group_of_ino(fs, ino);
 821         ext2fs_unmark_inode_bitmap(fs->inode_map, ino);
 822         ext2fs_mark_ib_dirty(fs);
 823         fs->group_desc[group].bg_free_inodes_count++;
 824         fs->super->s_free_inodes_count++;
 825         return;
 826
 827 err_out:
 828         pctx.errcode = retval;
 829         fix_problem(ctx, PR_0_ERR_MOVE_JOURNAL, &pctx);
 830         fs->super->s_state &= ~EXT2_VALID_FS;
 831         ext2fs_mark_super_dirty(fs);
 832         return;
 833 }
 834