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 #include "config.h"
  16 #ifdef HAVE_SYS_MOUNT_H
  17 #include <sys/param.h>
  18 #include <sys/mount.h>
  19 #define MNT_FL (MS_MGC_VAL | MS_RDONLY)
  20 #endif
  21 #ifdef HAVE_SYS_STAT_H
  22 #include <sys/stat.h>
  23 #endif
  24
  25 #define E2FSCK_INCLUDE_INLINE_FUNCS
  26 #include "jfs_user.h"
  27 #include "problem.h"
  28 #include "uuid/uuid.h"
  29
  30 #ifdef CONFIG_JBD_DEBUG         /* Enabled by configure --enable-jfs-debug */
  31 static int bh_count = 0;
  32 #endif
  33
  34 /*
  35  * Define USE_INODE_IO to use the inode_io.c / fileio.c codepaths.
  36  * This creates a larger static binary, and a smaller binary using
  37  * shared libraries.  It's also probably slightly less CPU-efficient,
  38  * which is why it's not on by default.  But, it's a good way of
  39  * testing the functions in inode_io.c and fileio.c.
  40  */
  41 #undef USE_INODE_IO
  42
  43 /* Kernel compatibility functions for handling the journal.  These allow us
  44  * to use the recovery.c file virtually unchanged from the kernel, so we
  45  * don't have to do much to keep kernel and user recovery in sync.
  46  */
  47 int journal_bmap(journal_t *journal, blk64_t block, unsigned long long *phys)
  48 {
  49 #ifdef USE_INODE_IO
  50         *phys = block;
  51         return 0;
  52 #else
  53         struct inode    *inode = journal->j_inode;
  54         errcode_t       retval;
  55         blk64_t         pblk;
  56
  57         if (!inode) {
  58                 *phys = block;
  59                 return 0;
  60         }
  61
  62         retval= ext2fs_bmap2(inode->i_ctx->fs, inode->i_ino,
  63                              &inode->i_ext2, NULL, 0, block, 0, &pblk);
  64         *phys = pblk;
  65         return (int) retval;
  66 #endif
  67 }
  68
  69 struct buffer_head *getblk(kdev_t kdev, blk64_t blocknr, int blocksize)
  70 {
  71         struct buffer_head *bh;
  72         int bufsize = sizeof(*bh) + kdev->k_ctx->fs->blocksize -
  73                 sizeof(bh->b_data);
  74
  75         bh = e2fsck_allocate_memory(kdev->k_ctx, bufsize, "block buffer");
  76         if (!bh)
  77                 return NULL;
  78
  79 #ifdef CONFIG_JBD_DEBUG
  80         if (journal_enable_debug >= 3)
  81                 bh_count++;
  82 #endif
  83         jfs_debug(4, "getblk for block %llu (%d bytes)(total %d)\n",
  84                   (unsigned long long) blocknr, blocksize, bh_count);
  85
  86         bh->b_ctx = kdev->k_ctx;
  87         if (kdev->k_dev == K_DEV_FS)
  88                 bh->b_io = kdev->k_ctx->fs->io;
  89         else
  90                 bh->b_io = kdev->k_ctx->journal_io;
  91         bh->b_size = blocksize;
  92         bh->b_blocknr = blocknr;
  93
  94         return bh;
  95 }
  96
  97 void sync_blockdev(kdev_t kdev)
  98 {
  99         io_channel      io;
 100
 101         if (kdev->k_dev == K_DEV_FS)
 102                 io = kdev->k_ctx->fs->io;
 103         else
 104                 io = kdev->k_ctx->journal_io;
 105
 106         io_channel_flush(io);
 107 }
 108
 109 void ll_rw_block(int rw, int nr, struct buffer_head *bhp[])
 110 {
 111         errcode_t retval;
 112         struct buffer_head *bh;
 113
 114         for (; nr > 0; --nr) {
 115                 bh = *bhp++;
 116                 if (rw == READ && !bh->b_uptodate) {
 117                         jfs_debug(3, "reading block %llu/%p\n",
 118                                   bh->b_blocknr, (void *) bh);
 119                         retval = io_channel_read_blk64(bh->b_io,
 120                                                      bh->b_blocknr,
 121                                                      1, bh->b_data);
 122                         if (retval) {
 123                                 com_err(bh->b_ctx->device_name, retval,
 124                                         "while reading block %llu\n",
 125                                         bh->b_blocknr);
 126                                 bh->b_err = (int) retval;
 127                                 continue;
 128                         }
 129                         bh->b_uptodate = 1;
 130                 } else if (rw == WRITE && bh->b_dirty) {
 131                         jfs_debug(3, "writing block %llu/%p\n",
 132                                   bh->b_blocknr,
 133                                   (void *) bh);
 134                         retval = io_channel_write_blk64(bh->b_io,
 135                                                       bh->b_blocknr,
 136                                                       1, bh->b_data);
 137                         if (retval) {
 138                                 com_err(bh->b_ctx->device_name, retval,
 139                                         "while writing block %llu\n",
 140                                         bh->b_blocknr);
 141                                 bh->b_err = (int) retval;
 142                                 continue;
 143                         }
 144                         bh->b_dirty = 0;
 145                         bh->b_uptodate = 1;
 146                 } else {
 147                         jfs_debug(3, "no-op %s for block %llu\n",
 148                                   rw == READ ? "read" : "write",
 149                                   bh->b_blocknr);
 150                 }
 151         }
 152 }
 153
 154 void mark_buffer_dirty(struct buffer_head *bh)
 155 {
 156         bh->b_dirty = 1;
 157 }
 158
 159 static void mark_buffer_clean(struct buffer_head * bh)
 160 {
 161         bh->b_dirty = 0;
 162 }
 163
 164 void brelse(struct buffer_head *bh)
 165 {
 166         if (bh->b_dirty)
 167                 ll_rw_block(WRITE, 1, &bh);
 168         jfs_debug(3, "freeing block %llu/%p (total %d)\n",
 169                   bh->b_blocknr, (void *) bh, --bh_count);
 170         ext2fs_free_mem(&bh);
 171 }
 172
 173 int buffer_uptodate(struct buffer_head *bh)
 174 {
 175         return bh->b_uptodate;
 176 }
 177
 178 void mark_buffer_uptodate(struct buffer_head *bh, int val)
 179 {
 180         bh->b_uptodate = val;
 181 }
 182
 183 void wait_on_buffer(struct buffer_head *bh)
 184 {
 185         if (!bh->b_uptodate)
 186                 ll_rw_block(READ, 1, &bh);
 187 }
 188
 189
 190 static void e2fsck_clear_recover(e2fsck_t ctx, int error)
 191 {
 192         ctx->fs->super->s_feature_incompat &= ~EXT3_FEATURE_INCOMPAT_RECOVER;
 193
 194         /* if we had an error doing journal recovery, we need a full fsck */
 195         if (error)
 196                 ctx->fs->super->s_state &= ~EXT2_VALID_FS;
 197         ext2fs_mark_super_dirty(ctx->fs);
 198 }
 199
 200 /*
 201  * This is a helper function to check the validity of the journal.
 202  */
 203 struct process_block_struct {
 204         e2_blkcnt_t     last_block;
 205 };
 206
 207 static int process_journal_block(ext2_filsys fs,
 208                                  blk64_t        *block_nr,
 209                                  e2_blkcnt_t blockcnt,
 210                                  blk64_t ref_block EXT2FS_ATTR((unused)),
 211                                  int ref_offset EXT2FS_ATTR((unused)),
 212                                  void *priv_data)
 213 {
 214         struct process_block_struct *p;
 215         blk64_t blk = *block_nr;
 216
 217         p = (struct process_block_struct *) priv_data;
 218
 219         if (!blk || blk < fs->super->s_first_data_block ||
 220             blk >= ext2fs_blocks_count(fs->super))
 221                 return BLOCK_ABORT;
 222
 223         if (blockcnt >= 0)
 224                 p->last_block = blockcnt;
 225         return 0;
 226 }
 227
 228 static errcode_t e2fsck_get_journal(e2fsck_t ctx, journal_t **ret_journal)
 229 {
 230         struct process_block_struct pb;
 231         struct ext2_super_block *sb = ctx->fs->super;
 232         struct ext2_super_block jsuper;
 233         struct problem_context  pctx;
 234         struct buffer_head      *bh;
 235         struct inode            *j_inode = NULL;
 236         struct kdev_s           *dev_fs = NULL, *dev_journal;
 237         const char              *journal_name = 0;
 238         journal_t               *journal = NULL;
 239         errcode_t               retval = 0;
 240         io_manager              io_ptr = 0;
 241         unsigned long long      start = 0;
 242         int                     ext_journal = 0;
 243         int                     tried_backup_jnl = 0;
 244         blk64_t                 maxlen;
 245
 246         clear_problem_context(&pctx);
 247
 248         journal = e2fsck_allocate_memory(ctx, sizeof(journal_t), "journal");
 249         if (!journal) {
 250                 return EXT2_ET_NO_MEMORY;
 251         }
 252
 253         dev_fs = e2fsck_allocate_memory(ctx, 2*sizeof(struct kdev_s), "kdev");
 254         if (!dev_fs) {
 255                 retval = EXT2_ET_NO_MEMORY;
 256                 goto errout;
 257         }
 258         dev_journal = dev_fs+1;
 259
 260         dev_fs->k_ctx = dev_journal->k_ctx = ctx;
 261         dev_fs->k_dev = K_DEV_FS;
 262         dev_journal->k_dev = K_DEV_JOURNAL;
 263
 264         journal->j_dev = dev_journal;
 265         journal->j_fs_dev = dev_fs;
 266         journal->j_inode = NULL;
 267         journal->j_blocksize = ctx->fs->blocksize;
 268
 269         if (uuid_is_null(sb->s_journal_uuid)) {
 270                 if (!sb->s_journal_inum) {
 271                         retval = EXT2_ET_BAD_INODE_NUM;
 272                         goto errout;
 273                 }
 274                 j_inode = e2fsck_allocate_memory(ctx, sizeof(*j_inode),
 275                                                  "journal inode");
 276                 if (!j_inode) {
 277                         retval = EXT2_ET_NO_MEMORY;
 278                         goto errout;
 279                 }
 280
 281                 j_inode->i_ctx = ctx;
 282                 j_inode->i_ino = sb->s_journal_inum;
 283
 284                 if ((retval = ext2fs_read_inode(ctx->fs,
 285                                                 sb->s_journal_inum,
 286                                                 &j_inode->i_ext2))) {
 287                 try_backup_journal:
 288                         if (sb->s_jnl_backup_type != EXT3_JNL_BACKUP_BLOCKS ||
 289                             tried_backup_jnl)
 290                                 goto errout;
 291                         memset(&j_inode->i_ext2, 0, sizeof(struct ext2_inode));
 292                         memcpy(&j_inode->i_ext2.i_block[0], sb->s_jnl_blocks,
 293                                EXT2_N_BLOCKS*4);
 294                         j_inode->i_ext2.i_size_high = sb->s_jnl_blocks[15];
 295                         j_inode->i_ext2.i_size = sb->s_jnl_blocks[16];
 296                         j_inode->i_ext2.i_links_count = 1;
 297                         j_inode->i_ext2.i_mode = LINUX_S_IFREG | 0600;
 298                         e2fsck_use_inode_shortcuts(ctx, 1);
 299                         ctx->stashed_ino = j_inode->i_ino;
 300                         ctx->stashed_inode = &j_inode->i_ext2;
 301                         tried_backup_jnl++;
 302                 }
 303                 if (!j_inode->i_ext2.i_links_count ||
 304                     !LINUX_S_ISREG(j_inode->i_ext2.i_mode)) {
 305                         retval = EXT2_ET_NO_JOURNAL;
 306                         goto try_backup_journal;
 307                 }
 308                 if (EXT2_I_SIZE(&j_inode->i_ext2) / journal->j_blocksize <
 309                     JFS_MIN_JOURNAL_BLOCKS) {
 310                         retval = EXT2_ET_JOURNAL_TOO_SMALL;
 311                         goto try_backup_journal;
 312                 }
 313                 pb.last_block = -1;
 314                 retval = ext2fs_block_iterate3(ctx->fs, j_inode->i_ino,
 315                                                BLOCK_FLAG_HOLE, 0,
 316                                                process_journal_block, &pb);
 317                 if ((pb.last_block + 1) * ctx->fs->blocksize <
 318                     (int) EXT2_I_SIZE(&j_inode->i_ext2)) {
 319                         retval = EXT2_ET_JOURNAL_TOO_SMALL;
 320                         goto try_backup_journal;
 321                 }
 322                 if (tried_backup_jnl && !(ctx->options & E2F_OPT_READONLY)) {
 323                         retval = ext2fs_write_inode(ctx->fs, sb->s_journal_inum,
 324                                                     &j_inode->i_ext2);
 325                         if (retval)
 326                                 goto errout;
 327                 }
 328
 329                 journal->j_maxlen = EXT2_I_SIZE(&j_inode->i_ext2) /
 330                         journal->j_blocksize;
 331
 332 #ifdef USE_INODE_IO
 333                 retval = ext2fs_inode_io_intern2(ctx->fs, sb->s_journal_inum,
 334                                                  &j_inode->i_ext2,
 335                                                  &journal_name);
 336                 if (retval)
 337                         goto errout;
 338
 339                 io_ptr = inode_io_manager;
 340 #else
 341                 journal->j_inode = j_inode;
 342                 ctx->journal_io = ctx->fs->io;
 343                 if ((retval = (errcode_t) journal_bmap(journal, 0, &start)) != 0)
 344                         goto errout;
 345 #endif
 346         } else {
 347                 ext_journal = 1;
 348                 if (!ctx->journal_name) {
 349                         char uuid[37];
 350
 351                         uuid_unparse(sb->s_journal_uuid, uuid);
 352                         ctx->journal_name = blkid_get_devname(ctx->blkid,
 353                                                               "UUID", uuid);
 354                         if (!ctx->journal_name)
 355                                 ctx->journal_name = blkid_devno_to_devname(sb->s_journal_dev);
 356                 }
 357                 journal_name = ctx->journal_name;
 358
 359                 if (!journal_name) {
 360                         fix_problem(ctx, PR_0_CANT_FIND_JOURNAL, &pctx);
 361                         retval = EXT2_ET_LOAD_EXT_JOURNAL;
 362                         goto errout;
 363                 }
 364
 365                 jfs_debug(1, "Using journal file %s\n", journal_name);
 366                 io_ptr = unix_io_manager;
 367         }
 368
 369 #if 0
 370         test_io_backing_manager = io_ptr;
 371         io_ptr = test_io_manager;
 372 #endif
 373 #ifndef USE_INODE_IO
 374         if (ext_journal)
 375 #endif
 376         {
 377                 int flags = IO_FLAG_RW;
 378                 if (!(ctx->mount_flags & EXT2_MF_ISROOT &&
 379                       ctx->mount_flags & EXT2_MF_READONLY))
 380                         flags |= IO_FLAG_EXCLUSIVE;
 381                 if ((ctx->mount_flags & EXT2_MF_READONLY) &&
 382                     (ctx->options & E2F_OPT_FORCE))
 383                         flags &= ~IO_FLAG_EXCLUSIVE;
 384
 385
 386                 retval = io_ptr->open(journal_name, flags,
 387                                       &ctx->journal_io);
 388         }
 389         if (retval)
 390                 goto errout;
 391
 392         io_channel_set_blksize(ctx->journal_io, ctx->fs->blocksize);
 393
 394         if (ext_journal) {
 395                 if (ctx->fs->blocksize == 1024)
 396                         start = 1;
 397                 bh = getblk(dev_journal, start, ctx->fs->blocksize);
 398                 if (!bh) {
 399                         retval = EXT2_ET_NO_MEMORY;
 400                         goto errout;
 401                 }
 402                 ll_rw_block(READ, 1, &bh);
 403                 if ((retval = bh->b_err) != 0) {
 404                         brelse(bh);
 405                         goto errout;
 406                 }
 407                 memcpy(&jsuper, start ? bh->b_data :  bh->b_data + 1024,
 408                        sizeof(jsuper));
 409                 brelse(bh);
 410 #ifdef WORDS_BIGENDIAN
 411                 if (jsuper.s_magic == ext2fs_swab16(EXT2_SUPER_MAGIC))
 412                         ext2fs_swap_super(&jsuper);
 413 #endif
 414                 if (jsuper.s_magic != EXT2_SUPER_MAGIC ||
 415                     !(jsuper.s_feature_incompat & EXT3_FEATURE_INCOMPAT_JOURNAL_DEV)) {
 416                         fix_problem(ctx, PR_0_EXT_JOURNAL_BAD_SUPER, &pctx);
 417                         retval = EXT2_ET_LOAD_EXT_JOURNAL;
 418                         goto errout;
 419                 }
 420                 /* Make sure the journal UUID is correct */
 421                 if (memcmp(jsuper.s_uuid, ctx->fs->super->s_journal_uuid,
 422                            sizeof(jsuper.s_uuid))) {
 423                         fix_problem(ctx, PR_0_JOURNAL_BAD_UUID, &pctx);
 424                         retval = EXT2_ET_LOAD_EXT_JOURNAL;
 425                         goto errout;
 426                 }
 427
 428                 maxlen = ext2fs_blocks_count(&jsuper);
 429                 if (maxlen > 1ULL << 32)
 430                         maxlen = (1ULL << 32) - 1;
 431                 journal->j_maxlen = maxlen;
 432                 start++;
 433         }
 434
 435         if (!(bh = getblk(dev_journal, start, journal->j_blocksize))) {
 436                 retval = EXT2_ET_NO_MEMORY;
 437                 goto errout;
 438         }
 439
 440         journal->j_sb_buffer = bh;
 441         journal->j_superblock = (journal_superblock_t *)bh->b_data;
 442
 443 #ifdef USE_INODE_IO
 444         if (j_inode)
 445                 ext2fs_free_mem(&j_inode);
 446 #endif
 447
 448         *ret_journal = journal;
 449         e2fsck_use_inode_shortcuts(ctx, 0);
 450         return 0;
 451
 452 errout:
 453         e2fsck_use_inode_shortcuts(ctx, 0);
 454         if (dev_fs)
 455                 ext2fs_free_mem(&dev_fs);
 456         if (j_inode)
 457                 ext2fs_free_mem(&j_inode);
 458         if (journal)
 459                 ext2fs_free_mem(&journal);
 460         return retval;
 461 }
 462
 463 static errcode_t e2fsck_journal_fix_bad_inode(e2fsck_t ctx,
 464                                               struct problem_context *pctx)
 465 {
 466         struct ext2_super_block *sb = ctx->fs->super;
 467         int recover = ctx->fs->super->s_feature_incompat &
 468                 EXT3_FEATURE_INCOMPAT_RECOVER;
 469         int has_journal = ctx->fs->super->s_feature_compat &
 470                 EXT3_FEATURE_COMPAT_HAS_JOURNAL;
 471
 472         if (has_journal || sb->s_journal_inum) {
 473                 /* The journal inode is bogus, remove and force full fsck */
 474                 pctx->ino = sb->s_journal_inum;
 475                 if (fix_problem(ctx, PR_0_JOURNAL_BAD_INODE, pctx)) {
 476                         if (has_journal && sb->s_journal_inum)
 477                                 printf("*** ext3 journal has been deleted - "
 478                                        "filesystem is now ext2 only ***\n\n");
 479                         sb->s_feature_compat &= ~EXT3_FEATURE_COMPAT_HAS_JOURNAL;
 480                         sb->s_journal_inum = 0;
 481                         ctx->flags |= E2F_FLAG_JOURNAL_INODE;
 482                         ctx->fs->flags &= ~EXT2_FLAG_MASTER_SB_ONLY;
 483                         e2fsck_clear_recover(ctx, 1);
 484                         return 0;
 485                 }
 486                 return EXT2_ET_BAD_INODE_NUM;
 487         } else if (recover) {
 488                 if (fix_problem(ctx, PR_0_JOURNAL_RECOVER_SET, pctx)) {
 489                         e2fsck_clear_recover(ctx, 1);
 490                         return 0;
 491                 }
 492                 return EXT2_ET_UNSUPP_FEATURE;
 493         }
 494         return 0;
 495 }
 496
 497 #define V1_SB_SIZE      0x0024
 498 static void clear_v2_journal_fields(journal_t *journal)
 499 {
 500         e2fsck_t ctx = journal->j_dev->k_ctx;
 501         struct problem_context pctx;
 502
 503         clear_problem_context(&pctx);
 504
 505         if (!fix_problem(ctx, PR_0_CLEAR_V2_JOURNAL, &pctx))
 506                 return;
 507
 508         memset(((char *) journal->j_superblock) + V1_SB_SIZE, 0,
 509                ctx->fs->blocksize-V1_SB_SIZE);
 510         mark_buffer_dirty(journal->j_sb_buffer);
 511 }
 512
 513
 514 static errcode_t e2fsck_journal_load(journal_t *journal)
 515 {
 516         e2fsck_t ctx = journal->j_dev->k_ctx;
 517         journal_superblock_t *jsb;
 518         struct buffer_head *jbh = journal->j_sb_buffer;
 519         struct problem_context pctx;
 520
 521         clear_problem_context(&pctx);
 522
 523         ll_rw_block(READ, 1, &jbh);
 524         if (jbh->b_err) {
 525                 com_err(ctx->device_name, jbh->b_err,
 526                         _("reading journal superblock\n"));
 527                 return jbh->b_err;
 528         }
 529
 530         jsb = journal->j_superblock;
 531         /* If we don't even have JFS_MAGIC, we probably have a wrong inode */
 532         if (jsb->s_header.h_magic != htonl(JFS_MAGIC_NUMBER))
 533                 return e2fsck_journal_fix_bad_inode(ctx, &pctx);
 534
 535         switch (ntohl(jsb->s_header.h_blocktype)) {
 536         case JFS_SUPERBLOCK_V1:
 537                 journal->j_format_version = 1;
 538                 if (jsb->s_feature_compat ||
 539                     jsb->s_feature_incompat ||
 540                     jsb->s_feature_ro_compat ||
 541                     jsb->s_nr_users)
 542                         clear_v2_journal_fields(journal);
 543                 break;
 544
 545         case JFS_SUPERBLOCK_V2:
 546                 journal->j_format_version = 2;
 547                 if (ntohl(jsb->s_nr_users) > 1 &&
 548                     uuid_is_null(ctx->fs->super->s_journal_uuid))
 549                         clear_v2_journal_fields(journal);
 550                 if (ntohl(jsb->s_nr_users) > 1) {
 551                         fix_problem(ctx, PR_0_JOURNAL_UNSUPP_MULTIFS, &pctx);
 552                         return EXT2_ET_JOURNAL_UNSUPP_VERSION;
 553                 }
 554                 break;
 555
 556         /*
 557          * These should never appear in a journal super block, so if
 558          * they do, the journal is badly corrupted.
 559          */
 560         case JFS_DESCRIPTOR_BLOCK:
 561         case JFS_COMMIT_BLOCK:
 562         case JFS_REVOKE_BLOCK:
 563                 return EXT2_ET_CORRUPT_SUPERBLOCK;
 564
 565         /* If we don't understand the superblock major type, but there
 566          * is a magic number, then it is likely to be a new format we
 567          * just don't understand, so leave it alone. */
 568         default:
 569                 return EXT2_ET_JOURNAL_UNSUPP_VERSION;
 570         }
 571
 572         if (JFS_HAS_INCOMPAT_FEATURE(journal, ~JFS_KNOWN_INCOMPAT_FEATURES))
 573                 return EXT2_ET_UNSUPP_FEATURE;
 574
 575         if (JFS_HAS_RO_COMPAT_FEATURE(journal, ~JFS_KNOWN_ROCOMPAT_FEATURES))
 576                 return EXT2_ET_RO_UNSUPP_FEATURE;
 577
 578         /* We have now checked whether we know enough about the journal
 579          * format to be able to proceed safely, so any other checks that
 580          * fail we should attempt to recover from. */
 581         if (jsb->s_blocksize != htonl(journal->j_blocksize)) {
 582                 com_err(ctx->program_name, EXT2_ET_CORRUPT_SUPERBLOCK,
 583                         _("%s: no valid journal superblock found\n"),
 584                         ctx->device_name);
 585                 return EXT2_ET_CORRUPT_SUPERBLOCK;
 586         }
 587
 588         if (ntohl(jsb->s_maxlen) < journal->j_maxlen)
 589                 journal->j_maxlen = ntohl(jsb->s_maxlen);
 590         else if (ntohl(jsb->s_maxlen) > journal->j_maxlen) {
 591                 com_err(ctx->program_name, EXT2_ET_CORRUPT_SUPERBLOCK,
 592                         _("%s: journal too short\n"),
 593                         ctx->device_name);
 594                 return EXT2_ET_CORRUPT_SUPERBLOCK;
 595         }
 596
 597         journal->j_tail_sequence = ntohl(jsb->s_sequence);
 598         journal->j_transaction_sequence = journal->j_tail_sequence;
 599         journal->j_tail = ntohl(jsb->s_start);
 600         journal->j_first = ntohl(jsb->s_first);
 601         journal->j_last = ntohl(jsb->s_maxlen);
 602
 603         return 0;
 604 }
 605
 606 static void e2fsck_journal_reset_super(e2fsck_t ctx, journal_superblock_t *jsb,
 607                                        journal_t *journal)
 608 {
 609         char *p;
 610         union {
 611                 uuid_t uuid;
 612                 __u32 val[4];
 613         } u;
 614         __u32 new_seq = 0;
 615         int i;
 616
 617         /* Leave a valid existing V1 superblock signature alone.
 618          * Anything unrecognisable we overwrite with a new V2
 619          * signature. */
 620
 621         if (jsb->s_header.h_magic != htonl(JFS_MAGIC_NUMBER) ||
 622             jsb->s_header.h_blocktype != htonl(JFS_SUPERBLOCK_V1)) {
 623                 jsb->s_header.h_magic = htonl(JFS_MAGIC_NUMBER);
 624                 jsb->s_header.h_blocktype = htonl(JFS_SUPERBLOCK_V2);
 625         }
 626
 627         /* Zero out everything else beyond the superblock header */
 628
 629         p = ((char *) jsb) + sizeof(journal_header_t);
 630         memset (p, 0, ctx->fs->blocksize-sizeof(journal_header_t));
 631
 632         jsb->s_blocksize = htonl(ctx->fs->blocksize);
 633         jsb->s_maxlen = htonl(journal->j_maxlen);
 634         jsb->s_first = htonl(1);
 635
 636         /* Initialize the journal sequence number so that there is "no"
 637          * chance we will find old "valid" transactions in the journal.
 638          * This avoids the need to zero the whole journal (slow to do,
 639          * and risky when we are just recovering the filesystem).
 640          */
 641         uuid_generate(u.uuid);
 642         for (i = 0; i < 4; i ++)
 643                 new_seq ^= u.val[i];
 644         jsb->s_sequence = htonl(new_seq);
 645
 646         mark_buffer_dirty(journal->j_sb_buffer);
 647         ll_rw_block(WRITE, 1, &journal->j_sb_buffer);
 648 }
 649
 650 static errcode_t e2fsck_journal_fix_corrupt_super(e2fsck_t ctx,
 651                                                   journal_t *journal,
 652                                                   struct problem_context *pctx)
 653 {
 654         struct ext2_super_block *sb = ctx->fs->super;
 655         int recover = ctx->fs->super->s_feature_incompat &
 656                 EXT3_FEATURE_INCOMPAT_RECOVER;
 657
 658         if (sb->s_feature_compat & EXT3_FEATURE_COMPAT_HAS_JOURNAL) {
 659                 if (fix_problem(ctx, PR_0_JOURNAL_BAD_SUPER, pctx)) {
 660                         e2fsck_journal_reset_super(ctx, journal->j_superblock,
 661                                                    journal);
 662                         journal->j_transaction_sequence = 1;
 663                         e2fsck_clear_recover(ctx, recover);
 664                         return 0;
 665                 }
 666                 return EXT2_ET_CORRUPT_SUPERBLOCK;
 667         } else if (e2fsck_journal_fix_bad_inode(ctx, pctx))
 668                 return EXT2_ET_CORRUPT_SUPERBLOCK;
 669
 670         return 0;
 671 }
 672
 673 static void e2fsck_journal_release(e2fsck_t ctx, journal_t *journal,
 674                                    int reset, int drop)
 675 {
 676         journal_superblock_t *jsb;
 677
 678         if (drop)
 679                 mark_buffer_clean(journal->j_sb_buffer);
 680         else if (!(ctx->options & E2F_OPT_READONLY)) {
 681                 jsb = journal->j_superblock;
 682                 jsb->s_sequence = htonl(journal->j_transaction_sequence);
 683                 if (reset)
 684                         jsb->s_start = 0; /* this marks the journal as empty */
 685                 mark_buffer_dirty(journal->j_sb_buffer);
 686         }
 687         brelse(journal->j_sb_buffer);
 688
 689         if (ctx->journal_io) {
 690                 if (ctx->fs && ctx->fs->io != ctx->journal_io)
 691                         io_channel_close(ctx->journal_io);
 692                 ctx->journal_io = 0;
 693         }
 694
 695 #ifndef USE_INODE_IO
 696         if (journal->j_inode)
 697                 ext2fs_free_mem(&journal->j_inode);
 698 #endif
 699         if (journal->j_fs_dev)
 700                 ext2fs_free_mem(&journal->j_fs_dev);
 701         ext2fs_free_mem(&journal);
 702 }
 703
 704 /*
 705  * This function makes sure that the superblock fields regarding the
 706  * journal are consistent.
 707  */
 708 errcode_t e2fsck_check_ext3_journal(e2fsck_t ctx)
 709 {
 710         struct ext2_super_block *sb = ctx->fs->super;
 711         journal_t *journal;
 712         int recover = ctx->fs->super->s_feature_incompat &
 713                 EXT3_FEATURE_INCOMPAT_RECOVER;
 714         struct problem_context pctx;
 715         problem_t problem;
 716         int reset = 0, force_fsck = 0;
 717         errcode_t retval;
 718
 719         /* If we don't have any journal features, don't do anything more */
 720         if (!(sb->s_feature_compat & EXT3_FEATURE_COMPAT_HAS_JOURNAL) &&
 721             !recover && sb->s_journal_inum == 0 && sb->s_journal_dev == 0 &&
 722             uuid_is_null(sb->s_journal_uuid))
 723                 return 0;
 724
 725         clear_problem_context(&pctx);
 726         pctx.num = sb->s_journal_inum;
 727
 728         retval = e2fsck_get_journal(ctx, &journal);
 729         if (retval) {
 730                 if ((retval == EXT2_ET_BAD_INODE_NUM) ||
 731                     (retval == EXT2_ET_BAD_BLOCK_NUM) ||
 732                     (retval == EXT2_ET_JOURNAL_TOO_SMALL) ||
 733                     (retval == EXT2_ET_NO_JOURNAL))
 734                         return e2fsck_journal_fix_bad_inode(ctx, &pctx);
 735                 return retval;
 736         }
 737
 738         retval = e2fsck_journal_load(journal);
 739         if (retval) {
 740                 if ((retval == EXT2_ET_CORRUPT_SUPERBLOCK) ||
 741                     ((retval == EXT2_ET_UNSUPP_FEATURE) &&
 742                     (!fix_problem(ctx, PR_0_JOURNAL_UNSUPP_INCOMPAT,
 743                                   &pctx))) ||
 744                     ((retval == EXT2_ET_RO_UNSUPP_FEATURE) &&
 745                     (!fix_problem(ctx, PR_0_JOURNAL_UNSUPP_ROCOMPAT,
 746                                   &pctx))) ||
 747                     ((retval == EXT2_ET_JOURNAL_UNSUPP_VERSION) &&
 748                     (!fix_problem(ctx, PR_0_JOURNAL_UNSUPP_VERSION, &pctx))))
 749                         retval = e2fsck_journal_fix_corrupt_super(ctx, journal,
 750                                                                   &pctx);
 751                 e2fsck_journal_release(ctx, journal, 0, 1);
 752                 return retval;
 753         }
 754
 755         /*
 756          * We want to make the flags consistent here.  We will not leave with
 757          * needs_recovery set but has_journal clear.  We can't get in a loop
 758          * with -y, -n, or -p, only if a user isn't making up their mind.
 759          */
 760 no_has_journal:
 761         if (!(sb->s_feature_compat & EXT3_FEATURE_COMPAT_HAS_JOURNAL)) {
 762                 recover = sb->s_feature_incompat & EXT3_FEATURE_INCOMPAT_RECOVER;
 763                 pctx.str = "inode";
 764                 if (fix_problem(ctx, PR_0_JOURNAL_HAS_JOURNAL, &pctx)) {
 765                         if (recover &&
 766                             !fix_problem(ctx, PR_0_JOURNAL_RECOVER_SET, &pctx))
 767                                 goto no_has_journal;
 768                         /*
 769                          * Need a full fsck if we are releasing a
 770                          * journal stored on a reserved inode.
 771                          */
 772                         force_fsck = recover ||
 773                                 (sb->s_journal_inum < EXT2_FIRST_INODE(sb));
 774                         /* Clear all of the journal fields */
 775                         sb->s_journal_inum = 0;
 776                         sb->s_journal_dev = 0;
 777                         memset(sb->s_journal_uuid, 0,
 778                                sizeof(sb->s_journal_uuid));
 779                         e2fsck_clear_recover(ctx, force_fsck);
 780                 } else if (!(ctx->options & E2F_OPT_READONLY)) {
 781                         sb->s_feature_compat |= EXT3_FEATURE_COMPAT_HAS_JOURNAL;
 782                         ctx->fs->flags &= ~EXT2_FLAG_MASTER_SB_ONLY;
 783                         ext2fs_mark_super_dirty(ctx->fs);
 784                 }
 785         }
 786
 787         if (sb->s_feature_compat & EXT3_FEATURE_COMPAT_HAS_JOURNAL &&
 788             !(sb->s_feature_incompat & EXT3_FEATURE_INCOMPAT_RECOVER) &&
 789             journal->j_superblock->s_start != 0) {
 790                 /* Print status information */
 791                 fix_problem(ctx, PR_0_JOURNAL_RECOVERY_CLEAR, &pctx);
 792                 if (ctx->superblock)
 793                         problem = PR_0_JOURNAL_RUN_DEFAULT;
 794                 else
 795                         problem = PR_0_JOURNAL_RUN;
 796                 if (fix_problem(ctx, problem, &pctx)) {
 797                         ctx->options |= E2F_OPT_FORCE;
 798                         sb->s_feature_incompat |=
 799                                 EXT3_FEATURE_INCOMPAT_RECOVER;
 800                         ext2fs_mark_super_dirty(ctx->fs);
 801                 } else if (fix_problem(ctx,
 802                                        PR_0_JOURNAL_RESET_JOURNAL, &pctx)) {
 803                         reset = 1;
 804                         sb->s_state &= ~EXT2_VALID_FS;
 805                         ext2fs_mark_super_dirty(ctx->fs);
 806                 }
 807                 /*
 808                  * If the user answers no to the above question, we
 809                  * ignore the fact that journal apparently has data;
 810                  * accidentally replaying over valid data would be far
 811                  * worse than skipping a questionable recovery.
 812                  *
 813                  * XXX should we abort with a fatal error here?  What
 814                  * will the ext3 kernel code do if a filesystem with
 815                  * !NEEDS_RECOVERY but with a non-zero
 816                  * journal->j_superblock->s_start is mounted?
 817                  */
 818         }
 819
 820         /*
 821          * If we don't need to do replay the journal, check to see if
 822          * the journal's errno is set; if so, we need to mark the file
 823          * system as being corrupt and clear the journal's s_errno.
 824          */
 825         if (!(sb->s_feature_incompat & EXT3_FEATURE_INCOMPAT_RECOVER) &&
 826             journal->j_superblock->s_errno) {
 827                 ctx->fs->super->s_state |= EXT2_ERROR_FS;
 828                 ext2fs_mark_super_dirty(ctx->fs);
 829                 journal->j_superblock->s_errno = 0;
 830                 mark_buffer_dirty(journal->j_sb_buffer);
 831         }
 832
 833         e2fsck_journal_release(ctx, journal, reset, 0);
 834         return retval;
 835 }
 836
 837 static errcode_t recover_ext3_journal(e2fsck_t ctx)
 838 {
 839         struct problem_context  pctx;
 840         journal_t *journal;
 841         errcode_t retval;
 842
 843         clear_problem_context(&pctx);
 844
 845         journal_init_revoke_caches();
 846         retval = e2fsck_get_journal(ctx, &journal);
 847         if (retval)
 848                 return retval;
 849
 850         retval = e2fsck_journal_load(journal);
 851         if (retval)
 852                 goto errout;
 853
 854         retval = journal_init_revoke(journal, 1024);
 855         if (retval)
 856                 goto errout;
 857
 858         retval = -journal_recover(journal);
 859         if (retval)
 860                 goto errout;
 861
 862         if (journal->j_failed_commit) {
 863                 pctx.ino = journal->j_failed_commit;
 864                 fix_problem(ctx, PR_0_JNL_TXN_CORRUPT, &pctx);
 865                 journal->j_superblock->s_errno = -EINVAL;
 866                 mark_buffer_dirty(journal->j_sb_buffer);
 867         }
 868
 869 errout:
 870         journal_destroy_revoke(journal);
 871         journal_destroy_revoke_caches();
 872         e2fsck_journal_release(ctx, journal, 1, 0);
 873         return retval;
 874 }
 875
 876 errcode_t e2fsck_run_ext3_journal(e2fsck_t ctx)
 877 {
 878         io_manager io_ptr = ctx->fs->io->manager;
 879         int blocksize = ctx->fs->blocksize;
 880         errcode_t       retval, recover_retval;
 881         io_stats        stats = 0;
 882         unsigned long long kbytes_written = 0;
 883
 884         printf(_("%s: recovering journal\n"), ctx->device_name);
 885         if (ctx->options & E2F_OPT_READONLY) {
 886                 printf(_("%s: won't do journal recovery while read-only\n"),
 887                        ctx->device_name);
 888                 return EXT2_ET_FILE_RO;
 889         }
 890
 891         if (ctx->fs->flags & EXT2_FLAG_DIRTY)
 892                 ext2fs_flush(ctx->fs);  /* Force out any modifications */
 893
 894         recover_retval = recover_ext3_journal(ctx);
 895
 896         /*
 897          * Reload the filesystem context to get up-to-date data from disk
 898          * because journal recovery will change the filesystem under us.
 899          */
 900         if (ctx->fs->super->s_kbytes_written &&
 901             ctx->fs->io->manager->get_stats)
 902                 ctx->fs->io->manager->get_stats(ctx->fs->io, &stats);
 903         if (stats && stats->bytes_written)
 904                 kbytes_written = stats->bytes_written >> 10;
 905
 906         ext2fs_mmp_stop(ctx->fs);
 907         ext2fs_free(ctx->fs);
 908         retval = ext2fs_open(ctx->filesystem_name, EXT2_FLAG_RW,
 909                              ctx->superblock, blocksize, io_ptr,
 910                              &ctx->fs);
 911         if (retval) {
 912                 com_err(ctx->program_name, retval,
 913                         _("while trying to re-open %s"),
 914                         ctx->device_name);
 915                 fatal_error(ctx, 0);
 916         }
 917         ctx->fs->priv_data = ctx;
 918         ctx->fs->now = ctx->now;
 919         ctx->fs->flags |= EXT2_FLAG_MASTER_SB_ONLY;
 920         ctx->fs->super->s_kbytes_written += kbytes_written;
 921
 922         /* Set the superblock flags */
 923         e2fsck_clear_recover(ctx, recover_retval != 0);
 924
 925         /*
 926          * Do one last sanity check, and propagate journal->s_errno to
 927          * the EXT2_ERROR_FS flag in the fs superblock if needed.
 928          */
 929         retval = e2fsck_check_ext3_journal(ctx);
 930         return retval ? retval : recover_retval;
 931 }
 932
 933 /*
 934  * This function will move the journal inode from a visible file in
 935  * the filesystem directory hierarchy to the reserved inode if necessary.
 936  */
 937 static const char * const journal_names[] = {
 938         ".journal", "journal", ".journal.dat", "journal.dat", 0 };
 939
 940 void e2fsck_move_ext3_journal(e2fsck_t ctx)
 941 {
 942         struct ext2_super_block *sb = ctx->fs->super;
 943         struct problem_context  pctx;
 944         struct ext2_inode       inode;
 945         ext2_filsys             fs = ctx->fs;
 946         ext2_ino_t              ino;
 947         errcode_t               retval;
 948         const char * const *    cpp;
 949         int                     group, mount_flags;
 950
 951         clear_problem_context(&pctx);
 952
 953         /*
 954          * If the filesystem is opened read-only, or there is no
 955          * journal, then do nothing.
 956          */
 957         if ((ctx->options & E2F_OPT_READONLY) ||
 958             (sb->s_journal_inum == 0) ||
 959             !(sb->s_feature_compat & EXT3_FEATURE_COMPAT_HAS_JOURNAL))
 960                 return;
 961
 962         /*
 963          * Read in the journal inode
 964          */
 965         if (ext2fs_read_inode(fs, sb->s_journal_inum, &inode) != 0)
 966                 return;
 967
 968         /*
 969          * If it's necessary to backup the journal inode, do so.
 970          */
 971         if ((sb->s_jnl_backup_type == 0) ||
 972             ((sb->s_jnl_backup_type == EXT3_JNL_BACKUP_BLOCKS) &&
 973              memcmp(inode.i_block, sb->s_jnl_blocks, EXT2_N_BLOCKS*4))) {
 974                 if (fix_problem(ctx, PR_0_BACKUP_JNL, &pctx)) {
 975                         memcpy(sb->s_jnl_blocks, inode.i_block,
 976                                EXT2_N_BLOCKS*4);
 977                         sb->s_jnl_blocks[15] = inode.i_size_high;
 978                         sb->s_jnl_blocks[16] = inode.i_size;
 979                         sb->s_jnl_backup_type = EXT3_JNL_BACKUP_BLOCKS;
 980                         ext2fs_mark_super_dirty(fs);
 981                         fs->flags &= ~EXT2_FLAG_MASTER_SB_ONLY;
 982                 }
 983         }
 984
 985         /*
 986          * If the journal is already the hidden inode, then do nothing
 987          */
 988         if (sb->s_journal_inum == EXT2_JOURNAL_INO)
 989                 return;
 990
 991         /*
 992          * The journal inode had better have only one link and not be readable.
 993          */
 994         if (inode.i_links_count != 1)
 995                 return;
 996
 997         /*
 998          * If the filesystem is mounted, or we can't tell whether
 999          * or not it's mounted, do nothing.
1000          */
1001         retval = ext2fs_check_if_mounted(ctx->filesystem_name, &mount_flags);
1002         if (retval || (mount_flags & EXT2_MF_MOUNTED))
1003                 return;
1004
1005         /*
1006          * If we can't find the name of the journal inode, then do
1007          * nothing.
1008          */
1009         for (cpp = journal_names; *cpp; cpp++) {
1010                 retval = ext2fs_lookup(fs, EXT2_ROOT_INO, *cpp,
1011                                        strlen(*cpp), 0, &ino);
1012                 if ((retval == 0) && (ino == sb->s_journal_inum))
1013                         break;
1014         }
1015         if (*cpp == 0)
1016                 return;
1017
1018         /* We need the inode bitmap to be loaded */
1019         retval = ext2fs_read_bitmaps(fs);
1020         if (retval)
1021                 return;
1022
1023         pctx.str = *cpp;
1024         if (!fix_problem(ctx, PR_0_MOVE_JOURNAL, &pctx))
1025                 return;
1026
1027         /*
1028          * OK, we've done all the checks, let's actually move the
1029          * journal inode.  Errors at this point mean we need to force
1030          * an ext2 filesystem check.
1031          */
1032         if ((retval = ext2fs_unlink(fs, EXT2_ROOT_INO, *cpp, ino, 0)) != 0)
1033                 goto err_out;
1034         if ((retval = ext2fs_write_inode(fs, EXT2_JOURNAL_INO, &inode)) != 0)
1035                 goto err_out;
1036         sb->s_journal_inum = EXT2_JOURNAL_INO;
1037         ext2fs_mark_super_dirty(fs);
1038         fs->flags &= ~EXT2_FLAG_MASTER_SB_ONLY;
1039         inode.i_links_count = 0;
1040         inode.i_dtime = ctx->now;
1041         if ((retval = ext2fs_write_inode(fs, ino, &inode)) != 0)
1042                 goto err_out;
1043
1044         group = ext2fs_group_of_ino(fs, ino);
1045         ext2fs_unmark_inode_bitmap2(fs->inode_map, ino);
1046         ext2fs_mark_ib_dirty(fs);
1047         ext2fs_bg_free_inodes_count_set(fs, group, ext2fs_bg_free_inodes_count(fs, group) + 1);
1048         ext2fs_group_desc_csum_set(fs, group);
1049         fs->super->s_free_inodes_count++;
1050         return;
1051
1052 err_out:
1053         pctx.errcode = retval;
1054         fix_problem(ctx, PR_0_ERR_MOVE_JOURNAL, &pctx);
1055         fs->super->s_state &= ~EXT2_VALID_FS;
1056         ext2fs_mark_super_dirty(fs);
1057         return;
1058 }
1059
1060 /*
1061  * This function makes sure the superblock hint for the external
1062  * journal is correct.
1063  */
1064 int e2fsck_fix_ext3_journal_hint(e2fsck_t ctx)
1065 {
1066         struct ext2_super_block *sb = ctx->fs->super;
1067         struct problem_context pctx;
1068         char uuid[37], *journal_name;
1069         struct stat st;
1070
1071         if (!(sb->s_feature_compat & EXT3_FEATURE_COMPAT_HAS_JOURNAL) ||
1072             uuid_is_null(sb->s_journal_uuid))
1073                 return 0;
1074
1075         uuid_unparse(sb->s_journal_uuid, uuid);
1076         journal_name = blkid_get_devname(ctx->blkid, "UUID", uuid);
1077         if (!journal_name)
1078                 return 0;
1079
1080         if (stat(journal_name, &st) < 0)
1081                 return 0;
1082
1083         if (st.st_rdev != sb->s_journal_dev) {
1084                 clear_problem_context(&pctx);
1085                 pctx.num = st.st_rdev;
1086                 if (fix_problem(ctx, PR_0_EXTERNAL_JOURNAL_HINT, &pctx)) {
1087                         sb->s_journal_dev = st.st_rdev;
1088                         ext2fs_mark_super_dirty(ctx->fs);
1089                 }
1090         }
1091
1092         free(journal_name);
1093         return 0;
1094 }