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
 245         clear_problem_context(&pctx);
 246
 247         journal = e2fsck_allocate_memory(ctx, sizeof(journal_t), "journal");
 248         if (!journal) {
 249                 return EXT2_ET_NO_MEMORY;
 250         }
 251
 252         dev_fs = e2fsck_allocate_memory(ctx, 2*sizeof(struct kdev_s), "kdev");
 253         if (!dev_fs) {
 254                 retval = EXT2_ET_NO_MEMORY;
 255                 goto errout;
 256         }
 257         dev_journal = dev_fs+1;
 258
 259         dev_fs->k_ctx = dev_journal->k_ctx = ctx;
 260         dev_fs->k_dev = K_DEV_FS;
 261         dev_journal->k_dev = K_DEV_JOURNAL;
 262
 263         journal->j_dev = dev_journal;
 264         journal->j_fs_dev = dev_fs;
 265         journal->j_inode = NULL;
 266         journal->j_blocksize = ctx->fs->blocksize;
 267
 268         if (uuid_is_null(sb->s_journal_uuid)) {
 269                 if (!sb->s_journal_inum) {
 270                         retval = EXT2_ET_BAD_INODE_NUM;
 271                         goto errout;
 272                 }
 273                 j_inode = e2fsck_allocate_memory(ctx, sizeof(*j_inode),
 274                                                  "journal inode");
 275                 if (!j_inode) {
 276                         retval = EXT2_ET_NO_MEMORY;
 277                         goto errout;
 278                 }
 279
 280                 j_inode->i_ctx = ctx;
 281                 j_inode->i_ino = sb->s_journal_inum;
 282
 283                 if ((retval = ext2fs_read_inode(ctx->fs,
 284                                                 sb->s_journal_inum,
 285                                                 &j_inode->i_ext2))) {
 286                 try_backup_journal:
 287                         if (sb->s_jnl_backup_type != EXT3_JNL_BACKUP_BLOCKS ||
 288                             tried_backup_jnl)
 289                                 goto errout;
 290                         memset(&j_inode->i_ext2, 0, sizeof(struct ext2_inode));
 291                         memcpy(&j_inode->i_ext2.i_block[0], sb->s_jnl_blocks,
 292                                EXT2_N_BLOCKS*4);
 293                         j_inode->i_ext2.i_size_high = sb->s_jnl_blocks[15];
 294                         j_inode->i_ext2.i_size = sb->s_jnl_blocks[16];
 295                         j_inode->i_ext2.i_links_count = 1;
 296                         j_inode->i_ext2.i_mode = LINUX_S_IFREG | 0600;
 297                         e2fsck_use_inode_shortcuts(ctx, 1);
 298                         ctx->stashed_ino = j_inode->i_ino;
 299                         ctx->stashed_inode = &j_inode->i_ext2;
 300                         tried_backup_jnl++;
 301                 }
 302                 if (!j_inode->i_ext2.i_links_count ||
 303                     !LINUX_S_ISREG(j_inode->i_ext2.i_mode)) {
 304                         retval = EXT2_ET_NO_JOURNAL;
 305                         goto try_backup_journal;
 306                 }
 307                 if (EXT2_I_SIZE(&j_inode->i_ext2) / journal->j_blocksize <
 308                     JFS_MIN_JOURNAL_BLOCKS) {
 309                         retval = EXT2_ET_JOURNAL_TOO_SMALL;
 310                         goto try_backup_journal;
 311                 }
 312                 pb.last_block = -1;
 313                 retval = ext2fs_block_iterate3(ctx->fs, j_inode->i_ino,
 314                                                BLOCK_FLAG_HOLE, 0,
 315                                                process_journal_block, &pb);
 316                 if ((pb.last_block + 1) * ctx->fs->blocksize <
 317                     (int) EXT2_I_SIZE(&j_inode->i_ext2)) {
 318                         retval = EXT2_ET_JOURNAL_TOO_SMALL;
 319                         goto try_backup_journal;
 320                 }
 321                 if (tried_backup_jnl && !(ctx->options & E2F_OPT_READONLY)) {
 322                         retval = ext2fs_write_inode(ctx->fs, sb->s_journal_inum,
 323                                                     &j_inode->i_ext2);
 324                         if (retval)
 325                                 goto errout;
 326                 }
 327
 328                 journal->j_maxlen = EXT2_I_SIZE(&j_inode->i_ext2) /
 329                         journal->j_blocksize;
 330
 331 #ifdef USE_INODE_IO
 332                 retval = ext2fs_inode_io_intern2(ctx->fs, sb->s_journal_inum,
 333                                                  &j_inode->i_ext2,
 334                                                  &journal_name);
 335                 if (retval)
 336                         goto errout;
 337
 338                 io_ptr = inode_io_manager;
 339 #else
 340                 journal->j_inode = j_inode;
 341                 ctx->journal_io = ctx->fs->io;
 342                 if ((retval = (errcode_t) journal_bmap(journal, 0, &start)) != 0)
 343                         goto errout;
 344 #endif
 345         } else {
 346                 ext_journal = 1;
 347                 if (!ctx->journal_name) {
 348                         char uuid[37];
 349
 350                         uuid_unparse(sb->s_journal_uuid, uuid);
 351                         ctx->journal_name = blkid_get_devname(ctx->blkid,
 352                                                               "UUID", uuid);
 353                         if (!ctx->journal_name)
 354                                 ctx->journal_name = blkid_devno_to_devname(sb->s_journal_dev);
 355                 }
 356                 journal_name = ctx->journal_name;
 357
 358                 if (!journal_name) {
 359                         fix_problem(ctx, PR_0_CANT_FIND_JOURNAL, &pctx);
 360                         retval = EXT2_ET_LOAD_EXT_JOURNAL;
 361                         goto errout;
 362                 }
 363
 364                 jfs_debug(1, "Using journal file %s\n", journal_name);
 365                 io_ptr = unix_io_manager;
 366         }
 367
 368 #if 0
 369         test_io_backing_manager = io_ptr;
 370         io_ptr = test_io_manager;
 371 #endif
 372 #ifndef USE_INODE_IO
 373         if (ext_journal)
 374 #endif
 375         {
 376                 int flags = IO_FLAG_RW;
 377                 if (!(ctx->mount_flags & EXT2_MF_ISROOT &&
 378                       ctx->mount_flags & EXT2_MF_READONLY))
 379                         flags |= IO_FLAG_EXCLUSIVE;
 380                 if ((ctx->mount_flags & EXT2_MF_READONLY) &&
 381                     (ctx->options & E2F_OPT_FORCE))
 382                         flags &= ~IO_FLAG_EXCLUSIVE;
 383
 384
 385                 retval = io_ptr->open(journal_name, flags,
 386                                       &ctx->journal_io);
 387         }
 388         if (retval)
 389                 goto errout;
 390
 391         io_channel_set_blksize(ctx->journal_io, ctx->fs->blocksize);
 392
 393         if (ext_journal) {
 394                 blk64_t maxlen;
 395
 396                 if (ctx->fs->blocksize == 1024)
 397                         start = 1;
 398                 bh = getblk(dev_journal, start, ctx->fs->blocksize);
 399                 if (!bh) {
 400                         retval = EXT2_ET_NO_MEMORY;
 401                         goto errout;
 402                 }
 403                 ll_rw_block(READ, 1, &bh);
 404                 if ((retval = bh->b_err) != 0) {
 405                         brelse(bh);
 406                         goto errout;
 407                 }
 408                 memcpy(&jsuper, start ? bh->b_data :  bh->b_data + 1024,
 409                        sizeof(jsuper));
 410                 brelse(bh);
 411 #ifdef WORDS_BIGENDIAN
 412                 if (jsuper.s_magic == ext2fs_swab16(EXT2_SUPER_MAGIC))
 413                         ext2fs_swap_super(&jsuper);
 414 #endif
 415                 if (jsuper.s_magic != EXT2_SUPER_MAGIC ||
 416                     !(jsuper.s_feature_incompat & EXT3_FEATURE_INCOMPAT_JOURNAL_DEV)) {
 417                         fix_problem(ctx, PR_0_EXT_JOURNAL_BAD_SUPER, &pctx);
 418                         retval = EXT2_ET_LOAD_EXT_JOURNAL;
 419                         goto errout;
 420                 }
 421                 /* Make sure the journal UUID is correct */
 422                 if (memcmp(jsuper.s_uuid, ctx->fs->super->s_journal_uuid,
 423                            sizeof(jsuper.s_uuid))) {
 424                         fix_problem(ctx, PR_0_JOURNAL_BAD_UUID, &pctx);
 425                         retval = EXT2_ET_LOAD_EXT_JOURNAL;
 426                         goto errout;
 427                 }
 428
 429                 maxlen = ext2fs_blocks_count(&jsuper);
 430                 journal->j_maxlen = (maxlen < 1ULL << 32) ? maxlen : (1ULL << 32) - 1;
 431                 start++;
 432         }
 433
 434         if (!(bh = getblk(dev_journal, start, journal->j_blocksize))) {
 435                 retval = EXT2_ET_NO_MEMORY;
 436                 goto errout;
 437         }
 438
 439         journal->j_sb_buffer = bh;
 440         journal->j_superblock = (journal_superblock_t *)bh->b_data;
 441
 442 #ifdef USE_INODE_IO
 443         if (j_inode)
 444                 ext2fs_free_mem(&j_inode);
 445 #endif
 446
 447         *ret_journal = journal;
 448         e2fsck_use_inode_shortcuts(ctx, 0);
 449         return 0;
 450
 451 errout:
 452         e2fsck_use_inode_shortcuts(ctx, 0);
 453         if (dev_fs)
 454                 ext2fs_free_mem(&dev_fs);
 455         if (j_inode)
 456                 ext2fs_free_mem(&j_inode);
 457         if (journal)
 458                 ext2fs_free_mem(&journal);
 459         return retval;
 460 }
 461
 462 static errcode_t e2fsck_journal_fix_bad_inode(e2fsck_t ctx,
 463                                               struct problem_context *pctx)
 464 {
 465         struct ext2_super_block *sb = ctx->fs->super;
 466         int recover = ctx->fs->super->s_feature_incompat &
 467                 EXT3_FEATURE_INCOMPAT_RECOVER;
 468         int has_journal = ctx->fs->super->s_feature_compat &
 469                 EXT3_FEATURE_COMPAT_HAS_JOURNAL;
 470
 471         if (has_journal || sb->s_journal_inum) {
 472                 /* The journal inode is bogus, remove and force full fsck */
 473                 pctx->ino = sb->s_journal_inum;
 474                 if (fix_problem(ctx, PR_0_JOURNAL_BAD_INODE, pctx)) {
 475                         if (has_journal && sb->s_journal_inum)
 476                                 printf("*** ext3 journal has been deleted - "
 477                                        "filesystem is now ext2 only ***\n\n");
 478                         sb->s_feature_compat &= ~EXT3_FEATURE_COMPAT_HAS_JOURNAL;
 479                         sb->s_journal_inum = 0;
 480                         ctx->flags |= E2F_FLAG_JOURNAL_INODE;
 481                         ctx->fs->flags &= ~EXT2_FLAG_MASTER_SB_ONLY;
 482                         e2fsck_clear_recover(ctx, 1);
 483                         return 0;
 484                 }
 485                 return EXT2_ET_BAD_INODE_NUM;
 486         } else if (recover) {
 487                 if (fix_problem(ctx, PR_0_JOURNAL_RECOVER_SET, pctx)) {
 488                         e2fsck_clear_recover(ctx, 1);
 489                         return 0;
 490                 }
 491                 return EXT2_ET_UNSUPP_FEATURE;
 492         }
 493         return 0;
 494 }
 495
 496 #define V1_SB_SIZE      0x0024
 497 static void clear_v2_journal_fields(journal_t *journal)
 498 {
 499         e2fsck_t ctx = journal->j_dev->k_ctx;
 500         struct problem_context pctx;
 501
 502         clear_problem_context(&pctx);
 503
 504         if (!fix_problem(ctx, PR_0_CLEAR_V2_JOURNAL, &pctx))
 505                 return;
 506
 507         memset(((char *) journal->j_superblock) + V1_SB_SIZE, 0,
 508                ctx->fs->blocksize-V1_SB_SIZE);
 509         mark_buffer_dirty(journal->j_sb_buffer);
 510 }
 511
 512
 513 static errcode_t e2fsck_journal_load(journal_t *journal)
 514 {
 515         e2fsck_t ctx = journal->j_dev->k_ctx;
 516         journal_superblock_t *jsb;
 517         struct buffer_head *jbh = journal->j_sb_buffer;
 518         struct problem_context pctx;
 519
 520         clear_problem_context(&pctx);
 521
 522         ll_rw_block(READ, 1, &jbh);
 523         if (jbh->b_err) {
 524                 com_err(ctx->device_name, jbh->b_err, "%s",
 525                         _("reading journal superblock\n"));
 526                 return jbh->b_err;
 527         }
 528
 529         jsb = journal->j_superblock;
 530         /* If we don't even have JFS_MAGIC, we probably have a wrong inode */
 531         if (jsb->s_header.h_magic != htonl(JFS_MAGIC_NUMBER))
 532                 return e2fsck_journal_fix_bad_inode(ctx, &pctx);
 533
 534         switch (ntohl(jsb->s_header.h_blocktype)) {
 535         case JFS_SUPERBLOCK_V1:
 536                 journal->j_format_version = 1;
 537                 if (jsb->s_feature_compat ||
 538                     jsb->s_feature_incompat ||
 539                     jsb->s_feature_ro_compat ||
 540                     jsb->s_nr_users)
 541                         clear_v2_journal_fields(journal);
 542                 break;
 543
 544         case JFS_SUPERBLOCK_V2:
 545                 journal->j_format_version = 2;
 546                 if (ntohl(jsb->s_nr_users) > 1 &&
 547                     uuid_is_null(ctx->fs->super->s_journal_uuid))
 548                         clear_v2_journal_fields(journal);
 549                 if (ntohl(jsb->s_nr_users) > 1) {
 550                         fix_problem(ctx, PR_0_JOURNAL_UNSUPP_MULTIFS, &pctx);
 551                         return EXT2_ET_JOURNAL_UNSUPP_VERSION;
 552                 }
 553                 break;
 554
 555         /*
 556          * These should never appear in a journal super block, so if
 557          * they do, the journal is badly corrupted.
 558          */
 559         case JFS_DESCRIPTOR_BLOCK:
 560         case JFS_COMMIT_BLOCK:
 561         case JFS_REVOKE_BLOCK:
 562                 return EXT2_ET_CORRUPT_SUPERBLOCK;
 563
 564         /* If we don't understand the superblock major type, but there
 565          * is a magic number, then it is likely to be a new format we
 566          * just don't understand, so leave it alone. */
 567         default:
 568                 return EXT2_ET_JOURNAL_UNSUPP_VERSION;
 569         }
 570
 571         if (JFS_HAS_INCOMPAT_FEATURE(journal, ~JFS_KNOWN_INCOMPAT_FEATURES))
 572                 return EXT2_ET_UNSUPP_FEATURE;
 573
 574         if (JFS_HAS_RO_COMPAT_FEATURE(journal, ~JFS_KNOWN_ROCOMPAT_FEATURES))
 575                 return EXT2_ET_RO_UNSUPP_FEATURE;
 576
 577         /* We have now checked whether we know enough about the journal
 578          * format to be able to proceed safely, so any other checks that
 579          * fail we should attempt to recover from. */
 580         if (jsb->s_blocksize != htonl(journal->j_blocksize)) {
 581                 com_err(ctx->program_name, EXT2_ET_CORRUPT_SUPERBLOCK,
 582                         _("%s: no valid journal superblock found\n"),
 583                         ctx->device_name);
 584                 return EXT2_ET_CORRUPT_SUPERBLOCK;
 585         }
 586
 587         if (ntohl(jsb->s_maxlen) < journal->j_maxlen)
 588                 journal->j_maxlen = ntohl(jsb->s_maxlen);
 589         else if (ntohl(jsb->s_maxlen) > journal->j_maxlen) {
 590                 com_err(ctx->program_name, EXT2_ET_CORRUPT_SUPERBLOCK,
 591                         _("%s: journal too short\n"),
 592                         ctx->device_name);
 593                 return EXT2_ET_CORRUPT_SUPERBLOCK;
 594         }
 595
 596         journal->j_tail_sequence = ntohl(jsb->s_sequence);
 597         journal->j_transaction_sequence = journal->j_tail_sequence;
 598         journal->j_tail = ntohl(jsb->s_start);
 599         journal->j_first = ntohl(jsb->s_first);
 600         journal->j_last = ntohl(jsb->s_maxlen);
 601
 602         return 0;
 603 }
 604
 605 static void e2fsck_journal_reset_super(e2fsck_t ctx, journal_superblock_t *jsb,
 606                                        journal_t *journal)
 607 {
 608         char *p;
 609         union {
 610                 uuid_t uuid;
 611                 __u32 val[4];
 612         } u;
 613         __u32 new_seq = 0;
 614         int i;
 615
 616         /* Leave a valid existing V1 superblock signature alone.
 617          * Anything unrecognisable we overwrite with a new V2
 618          * signature. */
 619
 620         if (jsb->s_header.h_magic != htonl(JFS_MAGIC_NUMBER) ||
 621             jsb->s_header.h_blocktype != htonl(JFS_SUPERBLOCK_V1)) {
 622                 jsb->s_header.h_magic = htonl(JFS_MAGIC_NUMBER);
 623                 jsb->s_header.h_blocktype = htonl(JFS_SUPERBLOCK_V2);
 624         }
 625
 626         /* Zero out everything else beyond the superblock header */
 627
 628         p = ((char *) jsb) + sizeof(journal_header_t);
 629         memset (p, 0, ctx->fs->blocksize-sizeof(journal_header_t));
 630
 631         jsb->s_blocksize = htonl(ctx->fs->blocksize);
 632         jsb->s_maxlen = htonl(journal->j_maxlen);
 633         jsb->s_first = htonl(1);
 634
 635         /* Initialize the journal sequence number so that there is "no"
 636          * chance we will find old "valid" transactions in the journal.
 637          * This avoids the need to zero the whole journal (slow to do,
 638          * and risky when we are just recovering the filesystem).
 639          */
 640         uuid_generate(u.uuid);
 641         for (i = 0; i < 4; i ++)
 642                 new_seq ^= u.val[i];
 643         jsb->s_sequence = htonl(new_seq);
 644
 645         mark_buffer_dirty(journal->j_sb_buffer);
 646         ll_rw_block(WRITE, 1, &journal->j_sb_buffer);
 647 }
 648
 649 static errcode_t e2fsck_journal_fix_corrupt_super(e2fsck_t ctx,
 650                                                   journal_t *journal,
 651                                                   struct problem_context *pctx)
 652 {
 653         struct ext2_super_block *sb = ctx->fs->super;
 654         int recover = ctx->fs->super->s_feature_incompat &
 655                 EXT3_FEATURE_INCOMPAT_RECOVER;
 656
 657         if (sb->s_feature_compat & EXT3_FEATURE_COMPAT_HAS_JOURNAL) {
 658                 if (fix_problem(ctx, PR_0_JOURNAL_BAD_SUPER, pctx)) {
 659                         e2fsck_journal_reset_super(ctx, journal->j_superblock,
 660                                                    journal);
 661                         journal->j_transaction_sequence = 1;
 662                         e2fsck_clear_recover(ctx, recover);
 663                         return 0;
 664                 }
 665                 return EXT2_ET_CORRUPT_SUPERBLOCK;
 666         } else if (e2fsck_journal_fix_bad_inode(ctx, pctx))
 667                 return EXT2_ET_CORRUPT_SUPERBLOCK;
 668
 669         return 0;
 670 }
 671
 672 static void e2fsck_journal_release(e2fsck_t ctx, journal_t *journal,
 673                                    int reset, int drop)
 674 {
 675         journal_superblock_t *jsb;
 676
 677         if (drop)
 678                 mark_buffer_clean(journal->j_sb_buffer);
 679         else if (!(ctx->options & E2F_OPT_READONLY)) {
 680                 jsb = journal->j_superblock;
 681                 jsb->s_sequence = htonl(journal->j_transaction_sequence);
 682                 if (reset)
 683                         jsb->s_start = 0; /* this marks the journal as empty */
 684                 mark_buffer_dirty(journal->j_sb_buffer);
 685         }
 686         brelse(journal->j_sb_buffer);
 687
 688         if (ctx->journal_io) {
 689                 if (ctx->fs && ctx->fs->io != ctx->journal_io)
 690                         io_channel_close(ctx->journal_io);
 691                 ctx->journal_io = 0;
 692         }
 693
 694 #ifndef USE_INODE_IO
 695         if (journal->j_inode)
 696                 ext2fs_free_mem(&journal->j_inode);
 697 #endif
 698         if (journal->j_fs_dev)
 699                 ext2fs_free_mem(&journal->j_fs_dev);
 700         ext2fs_free_mem(&journal);
 701 }
 702
 703 /*
 704  * This function makes sure that the superblock fields regarding the
 705  * journal are consistent.
 706  */
 707 errcode_t e2fsck_check_ext3_journal(e2fsck_t ctx)
 708 {
 709         struct ext2_super_block *sb = ctx->fs->super;
 710         journal_t *journal;
 711         int recover = ctx->fs->super->s_feature_incompat &
 712                 EXT3_FEATURE_INCOMPAT_RECOVER;
 713         struct problem_context pctx;
 714         problem_t problem;
 715         int reset = 0, force_fsck = 0;
 716         errcode_t retval;
 717
 718         /* If we don't have any journal features, don't do anything more */
 719         if (!(sb->s_feature_compat & EXT3_FEATURE_COMPAT_HAS_JOURNAL) &&
 720             !recover && sb->s_journal_inum == 0 && sb->s_journal_dev == 0 &&
 721             uuid_is_null(sb->s_journal_uuid))
 722                 return 0;
 723
 724         clear_problem_context(&pctx);
 725         pctx.num = sb->s_journal_inum;
 726
 727         retval = e2fsck_get_journal(ctx, &journal);
 728         if (retval) {
 729                 if ((retval == EXT2_ET_BAD_INODE_NUM) ||
 730                     (retval == EXT2_ET_BAD_BLOCK_NUM) ||
 731                     (retval == EXT2_ET_JOURNAL_TOO_SMALL) ||
 732                     (retval == EXT2_ET_NO_JOURNAL))
 733                         return e2fsck_journal_fix_bad_inode(ctx, &pctx);
 734                 return retval;
 735         }
 736
 737         retval = e2fsck_journal_load(journal);
 738         if (retval) {
 739                 if ((retval == EXT2_ET_CORRUPT_SUPERBLOCK) ||
 740                     ((retval == EXT2_ET_UNSUPP_FEATURE) &&
 741                     (!fix_problem(ctx, PR_0_JOURNAL_UNSUPP_INCOMPAT,
 742                                   &pctx))) ||
 743                     ((retval == EXT2_ET_RO_UNSUPP_FEATURE) &&
 744                     (!fix_problem(ctx, PR_0_JOURNAL_UNSUPP_ROCOMPAT,
 745                                   &pctx))) ||
 746                     ((retval == EXT2_ET_JOURNAL_UNSUPP_VERSION) &&
 747                     (!fix_problem(ctx, PR_0_JOURNAL_UNSUPP_VERSION, &pctx))))
 748                         retval = e2fsck_journal_fix_corrupt_super(ctx, journal,
 749                                                                   &pctx);
 750                 e2fsck_journal_release(ctx, journal, 0, 1);
 751                 return retval;
 752         }
 753
 754         /*
 755          * We want to make the flags consistent here.  We will not leave with
 756          * needs_recovery set but has_journal clear.  We can't get in a loop
 757          * with -y, -n, or -p, only if a user isn't making up their mind.
 758          */
 759 no_has_journal:
 760         if (!(sb->s_feature_compat & EXT3_FEATURE_COMPAT_HAS_JOURNAL)) {
 761                 recover = sb->s_feature_incompat & EXT3_FEATURE_INCOMPAT_RECOVER;
 762                 pctx.str = "inode";
 763                 if (fix_problem(ctx, PR_0_JOURNAL_HAS_JOURNAL, &pctx)) {
 764                         if (recover &&
 765                             !fix_problem(ctx, PR_0_JOURNAL_RECOVER_SET, &pctx))
 766                                 goto no_has_journal;
 767                         /*
 768                          * Need a full fsck if we are releasing a
 769                          * journal stored on a reserved inode.
 770                          */
 771                         force_fsck = recover ||
 772                                 (sb->s_journal_inum < EXT2_FIRST_INODE(sb));
 773                         /* Clear all of the journal fields */
 774                         sb->s_journal_inum = 0;
 775                         sb->s_journal_dev = 0;
 776                         memset(sb->s_journal_uuid, 0,
 777                                sizeof(sb->s_journal_uuid));
 778                         e2fsck_clear_recover(ctx, force_fsck);
 779                 } else if (!(ctx->options & E2F_OPT_READONLY)) {
 780                         sb->s_feature_compat |= EXT3_FEATURE_COMPAT_HAS_JOURNAL;
 781                         ctx->fs->flags &= ~EXT2_FLAG_MASTER_SB_ONLY;
 782                         ext2fs_mark_super_dirty(ctx->fs);
 783                 }
 784         }
 785
 786         if (sb->s_feature_compat & EXT3_FEATURE_COMPAT_HAS_JOURNAL &&
 787             !(sb->s_feature_incompat & EXT3_FEATURE_INCOMPAT_RECOVER) &&
 788             journal->j_superblock->s_start != 0) {
 789                 /* Print status information */
 790                 fix_problem(ctx, PR_0_JOURNAL_RECOVERY_CLEAR, &pctx);
 791                 if (ctx->superblock)
 792                         problem = PR_0_JOURNAL_RUN_DEFAULT;
 793                 else
 794                         problem = PR_0_JOURNAL_RUN;
 795                 if (fix_problem(ctx, problem, &pctx)) {
 796                         ctx->options |= E2F_OPT_FORCE;
 797                         sb->s_feature_incompat |=
 798                                 EXT3_FEATURE_INCOMPAT_RECOVER;
 799                         ext2fs_mark_super_dirty(ctx->fs);
 800                 } else if (fix_problem(ctx,
 801                                        PR_0_JOURNAL_RESET_JOURNAL, &pctx)) {
 802                         reset = 1;
 803                         sb->s_state &= ~EXT2_VALID_FS;
 804                         ext2fs_mark_super_dirty(ctx->fs);
 805                 }
 806                 /*
 807                  * If the user answers no to the above question, we
 808                  * ignore the fact that journal apparently has data;
 809                  * accidentally replaying over valid data would be far
 810                  * worse than skipping a questionable recovery.
 811                  *
 812                  * XXX should we abort with a fatal error here?  What
 813                  * will the ext3 kernel code do if a filesystem with
 814                  * !NEEDS_RECOVERY but with a non-zero
 815                  * journal->j_superblock->s_start is mounted?
 816                  */
 817         }
 818
 819         /*
 820          * If we don't need to do replay the journal, check to see if
 821          * the journal's errno is set; if so, we need to mark the file
 822          * system as being corrupt and clear the journal's s_errno.
 823          */
 824         if (!(sb->s_feature_incompat & EXT3_FEATURE_INCOMPAT_RECOVER) &&
 825             journal->j_superblock->s_errno) {
 826                 ctx->fs->super->s_state |= EXT2_ERROR_FS;
 827                 ext2fs_mark_super_dirty(ctx->fs);
 828                 journal->j_superblock->s_errno = 0;
 829                 mark_buffer_dirty(journal->j_sb_buffer);
 830         }
 831
 832         e2fsck_journal_release(ctx, journal, reset, 0);
 833         return retval;
 834 }
 835
 836 static errcode_t recover_ext3_journal(e2fsck_t ctx)
 837 {
 838         struct problem_context  pctx;
 839         journal_t *journal;
 840         errcode_t retval;
 841
 842         clear_problem_context(&pctx);
 843
 844         journal_init_revoke_caches();
 845         retval = e2fsck_get_journal(ctx, &journal);
 846         if (retval)
 847                 return retval;
 848
 849         retval = e2fsck_journal_load(journal);
 850         if (retval)
 851                 goto errout;
 852
 853         retval = journal_init_revoke(journal, 1024);
 854         if (retval)
 855                 goto errout;
 856
 857         retval = -journal_recover(journal);
 858         if (retval)
 859                 goto errout;
 860
 861         if (journal->j_failed_commit) {
 862                 pctx.ino = journal->j_failed_commit;
 863                 fix_problem(ctx, PR_0_JNL_TXN_CORRUPT, &pctx);
 864                 journal->j_superblock->s_errno = -EINVAL;
 865                 mark_buffer_dirty(journal->j_sb_buffer);
 866         }
 867
 868 errout:
 869         journal_destroy_revoke(journal);
 870         journal_destroy_revoke_caches();
 871         e2fsck_journal_release(ctx, journal, 1, 0);
 872         return retval;
 873 }
 874
 875 errcode_t e2fsck_run_ext3_journal(e2fsck_t ctx)
 876 {
 877         io_manager io_ptr = ctx->fs->io->manager;
 878         int blocksize = ctx->fs->blocksize;
 879         errcode_t       retval, recover_retval;
 880         io_stats        stats = 0;
 881         unsigned long long kbytes_written = 0;
 882
 883         printf(_("%s: recovering journal\n"), ctx->device_name);
 884         if (ctx->options & E2F_OPT_READONLY) {
 885                 printf(_("%s: won't do journal recovery while read-only\n"),
 886                        ctx->device_name);
 887                 return EXT2_ET_FILE_RO;
 888         }
 889
 890         if (ctx->fs->flags & EXT2_FLAG_DIRTY)
 891                 ext2fs_flush(ctx->fs);  /* Force out any modifications */
 892
 893         recover_retval = recover_ext3_journal(ctx);
 894
 895         /*
 896          * Reload the filesystem context to get up-to-date data from disk
 897          * because journal recovery will change the filesystem under us.
 898          */
 899         if (ctx->fs->super->s_kbytes_written &&
 900             ctx->fs->io->manager->get_stats)
 901                 ctx->fs->io->manager->get_stats(ctx->fs->io, &stats);
 902         if (stats && stats->bytes_written)
 903                 kbytes_written = stats->bytes_written >> 10;
 904
 905         ext2fs_mmp_stop(ctx->fs);
 906         ext2fs_free(ctx->fs);
 907         retval = ext2fs_open(ctx->filesystem_name, EXT2_FLAG_RW,
 908                              ctx->superblock, blocksize, io_ptr,
 909                              &ctx->fs);
 910         if (retval) {
 911                 com_err(ctx->program_name, retval,
 912                         _("while trying to re-open %s"),
 913                         ctx->device_name);
 914                 fatal_error(ctx, 0);
 915         }
 916         ctx->fs->priv_data = ctx;
 917         ctx->fs->now = ctx->now;
 918         ctx->fs->flags |= EXT2_FLAG_MASTER_SB_ONLY;
 919         ctx->fs->super->s_kbytes_written += kbytes_written;
 920
 921         /* Set the superblock flags */
 922         e2fsck_clear_recover(ctx, recover_retval != 0);
 923
 924         /*
 925          * Do one last sanity check, and propagate journal->s_errno to
 926          * the EXT2_ERROR_FS flag in the fs superblock if needed.
 927          */
 928         retval = e2fsck_check_ext3_journal(ctx);
 929         return retval ? retval : recover_retval;
 930 }
 931
 932 /*
 933  * This function will move the journal inode from a visible file in
 934  * the filesystem directory hierarchy to the reserved inode if necessary.
 935  */
 936 static const char * const journal_names[] = {
 937         ".journal", "journal", ".journal.dat", "journal.dat", 0 };
 938
 939 void e2fsck_move_ext3_journal(e2fsck_t ctx)
 940 {
 941         struct ext2_super_block *sb = ctx->fs->super;
 942         struct problem_context  pctx;
 943         struct ext2_inode       inode;
 944         ext2_filsys             fs = ctx->fs;
 945         ext2_ino_t              ino;
 946         errcode_t               retval;
 947         const char * const *    cpp;
 948         dgrp_t                  group;
 949         int                     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                 free(journal_name);
1082                 return 0;
1083         }
1084
1085         if (st.st_rdev != sb->s_journal_dev) {
1086                 clear_problem_context(&pctx);
1087                 pctx.num = st.st_rdev;
1088                 if (fix_problem(ctx, PR_0_EXTERNAL_JOURNAL_HINT, &pctx)) {
1089                         sb->s_journal_dev = st.st_rdev;
1090                         ext2fs_mark_super_dirty(ctx->fs);
1091                 }
1092         }
1093
1094         free(journal_name);
1095         return 0;
1096 }