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 (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         int 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 = 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 = 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 = 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                 if (ctx->fs->blocksize == 1024)
 395                         start = 1;
 396                 bh = getblk(dev_journal, start, ctx->fs->blocksize);
 397                 if (!bh) {
 398                         retval = EXT2_ET_NO_MEMORY;
 399                         goto errout;
 400                 }
 401                 ll_rw_block(READ, 1, &bh);
 402                 if ((retval = bh->b_err) != 0) {
 403                         brelse(bh);
 404                         goto errout;
 405                 }
 406                 memcpy(&jsuper, start ? bh->b_data :  bh->b_data + 1024,
 407                        sizeof(jsuper));
 408                 brelse(bh);
 409 #ifdef WORDS_BIGENDIAN
 410                 if (jsuper.s_magic == ext2fs_swab16(EXT2_SUPER_MAGIC))
 411                         ext2fs_swap_super(&jsuper);
 412 #endif
 413                 if (jsuper.s_magic != EXT2_SUPER_MAGIC ||
 414                     !(jsuper.s_feature_incompat & EXT3_FEATURE_INCOMPAT_JOURNAL_DEV)) {
 415                         fix_problem(ctx, PR_0_EXT_JOURNAL_BAD_SUPER, &pctx);
 416                         retval = EXT2_ET_LOAD_EXT_JOURNAL;
 417                         goto errout;
 418                 }
 419                 /* Make sure the journal UUID is correct */
 420                 if (memcmp(jsuper.s_uuid, ctx->fs->super->s_journal_uuid,
 421                            sizeof(jsuper.s_uuid))) {
 422                         fix_problem(ctx, PR_0_JOURNAL_BAD_UUID, &pctx);
 423                         retval = EXT2_ET_LOAD_EXT_JOURNAL;
 424                         goto errout;
 425                 }
 426
 427                 journal->j_maxlen = ext2fs_blocks_count(&jsuper);
 428                 start++;
 429         }
 430
 431         if (!(bh = getblk(dev_journal, start, journal->j_blocksize))) {
 432                 retval = EXT2_ET_NO_MEMORY;
 433                 goto errout;
 434         }
 435
 436         journal->j_sb_buffer = bh;
 437         journal->j_superblock = (journal_superblock_t *)bh->b_data;
 438
 439 #ifdef USE_INODE_IO
 440         if (j_inode)
 441                 ext2fs_free_mem(&j_inode);
 442 #endif
 443
 444         *ret_journal = journal;
 445         e2fsck_use_inode_shortcuts(ctx, 0);
 446         return 0;
 447
 448 errout:
 449         e2fsck_use_inode_shortcuts(ctx, 0);
 450         if (dev_fs)
 451                 ext2fs_free_mem(&dev_fs);
 452         if (j_inode)
 453                 ext2fs_free_mem(&j_inode);
 454         if (journal)
 455                 ext2fs_free_mem(&journal);
 456         return retval;
 457 }
 458
 459 static errcode_t e2fsck_journal_fix_bad_inode(e2fsck_t ctx,
 460                                               struct problem_context *pctx)
 461 {
 462         struct ext2_super_block *sb = ctx->fs->super;
 463         int recover = ctx->fs->super->s_feature_incompat &
 464                 EXT3_FEATURE_INCOMPAT_RECOVER;
 465         int has_journal = ctx->fs->super->s_feature_compat &
 466                 EXT3_FEATURE_COMPAT_HAS_JOURNAL;
 467
 468         if (has_journal || sb->s_journal_inum) {
 469                 /* The journal inode is bogus, remove and force full fsck */
 470                 pctx->ino = sb->s_journal_inum;
 471                 if (fix_problem(ctx, PR_0_JOURNAL_BAD_INODE, pctx)) {
 472                         if (has_journal && sb->s_journal_inum)
 473                                 printf("*** ext3 journal has been deleted - "
 474                                        "filesystem is now ext2 only ***\n\n");
 475                         sb->s_feature_compat &= ~EXT3_FEATURE_COMPAT_HAS_JOURNAL;
 476                         sb->s_journal_inum = 0;
 477                         ctx->flags |= E2F_FLAG_JOURNAL_INODE;
 478                         ctx->fs->flags &= ~EXT2_FLAG_MASTER_SB_ONLY;
 479                         e2fsck_clear_recover(ctx, 1);
 480                         return 0;
 481                 }
 482                 return EXT2_ET_BAD_INODE_NUM;
 483         } else if (recover) {
 484                 if (fix_problem(ctx, PR_0_JOURNAL_RECOVER_SET, pctx)) {
 485                         e2fsck_clear_recover(ctx, 1);
 486                         return 0;
 487                 }
 488                 return EXT2_ET_UNSUPP_FEATURE;
 489         }
 490         return 0;
 491 }
 492
 493 #define V1_SB_SIZE      0x0024
 494 static void clear_v2_journal_fields(journal_t *journal)
 495 {
 496         e2fsck_t ctx = journal->j_dev->k_ctx;
 497         struct problem_context pctx;
 498
 499         clear_problem_context(&pctx);
 500
 501         if (!fix_problem(ctx, PR_0_CLEAR_V2_JOURNAL, &pctx))
 502                 return;
 503
 504         memset(((char *) journal->j_superblock) + V1_SB_SIZE, 0,
 505                ctx->fs->blocksize-V1_SB_SIZE);
 506         mark_buffer_dirty(journal->j_sb_buffer);
 507 }
 508
 509
 510 static errcode_t e2fsck_journal_load(journal_t *journal)
 511 {
 512         e2fsck_t ctx = journal->j_dev->k_ctx;
 513         journal_superblock_t *jsb;
 514         struct buffer_head *jbh = journal->j_sb_buffer;
 515         struct problem_context pctx;
 516
 517         clear_problem_context(&pctx);
 518
 519         ll_rw_block(READ, 1, &jbh);
 520         if (jbh->b_err) {
 521                 com_err(ctx->device_name, jbh->b_err,
 522                         _("reading journal superblock\n"));
 523                 return jbh->b_err;
 524         }
 525
 526         jsb = journal->j_superblock;
 527         /* If we don't even have JFS_MAGIC, we probably have a wrong inode */
 528         if (jsb->s_header.h_magic != htonl(JFS_MAGIC_NUMBER))
 529                 return e2fsck_journal_fix_bad_inode(ctx, &pctx);
 530
 531         switch (ntohl(jsb->s_header.h_blocktype)) {
 532         case JFS_SUPERBLOCK_V1:
 533                 journal->j_format_version = 1;
 534                 if (jsb->s_feature_compat ||
 535                     jsb->s_feature_incompat ||
 536                     jsb->s_feature_ro_compat ||
 537                     jsb->s_nr_users)
 538                         clear_v2_journal_fields(journal);
 539                 break;
 540
 541         case JFS_SUPERBLOCK_V2:
 542                 journal->j_format_version = 2;
 543                 if (ntohl(jsb->s_nr_users) > 1 &&
 544                     uuid_is_null(ctx->fs->super->s_journal_uuid))
 545                         clear_v2_journal_fields(journal);
 546                 if (ntohl(jsb->s_nr_users) > 1) {
 547                         fix_problem(ctx, PR_0_JOURNAL_UNSUPP_MULTIFS, &pctx);
 548                         return EXT2_ET_JOURNAL_UNSUPP_VERSION;
 549                 }
 550                 break;
 551
 552         /*
 553          * These should never appear in a journal super block, so if
 554          * they do, the journal is badly corrupted.
 555          */
 556         case JFS_DESCRIPTOR_BLOCK:
 557         case JFS_COMMIT_BLOCK:
 558         case JFS_REVOKE_BLOCK:
 559                 return EXT2_ET_CORRUPT_SUPERBLOCK;
 560
 561         /* If we don't understand the superblock major type, but there
 562          * is a magic number, then it is likely to be a new format we
 563          * just don't understand, so leave it alone. */
 564         default:
 565                 return EXT2_ET_JOURNAL_UNSUPP_VERSION;
 566         }
 567
 568         if (JFS_HAS_INCOMPAT_FEATURE(journal, ~JFS_KNOWN_INCOMPAT_FEATURES))
 569                 return EXT2_ET_UNSUPP_FEATURE;
 570
 571         if (JFS_HAS_RO_COMPAT_FEATURE(journal, ~JFS_KNOWN_ROCOMPAT_FEATURES))
 572                 return EXT2_ET_RO_UNSUPP_FEATURE;
 573
 574         /* We have now checked whether we know enough about the journal
 575          * format to be able to proceed safely, so any other checks that
 576          * fail we should attempt to recover from. */
 577         if (jsb->s_blocksize != htonl(journal->j_blocksize)) {
 578                 com_err(ctx->program_name, EXT2_ET_CORRUPT_SUPERBLOCK,
 579                         _("%s: no valid journal superblock found\n"),
 580                         ctx->device_name);
 581                 return EXT2_ET_CORRUPT_SUPERBLOCK;
 582         }
 583
 584         if (ntohl(jsb->s_maxlen) < journal->j_maxlen)
 585                 journal->j_maxlen = ntohl(jsb->s_maxlen);
 586         else if (ntohl(jsb->s_maxlen) > journal->j_maxlen) {
 587                 com_err(ctx->program_name, EXT2_ET_CORRUPT_SUPERBLOCK,
 588                         _("%s: journal too short\n"),
 589                         ctx->device_name);
 590                 return EXT2_ET_CORRUPT_SUPERBLOCK;
 591         }
 592
 593         journal->j_tail_sequence = ntohl(jsb->s_sequence);
 594         journal->j_transaction_sequence = journal->j_tail_sequence;
 595         journal->j_tail = ntohl(jsb->s_start);
 596         journal->j_first = ntohl(jsb->s_first);
 597         journal->j_last = ntohl(jsb->s_maxlen);
 598
 599         return 0;
 600 }
 601
 602 static void e2fsck_journal_reset_super(e2fsck_t ctx, journal_superblock_t *jsb,
 603                                        journal_t *journal)
 604 {
 605         char *p;
 606         union {
 607                 uuid_t uuid;
 608                 __u32 val[4];
 609         } u;
 610         __u32 new_seq = 0;
 611         int i;
 612
 613         /* Leave a valid existing V1 superblock signature alone.
 614          * Anything unrecognisable we overwrite with a new V2
 615          * signature. */
 616
 617         if (jsb->s_header.h_magic != htonl(JFS_MAGIC_NUMBER) ||
 618             jsb->s_header.h_blocktype != htonl(JFS_SUPERBLOCK_V1)) {
 619                 jsb->s_header.h_magic = htonl(JFS_MAGIC_NUMBER);
 620                 jsb->s_header.h_blocktype = htonl(JFS_SUPERBLOCK_V2);
 621         }
 622
 623         /* Zero out everything else beyond the superblock header */
 624
 625         p = ((char *) jsb) + sizeof(journal_header_t);
 626         memset (p, 0, ctx->fs->blocksize-sizeof(journal_header_t));
 627
 628         jsb->s_blocksize = htonl(ctx->fs->blocksize);
 629         jsb->s_maxlen = htonl(journal->j_maxlen);
 630         jsb->s_first = htonl(1);
 631
 632         /* Initialize the journal sequence number so that there is "no"
 633          * chance we will find old "valid" transactions in the journal.
 634          * This avoids the need to zero the whole journal (slow to do,
 635          * and risky when we are just recovering the filesystem).
 636          */
 637         uuid_generate(u.uuid);
 638         for (i = 0; i < 4; i ++)
 639                 new_seq ^= u.val[i];
 640         jsb->s_sequence = htonl(new_seq);
 641
 642         mark_buffer_dirty(journal->j_sb_buffer);
 643         ll_rw_block(WRITE, 1, &journal->j_sb_buffer);
 644 }
 645
 646 static errcode_t e2fsck_journal_fix_corrupt_super(e2fsck_t ctx,
 647                                                   journal_t *journal,
 648                                                   struct problem_context *pctx)
 649 {
 650         struct ext2_super_block *sb = ctx->fs->super;
 651         int recover = ctx->fs->super->s_feature_incompat &
 652                 EXT3_FEATURE_INCOMPAT_RECOVER;
 653
 654         if (sb->s_feature_compat & EXT3_FEATURE_COMPAT_HAS_JOURNAL) {
 655                 if (fix_problem(ctx, PR_0_JOURNAL_BAD_SUPER, pctx)) {
 656                         e2fsck_journal_reset_super(ctx, journal->j_superblock,
 657                                                    journal);
 658                         journal->j_transaction_sequence = 1;
 659                         e2fsck_clear_recover(ctx, recover);
 660                         return 0;
 661                 }
 662                 return EXT2_ET_CORRUPT_SUPERBLOCK;
 663         } else if (e2fsck_journal_fix_bad_inode(ctx, pctx))
 664                 return EXT2_ET_CORRUPT_SUPERBLOCK;
 665
 666         return 0;
 667 }
 668
 669 static void e2fsck_journal_release(e2fsck_t ctx, journal_t *journal,
 670                                    int reset, int drop)
 671 {
 672         journal_superblock_t *jsb;
 673
 674         if (drop)
 675                 mark_buffer_clean(journal->j_sb_buffer);
 676         else if (!(ctx->options & E2F_OPT_READONLY)) {
 677                 jsb = journal->j_superblock;
 678                 jsb->s_sequence = htonl(journal->j_transaction_sequence);
 679                 if (reset)
 680                         jsb->s_start = 0; /* this marks the journal as empty */
 681                 mark_buffer_dirty(journal->j_sb_buffer);
 682         }
 683         brelse(journal->j_sb_buffer);
 684
 685         if (ctx->journal_io) {
 686                 if (ctx->fs && ctx->fs->io != ctx->journal_io)
 687                         io_channel_close(ctx->journal_io);
 688                 ctx->journal_io = 0;
 689         }
 690
 691 #ifndef USE_INODE_IO
 692         if (journal->j_inode)
 693                 ext2fs_free_mem(&journal->j_inode);
 694 #endif
 695         if (journal->j_fs_dev)
 696                 ext2fs_free_mem(&journal->j_fs_dev);
 697         ext2fs_free_mem(&journal);
 698 }
 699
 700 /*
 701  * This function makes sure that the superblock fields regarding the
 702  * journal are consistent.
 703  */
 704 int e2fsck_check_ext3_journal(e2fsck_t ctx)
 705 {
 706         struct ext2_super_block *sb = ctx->fs->super;
 707         journal_t *journal;
 708         int recover = ctx->fs->super->s_feature_incompat &
 709                 EXT3_FEATURE_INCOMPAT_RECOVER;
 710         struct problem_context pctx;
 711         problem_t problem;
 712         int reset = 0, force_fsck = 0;
 713         int retval;
 714
 715         /* If we don't have any journal features, don't do anything more */
 716         if (!(sb->s_feature_compat & EXT3_FEATURE_COMPAT_HAS_JOURNAL) &&
 717             !recover && sb->s_journal_inum == 0 && sb->s_journal_dev == 0 &&
 718             uuid_is_null(sb->s_journal_uuid))
 719                 return 0;
 720
 721         clear_problem_context(&pctx);
 722         pctx.num = sb->s_journal_inum;
 723
 724         retval = e2fsck_get_journal(ctx, &journal);
 725         if (retval) {
 726                 if ((retval == EXT2_ET_BAD_INODE_NUM) ||
 727                     (retval == EXT2_ET_BAD_BLOCK_NUM) ||
 728                     (retval == EXT2_ET_JOURNAL_TOO_SMALL) ||
 729                     (retval == EXT2_ET_NO_JOURNAL))
 730                         return e2fsck_journal_fix_bad_inode(ctx, &pctx);
 731                 return retval;
 732         }
 733
 734         retval = e2fsck_journal_load(journal);
 735         if (retval) {
 736                 if ((retval == EXT2_ET_CORRUPT_SUPERBLOCK) ||
 737                     ((retval == EXT2_ET_UNSUPP_FEATURE) &&
 738                     (!fix_problem(ctx, PR_0_JOURNAL_UNSUPP_INCOMPAT,
 739                                   &pctx))) ||
 740                     ((retval == EXT2_ET_RO_UNSUPP_FEATURE) &&
 741                     (!fix_problem(ctx, PR_0_JOURNAL_UNSUPP_ROCOMPAT,
 742                                   &pctx))) ||
 743                     ((retval == EXT2_ET_JOURNAL_UNSUPP_VERSION) &&
 744                     (!fix_problem(ctx, PR_0_JOURNAL_UNSUPP_VERSION, &pctx))))
 745                         retval = e2fsck_journal_fix_corrupt_super(ctx, journal,
 746                                                                   &pctx);
 747                 e2fsck_journal_release(ctx, journal, 0, 1);
 748                 return retval;
 749         }
 750
 751         /*
 752          * We want to make the flags consistent here.  We will not leave with
 753          * needs_recovery set but has_journal clear.  We can't get in a loop
 754          * with -y, -n, or -p, only if a user isn't making up their mind.
 755          */
 756 no_has_journal:
 757         if (!(sb->s_feature_compat & EXT3_FEATURE_COMPAT_HAS_JOURNAL)) {
 758                 recover = sb->s_feature_incompat & EXT3_FEATURE_INCOMPAT_RECOVER;
 759                 pctx.str = "inode";
 760                 if (fix_problem(ctx, PR_0_JOURNAL_HAS_JOURNAL, &pctx)) {
 761                         if (recover &&
 762                             !fix_problem(ctx, PR_0_JOURNAL_RECOVER_SET, &pctx))
 763                                 goto no_has_journal;
 764                         /*
 765                          * Need a full fsck if we are releasing a
 766                          * journal stored on a reserved inode.
 767                          */
 768                         force_fsck = recover ||
 769                                 (sb->s_journal_inum < EXT2_FIRST_INODE(sb));
 770                         /* Clear all of the journal fields */
 771                         sb->s_journal_inum = 0;
 772                         sb->s_journal_dev = 0;
 773                         memset(sb->s_journal_uuid, 0,
 774                                sizeof(sb->s_journal_uuid));
 775                         e2fsck_clear_recover(ctx, force_fsck);
 776                 } else if (!(ctx->options & E2F_OPT_READONLY)) {
 777                         sb->s_feature_compat |= EXT3_FEATURE_COMPAT_HAS_JOURNAL;
 778                         ctx->fs->flags &= ~EXT2_FLAG_MASTER_SB_ONLY;
 779                         ext2fs_mark_super_dirty(ctx->fs);
 780                 }
 781         }
 782
 783         if (sb->s_feature_compat & EXT3_FEATURE_COMPAT_HAS_JOURNAL &&
 784             !(sb->s_feature_incompat & EXT3_FEATURE_INCOMPAT_RECOVER) &&
 785             journal->j_superblock->s_start != 0) {
 786                 /* Print status information */
 787                 fix_problem(ctx, PR_0_JOURNAL_RECOVERY_CLEAR, &pctx);
 788                 if (ctx->superblock)
 789                         problem = PR_0_JOURNAL_RUN_DEFAULT;
 790                 else
 791                         problem = PR_0_JOURNAL_RUN;
 792                 if (fix_problem(ctx, problem, &pctx)) {
 793                         ctx->options |= E2F_OPT_FORCE;
 794                         sb->s_feature_incompat |=
 795                                 EXT3_FEATURE_INCOMPAT_RECOVER;
 796                         ext2fs_mark_super_dirty(ctx->fs);
 797                 } else if (fix_problem(ctx,
 798                                        PR_0_JOURNAL_RESET_JOURNAL, &pctx)) {
 799                         reset = 1;
 800                         sb->s_state &= ~EXT2_VALID_FS;
 801                         ext2fs_mark_super_dirty(ctx->fs);
 802                 }
 803                 /*
 804                  * If the user answers no to the above question, we
 805                  * ignore the fact that journal apparently has data;
 806                  * accidentally replaying over valid data would be far
 807                  * worse than skipping a questionable recovery.
 808                  *
 809                  * XXX should we abort with a fatal error here?  What
 810                  * will the ext3 kernel code do if a filesystem with
 811                  * !NEEDS_RECOVERY but with a non-zero
 812                  * journal->j_superblock->s_start is mounted?
 813                  */
 814         }
 815
 816         /*
 817          * If we don't need to do replay the journal, check to see if
 818          * the journal's errno is set; if so, we need to mark the file
 819          * system as being corrupt and clear the journal's s_errno.
 820          */
 821         if (!(sb->s_feature_incompat & EXT3_FEATURE_INCOMPAT_RECOVER) &&
 822             journal->j_superblock->s_errno) {
 823                 ctx->fs->super->s_state |= EXT2_ERROR_FS;
 824                 ext2fs_mark_super_dirty(ctx->fs);
 825                 journal->j_superblock->s_errno = 0;
 826                 mark_buffer_dirty(journal->j_sb_buffer);
 827         }
 828
 829         e2fsck_journal_release(ctx, journal, reset, 0);
 830         return retval;
 831 }
 832
 833 static errcode_t recover_ext3_journal(e2fsck_t ctx)
 834 {
 835         struct problem_context  pctx;
 836         journal_t *journal;
 837         int retval;
 838
 839         clear_problem_context(&pctx);
 840
 841         journal_init_revoke_caches();
 842         retval = e2fsck_get_journal(ctx, &journal);
 843         if (retval)
 844                 return retval;
 845
 846         retval = e2fsck_journal_load(journal);
 847         if (retval)
 848                 goto errout;
 849
 850         retval = journal_init_revoke(journal, 1024);
 851         if (retval)
 852                 goto errout;
 853
 854         retval = -journal_recover(journal);
 855         if (retval)
 856                 goto errout;
 857
 858         if (journal->j_failed_commit) {
 859                 pctx.ino = journal->j_failed_commit;
 860                 fix_problem(ctx, PR_0_JNL_TXN_CORRUPT, &pctx);
 861                 journal->j_superblock->s_errno = -EINVAL;
 862                 mark_buffer_dirty(journal->j_sb_buffer);
 863         }
 864
 865 errout:
 866         journal_destroy_revoke(journal);
 867         journal_destroy_revoke_caches();
 868         e2fsck_journal_release(ctx, journal, 1, 0);
 869         return retval;
 870 }
 871
 872 int e2fsck_run_ext3_journal(e2fsck_t ctx)
 873 {
 874         io_manager io_ptr = ctx->fs->io->manager;
 875         int blocksize = ctx->fs->blocksize;
 876         errcode_t       retval, recover_retval;
 877         io_stats        stats = 0;
 878         unsigned long long kbytes_written = 0;
 879
 880         printf(_("%s: recovering journal\n"), ctx->device_name);
 881         if (ctx->options & E2F_OPT_READONLY) {
 882                 printf(_("%s: won't do journal recovery while read-only\n"),
 883                        ctx->device_name);
 884                 return EXT2_ET_FILE_RO;
 885         }
 886
 887         if (ctx->fs->flags & EXT2_FLAG_DIRTY)
 888                 ext2fs_flush(ctx->fs);  /* Force out any modifications */
 889
 890         recover_retval = recover_ext3_journal(ctx);
 891
 892         /*
 893          * Reload the filesystem context to get up-to-date data from disk
 894          * because journal recovery will change the filesystem under us.
 895          */
 896         if (ctx->fs->super->s_kbytes_written &&
 897             ctx->fs->io->manager->get_stats)
 898                 ctx->fs->io->manager->get_stats(ctx->fs->io, &stats);
 899         if (stats && stats->bytes_written)
 900                 kbytes_written = stats->bytes_written >> 10;
 901
 902         ext2fs_mmp_stop(ctx->fs);
 903         ext2fs_free(ctx->fs);
 904         retval = ext2fs_open(ctx->filesystem_name, EXT2_FLAG_RW,
 905                              ctx->superblock, blocksize, io_ptr,
 906                              &ctx->fs);
 907         if (retval) {
 908                 com_err(ctx->program_name, retval,
 909                         _("while trying to re-open %s"),
 910                         ctx->device_name);
 911                 fatal_error(ctx, 0);
 912         }
 913         ctx->fs->priv_data = ctx;
 914         ctx->fs->now = ctx->now;
 915         ctx->fs->flags |= EXT2_FLAG_MASTER_SB_ONLY;
 916         ctx->fs->super->s_kbytes_written += kbytes_written;
 917
 918         /* Set the superblock flags */
 919         e2fsck_clear_recover(ctx, recover_retval);
 920
 921         /*
 922          * Do one last sanity check, and propagate journal->s_errno to
 923          * the EXT2_ERROR_FS flag in the fs superblock if needed.
 924          */
 925         retval = e2fsck_check_ext3_journal(ctx);
 926         return retval ? retval : recover_retval;
 927 }
 928
 929 /*
 930  * This function will move the journal inode from a visible file in
 931  * the filesystem directory hierarchy to the reserved inode if necessary.
 932  */
 933 static const char * const journal_names[] = {
 934         ".journal", "journal", ".journal.dat", "journal.dat", 0 };
 935
 936 void e2fsck_move_ext3_journal(e2fsck_t ctx)
 937 {
 938         struct ext2_super_block *sb = ctx->fs->super;
 939         struct problem_context  pctx;
 940         struct ext2_inode       inode;
 941         ext2_filsys             fs = ctx->fs;
 942         ext2_ino_t              ino;
 943         errcode_t               retval;
 944         const char * const *    cpp;
 945         int                     group, mount_flags;
 946
 947         clear_problem_context(&pctx);
 948
 949         /*
 950          * If the filesystem is opened read-only, or there is no
 951          * journal, then do nothing.
 952          */
 953         if ((ctx->options & E2F_OPT_READONLY) ||
 954             (sb->s_journal_inum == 0) ||
 955             !(sb->s_feature_compat & EXT3_FEATURE_COMPAT_HAS_JOURNAL))
 956                 return;
 957
 958         /*
 959          * Read in the journal inode
 960          */
 961         if (ext2fs_read_inode(fs, sb->s_journal_inum, &inode) != 0)
 962                 return;
 963
 964         /*
 965          * If it's necessary to backup the journal inode, do so.
 966          */
 967         if ((sb->s_jnl_backup_type == 0) ||
 968             ((sb->s_jnl_backup_type == EXT3_JNL_BACKUP_BLOCKS) &&
 969              memcmp(inode.i_block, sb->s_jnl_blocks, EXT2_N_BLOCKS*4))) {
 970                 if (fix_problem(ctx, PR_0_BACKUP_JNL, &pctx)) {
 971                         memcpy(sb->s_jnl_blocks, inode.i_block,
 972                                EXT2_N_BLOCKS*4);
 973                         sb->s_jnl_blocks[15] = inode.i_size_high;
 974                         sb->s_jnl_blocks[16] = inode.i_size;
 975                         sb->s_jnl_backup_type = EXT3_JNL_BACKUP_BLOCKS;
 976                         ext2fs_mark_super_dirty(fs);
 977                         fs->flags &= ~EXT2_FLAG_MASTER_SB_ONLY;
 978                 }
 979         }
 980
 981         /*
 982          * If the journal is already the hidden inode, then do nothing
 983          */
 984         if (sb->s_journal_inum == EXT2_JOURNAL_INO)
 985                 return;
 986
 987         /*
 988          * The journal inode had better have only one link and not be readable.
 989          */
 990         if (inode.i_links_count != 1)
 991                 return;
 992
 993         /*
 994          * If the filesystem is mounted, or we can't tell whether
 995          * or not it's mounted, do nothing.
 996          */
 997         retval = ext2fs_check_if_mounted(ctx->filesystem_name, &mount_flags);
 998         if (retval || (mount_flags & EXT2_MF_MOUNTED))
 999                 return;
1000
1001         /*
1002          * If we can't find the name of the journal inode, then do
1003          * nothing.
1004          */
1005         for (cpp = journal_names; *cpp; cpp++) {
1006                 retval = ext2fs_lookup(fs, EXT2_ROOT_INO, *cpp,
1007                                        strlen(*cpp), 0, &ino);
1008                 if ((retval == 0) && (ino == sb->s_journal_inum))
1009                         break;
1010         }
1011         if (*cpp == 0)
1012                 return;
1013
1014         /* We need the inode bitmap to be loaded */
1015         retval = ext2fs_read_bitmaps(fs);
1016         if (retval)
1017                 return;
1018
1019         pctx.str = *cpp;
1020         if (!fix_problem(ctx, PR_0_MOVE_JOURNAL, &pctx))
1021                 return;
1022
1023         /*
1024          * OK, we've done all the checks, let's actually move the
1025          * journal inode.  Errors at this point mean we need to force
1026          * an ext2 filesystem check.
1027          */
1028         if ((retval = ext2fs_unlink(fs, EXT2_ROOT_INO, *cpp, ino, 0)) != 0)
1029                 goto err_out;
1030         if ((retval = ext2fs_write_inode(fs, EXT2_JOURNAL_INO, &inode)) != 0)
1031                 goto err_out;
1032         sb->s_journal_inum = EXT2_JOURNAL_INO;
1033         ext2fs_mark_super_dirty(fs);
1034         fs->flags &= ~EXT2_FLAG_MASTER_SB_ONLY;
1035         inode.i_links_count = 0;
1036         inode.i_dtime = ctx->now;
1037         if ((retval = ext2fs_write_inode(fs, ino, &inode)) != 0)
1038                 goto err_out;
1039
1040         group = ext2fs_group_of_ino(fs, ino);
1041         ext2fs_unmark_inode_bitmap2(fs->inode_map, ino);
1042         ext2fs_mark_ib_dirty(fs);
1043         ext2fs_bg_free_inodes_count_set(fs, group, ext2fs_bg_free_inodes_count(fs, group) + 1);
1044         ext2fs_group_desc_csum_set(fs, group);
1045         fs->super->s_free_inodes_count++;
1046         return;
1047
1048 err_out:
1049         pctx.errcode = retval;
1050         fix_problem(ctx, PR_0_ERR_MOVE_JOURNAL, &pctx);
1051         fs->super->s_state &= ~EXT2_VALID_FS;
1052         ext2fs_mark_super_dirty(fs);
1053         return;
1054 }
1055
1056 /*
1057  * This function makes sure the superblock hint for the external
1058  * journal is correct.
1059  */
1060 int e2fsck_fix_ext3_journal_hint(e2fsck_t ctx)
1061 {
1062         struct ext2_super_block *sb = ctx->fs->super;
1063         struct problem_context pctx;
1064         char uuid[37], *journal_name;
1065         struct stat st;
1066
1067         if (!(sb->s_feature_compat & EXT3_FEATURE_COMPAT_HAS_JOURNAL) ||
1068             uuid_is_null(sb->s_journal_uuid))
1069                 return 0;
1070
1071         uuid_unparse(sb->s_journal_uuid, uuid);
1072         journal_name = blkid_get_devname(ctx->blkid, "UUID", uuid);
1073         if (!journal_name)
1074                 return 0;
1075
1076         if (stat(journal_name, &st) < 0)
1077                 return 0;
1078
1079         if (st.st_rdev != sb->s_journal_dev) {
1080                 clear_problem_context(&pctx);
1081                 pctx.num = st.st_rdev;
1082                 if (fix_problem(ctx, PR_0_EXTERNAL_JOURNAL_HINT, &pctx)) {
1083                         sb->s_journal_dev = st.st_rdev;
1084                         ext2fs_mark_super_dirty(ctx->fs);
1085                 }
1086         }
1087
1088         free(journal_name);
1089         return 0;
1090 }