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                     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                 retval = io_ptr->open(journal_name,
 376                                       IO_FLAG_RW | IO_FLAG_EXCLUSIVE,
 377                                       &ctx->journal_io);
 378         if (retval)
 379                 goto errout;
 380
 381         io_channel_set_blksize(ctx->journal_io, ctx->fs->blocksize);
 382
 383         if (ext_journal) {
 384                 if (ctx->fs->blocksize == 1024)
 385                         start = 1;
 386                 bh = getblk(dev_journal, start, ctx->fs->blocksize);
 387                 if (!bh) {
 388                         retval = EXT2_ET_NO_MEMORY;
 389                         goto errout;
 390                 }
 391                 ll_rw_block(READ, 1, &bh);
 392                 if ((retval = bh->b_err) != 0) {
 393                         brelse(bh);
 394                         goto errout;
 395                 }
 396                 memcpy(&jsuper, start ? bh->b_data :  bh->b_data + 1024,
 397                        sizeof(jsuper));
 398                 brelse(bh);
 399 #ifdef WORDS_BIGENDIAN
 400                 if (jsuper.s_magic == ext2fs_swab16(EXT2_SUPER_MAGIC))
 401                         ext2fs_swap_super(&jsuper);
 402 #endif
 403                 if (jsuper.s_magic != EXT2_SUPER_MAGIC ||
 404                     !(jsuper.s_feature_incompat & EXT3_FEATURE_INCOMPAT_JOURNAL_DEV)) {
 405                         fix_problem(ctx, PR_0_EXT_JOURNAL_BAD_SUPER, &pctx);
 406                         retval = EXT2_ET_LOAD_EXT_JOURNAL;
 407                         goto errout;
 408                 }
 409                 /* Make sure the journal UUID is correct */
 410                 if (memcmp(jsuper.s_uuid, ctx->fs->super->s_journal_uuid,
 411                            sizeof(jsuper.s_uuid))) {
 412                         fix_problem(ctx, PR_0_JOURNAL_BAD_UUID, &pctx);
 413                         retval = EXT2_ET_LOAD_EXT_JOURNAL;
 414                         goto errout;
 415                 }
 416
 417                 journal->j_maxlen = ext2fs_blocks_count(&jsuper);
 418                 start++;
 419         }
 420
 421         if (!(bh = getblk(dev_journal, start, journal->j_blocksize))) {
 422                 retval = EXT2_ET_NO_MEMORY;
 423                 goto errout;
 424         }
 425
 426         journal->j_sb_buffer = bh;
 427         journal->j_superblock = (journal_superblock_t *)bh->b_data;
 428
 429 #ifdef USE_INODE_IO
 430         if (j_inode)
 431                 ext2fs_free_mem(&j_inode);
 432 #endif
 433
 434         *ret_journal = journal;
 435         e2fsck_use_inode_shortcuts(ctx, 0);
 436         return 0;
 437
 438 errout:
 439         e2fsck_use_inode_shortcuts(ctx, 0);
 440         if (dev_fs)
 441                 ext2fs_free_mem(&dev_fs);
 442         if (j_inode)
 443                 ext2fs_free_mem(&j_inode);
 444         if (journal)
 445                 ext2fs_free_mem(&journal);
 446         return retval;
 447 }
 448
 449 static errcode_t e2fsck_journal_fix_bad_inode(e2fsck_t ctx,
 450                                               struct problem_context *pctx)
 451 {
 452         struct ext2_super_block *sb = ctx->fs->super;
 453         int recover = ctx->fs->super->s_feature_incompat &
 454                 EXT3_FEATURE_INCOMPAT_RECOVER;
 455         int has_journal = ctx->fs->super->s_feature_compat &
 456                 EXT3_FEATURE_COMPAT_HAS_JOURNAL;
 457
 458         if (has_journal || sb->s_journal_inum) {
 459                 /* The journal inode is bogus, remove and force full fsck */
 460                 pctx->ino = sb->s_journal_inum;
 461                 if (fix_problem(ctx, PR_0_JOURNAL_BAD_INODE, pctx)) {
 462                         if (has_journal && sb->s_journal_inum)
 463                                 printf("*** ext3 journal has been deleted - "
 464                                        "filesystem is now ext2 only ***\n\n");
 465                         sb->s_feature_compat &= ~EXT3_FEATURE_COMPAT_HAS_JOURNAL;
 466                         sb->s_journal_inum = 0;
 467                         ctx->flags |= E2F_FLAG_JOURNAL_INODE;
 468                         ctx->fs->flags &= ~EXT2_FLAG_MASTER_SB_ONLY;
 469                         e2fsck_clear_recover(ctx, 1);
 470                         return 0;
 471                 }
 472                 return EXT2_ET_BAD_INODE_NUM;
 473         } else if (recover) {
 474                 if (fix_problem(ctx, PR_0_JOURNAL_RECOVER_SET, pctx)) {
 475                         e2fsck_clear_recover(ctx, 1);
 476                         return 0;
 477                 }
 478                 return EXT2_ET_UNSUPP_FEATURE;
 479         }
 480         return 0;
 481 }
 482
 483 #define V1_SB_SIZE      0x0024
 484 static void clear_v2_journal_fields(journal_t *journal)
 485 {
 486         e2fsck_t ctx = journal->j_dev->k_ctx;
 487         struct problem_context pctx;
 488
 489         clear_problem_context(&pctx);
 490
 491         if (!fix_problem(ctx, PR_0_CLEAR_V2_JOURNAL, &pctx))
 492                 return;
 493
 494         memset(((char *) journal->j_superblock) + V1_SB_SIZE, 0,
 495                ctx->fs->blocksize-V1_SB_SIZE);
 496         mark_buffer_dirty(journal->j_sb_buffer);
 497 }
 498
 499
 500 static errcode_t e2fsck_journal_load(journal_t *journal)
 501 {
 502         e2fsck_t ctx = journal->j_dev->k_ctx;
 503         journal_superblock_t *jsb;
 504         struct buffer_head *jbh = journal->j_sb_buffer;
 505         struct problem_context pctx;
 506
 507         clear_problem_context(&pctx);
 508
 509         ll_rw_block(READ, 1, &jbh);
 510         if (jbh->b_err) {
 511                 com_err(ctx->device_name, jbh->b_err,
 512                         _("reading journal superblock\n"));
 513                 return jbh->b_err;
 514         }
 515
 516         jsb = journal->j_superblock;
 517         /* If we don't even have JFS_MAGIC, we probably have a wrong inode */
 518         if (jsb->s_header.h_magic != htonl(JFS_MAGIC_NUMBER))
 519                 return e2fsck_journal_fix_bad_inode(ctx, &pctx);
 520
 521         switch (ntohl(jsb->s_header.h_blocktype)) {
 522         case JFS_SUPERBLOCK_V1:
 523                 journal->j_format_version = 1;
 524                 if (jsb->s_feature_compat ||
 525                     jsb->s_feature_incompat ||
 526                     jsb->s_feature_ro_compat ||
 527                     jsb->s_nr_users)
 528                         clear_v2_journal_fields(journal);
 529                 break;
 530
 531         case JFS_SUPERBLOCK_V2:
 532                 journal->j_format_version = 2;
 533                 if (ntohl(jsb->s_nr_users) > 1 &&
 534                     uuid_is_null(ctx->fs->super->s_journal_uuid))
 535                         clear_v2_journal_fields(journal);
 536                 if (ntohl(jsb->s_nr_users) > 1) {
 537                         fix_problem(ctx, PR_0_JOURNAL_UNSUPP_MULTIFS, &pctx);
 538                         return EXT2_ET_JOURNAL_UNSUPP_VERSION;
 539                 }
 540                 break;
 541
 542         /*
 543          * These should never appear in a journal super block, so if
 544          * they do, the journal is badly corrupted.
 545          */
 546         case JFS_DESCRIPTOR_BLOCK:
 547         case JFS_COMMIT_BLOCK:
 548         case JFS_REVOKE_BLOCK:
 549                 return EXT2_ET_CORRUPT_SUPERBLOCK;
 550
 551         /* If we don't understand the superblock major type, but there
 552          * is a magic number, then it is likely to be a new format we
 553          * just don't understand, so leave it alone. */
 554         default:
 555                 return EXT2_ET_JOURNAL_UNSUPP_VERSION;
 556         }
 557
 558         if (JFS_HAS_INCOMPAT_FEATURE(journal, ~JFS_KNOWN_INCOMPAT_FEATURES))
 559                 return EXT2_ET_UNSUPP_FEATURE;
 560
 561         if (JFS_HAS_RO_COMPAT_FEATURE(journal, ~JFS_KNOWN_ROCOMPAT_FEATURES))
 562                 return EXT2_ET_RO_UNSUPP_FEATURE;
 563
 564         /* We have now checked whether we know enough about the journal
 565          * format to be able to proceed safely, so any other checks that
 566          * fail we should attempt to recover from. */
 567         if (jsb->s_blocksize != htonl(journal->j_blocksize)) {
 568                 com_err(ctx->program_name, EXT2_ET_CORRUPT_SUPERBLOCK,
 569                         _("%s: no valid journal superblock found\n"),
 570                         ctx->device_name);
 571                 return EXT2_ET_CORRUPT_SUPERBLOCK;
 572         }
 573
 574         if (ntohl(jsb->s_maxlen) < journal->j_maxlen)
 575                 journal->j_maxlen = ntohl(jsb->s_maxlen);
 576         else if (ntohl(jsb->s_maxlen) > journal->j_maxlen) {
 577                 com_err(ctx->program_name, EXT2_ET_CORRUPT_SUPERBLOCK,
 578                         _("%s: journal too short\n"),
 579                         ctx->device_name);
 580                 return EXT2_ET_CORRUPT_SUPERBLOCK;
 581         }
 582
 583         journal->j_tail_sequence = ntohl(jsb->s_sequence);
 584         journal->j_transaction_sequence = journal->j_tail_sequence;
 585         journal->j_tail = ntohl(jsb->s_start);
 586         journal->j_first = ntohl(jsb->s_first);
 587         journal->j_last = ntohl(jsb->s_maxlen);
 588
 589         return 0;
 590 }
 591
 592 static void e2fsck_journal_reset_super(e2fsck_t ctx, journal_superblock_t *jsb,
 593                                        journal_t *journal)
 594 {
 595         char *p;
 596         union {
 597                 uuid_t uuid;
 598                 __u32 val[4];
 599         } u;
 600         __u32 new_seq = 0;
 601         int i;
 602
 603         /* Leave a valid existing V1 superblock signature alone.
 604          * Anything unrecognisable we overwrite with a new V2
 605          * signature. */
 606
 607         if (jsb->s_header.h_magic != htonl(JFS_MAGIC_NUMBER) ||
 608             jsb->s_header.h_blocktype != htonl(JFS_SUPERBLOCK_V1)) {
 609                 jsb->s_header.h_magic = htonl(JFS_MAGIC_NUMBER);
 610                 jsb->s_header.h_blocktype = htonl(JFS_SUPERBLOCK_V2);
 611         }
 612
 613         /* Zero out everything else beyond the superblock header */
 614
 615         p = ((char *) jsb) + sizeof(journal_header_t);
 616         memset (p, 0, ctx->fs->blocksize-sizeof(journal_header_t));
 617
 618         jsb->s_blocksize = htonl(ctx->fs->blocksize);
 619         jsb->s_maxlen = htonl(journal->j_maxlen);
 620         jsb->s_first = htonl(1);
 621
 622         /* Initialize the journal sequence number so that there is "no"
 623          * chance we will find old "valid" transactions in the journal.
 624          * This avoids the need to zero the whole journal (slow to do,
 625          * and risky when we are just recovering the filesystem).
 626          */
 627         uuid_generate(u.uuid);
 628         for (i = 0; i < 4; i ++)
 629                 new_seq ^= u.val[i];
 630         jsb->s_sequence = htonl(new_seq);
 631
 632         mark_buffer_dirty(journal->j_sb_buffer);
 633         ll_rw_block(WRITE, 1, &journal->j_sb_buffer);
 634 }
 635
 636 static errcode_t e2fsck_journal_fix_corrupt_super(e2fsck_t ctx,
 637                                                   journal_t *journal,
 638                                                   struct problem_context *pctx)
 639 {
 640         struct ext2_super_block *sb = ctx->fs->super;
 641         int recover = ctx->fs->super->s_feature_incompat &
 642                 EXT3_FEATURE_INCOMPAT_RECOVER;
 643
 644         if (sb->s_feature_compat & EXT3_FEATURE_COMPAT_HAS_JOURNAL) {
 645                 if (fix_problem(ctx, PR_0_JOURNAL_BAD_SUPER, pctx)) {
 646                         e2fsck_journal_reset_super(ctx, journal->j_superblock,
 647                                                    journal);
 648                         journal->j_transaction_sequence = 1;
 649                         e2fsck_clear_recover(ctx, recover);
 650                         return 0;
 651                 }
 652                 return EXT2_ET_CORRUPT_SUPERBLOCK;
 653         } else if (e2fsck_journal_fix_bad_inode(ctx, pctx))
 654                 return EXT2_ET_CORRUPT_SUPERBLOCK;
 655
 656         return 0;
 657 }
 658
 659 static void e2fsck_journal_release(e2fsck_t ctx, journal_t *journal,
 660                                    int reset, int drop)
 661 {
 662         journal_superblock_t *jsb;
 663
 664         if (drop)
 665                 mark_buffer_clean(journal->j_sb_buffer);
 666         else if (!(ctx->options & E2F_OPT_READONLY)) {
 667                 jsb = journal->j_superblock;
 668                 jsb->s_sequence = htonl(journal->j_transaction_sequence);
 669                 if (reset)
 670                         jsb->s_start = 0; /* this marks the journal as empty */
 671                 mark_buffer_dirty(journal->j_sb_buffer);
 672         }
 673         brelse(journal->j_sb_buffer);
 674
 675         if (ctx->journal_io) {
 676                 if (ctx->fs && ctx->fs->io != ctx->journal_io)
 677                         io_channel_close(ctx->journal_io);
 678                 ctx->journal_io = 0;
 679         }
 680
 681 #ifndef USE_INODE_IO
 682         if (journal->j_inode)
 683                 ext2fs_free_mem(&journal->j_inode);
 684 #endif
 685         if (journal->j_fs_dev)
 686                 ext2fs_free_mem(&journal->j_fs_dev);
 687         ext2fs_free_mem(&journal);
 688 }
 689
 690 /*
 691  * This function makes sure that the superblock fields regarding the
 692  * journal are consistent.
 693  */
 694 int e2fsck_check_ext3_journal(e2fsck_t ctx)
 695 {
 696         struct ext2_super_block *sb = ctx->fs->super;
 697         journal_t *journal;
 698         int recover = ctx->fs->super->s_feature_incompat &
 699                 EXT3_FEATURE_INCOMPAT_RECOVER;
 700         struct problem_context pctx;
 701         problem_t problem;
 702         int reset = 0, force_fsck = 0;
 703         int retval;
 704
 705         /* If we don't have any journal features, don't do anything more */
 706         if (!(sb->s_feature_compat & EXT3_FEATURE_COMPAT_HAS_JOURNAL) &&
 707             !recover && sb->s_journal_inum == 0 && sb->s_journal_dev == 0 &&
 708             uuid_is_null(sb->s_journal_uuid))
 709                 return 0;
 710
 711         clear_problem_context(&pctx);
 712         pctx.num = sb->s_journal_inum;
 713
 714         retval = e2fsck_get_journal(ctx, &journal);
 715         if (retval) {
 716                 if ((retval == EXT2_ET_BAD_INODE_NUM) ||
 717                     (retval == EXT2_ET_BAD_BLOCK_NUM) ||
 718                     (retval == EXT2_ET_JOURNAL_TOO_SMALL) ||
 719                     (retval == EXT2_ET_NO_JOURNAL))
 720                         return e2fsck_journal_fix_bad_inode(ctx, &pctx);
 721                 return retval;
 722         }
 723
 724         retval = e2fsck_journal_load(journal);
 725         if (retval) {
 726                 if ((retval == EXT2_ET_CORRUPT_SUPERBLOCK) ||
 727                     ((retval == EXT2_ET_UNSUPP_FEATURE) &&
 728                     (!fix_problem(ctx, PR_0_JOURNAL_UNSUPP_INCOMPAT,
 729                                   &pctx))) ||
 730                     ((retval == EXT2_ET_RO_UNSUPP_FEATURE) &&
 731                     (!fix_problem(ctx, PR_0_JOURNAL_UNSUPP_ROCOMPAT,
 732                                   &pctx))) ||
 733                     ((retval == EXT2_ET_JOURNAL_UNSUPP_VERSION) &&
 734                     (!fix_problem(ctx, PR_0_JOURNAL_UNSUPP_VERSION, &pctx))))
 735                         retval = e2fsck_journal_fix_corrupt_super(ctx, journal,
 736                                                                   &pctx);
 737                 e2fsck_journal_release(ctx, journal, 0, 1);
 738                 return retval;
 739         }
 740
 741         /*
 742          * We want to make the flags consistent here.  We will not leave with
 743          * needs_recovery set but has_journal clear.  We can't get in a loop
 744          * with -y, -n, or -p, only if a user isn't making up their mind.
 745          */
 746 no_has_journal:
 747         if (!(sb->s_feature_compat & EXT3_FEATURE_COMPAT_HAS_JOURNAL)) {
 748                 recover = sb->s_feature_incompat & EXT3_FEATURE_INCOMPAT_RECOVER;
 749                 pctx.str = "inode";
 750                 if (fix_problem(ctx, PR_0_JOURNAL_HAS_JOURNAL, &pctx)) {
 751                         if (recover &&
 752                             !fix_problem(ctx, PR_0_JOURNAL_RECOVER_SET, &pctx))
 753                                 goto no_has_journal;
 754                         /*
 755                          * Need a full fsck if we are releasing a
 756                          * journal stored on a reserved inode.
 757                          */
 758                         force_fsck = recover ||
 759                                 (sb->s_journal_inum < EXT2_FIRST_INODE(sb));
 760                         /* Clear all of the journal fields */
 761                         sb->s_journal_inum = 0;
 762                         sb->s_journal_dev = 0;
 763                         memset(sb->s_journal_uuid, 0,
 764                                sizeof(sb->s_journal_uuid));
 765                         e2fsck_clear_recover(ctx, force_fsck);
 766                 } else if (!(ctx->options & E2F_OPT_READONLY)) {
 767                         sb->s_feature_compat |= EXT3_FEATURE_COMPAT_HAS_JOURNAL;
 768                         ctx->fs->flags &= ~EXT2_FLAG_MASTER_SB_ONLY;
 769                         ext2fs_mark_super_dirty(ctx->fs);
 770                 }
 771         }
 772
 773         if (sb->s_feature_compat & EXT3_FEATURE_COMPAT_HAS_JOURNAL &&
 774             !(sb->s_feature_incompat & EXT3_FEATURE_INCOMPAT_RECOVER) &&
 775             journal->j_superblock->s_start != 0) {
 776                 /* Print status information */
 777                 fix_problem(ctx, PR_0_JOURNAL_RECOVERY_CLEAR, &pctx);
 778                 if (ctx->superblock)
 779                         problem = PR_0_JOURNAL_RUN_DEFAULT;
 780                 else
 781                         problem = PR_0_JOURNAL_RUN;
 782                 if (fix_problem(ctx, problem, &pctx)) {
 783                         ctx->options |= E2F_OPT_FORCE;
 784                         sb->s_feature_incompat |=
 785                                 EXT3_FEATURE_INCOMPAT_RECOVER;
 786                         ext2fs_mark_super_dirty(ctx->fs);
 787                 } else if (fix_problem(ctx,
 788                                        PR_0_JOURNAL_RESET_JOURNAL, &pctx)) {
 789                         reset = 1;
 790                         sb->s_state &= ~EXT2_VALID_FS;
 791                         ext2fs_mark_super_dirty(ctx->fs);
 792                 }
 793                 /*
 794                  * If the user answers no to the above question, we
 795                  * ignore the fact that journal apparently has data;
 796                  * accidentally replaying over valid data would be far
 797                  * worse than skipping a questionable recovery.
 798                  *
 799                  * XXX should we abort with a fatal error here?  What
 800                  * will the ext3 kernel code do if a filesystem with
 801                  * !NEEDS_RECOVERY but with a non-zero
 802                  * journal->j_superblock->s_start is mounted?
 803                  */
 804         }
 805
 806         /*
 807          * If we don't need to do replay the journal, check to see if
 808          * the journal's errno is set; if so, we need to mark the file
 809          * system as being corrupt and clear the journal's s_errno.
 810          */
 811         if (!(sb->s_feature_incompat & EXT3_FEATURE_INCOMPAT_RECOVER) &&
 812             journal->j_superblock->s_errno) {
 813                 ctx->fs->super->s_state |= EXT2_ERROR_FS;
 814                 ext2fs_mark_super_dirty(ctx->fs);
 815                 journal->j_superblock->s_errno = 0;
 816                 mark_buffer_dirty(journal->j_sb_buffer);
 817         }
 818
 819         e2fsck_journal_release(ctx, journal, reset, 0);
 820         return retval;
 821 }
 822
 823 static errcode_t recover_ext3_journal(e2fsck_t ctx)
 824 {
 825         struct problem_context  pctx;
 826         journal_t *journal;
 827         int retval;
 828
 829         clear_problem_context(&pctx);
 830
 831         journal_init_revoke_caches();
 832         retval = e2fsck_get_journal(ctx, &journal);
 833         if (retval)
 834                 return retval;
 835
 836         retval = e2fsck_journal_load(journal);
 837         if (retval)
 838                 goto errout;
 839
 840         retval = journal_init_revoke(journal, 1024);
 841         if (retval)
 842                 goto errout;
 843
 844         retval = -journal_recover(journal);
 845         if (retval)
 846                 goto errout;
 847
 848         if (journal->j_failed_commit) {
 849                 pctx.ino = journal->j_failed_commit;
 850                 fix_problem(ctx, PR_0_JNL_TXN_CORRUPT, &pctx);
 851                 ctx->fs->super->s_state |= EXT2_ERROR_FS;
 852                 ext2fs_mark_super_dirty(ctx->fs);
 853         }
 854
 855
 856         if (journal->j_superblock->s_errno) {
 857                 ctx->fs->super->s_state |= EXT2_ERROR_FS;
 858                 ext2fs_mark_super_dirty(ctx->fs);
 859                 journal->j_superblock->s_errno = 0;
 860                 mark_buffer_dirty(journal->j_sb_buffer);
 861         }
 862
 863 errout:
 864         journal_destroy_revoke(journal);
 865         journal_destroy_revoke_caches();
 866         e2fsck_journal_release(ctx, journal, 1, 0);
 867         return retval;
 868 }
 869
 870 int e2fsck_run_ext3_journal(e2fsck_t ctx)
 871 {
 872         io_manager io_ptr = ctx->fs->io->manager;
 873         int blocksize = ctx->fs->blocksize;
 874         errcode_t       retval, recover_retval;
 875         io_stats        stats = 0;
 876         unsigned long long kbytes_written = 0;
 877
 878         printf(_("%s: recovering journal\n"), ctx->device_name);
 879         if (ctx->options & E2F_OPT_READONLY) {
 880                 printf(_("%s: won't do journal recovery while read-only\n"),
 881                        ctx->device_name);
 882                 return EXT2_ET_FILE_RO;
 883         }
 884
 885         if (ctx->fs->flags & EXT2_FLAG_DIRTY)
 886                 ext2fs_flush(ctx->fs);  /* Force out any modifications */
 887
 888         recover_retval = recover_ext3_journal(ctx);
 889
 890         /*
 891          * Reload the filesystem context to get up-to-date data from disk
 892          * because journal recovery will change the filesystem under us.
 893          */
 894         if (ctx->fs->super->s_kbytes_written &&
 895             ctx->fs->io->manager->get_stats)
 896                 ctx->fs->io->manager->get_stats(ctx->fs->io, &stats);
 897         if (stats && stats->bytes_written)
 898                 kbytes_written = stats->bytes_written >> 10;
 899
 900         ext2fs_mmp_stop(ctx->fs);
 901         ext2fs_free(ctx->fs);
 902         retval = ext2fs_open(ctx->filesystem_name, EXT2_FLAG_RW,
 903                              ctx->superblock, blocksize, io_ptr,
 904                              &ctx->fs);
 905         if (retval) {
 906                 com_err(ctx->program_name, retval,
 907                         _("while trying to re-open %s"),
 908                         ctx->device_name);
 909                 fatal_error(ctx, 0);
 910         }
 911         ctx->fs->priv_data = ctx;
 912         ctx->fs->now = ctx->now;
 913         ctx->fs->flags |= EXT2_FLAG_MASTER_SB_ONLY;
 914         ctx->fs->super->s_kbytes_written += kbytes_written;
 915
 916         /* Set the superblock flags */
 917         e2fsck_clear_recover(ctx, recover_retval);
 918         return recover_retval;
 919 }
 920
 921 /*
 922  * This function will move the journal inode from a visible file in
 923  * the filesystem directory hierarchy to the reserved inode if necessary.
 924  */
 925 static const char * const journal_names[] = {
 926         ".journal", "journal", ".journal.dat", "journal.dat", 0 };
 927
 928 void e2fsck_move_ext3_journal(e2fsck_t ctx)
 929 {
 930         struct ext2_super_block *sb = ctx->fs->super;
 931         struct problem_context  pctx;
 932         struct ext2_inode       inode;
 933         ext2_filsys             fs = ctx->fs;
 934         ext2_ino_t              ino;
 935         errcode_t               retval;
 936         const char * const *    cpp;
 937         int                     group, mount_flags;
 938
 939         clear_problem_context(&pctx);
 940
 941         /*
 942          * If the filesystem is opened read-only, or there is no
 943          * journal, then do nothing.
 944          */
 945         if ((ctx->options & E2F_OPT_READONLY) ||
 946             (sb->s_journal_inum == 0) ||
 947             !(sb->s_feature_compat & EXT3_FEATURE_COMPAT_HAS_JOURNAL))
 948                 return;
 949
 950         /*
 951          * Read in the journal inode
 952          */
 953         if (ext2fs_read_inode(fs, sb->s_journal_inum, &inode) != 0)
 954                 return;
 955
 956         /*
 957          * If it's necessary to backup the journal inode, do so.
 958          */
 959         if ((sb->s_jnl_backup_type == 0) ||
 960             ((sb->s_jnl_backup_type == EXT3_JNL_BACKUP_BLOCKS) &&
 961              memcmp(inode.i_block, sb->s_jnl_blocks, EXT2_N_BLOCKS*4))) {
 962                 if (fix_problem(ctx, PR_0_BACKUP_JNL, &pctx)) {
 963                         memcpy(sb->s_jnl_blocks, inode.i_block,
 964                                EXT2_N_BLOCKS*4);
 965                         sb->s_jnl_blocks[15] = inode.i_size_high;
 966                         sb->s_jnl_blocks[16] = inode.i_size;
 967                         sb->s_jnl_backup_type = EXT3_JNL_BACKUP_BLOCKS;
 968                         ext2fs_mark_super_dirty(fs);
 969                         fs->flags &= ~EXT2_FLAG_MASTER_SB_ONLY;
 970                 }
 971         }
 972
 973         /*
 974          * If the journal is already the hidden inode, then do nothing
 975          */
 976         if (sb->s_journal_inum == EXT2_JOURNAL_INO)
 977                 return;
 978
 979         /*
 980          * The journal inode had better have only one link and not be readable.
 981          */
 982         if (inode.i_links_count != 1)
 983                 return;
 984
 985         /*
 986          * If the filesystem is mounted, or we can't tell whether
 987          * or not it's mounted, do nothing.
 988          */
 989         retval = ext2fs_check_if_mounted(ctx->filesystem_name, &mount_flags);
 990         if (retval || (mount_flags & EXT2_MF_MOUNTED))
 991                 return;
 992
 993         /*
 994          * If we can't find the name of the journal inode, then do
 995          * nothing.
 996          */
 997         for (cpp = journal_names; *cpp; cpp++) {
 998                 retval = ext2fs_lookup(fs, EXT2_ROOT_INO, *cpp,
 999                                        strlen(*cpp), 0, &ino);
1000                 if ((retval == 0) && (ino == sb->s_journal_inum))
1001                         break;
1002         }
1003         if (*cpp == 0)
1004                 return;
1005
1006         /* We need the inode bitmap to be loaded */
1007         retval = ext2fs_read_bitmaps(fs);
1008         if (retval)
1009                 return;
1010
1011         pctx.str = *cpp;
1012         if (!fix_problem(ctx, PR_0_MOVE_JOURNAL, &pctx))
1013                 return;
1014
1015         /*
1016          * OK, we've done all the checks, let's actually move the
1017          * journal inode.  Errors at this point mean we need to force
1018          * an ext2 filesystem check.
1019          */
1020         if ((retval = ext2fs_unlink(fs, EXT2_ROOT_INO, *cpp, ino, 0)) != 0)
1021                 goto err_out;
1022         if ((retval = ext2fs_write_inode(fs, EXT2_JOURNAL_INO, &inode)) != 0)
1023                 goto err_out;
1024         sb->s_journal_inum = EXT2_JOURNAL_INO;
1025         ext2fs_mark_super_dirty(fs);
1026         fs->flags &= ~EXT2_FLAG_MASTER_SB_ONLY;
1027         inode.i_links_count = 0;
1028         inode.i_dtime = ctx->now;
1029         if ((retval = ext2fs_write_inode(fs, ino, &inode)) != 0)
1030                 goto err_out;
1031
1032         group = ext2fs_group_of_ino(fs, ino);
1033         ext2fs_unmark_inode_bitmap2(fs->inode_map, ino);
1034         ext2fs_mark_ib_dirty(fs);
1035         ext2fs_bg_free_inodes_count_set(fs, group, ext2fs_bg_free_inodes_count(fs, group) + 1);
1036         ext2fs_group_desc_csum_set(fs, group);
1037         fs->super->s_free_inodes_count++;
1038         return;
1039
1040 err_out:
1041         pctx.errcode = retval;
1042         fix_problem(ctx, PR_0_ERR_MOVE_JOURNAL, &pctx);
1043         fs->super->s_state &= ~EXT2_VALID_FS;
1044         ext2fs_mark_super_dirty(fs);
1045         return;
1046 }
1047
1048 /*
1049  * This function makes sure the superblock hint for the external
1050  * journal is correct.
1051  */
1052 int e2fsck_fix_ext3_journal_hint(e2fsck_t ctx)
1053 {
1054         struct ext2_super_block *sb = ctx->fs->super;
1055         struct problem_context pctx;
1056         char uuid[37], *journal_name;
1057         struct stat st;
1058
1059         if (!(sb->s_feature_compat & EXT3_FEATURE_COMPAT_HAS_JOURNAL) ||
1060             uuid_is_null(sb->s_journal_uuid))
1061                 return 0;
1062
1063         uuid_unparse(sb->s_journal_uuid, uuid);
1064         journal_name = blkid_get_devname(ctx->blkid, "UUID", uuid);
1065         if (!journal_name)
1066                 return 0;
1067
1068         if (stat(journal_name, &st) < 0)
1069                 return 0;
1070
1071         if (st.st_rdev != sb->s_journal_dev) {
1072                 clear_problem_context(&pctx);
1073                 pctx.num = st.st_rdev;
1074                 if (fix_problem(ctx, PR_0_EXTERNAL_JOURNAL_HINT, &pctx)) {
1075                         sb->s_journal_dev = st.st_rdev;
1076                         ext2fs_mark_super_dirty(ctx->fs);
1077                 }
1078         }
1079
1080         free(journal_name);
1081         return 0;
1082 }