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 static int bh_count = 0;
  31
  32 /*
  33  * Define USE_INODE_IO to use the inode_io.c / fileio.c codepaths.
  34  * This creates a larger static binary, and a smaller binary using
  35  * shared libraries.  It's also probably slightly less CPU-efficient,
  36  * which is why it's not on by default.  But, it's a good way of
  37  * testing the functions in inode_io.c and fileio.c.
  38  */
  39 #undef USE_INODE_IO
  40
  41 /* Checksumming functions */
  42 static int e2fsck_journal_verify_csum_type(journal_t *j,
  43                                            journal_superblock_t *jsb)
  44 {
  45         if (!jbd2_journal_has_csum_v2or3(j))
  46                 return 1;
  47
  48         return jsb->s_checksum_type == JBD2_CRC32C_CHKSUM;
  49 }
  50
  51 static __u32 e2fsck_journal_sb_csum(journal_superblock_t *jsb)
  52 {
  53         __u32 crc, old_crc;
  54
  55         old_crc = jsb->s_checksum;
  56         jsb->s_checksum = 0;
  57         crc = ext2fs_crc32c_le(~0, (unsigned char *)jsb,
  58                                sizeof(journal_superblock_t));
  59         jsb->s_checksum = old_crc;
  60
  61         return crc;
  62 }
  63
  64 static int e2fsck_journal_sb_csum_verify(journal_t *j,
  65                                          journal_superblock_t *jsb)
  66 {
  67         __u32 provided, calculated;
  68
  69         if (!jbd2_journal_has_csum_v2or3(j))
  70                 return 1;
  71
  72         provided = ext2fs_be32_to_cpu(jsb->s_checksum);
  73         calculated = e2fsck_journal_sb_csum(jsb);
  74
  75         return provided == calculated;
  76 }
  77
  78 static errcode_t e2fsck_journal_sb_csum_set(journal_t *j,
  79                                             journal_superblock_t *jsb)
  80 {
  81         __u32 crc;
  82
  83         if (!jbd2_journal_has_csum_v2or3(j))
  84                 return 0;
  85
  86         crc = e2fsck_journal_sb_csum(jsb);
  87         jsb->s_checksum = ext2fs_cpu_to_be32(crc);
  88         return 0;
  89 }
  90
  91 /* Kernel compatibility functions for handling the journal.  These allow us
  92  * to use the recovery.c file virtually unchanged from the kernel, so we
  93  * don't have to do much to keep kernel and user recovery in sync.
  94  */
  95 int jbd2_journal_bmap(journal_t *journal, unsigned long block,
  96                       unsigned long long *phys)
  97 {
  98 #ifdef USE_INODE_IO
  99         *phys = block;
 100         return 0;
 101 #else
 102         struct inode    *inode = journal->j_inode;
 103         errcode_t       retval;
 104         blk64_t         pblk;
 105
 106         if (!inode) {
 107                 *phys = block;
 108                 return 0;
 109         }
 110
 111         retval= ext2fs_bmap2(inode->i_ctx->fs, inode->i_ino,
 112                              &inode->i_ext2, NULL, 0, (blk64_t) block,
 113                              0, &pblk);
 114         *phys = pblk;
 115         return -1 * ((int) retval);
 116 #endif
 117 }
 118
 119 struct buffer_head *getblk(kdev_t kdev, unsigned long long blocknr,
 120                            int blocksize)
 121 {
 122         struct buffer_head *bh;
 123         int bufsize = sizeof(*bh) + kdev->k_ctx->fs->blocksize -
 124                 sizeof(bh->b_data);
 125
 126         bh = e2fsck_allocate_memory(kdev->k_ctx, bufsize, "block buffer");
 127         if (!bh)
 128                 return NULL;
 129
 130         if (journal_enable_debug >= 3)
 131                 bh_count++;
 132         jfs_debug(4, "getblk for block %llu (%d bytes)(total %d)\n",
 133                   blocknr, blocksize, bh_count);
 134
 135         bh->b_ctx = kdev->k_ctx;
 136         if (kdev->k_dev == K_DEV_FS)
 137                 bh->b_io = kdev->k_ctx->fs->io;
 138         else
 139                 bh->b_io = kdev->k_ctx->journal_io;
 140         bh->b_size = blocksize;
 141         bh->b_blocknr = blocknr;
 142
 143         return bh;
 144 }
 145
 146 int sync_blockdev(kdev_t kdev)
 147 {
 148         io_channel      io;
 149
 150         if (kdev->k_dev == K_DEV_FS)
 151                 io = kdev->k_ctx->fs->io;
 152         else
 153                 io = kdev->k_ctx->journal_io;
 154
 155         return io_channel_flush(io) ? -EIO : 0;
 156 }
 157
 158 void ll_rw_block(int rw, int op_flags EXT2FS_ATTR((unused)), int nr,
 159                  struct buffer_head *bhp[])
 160 {
 161         errcode_t retval;
 162         struct buffer_head *bh;
 163
 164         for (; nr > 0; --nr) {
 165                 bh = *bhp++;
 166                 if (rw == REQ_OP_READ && !bh->b_uptodate) {
 167                         jfs_debug(3, "reading block %llu/%p\n",
 168                                   bh->b_blocknr, (void *) bh);
 169                         retval = io_channel_read_blk64(bh->b_io,
 170                                                      bh->b_blocknr,
 171                                                      1, bh->b_data);
 172                         if (retval) {
 173                                 com_err(bh->b_ctx->device_name, retval,
 174                                         "while reading block %llu\n",
 175                                         bh->b_blocknr);
 176                                 bh->b_err = (int) retval;
 177                                 continue;
 178                         }
 179                         bh->b_uptodate = 1;
 180                 } else if (rw == REQ_OP_WRITE && bh->b_dirty) {
 181                         jfs_debug(3, "writing block %llu/%p\n",
 182                                   bh->b_blocknr,
 183                                   (void *) bh);
 184                         retval = io_channel_write_blk64(bh->b_io,
 185                                                       bh->b_blocknr,
 186                                                       1, bh->b_data);
 187                         if (retval) {
 188                                 com_err(bh->b_ctx->device_name, retval,
 189                                         "while writing block %llu\n",
 190                                         bh->b_blocknr);
 191                                 bh->b_err = (int) retval;
 192                                 continue;
 193                         }
 194                         bh->b_dirty = 0;
 195                         bh->b_uptodate = 1;
 196                 } else {
 197                         jfs_debug(3, "no-op %s for block %llu\n",
 198                                   rw == REQ_OP_READ ? "read" : "write",
 199                                   bh->b_blocknr);
 200                 }
 201         }
 202 }
 203
 204 void mark_buffer_dirty(struct buffer_head *bh)
 205 {
 206         bh->b_dirty = 1;
 207 }
 208
 209 static void mark_buffer_clean(struct buffer_head * bh)
 210 {
 211         bh->b_dirty = 0;
 212 }
 213
 214 void brelse(struct buffer_head *bh)
 215 {
 216         if (bh->b_dirty)
 217                 ll_rw_block(REQ_OP_WRITE, 0, 1, &bh);
 218         jfs_debug(3, "freeing block %llu/%p (total %d)\n",
 219                   bh->b_blocknr, (void *) bh, --bh_count);
 220         ext2fs_free_mem(&bh);
 221 }
 222
 223 int buffer_uptodate(struct buffer_head *bh)
 224 {
 225         return bh->b_uptodate;
 226 }
 227
 228 void mark_buffer_uptodate(struct buffer_head *bh, int val)
 229 {
 230         bh->b_uptodate = val;
 231 }
 232
 233 void wait_on_buffer(struct buffer_head *bh)
 234 {
 235         if (!bh->b_uptodate)
 236                 ll_rw_block(REQ_OP_READ, 0, 1, &bh);
 237 }
 238
 239
 240 static void e2fsck_clear_recover(e2fsck_t ctx, int error)
 241 {
 242         ext2fs_clear_feature_journal_needs_recovery(ctx->fs->super);
 243
 244         /* if we had an error doing journal recovery, we need a full fsck */
 245         if (error)
 246                 ctx->fs->super->s_state &= ~EXT2_VALID_FS;
 247         ext2fs_mark_super_dirty(ctx->fs);
 248 }
 249
 250 /*
 251  * This is a helper function to check the validity of the journal.
 252  */
 253 struct process_block_struct {
 254         e2_blkcnt_t     last_block;
 255 };
 256
 257 static int process_journal_block(ext2_filsys fs,
 258                                  blk64_t        *block_nr,
 259                                  e2_blkcnt_t blockcnt,
 260                                  blk64_t ref_block EXT2FS_ATTR((unused)),
 261                                  int ref_offset EXT2FS_ATTR((unused)),
 262                                  void *priv_data)
 263 {
 264         struct process_block_struct *p;
 265         blk64_t blk = *block_nr;
 266
 267         p = (struct process_block_struct *) priv_data;
 268
 269         if (!blk || blk < fs->super->s_first_data_block ||
 270             blk >= ext2fs_blocks_count(fs->super))
 271                 return BLOCK_ABORT;
 272
 273         if (blockcnt >= 0)
 274                 p->last_block = blockcnt;
 275         return 0;
 276 }
 277
 278 static int ext4_fc_replay_scan(journal_t *j, struct buffer_head *bh,
 279                                 int off, tid_t expected_tid)
 280 {
 281         e2fsck_t ctx = j->j_fs_dev->k_ctx;
 282         struct e2fsck_fc_replay_state *state;
 283         int ret = JBD2_FC_REPLAY_CONTINUE;
 284         struct ext4_fc_add_range ext;
 285         struct ext4_fc_tl tl;
 286         struct ext4_fc_tail tail;
 287         __u8 *start, *cur, *end, *val;
 288         struct ext4_fc_head head;
 289         struct ext2fs_extent ext2fs_ex = {0};
 290
 291         state = &ctx->fc_replay_state;
 292
 293         start = (__u8 *)bh->b_data;
 294         end = (__u8 *)bh->b_data + j->j_blocksize - 1;
 295
 296         jbd_debug(1, "Scan phase starting, expected %d", expected_tid);
 297         if (state->fc_replay_expected_off == 0) {
 298                 memset(state, 0, sizeof(*state));
 299                 /* Check if we can stop early */
 300                 if (le16_to_cpu(((struct ext4_fc_tl *)start)->fc_tag)
 301                         != EXT4_FC_TAG_HEAD) {
 302                         jbd_debug(1, "Ending early!, not a head tag");
 303                         return 0;
 304                 }
 305         }
 306
 307         if (off != state->fc_replay_expected_off) {
 308                 ret = -EFSCORRUPTED;
 309                 goto out_err;
 310         }
 311
 312         state->fc_replay_expected_off++;
 313         for (cur = start; cur < end; cur = cur + le16_to_cpu(tl.fc_len) + sizeof(tl)) {
 314                 memcpy(&tl, cur, sizeof(tl));
 315                 val = cur + sizeof(tl);
 316
 317                 jbd_debug(3, "Scan phase, tag:%s, blk %lld\n",
 318                           tag2str(le16_to_cpu(tl.fc_tag)), bh->b_blocknr);
 319                 switch (le16_to_cpu(tl.fc_tag)) {
 320                 case EXT4_FC_TAG_ADD_RANGE:
 321                         memcpy(&ext, val, sizeof(ext));
 322                         ret = ext2fs_decode_extent(&ext2fs_ex,
 323                                                    (void *)&ext.fc_ex,
 324                                                    sizeof(ext.fc_ex));
 325                         if (ret)
 326                                 ret = JBD2_FC_REPLAY_STOP;
 327                         else
 328                                 ret = JBD2_FC_REPLAY_CONTINUE;
 329                         /* fallthrough */
 330                 case EXT4_FC_TAG_DEL_RANGE:
 331                 case EXT4_FC_TAG_LINK:
 332                 case EXT4_FC_TAG_UNLINK:
 333                 case EXT4_FC_TAG_CREAT:
 334                 case EXT4_FC_TAG_INODE:
 335                 case EXT4_FC_TAG_PAD:
 336                         state->fc_cur_tag++;
 337                         state->fc_crc = jbd2_chksum(j, state->fc_crc, cur,
 338                                         sizeof(tl) + ext4_fc_tag_len(&tl));
 339                         break;
 340                 case EXT4_FC_TAG_TAIL:
 341                         state->fc_cur_tag++;
 342                         memcpy(&tail, val, sizeof(tail));
 343                         state->fc_crc = jbd2_chksum(j, state->fc_crc, cur,
 344                                                 sizeof(tl) +
 345                                                 offsetof(struct ext4_fc_tail,
 346                                                 fc_crc));
 347                         jbd_debug(1, "tail tid %d, expected %d\n",
 348                                   le32_to_cpu(tail.fc_tid), expected_tid);
 349                         if (le32_to_cpu(tail.fc_tid) == expected_tid &&
 350                             le32_to_cpu(tail.fc_crc) == state->fc_crc) {
 351                                 state->fc_replay_num_tags = state->fc_cur_tag;
 352                         } else {
 353                                 ret = state->fc_replay_num_tags ?
 354                                         JBD2_FC_REPLAY_STOP : -EFSBADCRC;
 355                         }
 356                         state->fc_crc = 0;
 357                         break;
 358                 case EXT4_FC_TAG_HEAD:
 359                         memcpy(&head, val, sizeof(head));
 360                         if (le32_to_cpu(head.fc_features) &
 361                             ~EXT4_FC_SUPPORTED_FEATURES) {
 362                                 ret = -EOPNOTSUPP;
 363                                 break;
 364                         }
 365                         if (le32_to_cpu(head.fc_tid) != expected_tid) {
 366                                 ret = -EINVAL;
 367                                 break;
 368                         }
 369                         state->fc_cur_tag++;
 370                         state->fc_crc = jbd2_chksum(j, state->fc_crc, cur,
 371                                         sizeof(tl) + ext4_fc_tag_len(&tl));
 372                         break;
 373                 default:
 374                         ret = state->fc_replay_num_tags ?
 375                                 JBD2_FC_REPLAY_STOP : -ECANCELED;
 376                 }
 377                 if (ret < 0 || ret == JBD2_FC_REPLAY_STOP)
 378                         break;
 379         }
 380
 381 out_err:
 382         return ret;
 383 }
 384
 385 static int __errcode_to_errno(errcode_t err, const char *func, int line)
 386 {
 387         if (err == 0)
 388                 return 0;
 389         fprintf(stderr, "Error \"%s\" encountered in function %s at line %d\n",
 390                 error_message(err), func, line);
 391         if (err <= 256)
 392                 return -err;
 393         return -EFAULT;
 394 }
 395
 396 #define errcode_to_errno(err)   __errcode_to_errno(err, __func__, __LINE__)
 397
 398 #define ex_end(__ex) ((__ex)->e_lblk + (__ex)->e_len - 1)
 399 #define ex_pend(__ex) ((__ex)->e_pblk + (__ex)->e_len - 1)
 400
 401 static int make_room(struct extent_list *list, int i)
 402 {
 403         int ret;
 404
 405         if (list->count == list->size) {
 406                 unsigned int new_size = (list->size + 341) *
 407                                         sizeof(struct ext2fs_extent);
 408                 ret = errcode_to_errno(ext2fs_resize_mem(0, new_size, &list->extents));
 409                 if (ret)
 410                         return ret;
 411                 list->size += 341;
 412         }
 413
 414         memmove(&list->extents[i + 1], &list->extents[i],
 415                         sizeof(list->extents[0]) * (list->count - i));
 416         list->count++;
 417         return 0;
 418 }
 419
 420 static int ex_compar(const void *arg1, const void *arg2)
 421 {
 422         const struct ext2fs_extent *ex1 = (const struct ext2fs_extent *)arg1;
 423         const struct ext2fs_extent *ex2 = (const struct ext2fs_extent *)arg2;
 424
 425         if (ex1->e_lblk < ex2->e_lblk)
 426                 return -1;
 427         if (ex1->e_lblk > ex2->e_lblk)
 428                 return 1;
 429         return ex1->e_len - ex2->e_len;
 430 }
 431
 432 static int ex_len_compar(const void *arg1, const void *arg2)
 433 {
 434         const struct ext2fs_extent *ex1 = (const struct ext2fs_extent *)arg1;
 435         const struct ext2fs_extent *ex2 = (const struct ext2fs_extent *)arg2;
 436
 437         if (ex1->e_len < ex2->e_len)
 438                 return 1;
 439
 440         if (ex1->e_lblk > ex2->e_lblk)
 441                 return -1;
 442
 443         return 0;
 444 }
 445
 446 static void ex_sort_and_merge(struct extent_list *list)
 447 {
 448         unsigned int i, j;
 449
 450         if (list->count < 2)
 451                 return;
 452
 453         /*
 454          * Reverse sort by length, that way we strip off all the 0 length
 455          * extents
 456          */
 457         qsort(list->extents, list->count, sizeof(struct ext2fs_extent),
 458                 ex_len_compar);
 459
 460         for (i = 0; i < list->count; i++) {
 461                 if (list->extents[i].e_len == 0) {
 462                         list->count = i;
 463                         break;
 464                 }
 465         }
 466
 467         if (list->count == 0)
 468                 return;
 469
 470         /* Now sort by logical offset */
 471         qsort(list->extents, list->count, sizeof(list->extents[0]),
 472                 ex_compar);
 473
 474         /* Merge adjacent extents if they are logically and physically contiguous */
 475         i = 0;
 476         while (i < list->count - 1) {
 477                 if (ex_end(&list->extents[i]) + 1 != list->extents[i + 1].e_lblk ||
 478                         ex_pend(&list->extents[i]) + 1 != list->extents[i + 1].e_pblk ||
 479                         (list->extents[i].e_flags & EXT2_EXTENT_FLAGS_UNINIT) !=
 480                                 (list->extents[i + 1].e_flags & EXT2_EXTENT_FLAGS_UNINIT)) {
 481                         i++;
 482                         continue;
 483                 }
 484
 485                 list->extents[i].e_len += list->extents[i + 1].e_len;
 486                 for (j = i + 1; j < list->count - 1; j++)
 487                         list->extents[j] = list->extents[j + 1];
 488                 list->count--;
 489         }
 490 }
 491
 492 /* must free blocks that are released */
 493 static int ext4_modify_extent_list(e2fsck_t ctx, struct extent_list *list,
 494                                         struct ext2fs_extent *ex, int del)
 495 {
 496         int ret, offset;
 497         unsigned int i;
 498         struct ext2fs_extent add_ex = *ex;
 499
 500         /* First let's create a hole from ex->e_lblk of length ex->e_len */
 501         for (i = 0; i < list->count; i++) {
 502                 if (ex_end(&list->extents[i]) < add_ex.e_lblk)
 503                         continue;
 504
 505                 /* Case 1: No overlap */
 506                 if (list->extents[i].e_lblk > ex_end(&add_ex))
 507                         break;
 508                 /*
 509                  * Unmark all the blocks in bb now. All the blocks get marked
 510                  * before we exit this function.
 511                  */
 512                 ext2fs_unmark_block_bitmap_range2(ctx->fs->block_map,
 513                         list->extents[i].e_pblk, list->extents[i].e_len);
 514                 /* Case 2: Split */
 515                 if (list->extents[i].e_lblk < add_ex.e_lblk &&
 516                         ex_end(&list->extents[i]) > ex_end(&add_ex)) {
 517                         ret = make_room(list, i + 1);
 518                         if (ret)
 519                                 return ret;
 520                         list->extents[i + 1] = list->extents[i];
 521                         offset = ex_end(&add_ex) + 1 - list->extents[i].e_lblk;
 522                         list->extents[i + 1].e_lblk += offset;
 523                         list->extents[i + 1].e_pblk += offset;
 524                         list->extents[i + 1].e_len -= offset;
 525                         list->extents[i].e_len =
 526                                 add_ex.e_lblk - list->extents[i].e_lblk;
 527                         break;
 528                 }
 529
 530                 /* Case 3: Exact overlap */
 531                 if (add_ex.e_lblk <= list->extents[i].e_lblk  &&
 532                         ex_end(&list->extents[i]) <= ex_end(&add_ex)) {
 533
 534                         list->extents[i].e_len = 0;
 535                         continue;
 536                 }
 537
 538                 /* Case 4: Partial overlap */
 539                 if (ex_end(&list->extents[i]) > ex_end(&add_ex)) {
 540                         offset = ex_end(&add_ex) + 1 - list->extents[i].e_lblk;
 541                         list->extents[i].e_lblk += offset;
 542                         list->extents[i].e_pblk += offset;
 543                         list->extents[i].e_len -= offset;
 544                         break;
 545                 }
 546
 547                 if (ex_end(&add_ex) >= ex_end(&list->extents[i]))
 548                         list->extents[i].e_len =
 549                                 add_ex.e_lblk > list->extents[i].e_lblk ?
 550                                 add_ex.e_lblk - list->extents[i].e_lblk : 0;
 551         }
 552
 553         if (add_ex.e_len && !del) {
 554                 make_room(list, list->count);
 555                 list->extents[list->count - 1] = add_ex;
 556         }
 557
 558         ex_sort_and_merge(list);
 559
 560         /* Mark all occupied blocks allocated */
 561         for (i = 0; i < list->count; i++)
 562                 ext2fs_mark_block_bitmap_range2(ctx->fs->block_map,
 563                         list->extents[i].e_pblk, list->extents[i].e_len);
 564         ext2fs_mark_bb_dirty(ctx->fs);
 565
 566         return 0;
 567 }
 568
 569 static int ext4_add_extent_to_list(e2fsck_t ctx, struct extent_list *list,
 570                                         struct ext2fs_extent *ex)
 571 {
 572         return ext4_modify_extent_list(ctx, list, ex, 0 /* add */);
 573 }
 574
 575 static int ext4_del_extent_from_list(e2fsck_t ctx, struct extent_list *list,
 576                                         struct ext2fs_extent *ex)
 577 {
 578         return ext4_modify_extent_list(ctx, list, ex, 1 /* delete */);
 579 }
 580
 581 static int ext4_fc_read_extents(e2fsck_t ctx, ino_t ino)
 582 {
 583         struct extent_list *extent_list = &ctx->fc_replay_state.fc_extent_list;
 584
 585         if (extent_list->ino == ino)
 586                 return 0;
 587
 588         extent_list->ino = ino;
 589         return errcode_to_errno(e2fsck_read_extents(ctx, extent_list));
 590 }
 591
 592 /*
 593  * Flush extents in replay state on disk. @ino is the inode that is going
 594  * to be processed next. So, we hold back flushing of the extent list
 595  * if the next inode that's going to be processed is same as the one with
 596  * cached extents in our replay state. That allows us to gather multiple extents
 597  * for the inode so that we can flush all of them at once and it also saves us
 598  * from continuously growing and shrinking the extent tree.
 599  */
 600 static void ext4_fc_flush_extents(e2fsck_t ctx, ino_t ino)
 601 {
 602         struct extent_list *extent_list = &ctx->fc_replay_state.fc_extent_list;
 603
 604         if (extent_list->ino == ino || extent_list->ino == 0)
 605                 return;
 606         e2fsck_rewrite_extent_tree(ctx, extent_list);
 607         ext2fs_free_mem(&extent_list->extents);
 608         memset(extent_list, 0, sizeof(*extent_list));
 609 }
 610
 611 /* Helper struct for dentry replay routines */
 612 struct dentry_info_args {
 613         ino_t parent_ino;
 614         int dname_len;
 615         ino_t ino;
 616         char *dname;
 617 };
 618
 619 static inline int tl_to_darg(struct dentry_info_args *darg,
 620                              struct  ext4_fc_tl *tl, __u8 *val)
 621 {
 622         struct ext4_fc_dentry_info fcd;
 623
 624         memcpy(&fcd, val, sizeof(fcd));
 625
 626         darg->parent_ino = le32_to_cpu(fcd.fc_parent_ino);
 627         darg->ino = le32_to_cpu(fcd.fc_ino);
 628         darg->dname_len = ext4_fc_tag_len(tl) -
 629                         sizeof(struct ext4_fc_dentry_info);
 630         darg->dname = malloc(darg->dname_len + 1);
 631         if (!darg->dname)
 632                 return -ENOMEM;
 633         memcpy(darg->dname,
 634                val + sizeof(struct ext4_fc_dentry_info),
 635                darg->dname_len);
 636         darg->dname[darg->dname_len] = 0;
 637         jbd_debug(1, "%s: %s, ino %lu, parent %lu\n",
 638                   le16_to_cpu(tl->fc_tag) == EXT4_FC_TAG_CREAT ? "create" :
 639                   (le16_to_cpu(tl->fc_tag) == EXT4_FC_TAG_LINK ? "link" :
 640                    (le16_to_cpu(tl->fc_tag) == EXT4_FC_TAG_UNLINK ? "unlink" :
 641                     "error")), darg->dname, darg->ino, darg->parent_ino);
 642         return 0;
 643 }
 644
 645 static int ext4_fc_handle_unlink(e2fsck_t ctx, struct ext4_fc_tl *tl, __u8 *val)
 646 {
 647         struct dentry_info_args darg;
 648         int ret;
 649
 650         ret = tl_to_darg(&darg, tl, val);
 651         if (ret)
 652                 return ret;
 653         ext4_fc_flush_extents(ctx, darg.ino);
 654         ret = errcode_to_errno(ext2fs_unlink(ctx->fs, darg.parent_ino,
 655                                              darg.dname, darg.ino, 0));
 656         /* It's okay if the above call fails */
 657         free(darg.dname);
 658
 659         return ret;
 660 }
 661
 662 static int ext4_fc_handle_link_and_create(e2fsck_t ctx, struct ext4_fc_tl *tl, __u8 *val)
 663 {
 664         struct dentry_info_args darg;
 665         ext2_filsys fs = ctx->fs;
 666         struct ext2_inode_large inode_large;
 667         int ret, filetype, mode;
 668
 669         ret = tl_to_darg(&darg, tl, val);
 670         if (ret)
 671                 return ret;
 672         ext4_fc_flush_extents(ctx, 0);
 673         ret = errcode_to_errno(ext2fs_read_inode(fs, darg.ino,
 674                                                  (struct ext2_inode *)&inode_large));
 675         if (ret)
 676                 goto out;
 677
 678         mode = inode_large.i_mode;
 679
 680         if (LINUX_S_ISREG(mode))
 681                 filetype = EXT2_FT_REG_FILE;
 682         else if (LINUX_S_ISDIR(mode))
 683                 filetype = EXT2_FT_DIR;
 684         else if (LINUX_S_ISCHR(mode))
 685                 filetype = EXT2_FT_CHRDEV;
 686         else if (LINUX_S_ISBLK(mode))
 687                 filetype = EXT2_FT_BLKDEV;
 688         else if (LINUX_S_ISLNK(mode))
 689                 return EXT2_FT_SYMLINK;
 690         else if (LINUX_S_ISFIFO(mode))
 691                 filetype = EXT2_FT_FIFO;
 692         else if (LINUX_S_ISSOCK(mode))
 693                 filetype = EXT2_FT_SOCK;
 694         else {
 695                 ret = -EINVAL;
 696                 goto out;
 697         }
 698
 699         /*
 700          * Forcefully unlink if the same name is present and ignore the error
 701          * if any, since this dirent might not exist
 702          */
 703         ext2fs_unlink(fs, darg.parent_ino, darg.dname, darg.ino,
 704                         EXT2FS_UNLINK_FORCE);
 705
 706         ret = errcode_to_errno(
 707                        ext2fs_link(fs, darg.parent_ino, darg.dname, darg.ino,
 708                                    filetype));
 709 out:
 710         free(darg.dname);
 711         return ret;
 712
 713 }
 714
 715 /* This function fixes the i_blocks field in the replayed indoe */
 716 static void ext4_fc_replay_fixup_iblocks(struct ext2_inode_large *ondisk_inode,
 717         struct ext2_inode_large *fc_inode)
 718 {
 719         if (ondisk_inode->i_flags & EXT4_EXTENTS_FL) {
 720                 struct ext3_extent_header *eh;
 721
 722                 eh = (struct ext3_extent_header *)(&ondisk_inode->i_block[0]);
 723                 if (le16_to_cpu(eh->eh_magic) != EXT3_EXT_MAGIC) {
 724                         memset(eh, 0, sizeof(*eh));
 725                         eh->eh_magic = cpu_to_le16(EXT3_EXT_MAGIC);
 726                         eh->eh_max = cpu_to_le16(
 727                                 (sizeof(ondisk_inode->i_block) -
 728                                         sizeof(struct ext3_extent_header)) /
 729                                 sizeof(struct ext3_extent));
 730                 }
 731         } else if (ondisk_inode->i_flags & EXT4_INLINE_DATA_FL) {
 732                 memcpy(ondisk_inode->i_block, fc_inode->i_block,
 733                         sizeof(fc_inode->i_block));
 734         }
 735 }
 736
 737 static int ext4_fc_handle_inode(e2fsck_t ctx, __u8 *val)
 738 {
 739         int ino, inode_len = EXT2_GOOD_OLD_INODE_SIZE;
 740         struct ext2_inode_large *inode = NULL, *fc_inode = NULL;
 741         __le32 fc_ino;
 742         __u8 *fc_raw_inode;
 743         errcode_t err;
 744         blk64_t blks;
 745
 746         memcpy(&fc_ino, val, sizeof(fc_ino));
 747         fc_raw_inode = val + sizeof(fc_ino);
 748         ino = le32_to_cpu(fc_ino);
 749
 750         if (EXT2_INODE_SIZE(ctx->fs->super) > EXT2_GOOD_OLD_INODE_SIZE) {
 751                 __u16 extra_isize = ext2fs_le16_to_cpu(
 752                         ((struct ext2_inode_large *)fc_raw_inode)->i_extra_isize);
 753
 754                 if ((extra_isize < (sizeof(inode->i_extra_isize) +
 755                                     sizeof(inode->i_checksum_hi))) ||
 756                     (extra_isize > (EXT2_INODE_SIZE(ctx->fs->super) -
 757                                     EXT2_GOOD_OLD_INODE_SIZE))) {
 758                         err = EFSCORRUPTED;
 759                         goto out;
 760                 }
 761                 inode_len += extra_isize;
 762         }
 763         err = ext2fs_get_mem(inode_len, &inode);
 764         if (err)
 765                 goto out;
 766         err = ext2fs_get_mem(inode_len, &fc_inode);
 767         if (err)
 768                 goto out;
 769         ext4_fc_flush_extents(ctx, ino);
 770
 771         err = ext2fs_read_inode_full(ctx->fs, ino, (struct ext2_inode *)inode,
 772                                         inode_len);
 773         if (err)
 774                 goto out;
 775         memcpy(fc_inode, fc_raw_inode, inode_len);
 776 #ifdef WORDS_BIGENDIAN
 777         ext2fs_swap_inode_full(ctx->fs, fc_inode, fc_inode, 0, inode_len);
 778 #endif
 779         memcpy(inode, fc_inode, offsetof(struct ext2_inode_large, i_block));
 780         memcpy(&inode->i_generation, &fc_inode->i_generation,
 781                 inode_len - offsetof(struct ext2_inode_large, i_generation));
 782         ext4_fc_replay_fixup_iblocks(inode, fc_inode);
 783         err = ext2fs_count_blocks(ctx->fs, ino, EXT2_INODE(inode), &blks);
 784         if (err)
 785                 goto out;
 786         ext2fs_iblk_set(ctx->fs, EXT2_INODE(inode), blks);
 787         ext2fs_inode_csum_set(ctx->fs, ino, inode);
 788
 789         err = ext2fs_write_inode_full(ctx->fs, ino, (struct ext2_inode *)inode,
 790                                         inode_len);
 791         if (err)
 792                 goto out;
 793         if (inode->i_links_count)
 794                 ext2fs_mark_inode_bitmap2(ctx->fs->inode_map, ino);
 795         else
 796                 ext2fs_unmark_inode_bitmap2(ctx->fs->inode_map, ino);
 797         ext2fs_mark_ib_dirty(ctx->fs);
 798
 799 out:
 800         ext2fs_free_mem(&inode);
 801         ext2fs_free_mem(&fc_inode);
 802         return errcode_to_errno(err);
 803 }
 804
 805 /*
 806  * Handle add extent replay tag.
 807  */
 808 static int ext4_fc_handle_add_extent(e2fsck_t ctx, __u8 *val)
 809 {
 810         struct ext2fs_extent extent;
 811         struct ext4_fc_add_range add_range;
 812         ino_t ino;
 813         int ret = 0;
 814
 815         memcpy(&add_range, val, sizeof(add_range));
 816         ino = le32_to_cpu(add_range.fc_ino);
 817         ext4_fc_flush_extents(ctx, ino);
 818
 819         ret = ext4_fc_read_extents(ctx, ino);
 820         if (ret)
 821                 return ret;
 822         memset(&extent, 0, sizeof(extent));
 823         ret = errcode_to_errno(ext2fs_decode_extent(
 824                         &extent, (void *)add_range.fc_ex,
 825                         sizeof(add_range.fc_ex)));
 826         if (ret)
 827                 return ret;
 828         return ext4_add_extent_to_list(ctx,
 829                 &ctx->fc_replay_state.fc_extent_list, &extent);
 830 }
 831
 832 /*
 833  * Handle delete logical range replay tag.
 834  */
 835 static int ext4_fc_handle_del_range(e2fsck_t ctx, __u8 *val)
 836 {
 837         struct ext2fs_extent extent;
 838         struct ext4_fc_del_range del_range;
 839         int ret, ino;
 840
 841         memcpy(&del_range, val, sizeof(del_range));
 842         ino = le32_to_cpu(del_range.fc_ino);
 843         ext4_fc_flush_extents(ctx, ino);
 844
 845         memset(&extent, 0, sizeof(extent));
 846         extent.e_lblk = le32_to_cpu(del_range.fc_lblk);
 847         extent.e_len = le32_to_cpu(del_range.fc_len);
 848         ret = ext4_fc_read_extents(ctx, ino);
 849         if (ret)
 850                 return ret;
 851         return ext4_del_extent_from_list(ctx,
 852                 &ctx->fc_replay_state.fc_extent_list, &extent);
 853 }
 854
 855 /*
 856  * Main recovery path entry point. This function returns JBD2_FC_REPLAY_CONTINUE
 857  * to indicate that it is expecting more fast commit blocks. It returns
 858  * JBD2_FC_REPLAY_STOP to indicate that replay is done.
 859  */
 860 static int ext4_fc_replay(journal_t *journal, struct buffer_head *bh,
 861                                 enum passtype pass, int off, tid_t expected_tid)
 862 {
 863         e2fsck_t ctx = journal->j_fs_dev->k_ctx;
 864         struct e2fsck_fc_replay_state *state = &ctx->fc_replay_state;
 865         int ret = JBD2_FC_REPLAY_CONTINUE;
 866         struct ext4_fc_tl tl;
 867         __u8 *start, *end, *cur, *val;
 868
 869         if (pass == PASS_SCAN) {
 870                 state->fc_current_pass = PASS_SCAN;
 871                 return ext4_fc_replay_scan(journal, bh, off, expected_tid);
 872         }
 873
 874         if (state->fc_replay_num_tags == 0)
 875                 goto replay_done;
 876
 877         if (state->fc_current_pass != pass) {
 878                 /* Starting replay phase */
 879                 state->fc_current_pass = pass;
 880                 /* We will reset checksums */
 881                 ctx->fs->flags |= EXT2_FLAG_IGNORE_CSUM_ERRORS;
 882                 ret = errcode_to_errno(ext2fs_read_bitmaps(ctx->fs));
 883                 if (ret) {
 884                         jbd_debug(1, "Error %d while reading bitmaps\n", ret);
 885                         return ret;
 886                 }
 887                 state->fc_super_state = ctx->fs->super->s_state;
 888                 /*
 889                  * Mark the file system to indicate it contains errors. That's
 890                  * because the updates performed by fast commit replay code are
 891                  * not atomic and may result in incosistent file system if it
 892                  * crashes before the replay is complete.
 893                  */
 894                 ctx->fs->super->s_state |= EXT2_ERROR_FS;
 895                 ctx->fs->super->s_state |= EXT4_FC_REPLAY;
 896                 ext2fs_mark_super_dirty(ctx->fs);
 897                 ext2fs_flush(ctx->fs);
 898         }
 899
 900         start = (__u8 *)bh->b_data;
 901         end = (__u8 *)bh->b_data + journal->j_blocksize - 1;
 902
 903         for (cur = start; cur < end; cur = cur + le16_to_cpu(tl.fc_len) + sizeof(tl)) {
 904                 memcpy(&tl, cur, sizeof(tl));
 905                 val = cur + sizeof(tl);
 906
 907                 if (state->fc_replay_num_tags == 0)
 908                         goto replay_done;
 909                 jbd_debug(3, "Replay phase processing %s tag\n",
 910                                 tag2str(le16_to_cpu(tl.fc_tag)));
 911                 state->fc_replay_num_tags--;
 912                 switch (le16_to_cpu(tl.fc_tag)) {
 913                 case EXT4_FC_TAG_CREAT:
 914                 case EXT4_FC_TAG_LINK:
 915                         ret = ext4_fc_handle_link_and_create(ctx, &tl, val);
 916                         break;
 917                 case EXT4_FC_TAG_UNLINK:
 918                         ret = ext4_fc_handle_unlink(ctx, &tl, val);
 919                         break;
 920                 case EXT4_FC_TAG_ADD_RANGE:
 921                         ret = ext4_fc_handle_add_extent(ctx, val);
 922                         break;
 923                 case EXT4_FC_TAG_DEL_RANGE:
 924                         ret = ext4_fc_handle_del_range(ctx, val);
 925                         break;
 926                 case EXT4_FC_TAG_INODE:
 927                         ret = ext4_fc_handle_inode(ctx, val);
 928                         break;
 929                 case EXT4_FC_TAG_TAIL:
 930                         ext4_fc_flush_extents(ctx, 0);
 931                 case EXT4_FC_TAG_PAD:
 932                 case EXT4_FC_TAG_HEAD:
 933                         break;
 934                 default:
 935                         ret = -ECANCELED;
 936                         break;
 937                 }
 938                 if (ret < 0)
 939                         break;
 940                 ret = JBD2_FC_REPLAY_CONTINUE;
 941         }
 942         return ret;
 943 replay_done:
 944         jbd_debug(1, "End of fast commit replay\n");
 945         if (state->fc_current_pass != pass)
 946                 return JBD2_FC_REPLAY_STOP;
 947
 948         ext2fs_calculate_summary_stats(ctx->fs, 0 /* update bg also */);
 949         ext2fs_write_block_bitmap(ctx->fs);
 950         ext2fs_write_inode_bitmap(ctx->fs);
 951         ext2fs_mark_super_dirty(ctx->fs);
 952         ext2fs_set_gdt_csum(ctx->fs);
 953         ctx->fs->super->s_state = state->fc_super_state;
 954         ext2fs_flush(ctx->fs);
 955
 956         return JBD2_FC_REPLAY_STOP;
 957 }
 958
 959 static errcode_t e2fsck_get_journal(e2fsck_t ctx, journal_t **ret_journal)
 960 {
 961         struct process_block_struct pb;
 962         struct ext2_super_block *sb = ctx->fs->super;
 963         struct ext2_super_block jsuper;
 964         struct problem_context  pctx;
 965         struct buffer_head      *bh;
 966         struct inode            *j_inode = NULL;
 967         struct kdev_s           *dev_fs = NULL, *dev_journal;
 968         const char              *journal_name = 0;
 969         journal_t               *journal = NULL;
 970         errcode_t               retval = 0;
 971         io_manager              io_ptr = 0;
 972         unsigned long long      start = 0;
 973         int                     ret;
 974         int                     ext_journal = 0;
 975         int                     tried_backup_jnl = 0;
 976
 977         clear_problem_context(&pctx);
 978
 979         journal = e2fsck_allocate_memory(ctx, sizeof(journal_t), "journal");
 980         if (!journal) {
 981                 return EXT2_ET_NO_MEMORY;
 982         }
 983
 984         dev_fs = e2fsck_allocate_memory(ctx, 2*sizeof(struct kdev_s), "kdev");
 985         if (!dev_fs) {
 986                 retval = EXT2_ET_NO_MEMORY;
 987                 goto errout;
 988         }
 989         dev_journal = dev_fs+1;
 990
 991         dev_fs->k_ctx = dev_journal->k_ctx = ctx;
 992         dev_fs->k_dev = K_DEV_FS;
 993         dev_journal->k_dev = K_DEV_JOURNAL;
 994
 995         journal->j_dev = dev_journal;
 996         journal->j_fs_dev = dev_fs;
 997         journal->j_inode = NULL;
 998         journal->j_blocksize = ctx->fs->blocksize;
 999
1000         if (uuid_is_null(sb->s_journal_uuid)) {
1001                 /*
1002                  * The full set of superblock sanity checks haven't
1003                  * been performed yet, so we need to do some basic
1004                  * checks here to avoid potential array overruns.
1005                  */
1006                 if (!sb->s_journal_inum ||
1007                     (sb->s_journal_inum >
1008                      (ctx->fs->group_desc_count * sb->s_inodes_per_group))) {
1009                         retval = EXT2_ET_BAD_INODE_NUM;
1010                         goto errout;
1011                 }
1012                 j_inode = e2fsck_allocate_memory(ctx, sizeof(*j_inode),
1013                                                  "journal inode");
1014                 if (!j_inode) {
1015                         retval = EXT2_ET_NO_MEMORY;
1016                         goto errout;
1017                 }
1018
1019                 j_inode->i_ctx = ctx;
1020                 j_inode->i_ino = sb->s_journal_inum;
1021
1022                 if ((retval = ext2fs_read_inode(ctx->fs,
1023                                                 sb->s_journal_inum,
1024                                                 &j_inode->i_ext2))) {
1025                 try_backup_journal:
1026                         if (sb->s_jnl_backup_type != EXT3_JNL_BACKUP_BLOCKS ||
1027                             tried_backup_jnl)
1028                                 goto errout;
1029                         memset(&j_inode->i_ext2, 0, sizeof(struct ext2_inode));
1030                         memcpy(&j_inode->i_ext2.i_block[0], sb->s_jnl_blocks,
1031                                EXT2_N_BLOCKS*4);
1032                         j_inode->i_ext2.i_size_high = sb->s_jnl_blocks[15];
1033                         j_inode->i_ext2.i_size = sb->s_jnl_blocks[16];
1034                         j_inode->i_ext2.i_links_count = 1;
1035                         j_inode->i_ext2.i_mode = LINUX_S_IFREG | 0600;
1036                         e2fsck_use_inode_shortcuts(ctx, 1);
1037                         ctx->stashed_ino = j_inode->i_ino;
1038                         ctx->stashed_inode = &j_inode->i_ext2;
1039                         tried_backup_jnl++;
1040                 }
1041                 if (!j_inode->i_ext2.i_links_count ||
1042                     !LINUX_S_ISREG(j_inode->i_ext2.i_mode)) {
1043                         retval = EXT2_ET_NO_JOURNAL;
1044                         goto try_backup_journal;
1045                 }
1046                 if (EXT2_I_SIZE(&j_inode->i_ext2) / journal->j_blocksize <
1047                     JBD2_MIN_JOURNAL_BLOCKS) {
1048                         retval = EXT2_ET_JOURNAL_TOO_SMALL;
1049                         goto try_backup_journal;
1050                 }
1051                 pb.last_block = -1;
1052                 retval = ext2fs_block_iterate3(ctx->fs, j_inode->i_ino,
1053                                                BLOCK_FLAG_HOLE, 0,
1054                                                process_journal_block, &pb);
1055                 if ((pb.last_block + 1) * ctx->fs->blocksize <
1056                     (int) EXT2_I_SIZE(&j_inode->i_ext2)) {
1057                         retval = EXT2_ET_JOURNAL_TOO_SMALL;
1058                         goto try_backup_journal;
1059                 }
1060                 if (tried_backup_jnl && !(ctx->options & E2F_OPT_READONLY)) {
1061                         retval = ext2fs_write_inode(ctx->fs, sb->s_journal_inum,
1062                                                     &j_inode->i_ext2);
1063                         if (retval)
1064                                 goto errout;
1065                 }
1066
1067                 journal->j_total_len = EXT2_I_SIZE(&j_inode->i_ext2) /
1068                         journal->j_blocksize;
1069
1070 #ifdef USE_INODE_IO
1071                 retval = ext2fs_inode_io_intern2(ctx->fs, sb->s_journal_inum,
1072                                                  &j_inode->i_ext2,
1073                                                  &journal_name);
1074                 if (retval)
1075                         goto errout;
1076
1077                 io_ptr = inode_io_manager;
1078 #else
1079                 journal->j_inode = j_inode;
1080                 ctx->journal_io = ctx->fs->io;
1081                 if ((ret = jbd2_journal_bmap(journal, 0, &start)) != 0) {
1082                         retval = (errcode_t) (-1 * ret);
1083                         goto errout;
1084                 }
1085 #endif
1086         } else {
1087                 ext_journal = 1;
1088                 if (!ctx->journal_name) {
1089                         char uuid[37];
1090
1091                         uuid_unparse(sb->s_journal_uuid, uuid);
1092                         ctx->journal_name = blkid_get_devname(ctx->blkid,
1093                                                               "UUID", uuid);
1094                         if (!ctx->journal_name)
1095                                 ctx->journal_name = blkid_devno_to_devname(sb->s_journal_dev);
1096                 }
1097                 journal_name = ctx->journal_name;
1098
1099                 if (!journal_name) {
1100                         fix_problem(ctx, PR_0_CANT_FIND_JOURNAL, &pctx);
1101                         retval = EXT2_ET_LOAD_EXT_JOURNAL;
1102                         goto errout;
1103                 }
1104
1105                 jfs_debug(1, "Using journal file %s\n", journal_name);
1106                 io_ptr = unix_io_manager;
1107         }
1108
1109 #if 0
1110         test_io_backing_manager = io_ptr;
1111         io_ptr = test_io_manager;
1112 #endif
1113 #ifndef USE_INODE_IO
1114         if (ext_journal)
1115 #endif
1116         {
1117                 int flags = IO_FLAG_RW;
1118                 if (!(ctx->mount_flags & EXT2_MF_ISROOT &&
1119                       ctx->mount_flags & EXT2_MF_READONLY))
1120                         flags |= IO_FLAG_EXCLUSIVE;
1121                 if ((ctx->mount_flags & EXT2_MF_READONLY) &&
1122                     (ctx->options & E2F_OPT_FORCE))
1123                         flags &= ~IO_FLAG_EXCLUSIVE;
1124
1125
1126                 retval = io_ptr->open(journal_name, flags,
1127                                       &ctx->journal_io);
1128         }
1129         if (retval)
1130                 goto errout;
1131
1132         io_channel_set_blksize(ctx->journal_io, ctx->fs->blocksize);
1133
1134         if (ext_journal) {
1135                 blk64_t maxlen;
1136
1137                 start = ext2fs_journal_sb_start(ctx->fs->blocksize) - 1;
1138                 bh = getblk(dev_journal, start, ctx->fs->blocksize);
1139                 if (!bh) {
1140                         retval = EXT2_ET_NO_MEMORY;
1141                         goto errout;
1142                 }
1143                 ll_rw_block(REQ_OP_READ, 0, 1, &bh);
1144                 if ((retval = bh->b_err) != 0) {
1145                         brelse(bh);
1146                         goto errout;
1147                 }
1148                 memcpy(&jsuper, start ? bh->b_data :  bh->b_data + SUPERBLOCK_OFFSET,
1149                        sizeof(jsuper));
1150 #ifdef WORDS_BIGENDIAN
1151                 if (jsuper.s_magic == ext2fs_swab16(EXT2_SUPER_MAGIC))
1152                         ext2fs_swap_super(&jsuper);
1153 #endif
1154                 if (jsuper.s_magic != EXT2_SUPER_MAGIC ||
1155                     !ext2fs_has_feature_journal_dev(&jsuper)) {
1156                         fix_problem(ctx, PR_0_EXT_JOURNAL_BAD_SUPER, &pctx);
1157                         retval = EXT2_ET_LOAD_EXT_JOURNAL;
1158                         brelse(bh);
1159                         goto errout;
1160                 }
1161                 /* Make sure the journal UUID is correct */
1162                 if (memcmp(jsuper.s_uuid, ctx->fs->super->s_journal_uuid,
1163                            sizeof(jsuper.s_uuid))) {
1164                         fix_problem(ctx, PR_0_JOURNAL_BAD_UUID, &pctx);
1165                         retval = EXT2_ET_LOAD_EXT_JOURNAL;
1166                         brelse(bh);
1167                         goto errout;
1168                 }
1169
1170                 /* Check the superblock checksum */
1171                 if (ext2fs_has_feature_metadata_csum(&jsuper)) {
1172                         struct struct_ext2_filsys fsx;
1173                         struct ext2_super_block superx;
1174                         void *p;
1175
1176                         p = start ? bh->b_data : bh->b_data + SUPERBLOCK_OFFSET;
1177                         memcpy(&fsx, ctx->fs, sizeof(fsx));
1178                         memcpy(&superx, ctx->fs->super, sizeof(superx));
1179                         fsx.super = &superx;
1180                         ext2fs_set_feature_metadata_csum(fsx.super);
1181                         if (!ext2fs_superblock_csum_verify(&fsx, p) &&
1182                             fix_problem(ctx, PR_0_EXT_JOURNAL_SUPER_CSUM_INVALID,
1183                                         &pctx)) {
1184                                 ext2fs_superblock_csum_set(&fsx, p);
1185                                 mark_buffer_dirty(bh);
1186                         }
1187                 }
1188                 brelse(bh);
1189
1190                 maxlen = ext2fs_blocks_count(&jsuper);
1191                 journal->j_total_len = (maxlen < 1ULL << 32) ? maxlen : (1ULL << 32) - 1;
1192                 start++;
1193         }
1194
1195         if (!(bh = getblk(dev_journal, start, journal->j_blocksize))) {
1196                 retval = EXT2_ET_NO_MEMORY;
1197                 goto errout;
1198         }
1199
1200         journal->j_sb_buffer = bh;
1201         journal->j_superblock = (journal_superblock_t *)bh->b_data;
1202         if (ext2fs_has_feature_fast_commit(ctx->fs->super))
1203                 journal->j_fc_replay_callback = ext4_fc_replay;
1204         else
1205                 journal->j_fc_replay_callback = NULL;
1206
1207 #ifdef USE_INODE_IO
1208         if (j_inode)
1209                 ext2fs_free_mem(&j_inode);
1210 #endif
1211
1212         *ret_journal = journal;
1213         e2fsck_use_inode_shortcuts(ctx, 0);
1214         return 0;
1215
1216 errout:
1217         e2fsck_use_inode_shortcuts(ctx, 0);
1218         if (dev_fs)
1219                 ext2fs_free_mem(&dev_fs);
1220         if (j_inode)
1221                 ext2fs_free_mem(&j_inode);
1222         if (journal)
1223                 ext2fs_free_mem(&journal);
1224         return retval;
1225 }
1226
1227 static errcode_t e2fsck_journal_fix_bad_inode(e2fsck_t ctx,
1228                                               struct problem_context *pctx)
1229 {
1230         struct ext2_super_block *sb = ctx->fs->super;
1231         int recover = ext2fs_has_feature_journal_needs_recovery(ctx->fs->super);
1232         int has_journal = ext2fs_has_feature_journal(ctx->fs->super);
1233
1234         if (has_journal || sb->s_journal_inum) {
1235                 /* The journal inode is bogus, remove and force full fsck */
1236                 pctx->ino = sb->s_journal_inum;
1237                 if (fix_problem(ctx, PR_0_JOURNAL_BAD_INODE, pctx)) {
1238                         if (has_journal && sb->s_journal_inum)
1239                                 printf("*** journal has been deleted ***\n\n");
1240                         ext2fs_clear_feature_journal(sb);
1241                         sb->s_journal_inum = 0;
1242                         memset(sb->s_jnl_blocks, 0, sizeof(sb->s_jnl_blocks));
1243                         ctx->flags |= E2F_FLAG_JOURNAL_INODE;
1244                         ctx->fs->flags &= ~EXT2_FLAG_MASTER_SB_ONLY;
1245                         e2fsck_clear_recover(ctx, 1);
1246                         return 0;
1247                 }
1248                 return EXT2_ET_CORRUPT_JOURNAL_SB;
1249         } else if (recover) {
1250                 if (fix_problem(ctx, PR_0_JOURNAL_RECOVER_SET, pctx)) {
1251                         e2fsck_clear_recover(ctx, 1);
1252                         return 0;
1253                 }
1254                 return EXT2_ET_UNSUPP_FEATURE;
1255         }
1256         return 0;
1257 }
1258
1259 #define V1_SB_SIZE      0x0024
1260 static void clear_v2_journal_fields(journal_t *journal)
1261 {
1262         e2fsck_t ctx = journal->j_dev->k_ctx;
1263         struct problem_context pctx;
1264
1265         clear_problem_context(&pctx);
1266
1267         if (!fix_problem(ctx, PR_0_CLEAR_V2_JOURNAL, &pctx))
1268                 return;
1269
1270         ctx->flags |= E2F_FLAG_PROBLEMS_FIXED;
1271         memset(((char *) journal->j_superblock) + V1_SB_SIZE, 0,
1272                ctx->fs->blocksize-V1_SB_SIZE);
1273         mark_buffer_dirty(journal->j_sb_buffer);
1274 }
1275
1276
1277 static errcode_t e2fsck_journal_load(journal_t *journal)
1278 {
1279         e2fsck_t ctx = journal->j_dev->k_ctx;
1280         journal_superblock_t *jsb;
1281         struct buffer_head *jbh = journal->j_sb_buffer;
1282         struct problem_context pctx;
1283
1284         clear_problem_context(&pctx);
1285
1286         ll_rw_block(REQ_OP_READ, 0, 1, &jbh);
1287         if (jbh->b_err) {
1288                 com_err(ctx->device_name, jbh->b_err, "%s",
1289                         _("reading journal superblock\n"));
1290                 return jbh->b_err;
1291         }
1292
1293         jsb = journal->j_superblock;
1294         /* If we don't even have JBD2_MAGIC, we probably have a wrong inode */
1295         if (jsb->s_header.h_magic != htonl(JBD2_MAGIC_NUMBER))
1296                 return e2fsck_journal_fix_bad_inode(ctx, &pctx);
1297
1298         switch (ntohl(jsb->s_header.h_blocktype)) {
1299         case JBD2_SUPERBLOCK_V1:
1300                 journal->j_format_version = 1;
1301                 if (jsb->s_feature_compat ||
1302                     jsb->s_feature_incompat ||
1303                     jsb->s_feature_ro_compat ||
1304                     jsb->s_nr_users)
1305                         clear_v2_journal_fields(journal);
1306                 break;
1307
1308         case JBD2_SUPERBLOCK_V2:
1309                 journal->j_format_version = 2;
1310                 if (ntohl(jsb->s_nr_users) > 1 &&
1311                     uuid_is_null(ctx->fs->super->s_journal_uuid))
1312                         clear_v2_journal_fields(journal);
1313                 if (ntohl(jsb->s_nr_users) > 1) {
1314                         fix_problem(ctx, PR_0_JOURNAL_UNSUPP_MULTIFS, &pctx);
1315                         return EXT2_ET_JOURNAL_UNSUPP_VERSION;
1316                 }
1317                 break;
1318
1319         /*
1320          * These should never appear in a journal super block, so if
1321          * they do, the journal is badly corrupted.
1322          */
1323         case JBD2_DESCRIPTOR_BLOCK:
1324         case JBD2_COMMIT_BLOCK:
1325         case JBD2_REVOKE_BLOCK:
1326                 return EXT2_ET_CORRUPT_JOURNAL_SB;
1327
1328         /* If we don't understand the superblock major type, but there
1329          * is a magic number, then it is likely to be a new format we
1330          * just don't understand, so leave it alone. */
1331         default:
1332                 return EXT2_ET_JOURNAL_UNSUPP_VERSION;
1333         }
1334
1335         if (JBD2_HAS_INCOMPAT_FEATURE(journal, ~JBD2_KNOWN_INCOMPAT_FEATURES))
1336                 return EXT2_ET_UNSUPP_FEATURE;
1337
1338         if (JBD2_HAS_RO_COMPAT_FEATURE(journal, ~JBD2_KNOWN_ROCOMPAT_FEATURES))
1339                 return EXT2_ET_RO_UNSUPP_FEATURE;
1340
1341         /* Checksum v1-3 are mutually exclusive features. */
1342         if (jbd2_has_feature_csum2(journal) && jbd2_has_feature_csum3(journal))
1343                 return EXT2_ET_CORRUPT_JOURNAL_SB;
1344
1345         if (jbd2_journal_has_csum_v2or3(journal) &&
1346             jbd2_has_feature_checksum(journal))
1347                 return EXT2_ET_CORRUPT_JOURNAL_SB;
1348
1349         if (!e2fsck_journal_verify_csum_type(journal, jsb) ||
1350             !e2fsck_journal_sb_csum_verify(journal, jsb))
1351                 return EXT2_ET_CORRUPT_JOURNAL_SB;
1352
1353         if (jbd2_journal_has_csum_v2or3(journal))
1354                 journal->j_csum_seed = jbd2_chksum(journal, ~0, jsb->s_uuid,
1355                                                    sizeof(jsb->s_uuid));
1356
1357         /* We have now checked whether we know enough about the journal
1358          * format to be able to proceed safely, so any other checks that
1359          * fail we should attempt to recover from. */
1360         if (jsb->s_blocksize != htonl(journal->j_blocksize)) {
1361                 com_err(ctx->program_name, EXT2_ET_CORRUPT_JOURNAL_SB,
1362                         _("%s: no valid journal superblock found\n"),
1363                         ctx->device_name);
1364                 return EXT2_ET_CORRUPT_JOURNAL_SB;
1365         }
1366
1367         if (ntohl(jsb->s_maxlen) < journal->j_total_len)
1368                 journal->j_total_len = ntohl(jsb->s_maxlen);
1369         else if (ntohl(jsb->s_maxlen) > journal->j_total_len) {
1370                 com_err(ctx->program_name, EXT2_ET_CORRUPT_JOURNAL_SB,
1371                         _("%s: journal too short\n"),
1372                         ctx->device_name);
1373                 return EXT2_ET_CORRUPT_JOURNAL_SB;
1374         }
1375
1376         journal->j_tail_sequence = ntohl(jsb->s_sequence);
1377         journal->j_transaction_sequence = journal->j_tail_sequence;
1378         journal->j_tail = ntohl(jsb->s_start);
1379         journal->j_first = ntohl(jsb->s_first);
1380         if (jbd2_has_feature_fast_commit(journal)) {
1381                 if (ntohl(jsb->s_maxlen) - jbd2_journal_get_num_fc_blks(jsb)
1382                         < JBD2_MIN_JOURNAL_BLOCKS) {
1383                         com_err(ctx->program_name, EXT2_ET_CORRUPT_JOURNAL_SB,
1384                                 _("%s: incorrect fast commit blocks\n"),
1385                                 ctx->device_name);
1386                         return EXT2_ET_CORRUPT_JOURNAL_SB;
1387                 }
1388                 journal->j_fc_last = ntohl(jsb->s_maxlen);
1389                 journal->j_last = journal->j_fc_last -
1390                                         jbd2_journal_get_num_fc_blks(jsb);
1391                 journal->j_fc_first = journal->j_last + 1;
1392         } else {
1393                 journal->j_last = ntohl(jsb->s_maxlen);
1394         }
1395
1396         return 0;
1397 }
1398
1399 static void e2fsck_journal_reset_super(e2fsck_t ctx, journal_superblock_t *jsb,
1400                                        journal_t *journal)
1401 {
1402         char *p;
1403         union {
1404                 uuid_t uuid;
1405                 __u32 val[4];
1406         } u;
1407         __u32 new_seq = 0;
1408         int i;
1409
1410         /* Leave a valid existing V1 superblock signature alone.
1411          * Anything unrecognisable we overwrite with a new V2
1412          * signature. */
1413
1414         if (jsb->s_header.h_magic != htonl(JBD2_MAGIC_NUMBER) ||
1415             jsb->s_header.h_blocktype != htonl(JBD2_SUPERBLOCK_V1)) {
1416                 jsb->s_header.h_magic = htonl(JBD2_MAGIC_NUMBER);
1417                 jsb->s_header.h_blocktype = htonl(JBD2_SUPERBLOCK_V2);
1418         }
1419
1420         /* Zero out everything else beyond the superblock header */
1421
1422         p = ((char *) jsb) + sizeof(journal_header_t);
1423         memset (p, 0, ctx->fs->blocksize-sizeof(journal_header_t));
1424
1425         jsb->s_blocksize = htonl(ctx->fs->blocksize);
1426         jsb->s_maxlen = htonl(journal->j_total_len);
1427         jsb->s_first = htonl(1);
1428
1429         /* Initialize the journal sequence number so that there is "no"
1430          * chance we will find old "valid" transactions in the journal.
1431          * This avoids the need to zero the whole journal (slow to do,
1432          * and risky when we are just recovering the filesystem).
1433          */
1434         uuid_generate(u.uuid);
1435         for (i = 0; i < 4; i ++)
1436                 new_seq ^= u.val[i];
1437         jsb->s_sequence = htonl(new_seq);
1438         e2fsck_journal_sb_csum_set(journal, jsb);
1439
1440         mark_buffer_dirty(journal->j_sb_buffer);
1441         ll_rw_block(REQ_OP_WRITE, 0, 1, &journal->j_sb_buffer);
1442 }
1443
1444 static errcode_t e2fsck_journal_fix_corrupt_super(e2fsck_t ctx,
1445                                                   journal_t *journal,
1446                                                   struct problem_context *pctx)
1447 {
1448         struct ext2_super_block *sb = ctx->fs->super;
1449         int recover = ext2fs_has_feature_journal_needs_recovery(ctx->fs->super);
1450
1451         if (ext2fs_has_feature_journal(sb)) {
1452                 if (fix_problem(ctx, PR_0_JOURNAL_BAD_SUPER, pctx)) {
1453                         e2fsck_journal_reset_super(ctx, journal->j_superblock,
1454                                                    journal);
1455                         journal->j_transaction_sequence = 1;
1456                         e2fsck_clear_recover(ctx, recover);
1457                         return 0;
1458                 }
1459                 return EXT2_ET_CORRUPT_JOURNAL_SB;
1460         } else if (e2fsck_journal_fix_bad_inode(ctx, pctx))
1461                 return EXT2_ET_CORRUPT_JOURNAL_SB;
1462
1463         return 0;
1464 }
1465
1466 static void e2fsck_journal_release(e2fsck_t ctx, journal_t *journal,
1467                                    int reset, int drop)
1468 {
1469         journal_superblock_t *jsb;
1470
1471         if (drop)
1472                 mark_buffer_clean(journal->j_sb_buffer);
1473         else if (!(ctx->options & E2F_OPT_READONLY)) {
1474                 jsb = journal->j_superblock;
1475                 jsb->s_sequence = htonl(journal->j_tail_sequence);
1476                 if (reset)
1477                         jsb->s_start = 0; /* this marks the journal as empty */
1478                 e2fsck_journal_sb_csum_set(journal, jsb);
1479                 mark_buffer_dirty(journal->j_sb_buffer);
1480         }
1481         brelse(journal->j_sb_buffer);
1482
1483         if (ctx->journal_io) {
1484                 if (ctx->fs && ctx->fs->io != ctx->journal_io)
1485                         io_channel_close(ctx->journal_io);
1486                 ctx->journal_io = 0;
1487         }
1488
1489 #ifndef USE_INODE_IO
1490         if (journal->j_inode)
1491                 ext2fs_free_mem(&journal->j_inode);
1492 #endif
1493         if (journal->j_fs_dev)
1494                 ext2fs_free_mem(&journal->j_fs_dev);
1495         ext2fs_free_mem(&journal);
1496 }
1497
1498 /*
1499  * This function makes sure that the superblock fields regarding the
1500  * journal are consistent.
1501  */
1502 errcode_t e2fsck_check_ext3_journal(e2fsck_t ctx)
1503 {
1504         struct ext2_super_block *sb = ctx->fs->super;
1505         journal_t *journal;
1506         int recover = ext2fs_has_feature_journal_needs_recovery(ctx->fs->super);
1507         struct problem_context pctx;
1508         problem_t problem;
1509         int reset = 0, force_fsck = 0;
1510         errcode_t retval;
1511
1512         /* If we don't have any journal features, don't do anything more */
1513         if (!ext2fs_has_feature_journal(sb) &&
1514             !recover && sb->s_journal_inum == 0 && sb->s_journal_dev == 0 &&
1515             uuid_is_null(sb->s_journal_uuid))
1516                 return 0;
1517
1518         clear_problem_context(&pctx);
1519         pctx.num = sb->s_journal_inum;
1520
1521         retval = e2fsck_get_journal(ctx, &journal);
1522         if (retval) {
1523                 if ((retval == EXT2_ET_BAD_INODE_NUM) ||
1524                     (retval == EXT2_ET_BAD_BLOCK_NUM) ||
1525                     (retval == EXT2_ET_JOURNAL_TOO_SMALL) ||
1526                     (retval == EXT2_ET_NO_JOURNAL))
1527                         return e2fsck_journal_fix_bad_inode(ctx, &pctx);
1528                 return retval;
1529         }
1530
1531         retval = e2fsck_journal_load(journal);
1532         if (retval) {
1533                 if ((retval == EXT2_ET_CORRUPT_JOURNAL_SB) ||
1534                     ((retval == EXT2_ET_UNSUPP_FEATURE) &&
1535                     (!fix_problem(ctx, PR_0_JOURNAL_UNSUPP_INCOMPAT,
1536                                   &pctx))) ||
1537                     ((retval == EXT2_ET_RO_UNSUPP_FEATURE) &&
1538                     (!fix_problem(ctx, PR_0_JOURNAL_UNSUPP_ROCOMPAT,
1539                                   &pctx))) ||
1540                     ((retval == EXT2_ET_JOURNAL_UNSUPP_VERSION) &&
1541                     (!fix_problem(ctx, PR_0_JOURNAL_UNSUPP_VERSION, &pctx))))
1542                         retval = e2fsck_journal_fix_corrupt_super(ctx, journal,
1543                                                                   &pctx);
1544                 e2fsck_journal_release(ctx, journal, 0, 1);
1545                 return retval;
1546         }
1547
1548         /*
1549          * We want to make the flags consistent here.  We will not leave with
1550          * needs_recovery set but has_journal clear.  We can't get in a loop
1551          * with -y, -n, or -p, only if a user isn't making up their mind.
1552          */
1553 no_has_journal:
1554         if (!ext2fs_has_feature_journal(sb)) {
1555                 recover = ext2fs_has_feature_journal_needs_recovery(sb);
1556                 if (fix_problem(ctx, PR_0_JOURNAL_HAS_JOURNAL, &pctx)) {
1557                         if (recover &&
1558                             !fix_problem(ctx, PR_0_JOURNAL_RECOVER_SET, &pctx))
1559                                 goto no_has_journal;
1560                         /*
1561                          * Need a full fsck if we are releasing a
1562                          * journal stored on a reserved inode.
1563                          */
1564                         force_fsck = recover ||
1565                                 (sb->s_journal_inum < EXT2_FIRST_INODE(sb));
1566                         /* Clear all of the journal fields */
1567                         sb->s_journal_inum = 0;
1568                         sb->s_journal_dev = 0;
1569                         memset(sb->s_journal_uuid, 0,
1570                                sizeof(sb->s_journal_uuid));
1571                         e2fsck_clear_recover(ctx, force_fsck);
1572                 } else if (!(ctx->options & E2F_OPT_READONLY)) {
1573                         ext2fs_set_feature_journal(sb);
1574                         ctx->fs->flags &= ~EXT2_FLAG_MASTER_SB_ONLY;
1575                         ext2fs_mark_super_dirty(ctx->fs);
1576                 }
1577         }
1578
1579         if (ext2fs_has_feature_journal(sb) &&
1580             !ext2fs_has_feature_journal_needs_recovery(sb) &&
1581             journal->j_superblock->s_start != 0) {
1582                 /* Print status information */
1583                 fix_problem(ctx, PR_0_JOURNAL_RECOVERY_CLEAR, &pctx);
1584                 if (ctx->superblock)
1585                         problem = PR_0_JOURNAL_RUN_DEFAULT;
1586                 else
1587                         problem = PR_0_JOURNAL_RUN;
1588                 if (fix_problem(ctx, problem, &pctx)) {
1589                         ctx->options |= E2F_OPT_FORCE;
1590                         ext2fs_set_feature_journal_needs_recovery(sb);
1591                         ext2fs_mark_super_dirty(ctx->fs);
1592                 } else if (fix_problem(ctx,
1593                                        PR_0_JOURNAL_RESET_JOURNAL, &pctx)) {
1594                         reset = 1;
1595                         sb->s_state &= ~EXT2_VALID_FS;
1596                         ext2fs_mark_super_dirty(ctx->fs);
1597                 }
1598                 /*
1599                  * If the user answers no to the above question, we
1600                  * ignore the fact that journal apparently has data;
1601                  * accidentally replaying over valid data would be far
1602                  * worse than skipping a questionable recovery.
1603                  *
1604                  * XXX should we abort with a fatal error here?  What
1605                  * will the ext3 kernel code do if a filesystem with
1606                  * !NEEDS_RECOVERY but with a non-zero
1607                  * journal->j_superblock->s_start is mounted?
1608                  */
1609         }
1610
1611         /*
1612          * If we don't need to do replay the journal, check to see if
1613          * the journal's errno is set; if so, we need to mark the file
1614          * system as being corrupt and clear the journal's s_errno.
1615          */
1616         if (!ext2fs_has_feature_journal_needs_recovery(sb) &&
1617             journal->j_superblock->s_errno) {
1618                 ctx->fs->super->s_state |= EXT2_ERROR_FS;
1619                 ext2fs_mark_super_dirty(ctx->fs);
1620                 journal->j_superblock->s_errno = 0;
1621                 e2fsck_journal_sb_csum_set(journal, journal->j_superblock);
1622                 mark_buffer_dirty(journal->j_sb_buffer);
1623         }
1624
1625         e2fsck_journal_release(ctx, journal, reset, 0);
1626         return retval;
1627 }
1628
1629 static errcode_t recover_ext3_journal(e2fsck_t ctx)
1630 {
1631         struct problem_context  pctx;
1632         journal_t *journal;
1633         errcode_t retval;
1634
1635         clear_problem_context(&pctx);
1636
1637         retval = jbd2_journal_init_revoke_record_cache();
1638         if (retval)
1639                 return retval;
1640
1641         retval = jbd2_journal_init_revoke_table_cache();
1642         if (retval)
1643                 return retval;
1644
1645         retval = e2fsck_get_journal(ctx, &journal);
1646         if (retval)
1647                 return retval;
1648
1649         retval = e2fsck_journal_load(journal);
1650         if (retval)
1651                 goto errout;
1652
1653         retval = jbd2_journal_init_revoke(journal, 1024);
1654         if (retval)
1655                 goto errout;
1656
1657         retval = -jbd2_journal_recover(journal);
1658         if (retval)
1659                 goto errout;
1660
1661         if (journal->j_failed_commit) {
1662                 pctx.ino = journal->j_failed_commit;
1663                 fix_problem(ctx, PR_0_JNL_TXN_CORRUPT, &pctx);
1664                 journal->j_superblock->s_errno = -EINVAL;
1665                 mark_buffer_dirty(journal->j_sb_buffer);
1666         }
1667
1668         journal->j_tail_sequence = journal->j_transaction_sequence;
1669
1670 errout:
1671         jbd2_journal_destroy_revoke(journal);
1672         jbd2_journal_destroy_revoke_record_cache();
1673         jbd2_journal_destroy_revoke_table_cache();
1674         e2fsck_journal_release(ctx, journal, 1, 0);
1675         return retval;
1676 }
1677
1678 errcode_t e2fsck_run_ext3_journal(e2fsck_t ctx)
1679 {
1680         io_manager io_ptr = ctx->fs->io->manager;
1681         int blocksize = ctx->fs->blocksize;
1682         errcode_t       retval, recover_retval;
1683         io_stats        stats = 0;
1684         unsigned long long kbytes_written = 0;
1685
1686         printf(_("%s: recovering journal\n"), ctx->device_name);
1687         if (ctx->options & E2F_OPT_READONLY) {
1688                 printf(_("%s: won't do journal recovery while read-only\n"),
1689                        ctx->device_name);
1690                 return EXT2_ET_FILE_RO;
1691         }
1692
1693         if (ctx->fs->flags & EXT2_FLAG_DIRTY)
1694                 ext2fs_flush(ctx->fs);  /* Force out any modifications */
1695
1696         recover_retval = recover_ext3_journal(ctx);
1697
1698         /*
1699          * Reload the filesystem context to get up-to-date data from disk
1700          * because journal recovery will change the filesystem under us.
1701          */
1702         if (ctx->fs->super->s_kbytes_written &&
1703             ctx->fs->io->manager->get_stats)
1704                 ctx->fs->io->manager->get_stats(ctx->fs->io, &stats);
1705         if (stats && stats->bytes_written)
1706                 kbytes_written = stats->bytes_written >> 10;
1707
1708         ext2fs_mmp_stop(ctx->fs);
1709         ext2fs_free(ctx->fs);
1710         retval = ext2fs_open(ctx->filesystem_name, ctx->openfs_flags,
1711                              ctx->superblock, blocksize, io_ptr,
1712                              &ctx->fs);
1713         if (retval) {
1714                 com_err(ctx->program_name, retval,
1715                         _("while trying to re-open %s"),
1716                         ctx->device_name);
1717                 fatal_error(ctx, 0);
1718         }
1719         ctx->fs->priv_data = ctx;
1720         ctx->fs->now = ctx->now;
1721         ctx->fs->flags |= EXT2_FLAG_MASTER_SB_ONLY;
1722         ctx->fs->super->s_kbytes_written += kbytes_written;
1723
1724         /* Set the superblock flags */
1725         e2fsck_clear_recover(ctx, recover_retval != 0);
1726
1727         /*
1728          * Do one last sanity check, and propagate journal->s_errno to
1729          * the EXT2_ERROR_FS flag in the fs superblock if needed.
1730          */
1731         retval = e2fsck_check_ext3_journal(ctx);
1732         return retval ? retval : recover_retval;
1733 }
1734
1735 /*
1736  * This function will move the journal inode from a visible file in
1737  * the filesystem directory hierarchy to the reserved inode if necessary.
1738  */
1739 static const char * const journal_names[] = {
1740         ".journal", "journal", ".journal.dat", "journal.dat", 0 };
1741
1742 void e2fsck_move_ext3_journal(e2fsck_t ctx)
1743 {
1744         struct ext2_super_block *sb = ctx->fs->super;
1745         struct problem_context  pctx;
1746         struct ext2_inode       inode;
1747         ext2_filsys             fs = ctx->fs;
1748         ext2_ino_t              ino;
1749         errcode_t               retval;
1750         const char * const *    cpp;
1751         dgrp_t                  group;
1752         int                     mount_flags;
1753
1754         clear_problem_context(&pctx);
1755
1756         /*
1757          * If the filesystem is opened read-only, or there is no
1758          * journal, then do nothing.
1759          */
1760         if ((ctx->options & E2F_OPT_READONLY) ||
1761             (sb->s_journal_inum == 0) ||
1762             !ext2fs_has_feature_journal(sb))
1763                 return;
1764
1765         /*
1766          * Read in the journal inode
1767          */
1768         if (ext2fs_read_inode(fs, sb->s_journal_inum, &inode) != 0)
1769                 return;
1770
1771         /*
1772          * If it's necessary to backup the journal inode, do so.
1773          */
1774         if ((sb->s_jnl_backup_type == 0) ||
1775             ((sb->s_jnl_backup_type == EXT3_JNL_BACKUP_BLOCKS) &&
1776              memcmp(inode.i_block, sb->s_jnl_blocks, EXT2_N_BLOCKS*4))) {
1777                 if (fix_problem(ctx, PR_0_BACKUP_JNL, &pctx)) {
1778                         memcpy(sb->s_jnl_blocks, inode.i_block,
1779                                EXT2_N_BLOCKS*4);
1780                         sb->s_jnl_blocks[15] = inode.i_size_high;
1781                         sb->s_jnl_blocks[16] = inode.i_size;
1782                         sb->s_jnl_backup_type = EXT3_JNL_BACKUP_BLOCKS;
1783                         ext2fs_mark_super_dirty(fs);
1784                         fs->flags &= ~EXT2_FLAG_MASTER_SB_ONLY;
1785                 }
1786         }
1787
1788         /*
1789          * If the journal is already the hidden inode, then do nothing
1790          */
1791         if (sb->s_journal_inum == EXT2_JOURNAL_INO)
1792                 return;
1793
1794         /*
1795          * The journal inode had better have only one link and not be readable.
1796          */
1797         if (inode.i_links_count != 1)
1798                 return;
1799
1800         /*
1801          * If the filesystem is mounted, or we can't tell whether
1802          * or not it's mounted, do nothing.
1803          */
1804         retval = ext2fs_check_if_mounted(ctx->filesystem_name, &mount_flags);
1805         if (retval || (mount_flags & EXT2_MF_MOUNTED))
1806                 return;
1807
1808         /*
1809          * If we can't find the name of the journal inode, then do
1810          * nothing.
1811          */
1812         for (cpp = journal_names; *cpp; cpp++) {
1813                 retval = ext2fs_lookup(fs, EXT2_ROOT_INO, *cpp,
1814                                        strlen(*cpp), 0, &ino);
1815                 if ((retval == 0) && (ino == sb->s_journal_inum))
1816                         break;
1817         }
1818         if (*cpp == 0)
1819                 return;
1820
1821         /* We need the inode bitmap to be loaded */
1822         retval = ext2fs_read_bitmaps(fs);
1823         if (retval)
1824                 return;
1825
1826         pctx.str = *cpp;
1827         if (!fix_problem(ctx, PR_0_MOVE_JOURNAL, &pctx))
1828                 return;
1829
1830         /*
1831          * OK, we've done all the checks, let's actually move the
1832          * journal inode.  Errors at this point mean we need to force
1833          * an ext2 filesystem check.
1834          */
1835         if ((retval = ext2fs_unlink(fs, EXT2_ROOT_INO, *cpp, ino, 0)) != 0)
1836                 goto err_out;
1837         if ((retval = ext2fs_write_inode(fs, EXT2_JOURNAL_INO, &inode)) != 0)
1838                 goto err_out;
1839         sb->s_journal_inum = EXT2_JOURNAL_INO;
1840         ext2fs_mark_super_dirty(fs);
1841         fs->flags &= ~EXT2_FLAG_MASTER_SB_ONLY;
1842         inode.i_links_count = 0;
1843         inode.i_dtime = ctx->now;
1844         if ((retval = ext2fs_write_inode(fs, ino, &inode)) != 0)
1845                 goto err_out;
1846
1847         group = ext2fs_group_of_ino(fs, ino);
1848         ext2fs_unmark_inode_bitmap2(fs->inode_map, ino);
1849         ext2fs_mark_ib_dirty(fs);
1850         ext2fs_bg_free_inodes_count_set(fs, group, ext2fs_bg_free_inodes_count(fs, group) + 1);
1851         ext2fs_group_desc_csum_set(fs, group);
1852         fs->super->s_free_inodes_count++;
1853         return;
1854
1855 err_out:
1856         pctx.errcode = retval;
1857         fix_problem(ctx, PR_0_ERR_MOVE_JOURNAL, &pctx);
1858         fs->super->s_state &= ~EXT2_VALID_FS;
1859         ext2fs_mark_super_dirty(fs);
1860         return;
1861 }
1862
1863 /*
1864  * This function makes sure the superblock hint for the external
1865  * journal is correct.
1866  */
1867 int e2fsck_fix_ext3_journal_hint(e2fsck_t ctx)
1868 {
1869         struct ext2_super_block *sb = ctx->fs->super;
1870         struct problem_context pctx;
1871         char uuid[37], *journal_name;
1872         struct stat st;
1873
1874         if (!ext2fs_has_feature_journal(sb) ||
1875             uuid_is_null(sb->s_journal_uuid))
1876                 return 0;
1877
1878         uuid_unparse(sb->s_journal_uuid, uuid);
1879         journal_name = blkid_get_devname(ctx->blkid, "UUID", uuid);
1880         if (!journal_name)
1881                 return 0;
1882
1883         if (stat(journal_name, &st) < 0) {
1884                 free(journal_name);
1885                 return 0;
1886         }
1887
1888         if (st.st_rdev != sb->s_journal_dev) {
1889                 clear_problem_context(&pctx);
1890                 pctx.num = st.st_rdev;
1891                 if (fix_problem(ctx, PR_0_EXTERNAL_JOURNAL_HINT, &pctx)) {
1892                         sb->s_journal_dev = st.st_rdev;
1893                         ext2fs_mark_super_dirty(ctx->fs);
1894                 }
1895         }
1896
1897         free(journal_name);
1898         return 0;
1899 }