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         int tag = le16_to_cpu(tl->fc_tag);
 624
 625         memcpy(&fcd, val, sizeof(fcd));
 626
 627         darg->parent_ino = le32_to_cpu(fcd.fc_parent_ino);
 628         darg->ino = le32_to_cpu(fcd.fc_ino);
 629         darg->dname_len = ext4_fc_tag_len(tl) -
 630                         sizeof(struct ext4_fc_dentry_info);
 631         darg->dname = malloc(darg->dname_len + 1);
 632         if (!darg->dname)
 633                 return -ENOMEM;
 634         memcpy(darg->dname,
 635                val + sizeof(struct ext4_fc_dentry_info),
 636                darg->dname_len);
 637         darg->dname[darg->dname_len] = 0;
 638         jbd_debug(1, "%s: %s, ino %lu, parent %lu\n",
 639                 tag == EXT4_FC_TAG_CREAT ? "create" :
 640                 (tag == EXT4_FC_TAG_LINK ? "link" :
 641                 (tag == EXT4_FC_TAG_UNLINK ? "unlink" : "error")),
 642                 darg->dname, darg->ino, darg->parent_ino);
 643         return 0;
 644 }
 645
 646 static int ext4_fc_handle_unlink(e2fsck_t ctx, struct ext4_fc_tl *tl, __u8 *val)
 647 {
 648         struct dentry_info_args darg;
 649         int ret;
 650
 651         ret = tl_to_darg(&darg, tl, val);
 652         if (ret)
 653                 return ret;
 654         ext4_fc_flush_extents(ctx, darg.ino);
 655         ret = errcode_to_errno(
 656                        ext2fs_unlink(ctx->fs, darg.parent_ino,
 657                                      darg.dname, darg.ino, 0));
 658         /* It's okay if the above call fails */
 659         free(darg.dname);
 660         return ret;
 661 }
 662
 663 static int ext4_fc_handle_link_and_create(e2fsck_t ctx, struct ext4_fc_tl *tl, __u8 *val)
 664 {
 665         struct dentry_info_args darg;
 666         ext2_filsys fs = ctx->fs;
 667         struct ext2_inode_large inode_large;
 668         int ret, filetype, mode;
 669
 670         ret = tl_to_darg(&darg, tl, val);
 671         if (ret)
 672                 return ret;
 673         ext4_fc_flush_extents(ctx, 0);
 674         ret = errcode_to_errno(ext2fs_read_inode(fs, darg.ino,
 675                                                  (struct ext2_inode *)&inode_large));
 676         if (ret)
 677                 goto out;
 678
 679         mode = inode_large.i_mode;
 680
 681         if (LINUX_S_ISREG(mode))
 682                 filetype = EXT2_FT_REG_FILE;
 683         else if (LINUX_S_ISDIR(mode))
 684                 filetype = EXT2_FT_DIR;
 685         else if (LINUX_S_ISCHR(mode))
 686                 filetype = EXT2_FT_CHRDEV;
 687         else if (LINUX_S_ISBLK(mode))
 688                 filetype = EXT2_FT_BLKDEV;
 689         else if (LINUX_S_ISLNK(mode))
 690                 return EXT2_FT_SYMLINK;
 691         else if (LINUX_S_ISFIFO(mode))
 692                 filetype = EXT2_FT_FIFO;
 693         else if (LINUX_S_ISSOCK(mode))
 694                 filetype = EXT2_FT_SOCK;
 695         else {
 696                 ret = -EINVAL;
 697                 goto out;
 698         }
 699
 700         /*
 701          * Forcefully unlink if the same name is present and ignore the error
 702          * if any, since this dirent might not exist
 703          */
 704         ext2fs_unlink(fs, darg.parent_ino, darg.dname, darg.ino,
 705                         EXT2FS_UNLINK_FORCE);
 706
 707         ret = errcode_to_errno(
 708                        ext2fs_link(fs, darg.parent_ino, darg.dname, darg.ino,
 709                                    filetype));
 710 out:
 711         free(darg.dname);
 712         return ret;
 713
 714 }
 715
 716 /* This function fixes the i_blocks field in the replayed indoe */
 717 static void ext4_fc_replay_fixup_iblocks(struct ext2_inode_large *ondisk_inode,
 718         struct ext2_inode_large *fc_inode)
 719 {
 720         if (ondisk_inode->i_flags & EXT4_EXTENTS_FL) {
 721                 struct ext3_extent_header *eh;
 722
 723                 eh = (struct ext3_extent_header *)(&ondisk_inode->i_block[0]);
 724                 if (le16_to_cpu(eh->eh_magic) != EXT3_EXT_MAGIC) {
 725                         memset(eh, 0, sizeof(*eh));
 726                         eh->eh_magic = cpu_to_le16(EXT3_EXT_MAGIC);
 727                         eh->eh_max = cpu_to_le16(
 728                                 (sizeof(ondisk_inode->i_block) -
 729                                         sizeof(struct ext3_extent_header)) /
 730                                 sizeof(struct ext3_extent));
 731                 }
 732         } else if (ondisk_inode->i_flags & EXT4_INLINE_DATA_FL) {
 733                 memcpy(ondisk_inode->i_block, fc_inode->i_block,
 734                         sizeof(fc_inode->i_block));
 735         }
 736 }
 737
 738 static int ext4_fc_handle_inode(e2fsck_t ctx, __u8 *val)
 739 {
 740         int ino, inode_len = EXT2_GOOD_OLD_INODE_SIZE;
 741         struct ext2_inode_large *inode = NULL, *fc_inode = NULL;
 742         __le32 fc_ino;
 743         __u8 *fc_raw_inode;
 744         errcode_t err;
 745         blk64_t blks;
 746
 747         memcpy(&fc_ino, val, sizeof(fc_ino));
 748         fc_raw_inode = val + sizeof(fc_ino);
 749         ino = le32_to_cpu(fc_ino);
 750
 751         if (EXT2_INODE_SIZE(ctx->fs->super) > EXT2_GOOD_OLD_INODE_SIZE)
 752                 inode_len += ext2fs_le16_to_cpu(
 753                         ((struct ext2_inode_large *)fc_raw_inode)->i_extra_isize);
 754         err = ext2fs_get_mem(inode_len, &inode);
 755         if (err)
 756                 goto out;
 757         err = ext2fs_get_mem(inode_len, &fc_inode);
 758         if (err)
 759                 goto out;
 760         ext4_fc_flush_extents(ctx, ino);
 761
 762         err = ext2fs_read_inode_full(ctx->fs, ino, (struct ext2_inode *)inode,
 763                                         inode_len);
 764         if (err)
 765                 goto out;
 766         memcpy(fc_inode, fc_raw_inode, inode_len);
 767 #ifdef WORDS_BIGENDIAN
 768         ext2fs_swap_inode_full(ctx->fs, fc_inode, fc_inode, 0, inode_len);
 769 #endif
 770         memcpy(inode, fc_inode, offsetof(struct ext2_inode_large, i_block));
 771         memcpy(&inode->i_generation, &fc_inode->i_generation,
 772                 inode_len - offsetof(struct ext2_inode_large, i_generation));
 773         ext4_fc_replay_fixup_iblocks(inode, fc_inode);
 774         err = ext2fs_count_blocks(ctx->fs, ino, EXT2_INODE(inode), &blks);
 775         if (err)
 776                 goto out;
 777         ext2fs_iblk_set(ctx->fs, EXT2_INODE(inode), blks);
 778         ext2fs_inode_csum_set(ctx->fs, ino, inode);
 779
 780         err = ext2fs_write_inode_full(ctx->fs, ino, (struct ext2_inode *)inode,
 781                                         inode_len);
 782         if (err)
 783                 goto out;
 784         if (inode->i_links_count)
 785                 ext2fs_mark_inode_bitmap2(ctx->fs->inode_map, ino);
 786         else
 787                 ext2fs_unmark_inode_bitmap2(ctx->fs->inode_map, ino);
 788         ext2fs_mark_ib_dirty(ctx->fs);
 789
 790 out:
 791         ext2fs_free_mem(&inode);
 792         ext2fs_free_mem(&fc_inode);
 793         return errcode_to_errno(err);
 794 }
 795
 796 /*
 797  * Handle add extent replay tag.
 798  */
 799 static int ext4_fc_handle_add_extent(e2fsck_t ctx, __u8 *val)
 800 {
 801         struct ext2fs_extent extent;
 802         struct ext4_fc_add_range add_range;
 803         ino_t ino;
 804         int ret = 0;
 805
 806         memcpy(&add_range, val, sizeof(add_range));
 807         ino = le32_to_cpu(add_range.fc_ino);
 808         ext4_fc_flush_extents(ctx, ino);
 809
 810         ret = ext4_fc_read_extents(ctx, ino);
 811         if (ret)
 812                 return ret;
 813         memset(&extent, 0, sizeof(extent));
 814         ret = errcode_to_errno(ext2fs_decode_extent(
 815                         &extent, (void *)add_range.fc_ex,
 816                         sizeof(add_range.fc_ex)));
 817         if (ret)
 818                 return ret;
 819         return ext4_add_extent_to_list(ctx,
 820                 &ctx->fc_replay_state.fc_extent_list, &extent);
 821 }
 822
 823 /*
 824  * Handle delete logical range replay tag.
 825  */
 826 static int ext4_fc_handle_del_range(e2fsck_t ctx, __u8 *val)
 827 {
 828         struct ext2fs_extent extent;
 829         struct ext4_fc_del_range del_range;
 830         int ret, ino;
 831
 832         memcpy(&del_range, val, sizeof(del_range));
 833         ino = le32_to_cpu(del_range.fc_ino);
 834         ext4_fc_flush_extents(ctx, ino);
 835
 836         memset(&extent, 0, sizeof(extent));
 837         extent.e_lblk = le32_to_cpu(del_range.fc_lblk);
 838         extent.e_len = le32_to_cpu(del_range.fc_len);
 839         ret = ext4_fc_read_extents(ctx, ino);
 840         if (ret)
 841                 return ret;
 842         return ext4_del_extent_from_list(ctx,
 843                 &ctx->fc_replay_state.fc_extent_list, &extent);
 844 }
 845
 846 /*
 847  * Main recovery path entry point. This function returns JBD2_FC_REPLAY_CONTINUE
 848  * to indicate that it is expecting more fast commit blocks. It returns
 849  * JBD2_FC_REPLAY_STOP to indicate that replay is done.
 850  */
 851 static int ext4_fc_replay(journal_t *journal, struct buffer_head *bh,
 852                                 enum passtype pass, int off, tid_t expected_tid)
 853 {
 854         e2fsck_t ctx = journal->j_fs_dev->k_ctx;
 855         struct e2fsck_fc_replay_state *state = &ctx->fc_replay_state;
 856         int ret = JBD2_FC_REPLAY_CONTINUE;
 857         struct ext4_fc_tl tl;
 858         __u8 *start, *end, *cur, *val;
 859
 860         if (pass == PASS_SCAN) {
 861                 state->fc_current_pass = PASS_SCAN;
 862                 return ext4_fc_replay_scan(journal, bh, off, expected_tid);
 863         }
 864
 865         if (state->fc_replay_num_tags == 0)
 866                 goto replay_done;
 867
 868         if (state->fc_current_pass != pass) {
 869                 /* Starting replay phase */
 870                 state->fc_current_pass = pass;
 871                 /* We will reset checksums */
 872                 ctx->fs->flags |= EXT2_FLAG_IGNORE_CSUM_ERRORS;
 873                 ret = errcode_to_errno(ext2fs_read_bitmaps(ctx->fs));
 874                 if (ret) {
 875                         jbd_debug(1, "Error %d while reading bitmaps\n", ret);
 876                         return ret;
 877                 }
 878                 state->fc_super_state = ctx->fs->super->s_state;
 879                 /*
 880                  * Mark the file system to indicate it contains errors. That's
 881                  * because the updates performed by fast commit replay code are
 882                  * not atomic and may result in incosistent file system if it
 883                  * crashes before the replay is complete.
 884                  */
 885                 ctx->fs->super->s_state |= EXT2_ERROR_FS;
 886                 ctx->fs->super->s_state |= EXT4_FC_REPLAY;
 887                 ext2fs_mark_super_dirty(ctx->fs);
 888                 ext2fs_flush(ctx->fs);
 889         }
 890
 891         start = (__u8 *)bh->b_data;
 892         end = (__u8 *)bh->b_data + journal->j_blocksize - 1;
 893
 894         for (cur = start; cur < end; cur = cur + le16_to_cpu(tl.fc_len) + sizeof(tl)) {
 895                 memcpy(&tl, cur, sizeof(tl));
 896                 val = cur + sizeof(tl);
 897
 898                 if (state->fc_replay_num_tags == 0)
 899                         goto replay_done;
 900                 jbd_debug(3, "Replay phase processing %s tag\n",
 901                                 tag2str(le16_to_cpu(tl.fc_tag)));
 902                 state->fc_replay_num_tags--;
 903                 switch (le16_to_cpu(tl.fc_tag)) {
 904                 case EXT4_FC_TAG_CREAT:
 905                 case EXT4_FC_TAG_LINK:
 906                         ret = ext4_fc_handle_link_and_create(ctx, &tl, val);
 907                         break;
 908                 case EXT4_FC_TAG_UNLINK:
 909                         ret = ext4_fc_handle_unlink(ctx, &tl, val);
 910                         break;
 911                 case EXT4_FC_TAG_ADD_RANGE:
 912                         ret = ext4_fc_handle_add_extent(ctx, val);
 913                         break;
 914                 case EXT4_FC_TAG_DEL_RANGE:
 915                         ret = ext4_fc_handle_del_range(ctx, val);
 916                         break;
 917                 case EXT4_FC_TAG_INODE:
 918                         ret = ext4_fc_handle_inode(ctx, val);
 919                         break;
 920                 case EXT4_FC_TAG_TAIL:
 921                         ext4_fc_flush_extents(ctx, 0);
 922                 case EXT4_FC_TAG_PAD:
 923                 case EXT4_FC_TAG_HEAD:
 924                         break;
 925                 default:
 926                         ret = -ECANCELED;
 927                         break;
 928                 }
 929                 if (ret < 0)
 930                         break;
 931                 ret = JBD2_FC_REPLAY_CONTINUE;
 932         }
 933         return ret;
 934 replay_done:
 935         jbd_debug(1, "End of fast commit replay\n");
 936         if (state->fc_current_pass != pass)
 937                 return JBD2_FC_REPLAY_STOP;
 938
 939         ext2fs_calculate_summary_stats(ctx->fs, 0 /* update bg also */);
 940         ext2fs_write_block_bitmap(ctx->fs);
 941         ext2fs_write_inode_bitmap(ctx->fs);
 942         ext2fs_mark_super_dirty(ctx->fs);
 943         ext2fs_set_gdt_csum(ctx->fs);
 944         ctx->fs->super->s_state = state->fc_super_state;
 945         ext2fs_flush(ctx->fs);
 946
 947         return JBD2_FC_REPLAY_STOP;
 948 }
 949
 950 static errcode_t e2fsck_get_journal(e2fsck_t ctx, journal_t **ret_journal)
 951 {
 952         struct process_block_struct pb;
 953         struct ext2_super_block *sb = ctx->fs->super;
 954         struct ext2_super_block jsuper;
 955         struct problem_context  pctx;
 956         struct buffer_head      *bh;
 957         struct inode            *j_inode = NULL;
 958         struct kdev_s           *dev_fs = NULL, *dev_journal;
 959         const char              *journal_name = 0;
 960         journal_t               *journal = NULL;
 961         errcode_t               retval = 0;
 962         io_manager              io_ptr = 0;
 963         unsigned long long      start = 0;
 964         int                     ret;
 965         int                     ext_journal = 0;
 966         int                     tried_backup_jnl = 0;
 967
 968         clear_problem_context(&pctx);
 969
 970         journal = e2fsck_allocate_memory(ctx, sizeof(journal_t), "journal");
 971         if (!journal) {
 972                 return EXT2_ET_NO_MEMORY;
 973         }
 974
 975         dev_fs = e2fsck_allocate_memory(ctx, 2*sizeof(struct kdev_s), "kdev");
 976         if (!dev_fs) {
 977                 retval = EXT2_ET_NO_MEMORY;
 978                 goto errout;
 979         }
 980         dev_journal = dev_fs+1;
 981
 982         dev_fs->k_ctx = dev_journal->k_ctx = ctx;
 983         dev_fs->k_dev = K_DEV_FS;
 984         dev_journal->k_dev = K_DEV_JOURNAL;
 985
 986         journal->j_dev = dev_journal;
 987         journal->j_fs_dev = dev_fs;
 988         journal->j_inode = NULL;
 989         journal->j_blocksize = ctx->fs->blocksize;
 990
 991         if (uuid_is_null(sb->s_journal_uuid)) {
 992                 /*
 993                  * The full set of superblock sanity checks haven't
 994                  * been performed yet, so we need to do some basic
 995                  * checks here to avoid potential array overruns.
 996                  */
 997                 if (!sb->s_journal_inum ||
 998                     (sb->s_journal_inum >
 999                      (ctx->fs->group_desc_count * sb->s_inodes_per_group))) {
1000                         retval = EXT2_ET_BAD_INODE_NUM;
1001                         goto errout;
1002                 }
1003                 j_inode = e2fsck_allocate_memory(ctx, sizeof(*j_inode),
1004                                                  "journal inode");
1005                 if (!j_inode) {
1006                         retval = EXT2_ET_NO_MEMORY;
1007                         goto errout;
1008                 }
1009
1010                 j_inode->i_ctx = ctx;
1011                 j_inode->i_ino = sb->s_journal_inum;
1012
1013                 if ((retval = ext2fs_read_inode(ctx->fs,
1014                                                 sb->s_journal_inum,
1015                                                 &j_inode->i_ext2))) {
1016                 try_backup_journal:
1017                         if (sb->s_jnl_backup_type != EXT3_JNL_BACKUP_BLOCKS ||
1018                             tried_backup_jnl)
1019                                 goto errout;
1020                         memset(&j_inode->i_ext2, 0, sizeof(struct ext2_inode));
1021                         memcpy(&j_inode->i_ext2.i_block[0], sb->s_jnl_blocks,
1022                                EXT2_N_BLOCKS*4);
1023                         j_inode->i_ext2.i_size_high = sb->s_jnl_blocks[15];
1024                         j_inode->i_ext2.i_size = sb->s_jnl_blocks[16];
1025                         j_inode->i_ext2.i_links_count = 1;
1026                         j_inode->i_ext2.i_mode = LINUX_S_IFREG | 0600;
1027                         e2fsck_use_inode_shortcuts(ctx, 1);
1028                         ctx->stashed_ino = j_inode->i_ino;
1029                         ctx->stashed_inode = &j_inode->i_ext2;
1030                         tried_backup_jnl++;
1031                 }
1032                 if (!j_inode->i_ext2.i_links_count ||
1033                     !LINUX_S_ISREG(j_inode->i_ext2.i_mode)) {
1034                         retval = EXT2_ET_NO_JOURNAL;
1035                         goto try_backup_journal;
1036                 }
1037                 if (EXT2_I_SIZE(&j_inode->i_ext2) / journal->j_blocksize <
1038                     JBD2_MIN_JOURNAL_BLOCKS) {
1039                         retval = EXT2_ET_JOURNAL_TOO_SMALL;
1040                         goto try_backup_journal;
1041                 }
1042                 pb.last_block = -1;
1043                 retval = ext2fs_block_iterate3(ctx->fs, j_inode->i_ino,
1044                                                BLOCK_FLAG_HOLE, 0,
1045                                                process_journal_block, &pb);
1046                 if ((pb.last_block + 1) * ctx->fs->blocksize <
1047                     (int) EXT2_I_SIZE(&j_inode->i_ext2)) {
1048                         retval = EXT2_ET_JOURNAL_TOO_SMALL;
1049                         goto try_backup_journal;
1050                 }
1051                 if (tried_backup_jnl && !(ctx->options & E2F_OPT_READONLY)) {
1052                         retval = ext2fs_write_inode(ctx->fs, sb->s_journal_inum,
1053                                                     &j_inode->i_ext2);
1054                         if (retval)
1055                                 goto errout;
1056                 }
1057
1058                 journal->j_total_len = EXT2_I_SIZE(&j_inode->i_ext2) /
1059                         journal->j_blocksize;
1060
1061 #ifdef USE_INODE_IO
1062                 retval = ext2fs_inode_io_intern2(ctx->fs, sb->s_journal_inum,
1063                                                  &j_inode->i_ext2,
1064                                                  &journal_name);
1065                 if (retval)
1066                         goto errout;
1067
1068                 io_ptr = inode_io_manager;
1069 #else
1070                 journal->j_inode = j_inode;
1071                 ctx->journal_io = ctx->fs->io;
1072                 if ((ret = jbd2_journal_bmap(journal, 0, &start)) != 0) {
1073                         retval = (errcode_t) (-1 * ret);
1074                         goto errout;
1075                 }
1076 #endif
1077         } else {
1078                 ext_journal = 1;
1079                 if (!ctx->journal_name) {
1080                         char uuid[37];
1081
1082                         uuid_unparse(sb->s_journal_uuid, uuid);
1083                         ctx->journal_name = blkid_get_devname(ctx->blkid,
1084                                                               "UUID", uuid);
1085                         if (!ctx->journal_name)
1086                                 ctx->journal_name = blkid_devno_to_devname(sb->s_journal_dev);
1087                 }
1088                 journal_name = ctx->journal_name;
1089
1090                 if (!journal_name) {
1091                         fix_problem(ctx, PR_0_CANT_FIND_JOURNAL, &pctx);
1092                         retval = EXT2_ET_LOAD_EXT_JOURNAL;
1093                         goto errout;
1094                 }
1095
1096                 jfs_debug(1, "Using journal file %s\n", journal_name);
1097                 io_ptr = unix_io_manager;
1098         }
1099
1100 #if 0
1101         test_io_backing_manager = io_ptr;
1102         io_ptr = test_io_manager;
1103 #endif
1104 #ifndef USE_INODE_IO
1105         if (ext_journal)
1106 #endif
1107         {
1108                 int flags = IO_FLAG_RW;
1109                 if (!(ctx->mount_flags & EXT2_MF_ISROOT &&
1110                       ctx->mount_flags & EXT2_MF_READONLY))
1111                         flags |= IO_FLAG_EXCLUSIVE;
1112                 if ((ctx->mount_flags & EXT2_MF_READONLY) &&
1113                     (ctx->options & E2F_OPT_FORCE))
1114                         flags &= ~IO_FLAG_EXCLUSIVE;
1115
1116
1117                 retval = io_ptr->open(journal_name, flags,
1118                                       &ctx->journal_io);
1119         }
1120         if (retval)
1121                 goto errout;
1122
1123         io_channel_set_blksize(ctx->journal_io, ctx->fs->blocksize);
1124
1125         if (ext_journal) {
1126                 blk64_t maxlen;
1127
1128                 start = ext2fs_journal_sb_start(ctx->fs->blocksize) - 1;
1129                 bh = getblk(dev_journal, start, ctx->fs->blocksize);
1130                 if (!bh) {
1131                         retval = EXT2_ET_NO_MEMORY;
1132                         goto errout;
1133                 }
1134                 ll_rw_block(REQ_OP_READ, 0, 1, &bh);
1135                 if ((retval = bh->b_err) != 0) {
1136                         brelse(bh);
1137                         goto errout;
1138                 }
1139                 memcpy(&jsuper, start ? bh->b_data :  bh->b_data + SUPERBLOCK_OFFSET,
1140                        sizeof(jsuper));
1141 #ifdef WORDS_BIGENDIAN
1142                 if (jsuper.s_magic == ext2fs_swab16(EXT2_SUPER_MAGIC))
1143                         ext2fs_swap_super(&jsuper);
1144 #endif
1145                 if (jsuper.s_magic != EXT2_SUPER_MAGIC ||
1146                     !ext2fs_has_feature_journal_dev(&jsuper)) {
1147                         fix_problem(ctx, PR_0_EXT_JOURNAL_BAD_SUPER, &pctx);
1148                         retval = EXT2_ET_LOAD_EXT_JOURNAL;
1149                         brelse(bh);
1150                         goto errout;
1151                 }
1152                 /* Make sure the journal UUID is correct */
1153                 if (memcmp(jsuper.s_uuid, ctx->fs->super->s_journal_uuid,
1154                            sizeof(jsuper.s_uuid))) {
1155                         fix_problem(ctx, PR_0_JOURNAL_BAD_UUID, &pctx);
1156                         retval = EXT2_ET_LOAD_EXT_JOURNAL;
1157                         brelse(bh);
1158                         goto errout;
1159                 }
1160
1161                 /* Check the superblock checksum */
1162                 if (ext2fs_has_feature_metadata_csum(&jsuper)) {
1163                         struct struct_ext2_filsys fsx;
1164                         struct ext2_super_block superx;
1165                         void *p;
1166
1167                         p = start ? bh->b_data : bh->b_data + SUPERBLOCK_OFFSET;
1168                         memcpy(&fsx, ctx->fs, sizeof(fsx));
1169                         memcpy(&superx, ctx->fs->super, sizeof(superx));
1170                         fsx.super = &superx;
1171                         ext2fs_set_feature_metadata_csum(fsx.super);
1172                         if (!ext2fs_superblock_csum_verify(&fsx, p) &&
1173                             fix_problem(ctx, PR_0_EXT_JOURNAL_SUPER_CSUM_INVALID,
1174                                         &pctx)) {
1175                                 ext2fs_superblock_csum_set(&fsx, p);
1176                                 mark_buffer_dirty(bh);
1177                         }
1178                 }
1179                 brelse(bh);
1180
1181                 maxlen = ext2fs_blocks_count(&jsuper);
1182                 journal->j_total_len = (maxlen < 1ULL << 32) ? maxlen : (1ULL << 32) - 1;
1183                 start++;
1184         }
1185
1186         if (!(bh = getblk(dev_journal, start, journal->j_blocksize))) {
1187                 retval = EXT2_ET_NO_MEMORY;
1188                 goto errout;
1189         }
1190
1191         journal->j_sb_buffer = bh;
1192         journal->j_superblock = (journal_superblock_t *)bh->b_data;
1193         if (ext2fs_has_feature_fast_commit(ctx->fs->super))
1194                 journal->j_fc_replay_callback = ext4_fc_replay;
1195         else
1196                 journal->j_fc_replay_callback = NULL;
1197
1198 #ifdef USE_INODE_IO
1199         if (j_inode)
1200                 ext2fs_free_mem(&j_inode);
1201 #endif
1202
1203         *ret_journal = journal;
1204         e2fsck_use_inode_shortcuts(ctx, 0);
1205         return 0;
1206
1207 errout:
1208         e2fsck_use_inode_shortcuts(ctx, 0);
1209         if (dev_fs)
1210                 ext2fs_free_mem(&dev_fs);
1211         if (j_inode)
1212                 ext2fs_free_mem(&j_inode);
1213         if (journal)
1214                 ext2fs_free_mem(&journal);
1215         return retval;
1216 }
1217
1218 static errcode_t e2fsck_journal_fix_bad_inode(e2fsck_t ctx,
1219                                               struct problem_context *pctx)
1220 {
1221         struct ext2_super_block *sb = ctx->fs->super;
1222         int recover = ext2fs_has_feature_journal_needs_recovery(ctx->fs->super);
1223         int has_journal = ext2fs_has_feature_journal(ctx->fs->super);
1224
1225         if (has_journal || sb->s_journal_inum) {
1226                 /* The journal inode is bogus, remove and force full fsck */
1227                 pctx->ino = sb->s_journal_inum;
1228                 if (fix_problem(ctx, PR_0_JOURNAL_BAD_INODE, pctx)) {
1229                         if (has_journal && sb->s_journal_inum)
1230                                 printf("*** journal has been deleted ***\n\n");
1231                         ext2fs_clear_feature_journal(sb);
1232                         sb->s_journal_inum = 0;
1233                         memset(sb->s_jnl_blocks, 0, sizeof(sb->s_jnl_blocks));
1234                         ctx->flags |= E2F_FLAG_JOURNAL_INODE;
1235                         ctx->fs->flags &= ~EXT2_FLAG_MASTER_SB_ONLY;
1236                         e2fsck_clear_recover(ctx, 1);
1237                         return 0;
1238                 }
1239                 return EXT2_ET_CORRUPT_JOURNAL_SB;
1240         } else if (recover) {
1241                 if (fix_problem(ctx, PR_0_JOURNAL_RECOVER_SET, pctx)) {
1242                         e2fsck_clear_recover(ctx, 1);
1243                         return 0;
1244                 }
1245                 return EXT2_ET_UNSUPP_FEATURE;
1246         }
1247         return 0;
1248 }
1249
1250 #define V1_SB_SIZE      0x0024
1251 static void clear_v2_journal_fields(journal_t *journal)
1252 {
1253         e2fsck_t ctx = journal->j_dev->k_ctx;
1254         struct problem_context pctx;
1255
1256         clear_problem_context(&pctx);
1257
1258         if (!fix_problem(ctx, PR_0_CLEAR_V2_JOURNAL, &pctx))
1259                 return;
1260
1261         ctx->flags |= E2F_FLAG_PROBLEMS_FIXED;
1262         memset(((char *) journal->j_superblock) + V1_SB_SIZE, 0,
1263                ctx->fs->blocksize-V1_SB_SIZE);
1264         mark_buffer_dirty(journal->j_sb_buffer);
1265 }
1266
1267
1268 static errcode_t e2fsck_journal_load(journal_t *journal)
1269 {
1270         e2fsck_t ctx = journal->j_dev->k_ctx;
1271         journal_superblock_t *jsb;
1272         struct buffer_head *jbh = journal->j_sb_buffer;
1273         struct problem_context pctx;
1274
1275         clear_problem_context(&pctx);
1276
1277         ll_rw_block(REQ_OP_READ, 0, 1, &jbh);
1278         if (jbh->b_err) {
1279                 com_err(ctx->device_name, jbh->b_err, "%s",
1280                         _("reading journal superblock\n"));
1281                 return jbh->b_err;
1282         }
1283
1284         jsb = journal->j_superblock;
1285         /* If we don't even have JBD2_MAGIC, we probably have a wrong inode */
1286         if (jsb->s_header.h_magic != htonl(JBD2_MAGIC_NUMBER))
1287                 return e2fsck_journal_fix_bad_inode(ctx, &pctx);
1288
1289         switch (ntohl(jsb->s_header.h_blocktype)) {
1290         case JBD2_SUPERBLOCK_V1:
1291                 journal->j_format_version = 1;
1292                 if (jsb->s_feature_compat ||
1293                     jsb->s_feature_incompat ||
1294                     jsb->s_feature_ro_compat ||
1295                     jsb->s_nr_users)
1296                         clear_v2_journal_fields(journal);
1297                 break;
1298
1299         case JBD2_SUPERBLOCK_V2:
1300                 journal->j_format_version = 2;
1301                 if (ntohl(jsb->s_nr_users) > 1 &&
1302                     uuid_is_null(ctx->fs->super->s_journal_uuid))
1303                         clear_v2_journal_fields(journal);
1304                 if (ntohl(jsb->s_nr_users) > 1) {
1305                         fix_problem(ctx, PR_0_JOURNAL_UNSUPP_MULTIFS, &pctx);
1306                         return EXT2_ET_JOURNAL_UNSUPP_VERSION;
1307                 }
1308                 break;
1309
1310         /*
1311          * These should never appear in a journal super block, so if
1312          * they do, the journal is badly corrupted.
1313          */
1314         case JBD2_DESCRIPTOR_BLOCK:
1315         case JBD2_COMMIT_BLOCK:
1316         case JBD2_REVOKE_BLOCK:
1317                 return EXT2_ET_CORRUPT_JOURNAL_SB;
1318
1319         /* If we don't understand the superblock major type, but there
1320          * is a magic number, then it is likely to be a new format we
1321          * just don't understand, so leave it alone. */
1322         default:
1323                 return EXT2_ET_JOURNAL_UNSUPP_VERSION;
1324         }
1325
1326         if (JBD2_HAS_INCOMPAT_FEATURE(journal, ~JBD2_KNOWN_INCOMPAT_FEATURES))
1327                 return EXT2_ET_UNSUPP_FEATURE;
1328
1329         if (JBD2_HAS_RO_COMPAT_FEATURE(journal, ~JBD2_KNOWN_ROCOMPAT_FEATURES))
1330                 return EXT2_ET_RO_UNSUPP_FEATURE;
1331
1332         /* Checksum v1-3 are mutually exclusive features. */
1333         if (jbd2_has_feature_csum2(journal) && jbd2_has_feature_csum3(journal))
1334                 return EXT2_ET_CORRUPT_JOURNAL_SB;
1335
1336         if (jbd2_journal_has_csum_v2or3(journal) &&
1337             jbd2_has_feature_checksum(journal))
1338                 return EXT2_ET_CORRUPT_JOURNAL_SB;
1339
1340         if (!e2fsck_journal_verify_csum_type(journal, jsb) ||
1341             !e2fsck_journal_sb_csum_verify(journal, jsb))
1342                 return EXT2_ET_CORRUPT_JOURNAL_SB;
1343
1344         if (jbd2_journal_has_csum_v2or3(journal))
1345                 journal->j_csum_seed = jbd2_chksum(journal, ~0, jsb->s_uuid,
1346                                                    sizeof(jsb->s_uuid));
1347
1348         /* We have now checked whether we know enough about the journal
1349          * format to be able to proceed safely, so any other checks that
1350          * fail we should attempt to recover from. */
1351         if (jsb->s_blocksize != htonl(journal->j_blocksize)) {
1352                 com_err(ctx->program_name, EXT2_ET_CORRUPT_JOURNAL_SB,
1353                         _("%s: no valid journal superblock found\n"),
1354                         ctx->device_name);
1355                 return EXT2_ET_CORRUPT_JOURNAL_SB;
1356         }
1357
1358         if (ntohl(jsb->s_maxlen) < journal->j_total_len)
1359                 journal->j_total_len = ntohl(jsb->s_maxlen);
1360         else if (ntohl(jsb->s_maxlen) > journal->j_total_len) {
1361                 com_err(ctx->program_name, EXT2_ET_CORRUPT_JOURNAL_SB,
1362                         _("%s: journal too short\n"),
1363                         ctx->device_name);
1364                 return EXT2_ET_CORRUPT_JOURNAL_SB;
1365         }
1366
1367         journal->j_tail_sequence = ntohl(jsb->s_sequence);
1368         journal->j_transaction_sequence = journal->j_tail_sequence;
1369         journal->j_tail = ntohl(jsb->s_start);
1370         journal->j_first = ntohl(jsb->s_first);
1371         if (jbd2_has_feature_fast_commit(journal)) {
1372                 if (ntohl(jsb->s_maxlen) - jbd2_journal_get_num_fc_blks(jsb)
1373                         < JBD2_MIN_JOURNAL_BLOCKS) {
1374                         com_err(ctx->program_name, EXT2_ET_CORRUPT_JOURNAL_SB,
1375                                 _("%s: incorrect fast commit blocks\n"),
1376                                 ctx->device_name);
1377                         return EXT2_ET_CORRUPT_JOURNAL_SB;
1378                 }
1379                 journal->j_fc_last = ntohl(jsb->s_maxlen);
1380                 journal->j_last = journal->j_fc_last -
1381                                         jbd2_journal_get_num_fc_blks(jsb);
1382                 journal->j_fc_first = journal->j_last + 1;
1383         } else {
1384                 journal->j_last = ntohl(jsb->s_maxlen);
1385         }
1386
1387         return 0;
1388 }
1389
1390 static void e2fsck_journal_reset_super(e2fsck_t ctx, journal_superblock_t *jsb,
1391                                        journal_t *journal)
1392 {
1393         char *p;
1394         union {
1395                 uuid_t uuid;
1396                 __u32 val[4];
1397         } u;
1398         __u32 new_seq = 0;
1399         int i;
1400
1401         /* Leave a valid existing V1 superblock signature alone.
1402          * Anything unrecognisable we overwrite with a new V2
1403          * signature. */
1404
1405         if (jsb->s_header.h_magic != htonl(JBD2_MAGIC_NUMBER) ||
1406             jsb->s_header.h_blocktype != htonl(JBD2_SUPERBLOCK_V1)) {
1407                 jsb->s_header.h_magic = htonl(JBD2_MAGIC_NUMBER);
1408                 jsb->s_header.h_blocktype = htonl(JBD2_SUPERBLOCK_V2);
1409         }
1410
1411         /* Zero out everything else beyond the superblock header */
1412
1413         p = ((char *) jsb) + sizeof(journal_header_t);
1414         memset (p, 0, ctx->fs->blocksize-sizeof(journal_header_t));
1415
1416         jsb->s_blocksize = htonl(ctx->fs->blocksize);
1417         jsb->s_maxlen = htonl(journal->j_total_len);
1418         jsb->s_first = htonl(1);
1419
1420         /* Initialize the journal sequence number so that there is "no"
1421          * chance we will find old "valid" transactions in the journal.
1422          * This avoids the need to zero the whole journal (slow to do,
1423          * and risky when we are just recovering the filesystem).
1424          */
1425         uuid_generate(u.uuid);
1426         for (i = 0; i < 4; i ++)
1427                 new_seq ^= u.val[i];
1428         jsb->s_sequence = htonl(new_seq);
1429         e2fsck_journal_sb_csum_set(journal, jsb);
1430
1431         mark_buffer_dirty(journal->j_sb_buffer);
1432         ll_rw_block(REQ_OP_WRITE, 0, 1, &journal->j_sb_buffer);
1433 }
1434
1435 static errcode_t e2fsck_journal_fix_corrupt_super(e2fsck_t ctx,
1436                                                   journal_t *journal,
1437                                                   struct problem_context *pctx)
1438 {
1439         struct ext2_super_block *sb = ctx->fs->super;
1440         int recover = ext2fs_has_feature_journal_needs_recovery(ctx->fs->super);
1441
1442         if (ext2fs_has_feature_journal(sb)) {
1443                 if (fix_problem(ctx, PR_0_JOURNAL_BAD_SUPER, pctx)) {
1444                         e2fsck_journal_reset_super(ctx, journal->j_superblock,
1445                                                    journal);
1446                         journal->j_transaction_sequence = 1;
1447                         e2fsck_clear_recover(ctx, recover);
1448                         return 0;
1449                 }
1450                 return EXT2_ET_CORRUPT_JOURNAL_SB;
1451         } else if (e2fsck_journal_fix_bad_inode(ctx, pctx))
1452                 return EXT2_ET_CORRUPT_JOURNAL_SB;
1453
1454         return 0;
1455 }
1456
1457 static void e2fsck_journal_release(e2fsck_t ctx, journal_t *journal,
1458                                    int reset, int drop)
1459 {
1460         journal_superblock_t *jsb;
1461
1462         if (drop)
1463                 mark_buffer_clean(journal->j_sb_buffer);
1464         else if (!(ctx->options & E2F_OPT_READONLY)) {
1465                 jsb = journal->j_superblock;
1466                 jsb->s_sequence = htonl(journal->j_tail_sequence);
1467                 if (reset)
1468                         jsb->s_start = 0; /* this marks the journal as empty */
1469                 e2fsck_journal_sb_csum_set(journal, jsb);
1470                 mark_buffer_dirty(journal->j_sb_buffer);
1471         }
1472         brelse(journal->j_sb_buffer);
1473
1474         if (ctx->journal_io) {
1475                 if (ctx->fs && ctx->fs->io != ctx->journal_io)
1476                         io_channel_close(ctx->journal_io);
1477                 ctx->journal_io = 0;
1478         }
1479
1480 #ifndef USE_INODE_IO
1481         if (journal->j_inode)
1482                 ext2fs_free_mem(&journal->j_inode);
1483 #endif
1484         if (journal->j_fs_dev)
1485                 ext2fs_free_mem(&journal->j_fs_dev);
1486         ext2fs_free_mem(&journal);
1487 }
1488
1489 /*
1490  * This function makes sure that the superblock fields regarding the
1491  * journal are consistent.
1492  */
1493 errcode_t e2fsck_check_ext3_journal(e2fsck_t ctx)
1494 {
1495         struct ext2_super_block *sb = ctx->fs->super;
1496         journal_t *journal;
1497         int recover = ext2fs_has_feature_journal_needs_recovery(ctx->fs->super);
1498         struct problem_context pctx;
1499         problem_t problem;
1500         int reset = 0, force_fsck = 0;
1501         errcode_t retval;
1502
1503         /* If we don't have any journal features, don't do anything more */
1504         if (!ext2fs_has_feature_journal(sb) &&
1505             !recover && sb->s_journal_inum == 0 && sb->s_journal_dev == 0 &&
1506             uuid_is_null(sb->s_journal_uuid))
1507                 return 0;
1508
1509         clear_problem_context(&pctx);
1510         pctx.num = sb->s_journal_inum;
1511
1512         retval = e2fsck_get_journal(ctx, &journal);
1513         if (retval) {
1514                 if ((retval == EXT2_ET_BAD_INODE_NUM) ||
1515                     (retval == EXT2_ET_BAD_BLOCK_NUM) ||
1516                     (retval == EXT2_ET_JOURNAL_TOO_SMALL) ||
1517                     (retval == EXT2_ET_NO_JOURNAL))
1518                         return e2fsck_journal_fix_bad_inode(ctx, &pctx);
1519                 return retval;
1520         }
1521
1522         retval = e2fsck_journal_load(journal);
1523         if (retval) {
1524                 if ((retval == EXT2_ET_CORRUPT_JOURNAL_SB) ||
1525                     ((retval == EXT2_ET_UNSUPP_FEATURE) &&
1526                     (!fix_problem(ctx, PR_0_JOURNAL_UNSUPP_INCOMPAT,
1527                                   &pctx))) ||
1528                     ((retval == EXT2_ET_RO_UNSUPP_FEATURE) &&
1529                     (!fix_problem(ctx, PR_0_JOURNAL_UNSUPP_ROCOMPAT,
1530                                   &pctx))) ||
1531                     ((retval == EXT2_ET_JOURNAL_UNSUPP_VERSION) &&
1532                     (!fix_problem(ctx, PR_0_JOURNAL_UNSUPP_VERSION, &pctx))))
1533                         retval = e2fsck_journal_fix_corrupt_super(ctx, journal,
1534                                                                   &pctx);
1535                 e2fsck_journal_release(ctx, journal, 0, 1);
1536                 return retval;
1537         }
1538
1539         /*
1540          * We want to make the flags consistent here.  We will not leave with
1541          * needs_recovery set but has_journal clear.  We can't get in a loop
1542          * with -y, -n, or -p, only if a user isn't making up their mind.
1543          */
1544 no_has_journal:
1545         if (!ext2fs_has_feature_journal(sb)) {
1546                 recover = ext2fs_has_feature_journal_needs_recovery(sb);
1547                 if (fix_problem(ctx, PR_0_JOURNAL_HAS_JOURNAL, &pctx)) {
1548                         if (recover &&
1549                             !fix_problem(ctx, PR_0_JOURNAL_RECOVER_SET, &pctx))
1550                                 goto no_has_journal;
1551                         /*
1552                          * Need a full fsck if we are releasing a
1553                          * journal stored on a reserved inode.
1554                          */
1555                         force_fsck = recover ||
1556                                 (sb->s_journal_inum < EXT2_FIRST_INODE(sb));
1557                         /* Clear all of the journal fields */
1558                         sb->s_journal_inum = 0;
1559                         sb->s_journal_dev = 0;
1560                         memset(sb->s_journal_uuid, 0,
1561                                sizeof(sb->s_journal_uuid));
1562                         e2fsck_clear_recover(ctx, force_fsck);
1563                 } else if (!(ctx->options & E2F_OPT_READONLY)) {
1564                         ext2fs_set_feature_journal(sb);
1565                         ctx->fs->flags &= ~EXT2_FLAG_MASTER_SB_ONLY;
1566                         ext2fs_mark_super_dirty(ctx->fs);
1567                 }
1568         }
1569
1570         if (ext2fs_has_feature_journal(sb) &&
1571             !ext2fs_has_feature_journal_needs_recovery(sb) &&
1572             journal->j_superblock->s_start != 0) {
1573                 /* Print status information */
1574                 fix_problem(ctx, PR_0_JOURNAL_RECOVERY_CLEAR, &pctx);
1575                 if (ctx->superblock)
1576                         problem = PR_0_JOURNAL_RUN_DEFAULT;
1577                 else
1578                         problem = PR_0_JOURNAL_RUN;
1579                 if (fix_problem(ctx, problem, &pctx)) {
1580                         ctx->options |= E2F_OPT_FORCE;
1581                         ext2fs_set_feature_journal_needs_recovery(sb);
1582                         ext2fs_mark_super_dirty(ctx->fs);
1583                 } else if (fix_problem(ctx,
1584                                        PR_0_JOURNAL_RESET_JOURNAL, &pctx)) {
1585                         reset = 1;
1586                         sb->s_state &= ~EXT2_VALID_FS;
1587                         ext2fs_mark_super_dirty(ctx->fs);
1588                 }
1589                 /*
1590                  * If the user answers no to the above question, we
1591                  * ignore the fact that journal apparently has data;
1592                  * accidentally replaying over valid data would be far
1593                  * worse than skipping a questionable recovery.
1594                  *
1595                  * XXX should we abort with a fatal error here?  What
1596                  * will the ext3 kernel code do if a filesystem with
1597                  * !NEEDS_RECOVERY but with a non-zero
1598                  * journal->j_superblock->s_start is mounted?
1599                  */
1600         }
1601
1602         /*
1603          * If we don't need to do replay the journal, check to see if
1604          * the journal's errno is set; if so, we need to mark the file
1605          * system as being corrupt and clear the journal's s_errno.
1606          */
1607         if (!ext2fs_has_feature_journal_needs_recovery(sb) &&
1608             journal->j_superblock->s_errno) {
1609                 ctx->fs->super->s_state |= EXT2_ERROR_FS;
1610                 ext2fs_mark_super_dirty(ctx->fs);
1611                 journal->j_superblock->s_errno = 0;
1612                 e2fsck_journal_sb_csum_set(journal, journal->j_superblock);
1613                 mark_buffer_dirty(journal->j_sb_buffer);
1614         }
1615
1616         e2fsck_journal_release(ctx, journal, reset, 0);
1617         return retval;
1618 }
1619
1620 static errcode_t recover_ext3_journal(e2fsck_t ctx)
1621 {
1622         struct problem_context  pctx;
1623         journal_t *journal;
1624         errcode_t retval;
1625
1626         clear_problem_context(&pctx);
1627
1628         retval = jbd2_journal_init_revoke_record_cache();
1629         if (retval)
1630                 return retval;
1631
1632         retval = jbd2_journal_init_revoke_table_cache();
1633         if (retval)
1634                 return retval;
1635
1636         retval = e2fsck_get_journal(ctx, &journal);
1637         if (retval)
1638                 return retval;
1639
1640         retval = e2fsck_journal_load(journal);
1641         if (retval)
1642                 goto errout;
1643
1644         retval = jbd2_journal_init_revoke(journal, 1024);
1645         if (retval)
1646                 goto errout;
1647
1648         retval = -jbd2_journal_recover(journal);
1649         if (retval)
1650                 goto errout;
1651
1652         if (journal->j_failed_commit) {
1653                 pctx.ino = journal->j_failed_commit;
1654                 fix_problem(ctx, PR_0_JNL_TXN_CORRUPT, &pctx);
1655                 journal->j_superblock->s_errno = -EINVAL;
1656                 mark_buffer_dirty(journal->j_sb_buffer);
1657         }
1658
1659         journal->j_tail_sequence = journal->j_transaction_sequence;
1660
1661 errout:
1662         jbd2_journal_destroy_revoke(journal);
1663         jbd2_journal_destroy_revoke_record_cache();
1664         jbd2_journal_destroy_revoke_table_cache();
1665         e2fsck_journal_release(ctx, journal, 1, 0);
1666         return retval;
1667 }
1668
1669 errcode_t e2fsck_run_ext3_journal(e2fsck_t ctx)
1670 {
1671         io_manager io_ptr = ctx->fs->io->manager;
1672         int blocksize = ctx->fs->blocksize;
1673         errcode_t       retval, recover_retval;
1674         io_stats        stats = 0;
1675         unsigned long long kbytes_written = 0;
1676
1677         printf(_("%s: recovering journal\n"), ctx->device_name);
1678         if (ctx->options & E2F_OPT_READONLY) {
1679                 printf(_("%s: won't do journal recovery while read-only\n"),
1680                        ctx->device_name);
1681                 return EXT2_ET_FILE_RO;
1682         }
1683
1684         if (ctx->fs->flags & EXT2_FLAG_DIRTY)
1685                 ext2fs_flush(ctx->fs);  /* Force out any modifications */
1686
1687         recover_retval = recover_ext3_journal(ctx);
1688
1689         /*
1690          * Reload the filesystem context to get up-to-date data from disk
1691          * because journal recovery will change the filesystem under us.
1692          */
1693         if (ctx->fs->super->s_kbytes_written &&
1694             ctx->fs->io->manager->get_stats)
1695                 ctx->fs->io->manager->get_stats(ctx->fs->io, &stats);
1696         if (stats && stats->bytes_written)
1697                 kbytes_written = stats->bytes_written >> 10;
1698
1699         ext2fs_mmp_stop(ctx->fs);
1700         ext2fs_free(ctx->fs);
1701         retval = ext2fs_open(ctx->filesystem_name, ctx->openfs_flags,
1702                              ctx->superblock, blocksize, io_ptr,
1703                              &ctx->fs);
1704         if (retval) {
1705                 com_err(ctx->program_name, retval,
1706                         _("while trying to re-open %s"),
1707                         ctx->device_name);
1708                 fatal_error(ctx, 0);
1709         }
1710         ctx->fs->priv_data = ctx;
1711         ctx->fs->now = ctx->now;
1712         ctx->fs->flags |= EXT2_FLAG_MASTER_SB_ONLY;
1713         ctx->fs->super->s_kbytes_written += kbytes_written;
1714
1715         /* Set the superblock flags */
1716         e2fsck_clear_recover(ctx, recover_retval != 0);
1717
1718         /*
1719          * Do one last sanity check, and propagate journal->s_errno to
1720          * the EXT2_ERROR_FS flag in the fs superblock if needed.
1721          */
1722         retval = e2fsck_check_ext3_journal(ctx);
1723         return retval ? retval : recover_retval;
1724 }
1725
1726 /*
1727  * This function will move the journal inode from a visible file in
1728  * the filesystem directory hierarchy to the reserved inode if necessary.
1729  */
1730 static const char * const journal_names[] = {
1731         ".journal", "journal", ".journal.dat", "journal.dat", 0 };
1732
1733 void e2fsck_move_ext3_journal(e2fsck_t ctx)
1734 {
1735         struct ext2_super_block *sb = ctx->fs->super;
1736         struct problem_context  pctx;
1737         struct ext2_inode       inode;
1738         ext2_filsys             fs = ctx->fs;
1739         ext2_ino_t              ino;
1740         errcode_t               retval;
1741         const char * const *    cpp;
1742         dgrp_t                  group;
1743         int                     mount_flags;
1744
1745         clear_problem_context(&pctx);
1746
1747         /*
1748          * If the filesystem is opened read-only, or there is no
1749          * journal, then do nothing.
1750          */
1751         if ((ctx->options & E2F_OPT_READONLY) ||
1752             (sb->s_journal_inum == 0) ||
1753             !ext2fs_has_feature_journal(sb))
1754                 return;
1755
1756         /*
1757          * Read in the journal inode
1758          */
1759         if (ext2fs_read_inode(fs, sb->s_journal_inum, &inode) != 0)
1760                 return;
1761
1762         /*
1763          * If it's necessary to backup the journal inode, do so.
1764          */
1765         if ((sb->s_jnl_backup_type == 0) ||
1766             ((sb->s_jnl_backup_type == EXT3_JNL_BACKUP_BLOCKS) &&
1767              memcmp(inode.i_block, sb->s_jnl_blocks, EXT2_N_BLOCKS*4))) {
1768                 if (fix_problem(ctx, PR_0_BACKUP_JNL, &pctx)) {
1769                         memcpy(sb->s_jnl_blocks, inode.i_block,
1770                                EXT2_N_BLOCKS*4);
1771                         sb->s_jnl_blocks[15] = inode.i_size_high;
1772                         sb->s_jnl_blocks[16] = inode.i_size;
1773                         sb->s_jnl_backup_type = EXT3_JNL_BACKUP_BLOCKS;
1774                         ext2fs_mark_super_dirty(fs);
1775                         fs->flags &= ~EXT2_FLAG_MASTER_SB_ONLY;
1776                 }
1777         }
1778
1779         /*
1780          * If the journal is already the hidden inode, then do nothing
1781          */
1782         if (sb->s_journal_inum == EXT2_JOURNAL_INO)
1783                 return;
1784
1785         /*
1786          * The journal inode had better have only one link and not be readable.
1787          */
1788         if (inode.i_links_count != 1)
1789                 return;
1790
1791         /*
1792          * If the filesystem is mounted, or we can't tell whether
1793          * or not it's mounted, do nothing.
1794          */
1795         retval = ext2fs_check_if_mounted(ctx->filesystem_name, &mount_flags);
1796         if (retval || (mount_flags & EXT2_MF_MOUNTED))
1797                 return;
1798
1799         /*
1800          * If we can't find the name of the journal inode, then do
1801          * nothing.
1802          */
1803         for (cpp = journal_names; *cpp; cpp++) {
1804                 retval = ext2fs_lookup(fs, EXT2_ROOT_INO, *cpp,
1805                                        strlen(*cpp), 0, &ino);
1806                 if ((retval == 0) && (ino == sb->s_journal_inum))
1807                         break;
1808         }
1809         if (*cpp == 0)
1810                 return;
1811
1812         /* We need the inode bitmap to be loaded */
1813         retval = ext2fs_read_bitmaps(fs);
1814         if (retval)
1815                 return;
1816
1817         pctx.str = *cpp;
1818         if (!fix_problem(ctx, PR_0_MOVE_JOURNAL, &pctx))
1819                 return;
1820
1821         /*
1822          * OK, we've done all the checks, let's actually move the
1823          * journal inode.  Errors at this point mean we need to force
1824          * an ext2 filesystem check.
1825          */
1826         if ((retval = ext2fs_unlink(fs, EXT2_ROOT_INO, *cpp, ino, 0)) != 0)
1827                 goto err_out;
1828         if ((retval = ext2fs_write_inode(fs, EXT2_JOURNAL_INO, &inode)) != 0)
1829                 goto err_out;
1830         sb->s_journal_inum = EXT2_JOURNAL_INO;
1831         ext2fs_mark_super_dirty(fs);
1832         fs->flags &= ~EXT2_FLAG_MASTER_SB_ONLY;
1833         inode.i_links_count = 0;
1834         inode.i_dtime = ctx->now;
1835         if ((retval = ext2fs_write_inode(fs, ino, &inode)) != 0)
1836                 goto err_out;
1837
1838         group = ext2fs_group_of_ino(fs, ino);
1839         ext2fs_unmark_inode_bitmap2(fs->inode_map, ino);
1840         ext2fs_mark_ib_dirty(fs);
1841         ext2fs_bg_free_inodes_count_set(fs, group, ext2fs_bg_free_inodes_count(fs, group) + 1);
1842         ext2fs_group_desc_csum_set(fs, group);
1843         fs->super->s_free_inodes_count++;
1844         return;
1845
1846 err_out:
1847         pctx.errcode = retval;
1848         fix_problem(ctx, PR_0_ERR_MOVE_JOURNAL, &pctx);
1849         fs->super->s_state &= ~EXT2_VALID_FS;
1850         ext2fs_mark_super_dirty(fs);
1851         return;
1852 }
1853
1854 /*
1855  * This function makes sure the superblock hint for the external
1856  * journal is correct.
1857  */
1858 int e2fsck_fix_ext3_journal_hint(e2fsck_t ctx)
1859 {
1860         struct ext2_super_block *sb = ctx->fs->super;
1861         struct problem_context pctx;
1862         char uuid[37], *journal_name;
1863         struct stat st;
1864
1865         if (!ext2fs_has_feature_journal(sb) ||
1866             uuid_is_null(sb->s_journal_uuid))
1867                 return 0;
1868
1869         uuid_unparse(sb->s_journal_uuid, uuid);
1870         journal_name = blkid_get_devname(ctx->blkid, "UUID", uuid);
1871         if (!journal_name)
1872                 return 0;
1873
1874         if (stat(journal_name, &st) < 0) {
1875                 free(journal_name);
1876                 return 0;
1877         }
1878
1879         if (st.st_rdev != sb->s_journal_dev) {
1880                 clear_problem_context(&pctx);
1881                 pctx.num = st.st_rdev;
1882                 if (fix_problem(ctx, PR_0_EXTERNAL_JOURNAL_HINT, &pctx)) {
1883                         sb->s_journal_dev = st.st_rdev;
1884                         ext2fs_mark_super_dirty(ctx->fs);
1885                 }
1886         }
1887
1888         free(journal_name);
1889         return 0;
1890 }