2 * journal.c --- code for handling the "ext3" journal
4 * Copyright (C) 2000 Andreas Dilger
5 * Copyright (C) 2000 Theodore Ts'o
7 * Parts of the code are based on fs/jfs/journal.c by Stephen C. Tweedie
8 * Copyright (C) 1999 Red Hat Software
10 * This file may be redistributed under the terms of the
11 * GNU General Public License version 2 or at your discretion
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)
21 #ifdef HAVE_SYS_STAT_H
25 #define E2FSCK_INCLUDE_INLINE_FUNCS
28 #include "uuid/uuid.h"
30 static int bh_count = 0;
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.
41 /* Checksumming functions */
42 static int e2fsck_journal_verify_csum_type(journal_t *j,
43 journal_superblock_t *jsb)
45 if (!jbd2_journal_has_csum_v2or3(j))
48 return jsb->s_checksum_type == JBD2_CRC32C_CHKSUM;
51 static __u32 e2fsck_journal_sb_csum(journal_superblock_t *jsb)
55 old_crc = jsb->s_checksum;
57 crc = ext2fs_crc32c_le(~0, (unsigned char *)jsb,
58 sizeof(journal_superblock_t));
59 jsb->s_checksum = old_crc;
64 static int e2fsck_journal_sb_csum_verify(journal_t *j,
65 journal_superblock_t *jsb)
67 __u32 provided, calculated;
69 if (!jbd2_journal_has_csum_v2or3(j))
72 provided = ext2fs_be32_to_cpu(jsb->s_checksum);
73 calculated = e2fsck_journal_sb_csum(jsb);
75 return provided == calculated;
78 static errcode_t e2fsck_journal_sb_csum_set(journal_t *j,
79 journal_superblock_t *jsb)
83 if (!jbd2_journal_has_csum_v2or3(j))
86 crc = e2fsck_journal_sb_csum(jsb);
87 jsb->s_checksum = ext2fs_cpu_to_be32(crc);
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.
95 int jbd2_journal_bmap(journal_t *journal, unsigned long block,
96 unsigned long long *phys)
102 struct inode *inode = journal->j_inode;
111 retval= ext2fs_bmap2(inode->i_ctx->fs, inode->i_ino,
112 &inode->i_ext2, NULL, 0, (blk64_t) block,
115 return -1 * ((int) retval);
119 struct buffer_head *getblk(kdev_t kdev, unsigned long long blocknr,
122 struct buffer_head *bh;
123 int bufsize = sizeof(*bh) + kdev->k_ctx->fs->blocksize -
126 bh = e2fsck_allocate_memory(kdev->k_ctx, bufsize, "block buffer");
130 if (journal_enable_debug >= 3)
132 jfs_debug(4, "getblk for block %llu (%d bytes)(total %d)\n",
133 blocknr, blocksize, bh_count);
135 bh->b_ctx = kdev->k_ctx;
136 if (kdev->k_dev == K_DEV_FS)
137 bh->b_io = kdev->k_ctx->fs->io;
139 bh->b_io = kdev->k_ctx->journal_io;
140 bh->b_size = blocksize;
141 bh->b_blocknr = blocknr;
146 int sync_blockdev(kdev_t kdev)
150 if (kdev->k_dev == K_DEV_FS)
151 io = kdev->k_ctx->fs->io;
153 io = kdev->k_ctx->journal_io;
155 return io_channel_flush(io) ? -EIO : 0;
158 void ll_rw_block(int rw, int op_flags EXT2FS_ATTR((unused)), int nr,
159 struct buffer_head *bhp[])
162 struct buffer_head *bh;
164 for (; nr > 0; --nr) {
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,
173 com_err(bh->b_ctx->device_name, retval,
174 "while reading block %llu\n",
176 bh->b_err = (int) retval;
180 } else if (rw == REQ_OP_WRITE && bh->b_dirty) {
181 jfs_debug(3, "writing block %llu/%p\n",
184 retval = io_channel_write_blk64(bh->b_io,
188 com_err(bh->b_ctx->device_name, retval,
189 "while writing block %llu\n",
191 bh->b_err = (int) retval;
197 jfs_debug(3, "no-op %s for block %llu\n",
198 rw == REQ_OP_READ ? "read" : "write",
204 void mark_buffer_dirty(struct buffer_head *bh)
209 static void mark_buffer_clean(struct buffer_head * bh)
214 void brelse(struct buffer_head *bh)
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);
223 int buffer_uptodate(struct buffer_head *bh)
225 return bh->b_uptodate;
228 void mark_buffer_uptodate(struct buffer_head *bh, int val)
230 bh->b_uptodate = val;
233 void wait_on_buffer(struct buffer_head *bh)
236 ll_rw_block(REQ_OP_READ, 0, 1, &bh);
240 static void e2fsck_clear_recover(e2fsck_t ctx, int error)
242 ext2fs_clear_feature_journal_needs_recovery(ctx->fs->super);
244 /* if we had an error doing journal recovery, we need a full fsck */
246 ctx->fs->super->s_state &= ~EXT2_VALID_FS;
247 ext2fs_mark_super_dirty(ctx->fs);
251 * This is a helper function to check the validity of the journal.
253 struct process_block_struct {
254 e2_blkcnt_t last_block;
257 static int process_journal_block(ext2_filsys fs,
259 e2_blkcnt_t blockcnt,
260 blk64_t ref_block EXT2FS_ATTR((unused)),
261 int ref_offset EXT2FS_ATTR((unused)),
264 struct process_block_struct *p;
265 blk64_t blk = *block_nr;
267 p = (struct process_block_struct *) priv_data;
269 if (!blk || blk < fs->super->s_first_data_block ||
270 blk >= ext2fs_blocks_count(fs->super))
274 p->last_block = blockcnt;
278 static int ext4_fc_replay_scan(journal_t *j, struct buffer_head *bh,
279 int off, tid_t expected_tid)
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};
291 state = &ctx->fc_replay_state;
293 start = (__u8 *)bh->b_data;
294 end = (__u8 *)bh->b_data + j->j_blocksize - 1;
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");
307 if (off != state->fc_replay_expected_off) {
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);
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,
326 ret = JBD2_FC_REPLAY_STOP;
328 ret = JBD2_FC_REPLAY_CONTINUE;
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:
337 state->fc_crc = jbd2_chksum(j, state->fc_crc, cur,
338 sizeof(tl) + ext4_fc_tag_len(&tl));
340 case EXT4_FC_TAG_TAIL:
342 memcpy(&tail, val, sizeof(tail));
343 state->fc_crc = jbd2_chksum(j, state->fc_crc, cur,
345 offsetof(struct ext4_fc_tail,
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;
353 ret = state->fc_replay_num_tags ?
354 JBD2_FC_REPLAY_STOP : -EFSBADCRC;
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) {
365 if (le32_to_cpu(head.fc_tid) != expected_tid) {
370 state->fc_crc = jbd2_chksum(j, state->fc_crc, cur,
371 sizeof(tl) + ext4_fc_tag_len(&tl));
374 ret = state->fc_replay_num_tags ?
375 JBD2_FC_REPLAY_STOP : -ECANCELED;
377 if (ret < 0 || ret == JBD2_FC_REPLAY_STOP)
385 static int __errcode_to_errno(errcode_t err, const char *func, int line)
389 fprintf(stderr, "Error \"%s\" encountered in function %s at line %d\n",
390 error_message(err), func, line);
396 #define errcode_to_errno(err) __errcode_to_errno(err, __func__, __LINE__)
398 #define ex_end(__ex) ((__ex)->e_lblk + (__ex)->e_len - 1)
399 #define ex_pend(__ex) ((__ex)->e_pblk + (__ex)->e_len - 1)
401 static int make_room(struct extent_list *list, int i)
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));
414 memmove(&list->extents[i + 1], &list->extents[i],
415 sizeof(list->extents[0]) * (list->count - i));
420 static int ex_compar(const void *arg1, const void *arg2)
422 const struct ext2fs_extent *ex1 = (const struct ext2fs_extent *)arg1;
423 const struct ext2fs_extent *ex2 = (const struct ext2fs_extent *)arg2;
425 if (ex1->e_lblk < ex2->e_lblk)
427 if (ex1->e_lblk > ex2->e_lblk)
429 return ex1->e_len - ex2->e_len;
432 static int ex_len_compar(const void *arg1, const void *arg2)
434 const struct ext2fs_extent *ex1 = (const struct ext2fs_extent *)arg1;
435 const struct ext2fs_extent *ex2 = (const struct ext2fs_extent *)arg2;
437 if (ex1->e_len < ex2->e_len)
440 if (ex1->e_lblk > ex2->e_lblk)
446 static void ex_sort_and_merge(struct extent_list *list)
454 * Reverse sort by length, that way we strip off all the 0 length
457 qsort(list->extents, list->count, sizeof(struct ext2fs_extent),
460 for (i = 0; i < list->count; i++) {
461 if (list->extents[i].e_len == 0) {
467 if (list->count == 0)
470 /* Now sort by logical offset */
471 qsort(list->extents, list->count, sizeof(list->extents[0]),
474 /* Merge adjacent extents if they are logically and physically contiguous */
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)) {
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];
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)
498 struct ext2fs_extent add_ex = *ex;
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)
505 /* Case 1: No overlap */
506 if (list->extents[i].e_lblk > ex_end(&add_ex))
509 * Unmark all the blocks in bb now. All the blocks get marked
510 * before we exit this function.
512 ext2fs_unmark_block_bitmap_range2(ctx->fs->block_map,
513 list->extents[i].e_pblk, list->extents[i].e_len);
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);
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;
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)) {
534 list->extents[i].e_len = 0;
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;
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;
553 if (add_ex.e_len && !del) {
554 make_room(list, list->count);
555 list->extents[list->count - 1] = add_ex;
558 ex_sort_and_merge(list);
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);
569 static int ext4_add_extent_to_list(e2fsck_t ctx, struct extent_list *list,
570 struct ext2fs_extent *ex)
572 return ext4_modify_extent_list(ctx, list, ex, 0 /* add */);
575 static int ext4_del_extent_from_list(e2fsck_t ctx, struct extent_list *list,
576 struct ext2fs_extent *ex)
578 return ext4_modify_extent_list(ctx, list, ex, 1 /* delete */);
581 static int ext4_fc_read_extents(e2fsck_t ctx, ino_t ino)
583 struct extent_list *extent_list = &ctx->fc_replay_state.fc_extent_list;
585 if (extent_list->ino == ino)
588 extent_list->ino = ino;
589 return errcode_to_errno(e2fsck_read_extents(ctx, extent_list));
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.
600 static void ext4_fc_flush_extents(e2fsck_t ctx, ino_t ino)
602 struct extent_list *extent_list = &ctx->fc_replay_state.fc_extent_list;
604 if (extent_list->ino == ino || extent_list->ino == 0)
606 e2fsck_rewrite_extent_tree(ctx, extent_list);
607 ext2fs_free_mem(&extent_list->extents);
608 memset(extent_list, 0, sizeof(*extent_list));
611 /* Helper struct for dentry replay routines */
612 struct dentry_info_args {
619 static inline int tl_to_darg(struct dentry_info_args *darg,
620 struct ext4_fc_tl *tl, __u8 *val)
622 struct ext4_fc_dentry_info fcd;
624 memcpy(&fcd, val, sizeof(fcd));
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);
634 val + sizeof(struct ext4_fc_dentry_info),
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);
645 static int ext4_fc_handle_unlink(e2fsck_t ctx, struct ext4_fc_tl *tl, __u8 *val)
647 struct dentry_info_args darg;
650 ret = tl_to_darg(&darg, tl, val);
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 */
662 static int ext4_fc_handle_link_and_create(e2fsck_t ctx, struct ext4_fc_tl *tl, __u8 *val)
664 struct dentry_info_args darg;
665 ext2_filsys fs = ctx->fs;
666 struct ext2_inode_large inode_large;
667 int ret, filetype, mode;
669 ret = tl_to_darg(&darg, tl, val);
672 ext4_fc_flush_extents(ctx, 0);
673 ret = errcode_to_errno(ext2fs_read_inode(fs, darg.ino,
674 (struct ext2_inode *)&inode_large));
678 mode = inode_large.i_mode;
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;
700 * Forcefully unlink if the same name is present and ignore the error
701 * if any, since this dirent might not exist
703 ext2fs_unlink(fs, darg.parent_ino, darg.dname, darg.ino,
704 EXT2FS_UNLINK_FORCE);
706 ret = errcode_to_errno(
707 ext2fs_link(fs, darg.parent_ino, darg.dname, darg.ino,
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)
719 if (ondisk_inode->i_flags & EXT4_EXTENTS_FL) {
720 struct ext3_extent_header *eh;
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));
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));
737 static int ext4_fc_handle_inode(e2fsck_t ctx, __u8 *val)
739 int ino, inode_len = EXT2_GOOD_OLD_INODE_SIZE;
740 struct ext2_inode_large *inode = NULL, *fc_inode = NULL;
746 memcpy(&fc_ino, val, sizeof(fc_ino));
747 fc_raw_inode = val + sizeof(fc_ino);
748 ino = le32_to_cpu(fc_ino);
750 if (EXT2_INODE_SIZE(ctx->fs->super) > EXT2_GOOD_OLD_INODE_SIZE)
751 inode_len += ext2fs_le16_to_cpu(
752 ((struct ext2_inode_large *)fc_raw_inode)->i_extra_isize);
753 err = ext2fs_get_mem(inode_len, &inode);
756 err = ext2fs_get_mem(inode_len, &fc_inode);
759 ext4_fc_flush_extents(ctx, ino);
761 err = ext2fs_read_inode_full(ctx->fs, ino, (struct ext2_inode *)inode,
765 memcpy(fc_inode, fc_raw_inode, inode_len);
766 #ifdef WORDS_BIGENDIAN
767 ext2fs_swap_inode_full(ctx->fs, fc_inode, fc_inode, 0, inode_len);
769 memcpy(inode, fc_inode, offsetof(struct ext2_inode_large, i_block));
770 memcpy(&inode->i_generation, &fc_inode->i_generation,
771 inode_len - offsetof(struct ext2_inode_large, i_generation));
772 ext4_fc_replay_fixup_iblocks(inode, fc_inode);
773 err = ext2fs_count_blocks(ctx->fs, ino, EXT2_INODE(inode), &blks);
776 ext2fs_iblk_set(ctx->fs, EXT2_INODE(inode), blks);
777 ext2fs_inode_csum_set(ctx->fs, ino, inode);
779 err = ext2fs_write_inode_full(ctx->fs, ino, (struct ext2_inode *)inode,
783 if (inode->i_links_count)
784 ext2fs_mark_inode_bitmap2(ctx->fs->inode_map, ino);
786 ext2fs_unmark_inode_bitmap2(ctx->fs->inode_map, ino);
787 ext2fs_mark_ib_dirty(ctx->fs);
790 ext2fs_free_mem(&inode);
791 ext2fs_free_mem(&fc_inode);
792 return errcode_to_errno(err);
796 * Handle add extent replay tag.
798 static int ext4_fc_handle_add_extent(e2fsck_t ctx, __u8 *val)
800 struct ext2fs_extent extent;
801 struct ext4_fc_add_range add_range;
805 memcpy(&add_range, val, sizeof(add_range));
806 ino = le32_to_cpu(add_range.fc_ino);
807 ext4_fc_flush_extents(ctx, ino);
809 ret = ext4_fc_read_extents(ctx, ino);
812 memset(&extent, 0, sizeof(extent));
813 ret = errcode_to_errno(ext2fs_decode_extent(
814 &extent, (void *)add_range.fc_ex,
815 sizeof(add_range.fc_ex)));
818 return ext4_add_extent_to_list(ctx,
819 &ctx->fc_replay_state.fc_extent_list, &extent);
823 * Handle delete logical range replay tag.
825 static int ext4_fc_handle_del_range(e2fsck_t ctx, __u8 *val)
827 struct ext2fs_extent extent;
828 struct ext4_fc_del_range del_range;
831 memcpy(&del_range, val, sizeof(del_range));
832 ino = le32_to_cpu(del_range.fc_ino);
833 ext4_fc_flush_extents(ctx, ino);
835 memset(&extent, 0, sizeof(extent));
836 extent.e_lblk = le32_to_cpu(del_range.fc_lblk);
837 extent.e_len = le32_to_cpu(del_range.fc_len);
838 ret = ext4_fc_read_extents(ctx, ino);
841 return ext4_del_extent_from_list(ctx,
842 &ctx->fc_replay_state.fc_extent_list, &extent);
846 * Main recovery path entry point. This function returns JBD2_FC_REPLAY_CONTINUE
847 * to indicate that it is expecting more fast commit blocks. It returns
848 * JBD2_FC_REPLAY_STOP to indicate that replay is done.
850 static int ext4_fc_replay(journal_t *journal, struct buffer_head *bh,
851 enum passtype pass, int off, tid_t expected_tid)
853 e2fsck_t ctx = journal->j_fs_dev->k_ctx;
854 struct e2fsck_fc_replay_state *state = &ctx->fc_replay_state;
855 int ret = JBD2_FC_REPLAY_CONTINUE;
856 struct ext4_fc_tl tl;
857 __u8 *start, *end, *cur, *val;
859 if (pass == PASS_SCAN) {
860 state->fc_current_pass = PASS_SCAN;
861 return ext4_fc_replay_scan(journal, bh, off, expected_tid);
864 if (state->fc_replay_num_tags == 0)
867 if (state->fc_current_pass != pass) {
868 /* Starting replay phase */
869 state->fc_current_pass = pass;
870 /* We will reset checksums */
871 ctx->fs->flags |= EXT2_FLAG_IGNORE_CSUM_ERRORS;
872 ret = errcode_to_errno(ext2fs_read_bitmaps(ctx->fs));
874 jbd_debug(1, "Error %d while reading bitmaps\n", ret);
877 state->fc_super_state = ctx->fs->super->s_state;
879 * Mark the file system to indicate it contains errors. That's
880 * because the updates performed by fast commit replay code are
881 * not atomic and may result in incosistent file system if it
882 * crashes before the replay is complete.
884 ctx->fs->super->s_state |= EXT2_ERROR_FS;
885 ctx->fs->super->s_state |= EXT4_FC_REPLAY;
886 ext2fs_mark_super_dirty(ctx->fs);
887 ext2fs_flush(ctx->fs);
890 start = (__u8 *)bh->b_data;
891 end = (__u8 *)bh->b_data + journal->j_blocksize - 1;
893 for (cur = start; cur < end; cur = cur + le16_to_cpu(tl.fc_len) + sizeof(tl)) {
894 memcpy(&tl, cur, sizeof(tl));
895 val = cur + sizeof(tl);
897 if (state->fc_replay_num_tags == 0)
899 jbd_debug(3, "Replay phase processing %s tag\n",
900 tag2str(le16_to_cpu(tl.fc_tag)));
901 state->fc_replay_num_tags--;
902 switch (le16_to_cpu(tl.fc_tag)) {
903 case EXT4_FC_TAG_CREAT:
904 case EXT4_FC_TAG_LINK:
905 ret = ext4_fc_handle_link_and_create(ctx, &tl, val);
907 case EXT4_FC_TAG_UNLINK:
908 ret = ext4_fc_handle_unlink(ctx, &tl, val);
910 case EXT4_FC_TAG_ADD_RANGE:
911 ret = ext4_fc_handle_add_extent(ctx, val);
913 case EXT4_FC_TAG_DEL_RANGE:
914 ret = ext4_fc_handle_del_range(ctx, val);
916 case EXT4_FC_TAG_INODE:
917 ret = ext4_fc_handle_inode(ctx, val);
919 case EXT4_FC_TAG_TAIL:
920 ext4_fc_flush_extents(ctx, 0);
921 case EXT4_FC_TAG_PAD:
922 case EXT4_FC_TAG_HEAD:
930 ret = JBD2_FC_REPLAY_CONTINUE;
934 jbd_debug(1, "End of fast commit replay\n");
935 if (state->fc_current_pass != pass)
936 return JBD2_FC_REPLAY_STOP;
938 ext2fs_calculate_summary_stats(ctx->fs, 0 /* update bg also */);
939 ext2fs_write_block_bitmap(ctx->fs);
940 ext2fs_write_inode_bitmap(ctx->fs);
941 ext2fs_mark_super_dirty(ctx->fs);
942 ext2fs_set_gdt_csum(ctx->fs);
943 ctx->fs->super->s_state = state->fc_super_state;
944 ext2fs_flush(ctx->fs);
946 return JBD2_FC_REPLAY_STOP;
949 static errcode_t e2fsck_get_journal(e2fsck_t ctx, journal_t **ret_journal)
951 struct process_block_struct pb;
952 struct ext2_super_block *sb = ctx->fs->super;
953 struct ext2_super_block jsuper;
954 struct problem_context pctx;
955 struct buffer_head *bh;
956 struct inode *j_inode = NULL;
957 struct kdev_s *dev_fs = NULL, *dev_journal;
958 const char *journal_name = 0;
959 journal_t *journal = NULL;
960 errcode_t retval = 0;
961 io_manager io_ptr = 0;
962 unsigned long long start = 0;
965 int tried_backup_jnl = 0;
967 clear_problem_context(&pctx);
969 journal = e2fsck_allocate_memory(ctx, sizeof(journal_t), "journal");
971 return EXT2_ET_NO_MEMORY;
974 dev_fs = e2fsck_allocate_memory(ctx, 2*sizeof(struct kdev_s), "kdev");
976 retval = EXT2_ET_NO_MEMORY;
979 dev_journal = dev_fs+1;
981 dev_fs->k_ctx = dev_journal->k_ctx = ctx;
982 dev_fs->k_dev = K_DEV_FS;
983 dev_journal->k_dev = K_DEV_JOURNAL;
985 journal->j_dev = dev_journal;
986 journal->j_fs_dev = dev_fs;
987 journal->j_inode = NULL;
988 journal->j_blocksize = ctx->fs->blocksize;
990 if (uuid_is_null(sb->s_journal_uuid)) {
992 * The full set of superblock sanity checks haven't
993 * been performed yet, so we need to do some basic
994 * checks here to avoid potential array overruns.
996 if (!sb->s_journal_inum ||
997 (sb->s_journal_inum >
998 (ctx->fs->group_desc_count * sb->s_inodes_per_group))) {
999 retval = EXT2_ET_BAD_INODE_NUM;
1002 j_inode = e2fsck_allocate_memory(ctx, sizeof(*j_inode),
1005 retval = EXT2_ET_NO_MEMORY;
1009 j_inode->i_ctx = ctx;
1010 j_inode->i_ino = sb->s_journal_inum;
1012 if ((retval = ext2fs_read_inode(ctx->fs,
1014 &j_inode->i_ext2))) {
1016 if (sb->s_jnl_backup_type != EXT3_JNL_BACKUP_BLOCKS ||
1019 memset(&j_inode->i_ext2, 0, sizeof(struct ext2_inode));
1020 memcpy(&j_inode->i_ext2.i_block[0], sb->s_jnl_blocks,
1022 j_inode->i_ext2.i_size_high = sb->s_jnl_blocks[15];
1023 j_inode->i_ext2.i_size = sb->s_jnl_blocks[16];
1024 j_inode->i_ext2.i_links_count = 1;
1025 j_inode->i_ext2.i_mode = LINUX_S_IFREG | 0600;
1026 e2fsck_use_inode_shortcuts(ctx, 1);
1027 ctx->stashed_ino = j_inode->i_ino;
1028 ctx->stashed_inode = &j_inode->i_ext2;
1031 if (!j_inode->i_ext2.i_links_count ||
1032 !LINUX_S_ISREG(j_inode->i_ext2.i_mode)) {
1033 retval = EXT2_ET_NO_JOURNAL;
1034 goto try_backup_journal;
1036 if (EXT2_I_SIZE(&j_inode->i_ext2) / journal->j_blocksize <
1037 JBD2_MIN_JOURNAL_BLOCKS) {
1038 retval = EXT2_ET_JOURNAL_TOO_SMALL;
1039 goto try_backup_journal;
1042 retval = ext2fs_block_iterate3(ctx->fs, j_inode->i_ino,
1044 process_journal_block, &pb);
1045 if ((pb.last_block + 1) * ctx->fs->blocksize <
1046 (int) EXT2_I_SIZE(&j_inode->i_ext2)) {
1047 retval = EXT2_ET_JOURNAL_TOO_SMALL;
1048 goto try_backup_journal;
1050 if (tried_backup_jnl && !(ctx->options & E2F_OPT_READONLY)) {
1051 retval = ext2fs_write_inode(ctx->fs, sb->s_journal_inum,
1057 journal->j_total_len = EXT2_I_SIZE(&j_inode->i_ext2) /
1058 journal->j_blocksize;
1061 retval = ext2fs_inode_io_intern2(ctx->fs, sb->s_journal_inum,
1067 io_ptr = inode_io_manager;
1069 journal->j_inode = j_inode;
1070 ctx->journal_io = ctx->fs->io;
1071 if ((ret = jbd2_journal_bmap(journal, 0, &start)) != 0) {
1072 retval = (errcode_t) (-1 * ret);
1078 if (!ctx->journal_name) {
1081 uuid_unparse(sb->s_journal_uuid, uuid);
1082 ctx->journal_name = blkid_get_devname(ctx->blkid,
1084 if (!ctx->journal_name)
1085 ctx->journal_name = blkid_devno_to_devname(sb->s_journal_dev);
1087 journal_name = ctx->journal_name;
1089 if (!journal_name) {
1090 fix_problem(ctx, PR_0_CANT_FIND_JOURNAL, &pctx);
1091 retval = EXT2_ET_LOAD_EXT_JOURNAL;
1095 jfs_debug(1, "Using journal file %s\n", journal_name);
1096 io_ptr = unix_io_manager;
1100 test_io_backing_manager = io_ptr;
1101 io_ptr = test_io_manager;
1103 #ifndef USE_INODE_IO
1107 int flags = IO_FLAG_RW;
1108 if (!(ctx->mount_flags & EXT2_MF_ISROOT &&
1109 ctx->mount_flags & EXT2_MF_READONLY))
1110 flags |= IO_FLAG_EXCLUSIVE;
1111 if ((ctx->mount_flags & EXT2_MF_READONLY) &&
1112 (ctx->options & E2F_OPT_FORCE))
1113 flags &= ~IO_FLAG_EXCLUSIVE;
1116 retval = io_ptr->open(journal_name, flags,
1122 io_channel_set_blksize(ctx->journal_io, ctx->fs->blocksize);
1127 start = ext2fs_journal_sb_start(ctx->fs->blocksize) - 1;
1128 bh = getblk(dev_journal, start, ctx->fs->blocksize);
1130 retval = EXT2_ET_NO_MEMORY;
1133 ll_rw_block(REQ_OP_READ, 0, 1, &bh);
1134 if ((retval = bh->b_err) != 0) {
1138 memcpy(&jsuper, start ? bh->b_data : bh->b_data + SUPERBLOCK_OFFSET,
1140 #ifdef WORDS_BIGENDIAN
1141 if (jsuper.s_magic == ext2fs_swab16(EXT2_SUPER_MAGIC))
1142 ext2fs_swap_super(&jsuper);
1144 if (jsuper.s_magic != EXT2_SUPER_MAGIC ||
1145 !ext2fs_has_feature_journal_dev(&jsuper)) {
1146 fix_problem(ctx, PR_0_EXT_JOURNAL_BAD_SUPER, &pctx);
1147 retval = EXT2_ET_LOAD_EXT_JOURNAL;
1151 /* Make sure the journal UUID is correct */
1152 if (memcmp(jsuper.s_uuid, ctx->fs->super->s_journal_uuid,
1153 sizeof(jsuper.s_uuid))) {
1154 fix_problem(ctx, PR_0_JOURNAL_BAD_UUID, &pctx);
1155 retval = EXT2_ET_LOAD_EXT_JOURNAL;
1160 /* Check the superblock checksum */
1161 if (ext2fs_has_feature_metadata_csum(&jsuper)) {
1162 struct struct_ext2_filsys fsx;
1163 struct ext2_super_block superx;
1166 p = start ? bh->b_data : bh->b_data + SUPERBLOCK_OFFSET;
1167 memcpy(&fsx, ctx->fs, sizeof(fsx));
1168 memcpy(&superx, ctx->fs->super, sizeof(superx));
1169 fsx.super = &superx;
1170 ext2fs_set_feature_metadata_csum(fsx.super);
1171 if (!ext2fs_superblock_csum_verify(&fsx, p) &&
1172 fix_problem(ctx, PR_0_EXT_JOURNAL_SUPER_CSUM_INVALID,
1174 ext2fs_superblock_csum_set(&fsx, p);
1175 mark_buffer_dirty(bh);
1180 maxlen = ext2fs_blocks_count(&jsuper);
1181 journal->j_total_len = (maxlen < 1ULL << 32) ? maxlen : (1ULL << 32) - 1;
1185 if (!(bh = getblk(dev_journal, start, journal->j_blocksize))) {
1186 retval = EXT2_ET_NO_MEMORY;
1190 journal->j_sb_buffer = bh;
1191 journal->j_superblock = (journal_superblock_t *)bh->b_data;
1192 if (ext2fs_has_feature_fast_commit(ctx->fs->super))
1193 journal->j_fc_replay_callback = ext4_fc_replay;
1195 journal->j_fc_replay_callback = NULL;
1199 ext2fs_free_mem(&j_inode);
1202 *ret_journal = journal;
1203 e2fsck_use_inode_shortcuts(ctx, 0);
1207 e2fsck_use_inode_shortcuts(ctx, 0);
1209 ext2fs_free_mem(&dev_fs);
1211 ext2fs_free_mem(&j_inode);
1213 ext2fs_free_mem(&journal);
1217 static errcode_t e2fsck_journal_fix_bad_inode(e2fsck_t ctx,
1218 struct problem_context *pctx)
1220 struct ext2_super_block *sb = ctx->fs->super;
1221 int recover = ext2fs_has_feature_journal_needs_recovery(ctx->fs->super);
1222 int has_journal = ext2fs_has_feature_journal(ctx->fs->super);
1224 if (has_journal || sb->s_journal_inum) {
1225 /* The journal inode is bogus, remove and force full fsck */
1226 pctx->ino = sb->s_journal_inum;
1227 if (fix_problem(ctx, PR_0_JOURNAL_BAD_INODE, pctx)) {
1228 if (has_journal && sb->s_journal_inum)
1229 printf("*** journal has been deleted ***\n\n");
1230 ext2fs_clear_feature_journal(sb);
1231 sb->s_journal_inum = 0;
1232 memset(sb->s_jnl_blocks, 0, sizeof(sb->s_jnl_blocks));
1233 ctx->flags |= E2F_FLAG_JOURNAL_INODE;
1234 ctx->fs->flags &= ~EXT2_FLAG_MASTER_SB_ONLY;
1235 e2fsck_clear_recover(ctx, 1);
1238 return EXT2_ET_CORRUPT_JOURNAL_SB;
1239 } else if (recover) {
1240 if (fix_problem(ctx, PR_0_JOURNAL_RECOVER_SET, pctx)) {
1241 e2fsck_clear_recover(ctx, 1);
1244 return EXT2_ET_UNSUPP_FEATURE;
1249 #define V1_SB_SIZE 0x0024
1250 static void clear_v2_journal_fields(journal_t *journal)
1252 e2fsck_t ctx = journal->j_dev->k_ctx;
1253 struct problem_context pctx;
1255 clear_problem_context(&pctx);
1257 if (!fix_problem(ctx, PR_0_CLEAR_V2_JOURNAL, &pctx))
1260 ctx->flags |= E2F_FLAG_PROBLEMS_FIXED;
1261 memset(((char *) journal->j_superblock) + V1_SB_SIZE, 0,
1262 ctx->fs->blocksize-V1_SB_SIZE);
1263 mark_buffer_dirty(journal->j_sb_buffer);
1267 static errcode_t e2fsck_journal_load(journal_t *journal)
1269 e2fsck_t ctx = journal->j_dev->k_ctx;
1270 journal_superblock_t *jsb;
1271 struct buffer_head *jbh = journal->j_sb_buffer;
1272 struct problem_context pctx;
1274 clear_problem_context(&pctx);
1276 ll_rw_block(REQ_OP_READ, 0, 1, &jbh);
1278 com_err(ctx->device_name, jbh->b_err, "%s",
1279 _("reading journal superblock\n"));
1283 jsb = journal->j_superblock;
1284 /* If we don't even have JBD2_MAGIC, we probably have a wrong inode */
1285 if (jsb->s_header.h_magic != htonl(JBD2_MAGIC_NUMBER))
1286 return e2fsck_journal_fix_bad_inode(ctx, &pctx);
1288 switch (ntohl(jsb->s_header.h_blocktype)) {
1289 case JBD2_SUPERBLOCK_V1:
1290 journal->j_format_version = 1;
1291 if (jsb->s_feature_compat ||
1292 jsb->s_feature_incompat ||
1293 jsb->s_feature_ro_compat ||
1295 clear_v2_journal_fields(journal);
1298 case JBD2_SUPERBLOCK_V2:
1299 journal->j_format_version = 2;
1300 if (ntohl(jsb->s_nr_users) > 1 &&
1301 uuid_is_null(ctx->fs->super->s_journal_uuid))
1302 clear_v2_journal_fields(journal);
1303 if (ntohl(jsb->s_nr_users) > 1) {
1304 fix_problem(ctx, PR_0_JOURNAL_UNSUPP_MULTIFS, &pctx);
1305 return EXT2_ET_JOURNAL_UNSUPP_VERSION;
1310 * These should never appear in a journal super block, so if
1311 * they do, the journal is badly corrupted.
1313 case JBD2_DESCRIPTOR_BLOCK:
1314 case JBD2_COMMIT_BLOCK:
1315 case JBD2_REVOKE_BLOCK:
1316 return EXT2_ET_CORRUPT_JOURNAL_SB;
1318 /* If we don't understand the superblock major type, but there
1319 * is a magic number, then it is likely to be a new format we
1320 * just don't understand, so leave it alone. */
1322 return EXT2_ET_JOURNAL_UNSUPP_VERSION;
1325 if (JBD2_HAS_INCOMPAT_FEATURE(journal, ~JBD2_KNOWN_INCOMPAT_FEATURES))
1326 return EXT2_ET_UNSUPP_FEATURE;
1328 if (JBD2_HAS_RO_COMPAT_FEATURE(journal, ~JBD2_KNOWN_ROCOMPAT_FEATURES))
1329 return EXT2_ET_RO_UNSUPP_FEATURE;
1331 /* Checksum v1-3 are mutually exclusive features. */
1332 if (jbd2_has_feature_csum2(journal) && jbd2_has_feature_csum3(journal))
1333 return EXT2_ET_CORRUPT_JOURNAL_SB;
1335 if (jbd2_journal_has_csum_v2or3(journal) &&
1336 jbd2_has_feature_checksum(journal))
1337 return EXT2_ET_CORRUPT_JOURNAL_SB;
1339 if (!e2fsck_journal_verify_csum_type(journal, jsb) ||
1340 !e2fsck_journal_sb_csum_verify(journal, jsb))
1341 return EXT2_ET_CORRUPT_JOURNAL_SB;
1343 if (jbd2_journal_has_csum_v2or3(journal))
1344 journal->j_csum_seed = jbd2_chksum(journal, ~0, jsb->s_uuid,
1345 sizeof(jsb->s_uuid));
1347 /* We have now checked whether we know enough about the journal
1348 * format to be able to proceed safely, so any other checks that
1349 * fail we should attempt to recover from. */
1350 if (jsb->s_blocksize != htonl(journal->j_blocksize)) {
1351 com_err(ctx->program_name, EXT2_ET_CORRUPT_JOURNAL_SB,
1352 _("%s: no valid journal superblock found\n"),
1354 return EXT2_ET_CORRUPT_JOURNAL_SB;
1357 if (ntohl(jsb->s_maxlen) < journal->j_total_len)
1358 journal->j_total_len = ntohl(jsb->s_maxlen);
1359 else if (ntohl(jsb->s_maxlen) > journal->j_total_len) {
1360 com_err(ctx->program_name, EXT2_ET_CORRUPT_JOURNAL_SB,
1361 _("%s: journal too short\n"),
1363 return EXT2_ET_CORRUPT_JOURNAL_SB;
1366 journal->j_tail_sequence = ntohl(jsb->s_sequence);
1367 journal->j_transaction_sequence = journal->j_tail_sequence;
1368 journal->j_tail = ntohl(jsb->s_start);
1369 journal->j_first = ntohl(jsb->s_first);
1370 if (jbd2_has_feature_fast_commit(journal)) {
1371 if (ntohl(jsb->s_maxlen) - jbd2_journal_get_num_fc_blks(jsb)
1372 < JBD2_MIN_JOURNAL_BLOCKS) {
1373 com_err(ctx->program_name, EXT2_ET_CORRUPT_JOURNAL_SB,
1374 _("%s: incorrect fast commit blocks\n"),
1376 return EXT2_ET_CORRUPT_JOURNAL_SB;
1378 journal->j_fc_last = ntohl(jsb->s_maxlen);
1379 journal->j_last = journal->j_fc_last -
1380 jbd2_journal_get_num_fc_blks(jsb);
1381 journal->j_fc_first = journal->j_last + 1;
1383 journal->j_last = ntohl(jsb->s_maxlen);
1389 static void e2fsck_journal_reset_super(e2fsck_t ctx, journal_superblock_t *jsb,
1400 /* Leave a valid existing V1 superblock signature alone.
1401 * Anything unrecognisable we overwrite with a new V2
1404 if (jsb->s_header.h_magic != htonl(JBD2_MAGIC_NUMBER) ||
1405 jsb->s_header.h_blocktype != htonl(JBD2_SUPERBLOCK_V1)) {
1406 jsb->s_header.h_magic = htonl(JBD2_MAGIC_NUMBER);
1407 jsb->s_header.h_blocktype = htonl(JBD2_SUPERBLOCK_V2);
1410 /* Zero out everything else beyond the superblock header */
1412 p = ((char *) jsb) + sizeof(journal_header_t);
1413 memset (p, 0, ctx->fs->blocksize-sizeof(journal_header_t));
1415 jsb->s_blocksize = htonl(ctx->fs->blocksize);
1416 jsb->s_maxlen = htonl(journal->j_total_len);
1417 jsb->s_first = htonl(1);
1419 /* Initialize the journal sequence number so that there is "no"
1420 * chance we will find old "valid" transactions in the journal.
1421 * This avoids the need to zero the whole journal (slow to do,
1422 * and risky when we are just recovering the filesystem).
1424 uuid_generate(u.uuid);
1425 for (i = 0; i < 4; i ++)
1426 new_seq ^= u.val[i];
1427 jsb->s_sequence = htonl(new_seq);
1428 e2fsck_journal_sb_csum_set(journal, jsb);
1430 mark_buffer_dirty(journal->j_sb_buffer);
1431 ll_rw_block(REQ_OP_WRITE, 0, 1, &journal->j_sb_buffer);
1434 static errcode_t e2fsck_journal_fix_corrupt_super(e2fsck_t ctx,
1436 struct problem_context *pctx)
1438 struct ext2_super_block *sb = ctx->fs->super;
1439 int recover = ext2fs_has_feature_journal_needs_recovery(ctx->fs->super);
1441 if (ext2fs_has_feature_journal(sb)) {
1442 if (fix_problem(ctx, PR_0_JOURNAL_BAD_SUPER, pctx)) {
1443 e2fsck_journal_reset_super(ctx, journal->j_superblock,
1445 journal->j_transaction_sequence = 1;
1446 e2fsck_clear_recover(ctx, recover);
1449 return EXT2_ET_CORRUPT_JOURNAL_SB;
1450 } else if (e2fsck_journal_fix_bad_inode(ctx, pctx))
1451 return EXT2_ET_CORRUPT_JOURNAL_SB;
1456 static void e2fsck_journal_release(e2fsck_t ctx, journal_t *journal,
1457 int reset, int drop)
1459 journal_superblock_t *jsb;
1462 mark_buffer_clean(journal->j_sb_buffer);
1463 else if (!(ctx->options & E2F_OPT_READONLY)) {
1464 jsb = journal->j_superblock;
1465 jsb->s_sequence = htonl(journal->j_tail_sequence);
1467 jsb->s_start = 0; /* this marks the journal as empty */
1468 e2fsck_journal_sb_csum_set(journal, jsb);
1469 mark_buffer_dirty(journal->j_sb_buffer);
1471 brelse(journal->j_sb_buffer);
1473 if (ctx->journal_io) {
1474 if (ctx->fs && ctx->fs->io != ctx->journal_io)
1475 io_channel_close(ctx->journal_io);
1476 ctx->journal_io = 0;
1479 #ifndef USE_INODE_IO
1480 if (journal->j_inode)
1481 ext2fs_free_mem(&journal->j_inode);
1483 if (journal->j_fs_dev)
1484 ext2fs_free_mem(&journal->j_fs_dev);
1485 ext2fs_free_mem(&journal);
1489 * This function makes sure that the superblock fields regarding the
1490 * journal are consistent.
1492 errcode_t e2fsck_check_ext3_journal(e2fsck_t ctx)
1494 struct ext2_super_block *sb = ctx->fs->super;
1496 int recover = ext2fs_has_feature_journal_needs_recovery(ctx->fs->super);
1497 struct problem_context pctx;
1499 int reset = 0, force_fsck = 0;
1502 /* If we don't have any journal features, don't do anything more */
1503 if (!ext2fs_has_feature_journal(sb) &&
1504 !recover && sb->s_journal_inum == 0 && sb->s_journal_dev == 0 &&
1505 uuid_is_null(sb->s_journal_uuid))
1508 clear_problem_context(&pctx);
1509 pctx.num = sb->s_journal_inum;
1511 retval = e2fsck_get_journal(ctx, &journal);
1513 if ((retval == EXT2_ET_BAD_INODE_NUM) ||
1514 (retval == EXT2_ET_BAD_BLOCK_NUM) ||
1515 (retval == EXT2_ET_JOURNAL_TOO_SMALL) ||
1516 (retval == EXT2_ET_NO_JOURNAL))
1517 return e2fsck_journal_fix_bad_inode(ctx, &pctx);
1521 retval = e2fsck_journal_load(journal);
1523 if ((retval == EXT2_ET_CORRUPT_JOURNAL_SB) ||
1524 ((retval == EXT2_ET_UNSUPP_FEATURE) &&
1525 (!fix_problem(ctx, PR_0_JOURNAL_UNSUPP_INCOMPAT,
1527 ((retval == EXT2_ET_RO_UNSUPP_FEATURE) &&
1528 (!fix_problem(ctx, PR_0_JOURNAL_UNSUPP_ROCOMPAT,
1530 ((retval == EXT2_ET_JOURNAL_UNSUPP_VERSION) &&
1531 (!fix_problem(ctx, PR_0_JOURNAL_UNSUPP_VERSION, &pctx))))
1532 retval = e2fsck_journal_fix_corrupt_super(ctx, journal,
1534 e2fsck_journal_release(ctx, journal, 0, 1);
1539 * We want to make the flags consistent here. We will not leave with
1540 * needs_recovery set but has_journal clear. We can't get in a loop
1541 * with -y, -n, or -p, only if a user isn't making up their mind.
1544 if (!ext2fs_has_feature_journal(sb)) {
1545 recover = ext2fs_has_feature_journal_needs_recovery(sb);
1546 if (fix_problem(ctx, PR_0_JOURNAL_HAS_JOURNAL, &pctx)) {
1548 !fix_problem(ctx, PR_0_JOURNAL_RECOVER_SET, &pctx))
1549 goto no_has_journal;
1551 * Need a full fsck if we are releasing a
1552 * journal stored on a reserved inode.
1554 force_fsck = recover ||
1555 (sb->s_journal_inum < EXT2_FIRST_INODE(sb));
1556 /* Clear all of the journal fields */
1557 sb->s_journal_inum = 0;
1558 sb->s_journal_dev = 0;
1559 memset(sb->s_journal_uuid, 0,
1560 sizeof(sb->s_journal_uuid));
1561 e2fsck_clear_recover(ctx, force_fsck);
1562 } else if (!(ctx->options & E2F_OPT_READONLY)) {
1563 ext2fs_set_feature_journal(sb);
1564 ctx->fs->flags &= ~EXT2_FLAG_MASTER_SB_ONLY;
1565 ext2fs_mark_super_dirty(ctx->fs);
1569 if (ext2fs_has_feature_journal(sb) &&
1570 !ext2fs_has_feature_journal_needs_recovery(sb) &&
1571 journal->j_superblock->s_start != 0) {
1572 /* Print status information */
1573 fix_problem(ctx, PR_0_JOURNAL_RECOVERY_CLEAR, &pctx);
1574 if (ctx->superblock)
1575 problem = PR_0_JOURNAL_RUN_DEFAULT;
1577 problem = PR_0_JOURNAL_RUN;
1578 if (fix_problem(ctx, problem, &pctx)) {
1579 ctx->options |= E2F_OPT_FORCE;
1580 ext2fs_set_feature_journal_needs_recovery(sb);
1581 ext2fs_mark_super_dirty(ctx->fs);
1582 } else if (fix_problem(ctx,
1583 PR_0_JOURNAL_RESET_JOURNAL, &pctx)) {
1585 sb->s_state &= ~EXT2_VALID_FS;
1586 ext2fs_mark_super_dirty(ctx->fs);
1589 * If the user answers no to the above question, we
1590 * ignore the fact that journal apparently has data;
1591 * accidentally replaying over valid data would be far
1592 * worse than skipping a questionable recovery.
1594 * XXX should we abort with a fatal error here? What
1595 * will the ext3 kernel code do if a filesystem with
1596 * !NEEDS_RECOVERY but with a non-zero
1597 * journal->j_superblock->s_start is mounted?
1602 * If we don't need to do replay the journal, check to see if
1603 * the journal's errno is set; if so, we need to mark the file
1604 * system as being corrupt and clear the journal's s_errno.
1606 if (!ext2fs_has_feature_journal_needs_recovery(sb) &&
1607 journal->j_superblock->s_errno) {
1608 ctx->fs->super->s_state |= EXT2_ERROR_FS;
1609 ext2fs_mark_super_dirty(ctx->fs);
1610 journal->j_superblock->s_errno = 0;
1611 e2fsck_journal_sb_csum_set(journal, journal->j_superblock);
1612 mark_buffer_dirty(journal->j_sb_buffer);
1615 e2fsck_journal_release(ctx, journal, reset, 0);
1619 static errcode_t recover_ext3_journal(e2fsck_t ctx)
1621 struct problem_context pctx;
1625 clear_problem_context(&pctx);
1627 retval = jbd2_journal_init_revoke_record_cache();
1631 retval = jbd2_journal_init_revoke_table_cache();
1635 retval = e2fsck_get_journal(ctx, &journal);
1639 retval = e2fsck_journal_load(journal);
1643 retval = jbd2_journal_init_revoke(journal, 1024);
1647 retval = -jbd2_journal_recover(journal);
1651 if (journal->j_failed_commit) {
1652 pctx.ino = journal->j_failed_commit;
1653 fix_problem(ctx, PR_0_JNL_TXN_CORRUPT, &pctx);
1654 journal->j_superblock->s_errno = -EINVAL;
1655 mark_buffer_dirty(journal->j_sb_buffer);
1658 journal->j_tail_sequence = journal->j_transaction_sequence;
1661 jbd2_journal_destroy_revoke(journal);
1662 jbd2_journal_destroy_revoke_record_cache();
1663 jbd2_journal_destroy_revoke_table_cache();
1664 e2fsck_journal_release(ctx, journal, 1, 0);
1668 errcode_t e2fsck_run_ext3_journal(e2fsck_t ctx)
1670 io_manager io_ptr = ctx->fs->io->manager;
1671 int blocksize = ctx->fs->blocksize;
1672 errcode_t retval, recover_retval;
1674 unsigned long long kbytes_written = 0;
1676 printf(_("%s: recovering journal\n"), ctx->device_name);
1677 if (ctx->options & E2F_OPT_READONLY) {
1678 printf(_("%s: won't do journal recovery while read-only\n"),
1680 return EXT2_ET_FILE_RO;
1683 if (ctx->fs->flags & EXT2_FLAG_DIRTY)
1684 ext2fs_flush(ctx->fs); /* Force out any modifications */
1686 recover_retval = recover_ext3_journal(ctx);
1689 * Reload the filesystem context to get up-to-date data from disk
1690 * because journal recovery will change the filesystem under us.
1692 if (ctx->fs->super->s_kbytes_written &&
1693 ctx->fs->io->manager->get_stats)
1694 ctx->fs->io->manager->get_stats(ctx->fs->io, &stats);
1695 if (stats && stats->bytes_written)
1696 kbytes_written = stats->bytes_written >> 10;
1698 ext2fs_mmp_stop(ctx->fs);
1699 ext2fs_free(ctx->fs);
1700 retval = ext2fs_open(ctx->filesystem_name, ctx->openfs_flags,
1701 ctx->superblock, blocksize, io_ptr,
1704 com_err(ctx->program_name, retval,
1705 _("while trying to re-open %s"),
1707 fatal_error(ctx, 0);
1709 ctx->fs->priv_data = ctx;
1710 ctx->fs->now = ctx->now;
1711 ctx->fs->flags |= EXT2_FLAG_MASTER_SB_ONLY;
1712 ctx->fs->super->s_kbytes_written += kbytes_written;
1714 /* Set the superblock flags */
1715 e2fsck_clear_recover(ctx, recover_retval != 0);
1718 * Do one last sanity check, and propagate journal->s_errno to
1719 * the EXT2_ERROR_FS flag in the fs superblock if needed.
1721 retval = e2fsck_check_ext3_journal(ctx);
1722 return retval ? retval : recover_retval;
1726 * This function will move the journal inode from a visible file in
1727 * the filesystem directory hierarchy to the reserved inode if necessary.
1729 static const char * const journal_names[] = {
1730 ".journal", "journal", ".journal.dat", "journal.dat", 0 };
1732 void e2fsck_move_ext3_journal(e2fsck_t ctx)
1734 struct ext2_super_block *sb = ctx->fs->super;
1735 struct problem_context pctx;
1736 struct ext2_inode inode;
1737 ext2_filsys fs = ctx->fs;
1740 const char * const * cpp;
1744 clear_problem_context(&pctx);
1747 * If the filesystem is opened read-only, or there is no
1748 * journal, then do nothing.
1750 if ((ctx->options & E2F_OPT_READONLY) ||
1751 (sb->s_journal_inum == 0) ||
1752 !ext2fs_has_feature_journal(sb))
1756 * Read in the journal inode
1758 if (ext2fs_read_inode(fs, sb->s_journal_inum, &inode) != 0)
1762 * If it's necessary to backup the journal inode, do so.
1764 if ((sb->s_jnl_backup_type == 0) ||
1765 ((sb->s_jnl_backup_type == EXT3_JNL_BACKUP_BLOCKS) &&
1766 memcmp(inode.i_block, sb->s_jnl_blocks, EXT2_N_BLOCKS*4))) {
1767 if (fix_problem(ctx, PR_0_BACKUP_JNL, &pctx)) {
1768 memcpy(sb->s_jnl_blocks, inode.i_block,
1770 sb->s_jnl_blocks[15] = inode.i_size_high;
1771 sb->s_jnl_blocks[16] = inode.i_size;
1772 sb->s_jnl_backup_type = EXT3_JNL_BACKUP_BLOCKS;
1773 ext2fs_mark_super_dirty(fs);
1774 fs->flags &= ~EXT2_FLAG_MASTER_SB_ONLY;
1779 * If the journal is already the hidden inode, then do nothing
1781 if (sb->s_journal_inum == EXT2_JOURNAL_INO)
1785 * The journal inode had better have only one link and not be readable.
1787 if (inode.i_links_count != 1)
1791 * If the filesystem is mounted, or we can't tell whether
1792 * or not it's mounted, do nothing.
1794 retval = ext2fs_check_if_mounted(ctx->filesystem_name, &mount_flags);
1795 if (retval || (mount_flags & EXT2_MF_MOUNTED))
1799 * If we can't find the name of the journal inode, then do
1802 for (cpp = journal_names; *cpp; cpp++) {
1803 retval = ext2fs_lookup(fs, EXT2_ROOT_INO, *cpp,
1804 strlen(*cpp), 0, &ino);
1805 if ((retval == 0) && (ino == sb->s_journal_inum))
1811 /* We need the inode bitmap to be loaded */
1812 retval = ext2fs_read_bitmaps(fs);
1817 if (!fix_problem(ctx, PR_0_MOVE_JOURNAL, &pctx))
1821 * OK, we've done all the checks, let's actually move the
1822 * journal inode. Errors at this point mean we need to force
1823 * an ext2 filesystem check.
1825 if ((retval = ext2fs_unlink(fs, EXT2_ROOT_INO, *cpp, ino, 0)) != 0)
1827 if ((retval = ext2fs_write_inode(fs, EXT2_JOURNAL_INO, &inode)) != 0)
1829 sb->s_journal_inum = EXT2_JOURNAL_INO;
1830 ext2fs_mark_super_dirty(fs);
1831 fs->flags &= ~EXT2_FLAG_MASTER_SB_ONLY;
1832 inode.i_links_count = 0;
1833 inode.i_dtime = ctx->now;
1834 if ((retval = ext2fs_write_inode(fs, ino, &inode)) != 0)
1837 group = ext2fs_group_of_ino(fs, ino);
1838 ext2fs_unmark_inode_bitmap2(fs->inode_map, ino);
1839 ext2fs_mark_ib_dirty(fs);
1840 ext2fs_bg_free_inodes_count_set(fs, group, ext2fs_bg_free_inodes_count(fs, group) + 1);
1841 ext2fs_group_desc_csum_set(fs, group);
1842 fs->super->s_free_inodes_count++;
1846 pctx.errcode = retval;
1847 fix_problem(ctx, PR_0_ERR_MOVE_JOURNAL, &pctx);
1848 fs->super->s_state &= ~EXT2_VALID_FS;
1849 ext2fs_mark_super_dirty(fs);
1854 * This function makes sure the superblock hint for the external
1855 * journal is correct.
1857 int e2fsck_fix_ext3_journal_hint(e2fsck_t ctx)
1859 struct ext2_super_block *sb = ctx->fs->super;
1860 struct problem_context pctx;
1861 char uuid[37], *journal_name;
1864 if (!ext2fs_has_feature_journal(sb) ||
1865 uuid_is_null(sb->s_journal_uuid))
1868 uuid_unparse(sb->s_journal_uuid, uuid);
1869 journal_name = blkid_get_devname(ctx->blkid, "UUID", uuid);
1873 if (stat(journal_name, &st) < 0) {
1878 if (st.st_rdev != sb->s_journal_dev) {
1879 clear_problem_context(&pctx);
1880 pctx.num = st.st_rdev;
1881 if (fix_problem(ctx, PR_0_EXTERNAL_JOURNAL_HINT, &pctx)) {
1882 sb->s_journal_dev = st.st_rdev;
1883 ext2fs_mark_super_dirty(ctx->fs);