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 #ifdef CONFIG_JBD_DEBUG /* Enabled by configure --enable-jfs-debug */
31 static int bh_count = 0;
35 * Define USE_INODE_IO to use the inode_io.c / fileio.c codepaths.
36 * This creates a larger static binary, and a smaller binary using
37 * shared libraries. It's also probably slightly less CPU-efficient,
38 * which is why it's not on by default. But, it's a good way of
39 * testing the functions in inode_io.c and fileio.c.
43 /* Kernel compatibility functions for handling the journal. These allow us
44 * to use the recovery.c file virtually unchanged from the kernel, so we
45 * don't have to do much to keep kernel and user recovery in sync.
47 int journal_bmap(journal_t *journal, blk64_t block, unsigned long long *phys)
53 struct inode *inode = journal->j_inode;
62 retval= ext2fs_bmap2(inode->i_ctx->fs, inode->i_ino,
63 &inode->i_ext2, NULL, 0, block, 0, &pblk);
69 struct buffer_head *getblk(kdev_t kdev, blk64_t blocknr, int blocksize)
71 struct buffer_head *bh;
72 int bufsize = sizeof(*bh) + kdev->k_ctx->fs->blocksize -
75 bh = e2fsck_allocate_memory(kdev->k_ctx, bufsize, "block buffer");
79 #ifdef CONFIG_JBD_DEBUG
80 if (journal_enable_debug >= 3)
83 jfs_debug(4, "getblk for block %llu (%d bytes)(total %d)\n",
84 (unsigned long long) blocknr, blocksize, bh_count);
86 bh->b_ctx = kdev->k_ctx;
87 if (kdev->k_dev == K_DEV_FS)
88 bh->b_io = kdev->k_ctx->fs->io;
90 bh->b_io = kdev->k_ctx->journal_io;
91 bh->b_size = blocksize;
92 bh->b_blocknr = blocknr;
97 void sync_blockdev(kdev_t kdev)
101 if (kdev->k_dev == K_DEV_FS)
102 io = kdev->k_ctx->fs->io;
104 io = kdev->k_ctx->journal_io;
106 io_channel_flush(io);
109 void ll_rw_block(int rw, int nr, struct buffer_head *bhp[])
112 struct buffer_head *bh;
114 for (; nr > 0; --nr) {
116 if (rw == READ && !bh->b_uptodate) {
117 jfs_debug(3, "reading block %llu/%p\n",
118 bh->b_blocknr, (void *) bh);
119 retval = io_channel_read_blk64(bh->b_io,
123 com_err(bh->b_ctx->device_name, retval,
124 "while reading block %llu\n",
130 } else if (rw == WRITE && bh->b_dirty) {
131 jfs_debug(3, "writing block %llu/%p\n",
134 retval = io_channel_write_blk64(bh->b_io,
138 com_err(bh->b_ctx->device_name, retval,
139 "while writing block %llu\n",
147 jfs_debug(3, "no-op %s for block %llu\n",
148 rw == READ ? "read" : "write",
154 void mark_buffer_dirty(struct buffer_head *bh)
159 static void mark_buffer_clean(struct buffer_head * bh)
164 void brelse(struct buffer_head *bh)
167 ll_rw_block(WRITE, 1, &bh);
168 jfs_debug(3, "freeing block %llu/%p (total %d)\n",
169 bh->b_blocknr, (void *) bh, --bh_count);
170 ext2fs_free_mem(&bh);
173 int buffer_uptodate(struct buffer_head *bh)
175 return bh->b_uptodate;
178 void mark_buffer_uptodate(struct buffer_head *bh, int val)
180 bh->b_uptodate = val;
183 void wait_on_buffer(struct buffer_head *bh)
186 ll_rw_block(READ, 1, &bh);
190 static void e2fsck_clear_recover(e2fsck_t ctx, int error)
192 ctx->fs->super->s_feature_incompat &= ~EXT3_FEATURE_INCOMPAT_RECOVER;
194 /* if we had an error doing journal recovery, we need a full fsck */
196 ctx->fs->super->s_state &= ~EXT2_VALID_FS;
197 ext2fs_mark_super_dirty(ctx->fs);
201 * This is a helper function to check the validity of the journal.
203 struct process_block_struct {
204 e2_blkcnt_t last_block;
207 static int process_journal_block(ext2_filsys fs,
209 e2_blkcnt_t blockcnt,
210 blk64_t ref_block EXT2FS_ATTR((unused)),
211 int ref_offset EXT2FS_ATTR((unused)),
214 struct process_block_struct *p;
215 blk64_t blk = *block_nr;
217 p = (struct process_block_struct *) priv_data;
219 if (!blk || blk < fs->super->s_first_data_block ||
220 blk >= ext2fs_blocks_count(fs->super))
224 p->last_block = blockcnt;
228 static errcode_t e2fsck_get_journal(e2fsck_t ctx, journal_t **ret_journal)
230 struct process_block_struct pb;
231 struct ext2_super_block *sb = ctx->fs->super;
232 struct ext2_super_block jsuper;
233 struct problem_context pctx;
234 struct buffer_head *bh;
235 struct inode *j_inode = NULL;
236 struct kdev_s *dev_fs = NULL, *dev_journal;
237 const char *journal_name = 0;
238 journal_t *journal = NULL;
239 errcode_t retval = 0;
240 io_manager io_ptr = 0;
241 unsigned long long start = 0;
243 int tried_backup_jnl = 0;
245 clear_problem_context(&pctx);
247 journal = e2fsck_allocate_memory(ctx, sizeof(journal_t), "journal");
249 return EXT2_ET_NO_MEMORY;
252 dev_fs = e2fsck_allocate_memory(ctx, 2*sizeof(struct kdev_s), "kdev");
254 retval = EXT2_ET_NO_MEMORY;
257 dev_journal = dev_fs+1;
259 dev_fs->k_ctx = dev_journal->k_ctx = ctx;
260 dev_fs->k_dev = K_DEV_FS;
261 dev_journal->k_dev = K_DEV_JOURNAL;
263 journal->j_dev = dev_journal;
264 journal->j_fs_dev = dev_fs;
265 journal->j_inode = NULL;
266 journal->j_blocksize = ctx->fs->blocksize;
268 if (uuid_is_null(sb->s_journal_uuid)) {
269 if (!sb->s_journal_inum) {
270 retval = EXT2_ET_BAD_INODE_NUM;
273 j_inode = e2fsck_allocate_memory(ctx, sizeof(*j_inode),
276 retval = EXT2_ET_NO_MEMORY;
280 j_inode->i_ctx = ctx;
281 j_inode->i_ino = sb->s_journal_inum;
283 if ((retval = ext2fs_read_inode(ctx->fs,
285 &j_inode->i_ext2))) {
287 if (sb->s_jnl_backup_type != EXT3_JNL_BACKUP_BLOCKS ||
290 memset(&j_inode->i_ext2, 0, sizeof(struct ext2_inode));
291 memcpy(&j_inode->i_ext2.i_block[0], sb->s_jnl_blocks,
293 j_inode->i_ext2.i_size_high = sb->s_jnl_blocks[15];
294 j_inode->i_ext2.i_size = sb->s_jnl_blocks[16];
295 j_inode->i_ext2.i_links_count = 1;
296 j_inode->i_ext2.i_mode = LINUX_S_IFREG | 0600;
297 e2fsck_use_inode_shortcuts(ctx, 1);
298 ctx->stashed_ino = j_inode->i_ino;
299 ctx->stashed_inode = &j_inode->i_ext2;
302 if (!j_inode->i_ext2.i_links_count ||
303 !LINUX_S_ISREG(j_inode->i_ext2.i_mode)) {
304 retval = EXT2_ET_NO_JOURNAL;
305 goto try_backup_journal;
307 if (EXT2_I_SIZE(&j_inode->i_ext2) / journal->j_blocksize <
308 JFS_MIN_JOURNAL_BLOCKS) {
309 retval = EXT2_ET_JOURNAL_TOO_SMALL;
310 goto try_backup_journal;
313 retval = ext2fs_block_iterate3(ctx->fs, j_inode->i_ino,
315 process_journal_block, &pb);
316 if ((pb.last_block + 1) * ctx->fs->blocksize <
317 (int) EXT2_I_SIZE(&j_inode->i_ext2)) {
318 retval = EXT2_ET_JOURNAL_TOO_SMALL;
319 goto try_backup_journal;
321 if (tried_backup_jnl && !(ctx->options & E2F_OPT_READONLY)) {
322 retval = ext2fs_write_inode(ctx->fs, sb->s_journal_inum,
328 journal->j_maxlen = EXT2_I_SIZE(&j_inode->i_ext2) /
329 journal->j_blocksize;
332 retval = ext2fs_inode_io_intern2(ctx->fs, sb->s_journal_inum,
338 io_ptr = inode_io_manager;
340 journal->j_inode = j_inode;
341 ctx->journal_io = ctx->fs->io;
342 if ((retval = journal_bmap(journal, 0, &start)) != 0)
347 if (!ctx->journal_name) {
350 uuid_unparse(sb->s_journal_uuid, uuid);
351 ctx->journal_name = blkid_get_devname(ctx->blkid,
353 if (!ctx->journal_name)
354 ctx->journal_name = blkid_devno_to_devname(sb->s_journal_dev);
356 journal_name = ctx->journal_name;
359 fix_problem(ctx, PR_0_CANT_FIND_JOURNAL, &pctx);
360 retval = EXT2_ET_LOAD_EXT_JOURNAL;
364 jfs_debug(1, "Using journal file %s\n", journal_name);
365 io_ptr = unix_io_manager;
369 test_io_backing_manager = io_ptr;
370 io_ptr = test_io_manager;
376 int flags = IO_FLAG_RW;
377 if (!(ctx->mount_flags & EXT2_MF_ISROOT &&
378 ctx->mount_flags & EXT2_MF_READONLY))
379 flags |= IO_FLAG_EXCLUSIVE;
380 if ((ctx->mount_flags & EXT2_MF_READONLY) &&
381 (ctx->options & E2F_OPT_FORCE))
382 flags &= ~IO_FLAG_EXCLUSIVE;
385 retval = io_ptr->open(journal_name, flags,
391 io_channel_set_blksize(ctx->journal_io, ctx->fs->blocksize);
394 if (ctx->fs->blocksize == 1024)
396 bh = getblk(dev_journal, start, ctx->fs->blocksize);
398 retval = EXT2_ET_NO_MEMORY;
401 ll_rw_block(READ, 1, &bh);
402 if ((retval = bh->b_err) != 0) {
406 memcpy(&jsuper, start ? bh->b_data : bh->b_data + 1024,
409 #ifdef WORDS_BIGENDIAN
410 if (jsuper.s_magic == ext2fs_swab16(EXT2_SUPER_MAGIC))
411 ext2fs_swap_super(&jsuper);
413 if (jsuper.s_magic != EXT2_SUPER_MAGIC ||
414 !(jsuper.s_feature_incompat & EXT3_FEATURE_INCOMPAT_JOURNAL_DEV)) {
415 fix_problem(ctx, PR_0_EXT_JOURNAL_BAD_SUPER, &pctx);
416 retval = EXT2_ET_LOAD_EXT_JOURNAL;
419 /* Make sure the journal UUID is correct */
420 if (memcmp(jsuper.s_uuid, ctx->fs->super->s_journal_uuid,
421 sizeof(jsuper.s_uuid))) {
422 fix_problem(ctx, PR_0_JOURNAL_BAD_UUID, &pctx);
423 retval = EXT2_ET_LOAD_EXT_JOURNAL;
427 journal->j_maxlen = ext2fs_blocks_count(&jsuper);
431 if (!(bh = getblk(dev_journal, start, journal->j_blocksize))) {
432 retval = EXT2_ET_NO_MEMORY;
436 journal->j_sb_buffer = bh;
437 journal->j_superblock = (journal_superblock_t *)bh->b_data;
441 ext2fs_free_mem(&j_inode);
444 *ret_journal = journal;
445 e2fsck_use_inode_shortcuts(ctx, 0);
449 e2fsck_use_inode_shortcuts(ctx, 0);
451 ext2fs_free_mem(&dev_fs);
453 ext2fs_free_mem(&j_inode);
455 ext2fs_free_mem(&journal);
459 static errcode_t e2fsck_journal_fix_bad_inode(e2fsck_t ctx,
460 struct problem_context *pctx)
462 struct ext2_super_block *sb = ctx->fs->super;
463 int recover = ctx->fs->super->s_feature_incompat &
464 EXT3_FEATURE_INCOMPAT_RECOVER;
465 int has_journal = ctx->fs->super->s_feature_compat &
466 EXT3_FEATURE_COMPAT_HAS_JOURNAL;
468 if (has_journal || sb->s_journal_inum) {
469 /* The journal inode is bogus, remove and force full fsck */
470 pctx->ino = sb->s_journal_inum;
471 if (fix_problem(ctx, PR_0_JOURNAL_BAD_INODE, pctx)) {
472 if (has_journal && sb->s_journal_inum)
473 printf("*** ext3 journal has been deleted - "
474 "filesystem is now ext2 only ***\n\n");
475 sb->s_feature_compat &= ~EXT3_FEATURE_COMPAT_HAS_JOURNAL;
476 sb->s_journal_inum = 0;
477 ctx->flags |= E2F_FLAG_JOURNAL_INODE;
478 ctx->fs->flags &= ~EXT2_FLAG_MASTER_SB_ONLY;
479 e2fsck_clear_recover(ctx, 1);
482 return EXT2_ET_BAD_INODE_NUM;
483 } else if (recover) {
484 if (fix_problem(ctx, PR_0_JOURNAL_RECOVER_SET, pctx)) {
485 e2fsck_clear_recover(ctx, 1);
488 return EXT2_ET_UNSUPP_FEATURE;
493 #define V1_SB_SIZE 0x0024
494 static void clear_v2_journal_fields(journal_t *journal)
496 e2fsck_t ctx = journal->j_dev->k_ctx;
497 struct problem_context pctx;
499 clear_problem_context(&pctx);
501 if (!fix_problem(ctx, PR_0_CLEAR_V2_JOURNAL, &pctx))
504 memset(((char *) journal->j_superblock) + V1_SB_SIZE, 0,
505 ctx->fs->blocksize-V1_SB_SIZE);
506 mark_buffer_dirty(journal->j_sb_buffer);
510 static errcode_t e2fsck_journal_load(journal_t *journal)
512 e2fsck_t ctx = journal->j_dev->k_ctx;
513 journal_superblock_t *jsb;
514 struct buffer_head *jbh = journal->j_sb_buffer;
515 struct problem_context pctx;
517 clear_problem_context(&pctx);
519 ll_rw_block(READ, 1, &jbh);
521 com_err(ctx->device_name, jbh->b_err,
522 _("reading journal superblock\n"));
526 jsb = journal->j_superblock;
527 /* If we don't even have JFS_MAGIC, we probably have a wrong inode */
528 if (jsb->s_header.h_magic != htonl(JFS_MAGIC_NUMBER))
529 return e2fsck_journal_fix_bad_inode(ctx, &pctx);
531 switch (ntohl(jsb->s_header.h_blocktype)) {
532 case JFS_SUPERBLOCK_V1:
533 journal->j_format_version = 1;
534 if (jsb->s_feature_compat ||
535 jsb->s_feature_incompat ||
536 jsb->s_feature_ro_compat ||
538 clear_v2_journal_fields(journal);
541 case JFS_SUPERBLOCK_V2:
542 journal->j_format_version = 2;
543 if (ntohl(jsb->s_nr_users) > 1 &&
544 uuid_is_null(ctx->fs->super->s_journal_uuid))
545 clear_v2_journal_fields(journal);
546 if (ntohl(jsb->s_nr_users) > 1) {
547 fix_problem(ctx, PR_0_JOURNAL_UNSUPP_MULTIFS, &pctx);
548 return EXT2_ET_JOURNAL_UNSUPP_VERSION;
553 * These should never appear in a journal super block, so if
554 * they do, the journal is badly corrupted.
556 case JFS_DESCRIPTOR_BLOCK:
557 case JFS_COMMIT_BLOCK:
558 case JFS_REVOKE_BLOCK:
559 return EXT2_ET_CORRUPT_SUPERBLOCK;
561 /* If we don't understand the superblock major type, but there
562 * is a magic number, then it is likely to be a new format we
563 * just don't understand, so leave it alone. */
565 return EXT2_ET_JOURNAL_UNSUPP_VERSION;
568 if (JFS_HAS_INCOMPAT_FEATURE(journal, ~JFS_KNOWN_INCOMPAT_FEATURES))
569 return EXT2_ET_UNSUPP_FEATURE;
571 if (JFS_HAS_RO_COMPAT_FEATURE(journal, ~JFS_KNOWN_ROCOMPAT_FEATURES))
572 return EXT2_ET_RO_UNSUPP_FEATURE;
574 /* We have now checked whether we know enough about the journal
575 * format to be able to proceed safely, so any other checks that
576 * fail we should attempt to recover from. */
577 if (jsb->s_blocksize != htonl(journal->j_blocksize)) {
578 com_err(ctx->program_name, EXT2_ET_CORRUPT_SUPERBLOCK,
579 _("%s: no valid journal superblock found\n"),
581 return EXT2_ET_CORRUPT_SUPERBLOCK;
584 if (ntohl(jsb->s_maxlen) < journal->j_maxlen)
585 journal->j_maxlen = ntohl(jsb->s_maxlen);
586 else if (ntohl(jsb->s_maxlen) > journal->j_maxlen) {
587 com_err(ctx->program_name, EXT2_ET_CORRUPT_SUPERBLOCK,
588 _("%s: journal too short\n"),
590 return EXT2_ET_CORRUPT_SUPERBLOCK;
593 journal->j_tail_sequence = ntohl(jsb->s_sequence);
594 journal->j_transaction_sequence = journal->j_tail_sequence;
595 journal->j_tail = ntohl(jsb->s_start);
596 journal->j_first = ntohl(jsb->s_first);
597 journal->j_last = ntohl(jsb->s_maxlen);
602 static void e2fsck_journal_reset_super(e2fsck_t ctx, journal_superblock_t *jsb,
613 /* Leave a valid existing V1 superblock signature alone.
614 * Anything unrecognisable we overwrite with a new V2
617 if (jsb->s_header.h_magic != htonl(JFS_MAGIC_NUMBER) ||
618 jsb->s_header.h_blocktype != htonl(JFS_SUPERBLOCK_V1)) {
619 jsb->s_header.h_magic = htonl(JFS_MAGIC_NUMBER);
620 jsb->s_header.h_blocktype = htonl(JFS_SUPERBLOCK_V2);
623 /* Zero out everything else beyond the superblock header */
625 p = ((char *) jsb) + sizeof(journal_header_t);
626 memset (p, 0, ctx->fs->blocksize-sizeof(journal_header_t));
628 jsb->s_blocksize = htonl(ctx->fs->blocksize);
629 jsb->s_maxlen = htonl(journal->j_maxlen);
630 jsb->s_first = htonl(1);
632 /* Initialize the journal sequence number so that there is "no"
633 * chance we will find old "valid" transactions in the journal.
634 * This avoids the need to zero the whole journal (slow to do,
635 * and risky when we are just recovering the filesystem).
637 uuid_generate(u.uuid);
638 for (i = 0; i < 4; i ++)
640 jsb->s_sequence = htonl(new_seq);
642 mark_buffer_dirty(journal->j_sb_buffer);
643 ll_rw_block(WRITE, 1, &journal->j_sb_buffer);
646 static errcode_t e2fsck_journal_fix_corrupt_super(e2fsck_t ctx,
648 struct problem_context *pctx)
650 struct ext2_super_block *sb = ctx->fs->super;
651 int recover = ctx->fs->super->s_feature_incompat &
652 EXT3_FEATURE_INCOMPAT_RECOVER;
654 if (sb->s_feature_compat & EXT3_FEATURE_COMPAT_HAS_JOURNAL) {
655 if (fix_problem(ctx, PR_0_JOURNAL_BAD_SUPER, pctx)) {
656 e2fsck_journal_reset_super(ctx, journal->j_superblock,
658 journal->j_transaction_sequence = 1;
659 e2fsck_clear_recover(ctx, recover);
662 return EXT2_ET_CORRUPT_SUPERBLOCK;
663 } else if (e2fsck_journal_fix_bad_inode(ctx, pctx))
664 return EXT2_ET_CORRUPT_SUPERBLOCK;
669 static void e2fsck_journal_release(e2fsck_t ctx, journal_t *journal,
672 journal_superblock_t *jsb;
675 mark_buffer_clean(journal->j_sb_buffer);
676 else if (!(ctx->options & E2F_OPT_READONLY)) {
677 jsb = journal->j_superblock;
678 jsb->s_sequence = htonl(journal->j_transaction_sequence);
680 jsb->s_start = 0; /* this marks the journal as empty */
681 mark_buffer_dirty(journal->j_sb_buffer);
683 brelse(journal->j_sb_buffer);
685 if (ctx->journal_io) {
686 if (ctx->fs && ctx->fs->io != ctx->journal_io)
687 io_channel_close(ctx->journal_io);
692 if (journal->j_inode)
693 ext2fs_free_mem(&journal->j_inode);
695 if (journal->j_fs_dev)
696 ext2fs_free_mem(&journal->j_fs_dev);
697 ext2fs_free_mem(&journal);
701 * This function makes sure that the superblock fields regarding the
702 * journal are consistent.
704 int e2fsck_check_ext3_journal(e2fsck_t ctx)
706 struct ext2_super_block *sb = ctx->fs->super;
708 int recover = ctx->fs->super->s_feature_incompat &
709 EXT3_FEATURE_INCOMPAT_RECOVER;
710 struct problem_context pctx;
712 int reset = 0, force_fsck = 0;
715 /* If we don't have any journal features, don't do anything more */
716 if (!(sb->s_feature_compat & EXT3_FEATURE_COMPAT_HAS_JOURNAL) &&
717 !recover && sb->s_journal_inum == 0 && sb->s_journal_dev == 0 &&
718 uuid_is_null(sb->s_journal_uuid))
721 clear_problem_context(&pctx);
722 pctx.num = sb->s_journal_inum;
724 retval = e2fsck_get_journal(ctx, &journal);
726 if ((retval == EXT2_ET_BAD_INODE_NUM) ||
727 (retval == EXT2_ET_BAD_BLOCK_NUM) ||
728 (retval == EXT2_ET_JOURNAL_TOO_SMALL) ||
729 (retval == EXT2_ET_NO_JOURNAL))
730 return e2fsck_journal_fix_bad_inode(ctx, &pctx);
734 retval = e2fsck_journal_load(journal);
736 if ((retval == EXT2_ET_CORRUPT_SUPERBLOCK) ||
737 ((retval == EXT2_ET_UNSUPP_FEATURE) &&
738 (!fix_problem(ctx, PR_0_JOURNAL_UNSUPP_INCOMPAT,
740 ((retval == EXT2_ET_RO_UNSUPP_FEATURE) &&
741 (!fix_problem(ctx, PR_0_JOURNAL_UNSUPP_ROCOMPAT,
743 ((retval == EXT2_ET_JOURNAL_UNSUPP_VERSION) &&
744 (!fix_problem(ctx, PR_0_JOURNAL_UNSUPP_VERSION, &pctx))))
745 retval = e2fsck_journal_fix_corrupt_super(ctx, journal,
747 e2fsck_journal_release(ctx, journal, 0, 1);
752 * We want to make the flags consistent here. We will not leave with
753 * needs_recovery set but has_journal clear. We can't get in a loop
754 * with -y, -n, or -p, only if a user isn't making up their mind.
757 if (!(sb->s_feature_compat & EXT3_FEATURE_COMPAT_HAS_JOURNAL)) {
758 recover = sb->s_feature_incompat & EXT3_FEATURE_INCOMPAT_RECOVER;
760 if (fix_problem(ctx, PR_0_JOURNAL_HAS_JOURNAL, &pctx)) {
762 !fix_problem(ctx, PR_0_JOURNAL_RECOVER_SET, &pctx))
765 * Need a full fsck if we are releasing a
766 * journal stored on a reserved inode.
768 force_fsck = recover ||
769 (sb->s_journal_inum < EXT2_FIRST_INODE(sb));
770 /* Clear all of the journal fields */
771 sb->s_journal_inum = 0;
772 sb->s_journal_dev = 0;
773 memset(sb->s_journal_uuid, 0,
774 sizeof(sb->s_journal_uuid));
775 e2fsck_clear_recover(ctx, force_fsck);
776 } else if (!(ctx->options & E2F_OPT_READONLY)) {
777 sb->s_feature_compat |= EXT3_FEATURE_COMPAT_HAS_JOURNAL;
778 ctx->fs->flags &= ~EXT2_FLAG_MASTER_SB_ONLY;
779 ext2fs_mark_super_dirty(ctx->fs);
783 if (sb->s_feature_compat & EXT3_FEATURE_COMPAT_HAS_JOURNAL &&
784 !(sb->s_feature_incompat & EXT3_FEATURE_INCOMPAT_RECOVER) &&
785 journal->j_superblock->s_start != 0) {
786 /* Print status information */
787 fix_problem(ctx, PR_0_JOURNAL_RECOVERY_CLEAR, &pctx);
789 problem = PR_0_JOURNAL_RUN_DEFAULT;
791 problem = PR_0_JOURNAL_RUN;
792 if (fix_problem(ctx, problem, &pctx)) {
793 ctx->options |= E2F_OPT_FORCE;
794 sb->s_feature_incompat |=
795 EXT3_FEATURE_INCOMPAT_RECOVER;
796 ext2fs_mark_super_dirty(ctx->fs);
797 } else if (fix_problem(ctx,
798 PR_0_JOURNAL_RESET_JOURNAL, &pctx)) {
800 sb->s_state &= ~EXT2_VALID_FS;
801 ext2fs_mark_super_dirty(ctx->fs);
804 * If the user answers no to the above question, we
805 * ignore the fact that journal apparently has data;
806 * accidentally replaying over valid data would be far
807 * worse than skipping a questionable recovery.
809 * XXX should we abort with a fatal error here? What
810 * will the ext3 kernel code do if a filesystem with
811 * !NEEDS_RECOVERY but with a non-zero
812 * journal->j_superblock->s_start is mounted?
817 * If we don't need to do replay the journal, check to see if
818 * the journal's errno is set; if so, we need to mark the file
819 * system as being corrupt and clear the journal's s_errno.
821 if (!(sb->s_feature_incompat & EXT3_FEATURE_INCOMPAT_RECOVER) &&
822 journal->j_superblock->s_errno) {
823 ctx->fs->super->s_state |= EXT2_ERROR_FS;
824 ext2fs_mark_super_dirty(ctx->fs);
825 journal->j_superblock->s_errno = 0;
826 mark_buffer_dirty(journal->j_sb_buffer);
829 e2fsck_journal_release(ctx, journal, reset, 0);
833 static errcode_t recover_ext3_journal(e2fsck_t ctx)
835 struct problem_context pctx;
839 clear_problem_context(&pctx);
841 journal_init_revoke_caches();
842 retval = e2fsck_get_journal(ctx, &journal);
846 retval = e2fsck_journal_load(journal);
850 retval = journal_init_revoke(journal, 1024);
854 retval = -journal_recover(journal);
858 if (journal->j_failed_commit) {
859 pctx.ino = journal->j_failed_commit;
860 fix_problem(ctx, PR_0_JNL_TXN_CORRUPT, &pctx);
861 journal->j_superblock->s_errno = -EINVAL;
862 mark_buffer_dirty(journal->j_sb_buffer);
866 journal_destroy_revoke(journal);
867 journal_destroy_revoke_caches();
868 e2fsck_journal_release(ctx, journal, 1, 0);
872 int e2fsck_run_ext3_journal(e2fsck_t ctx)
874 io_manager io_ptr = ctx->fs->io->manager;
875 int blocksize = ctx->fs->blocksize;
876 errcode_t retval, recover_retval;
878 unsigned long long kbytes_written = 0;
880 printf(_("%s: recovering journal\n"), ctx->device_name);
881 if (ctx->options & E2F_OPT_READONLY) {
882 printf(_("%s: won't do journal recovery while read-only\n"),
884 return EXT2_ET_FILE_RO;
887 if (ctx->fs->flags & EXT2_FLAG_DIRTY)
888 ext2fs_flush(ctx->fs); /* Force out any modifications */
890 recover_retval = recover_ext3_journal(ctx);
893 * Reload the filesystem context to get up-to-date data from disk
894 * because journal recovery will change the filesystem under us.
896 if (ctx->fs->super->s_kbytes_written &&
897 ctx->fs->io->manager->get_stats)
898 ctx->fs->io->manager->get_stats(ctx->fs->io, &stats);
899 if (stats && stats->bytes_written)
900 kbytes_written = stats->bytes_written >> 10;
902 ext2fs_mmp_stop(ctx->fs);
903 ext2fs_free(ctx->fs);
904 retval = ext2fs_open(ctx->filesystem_name, EXT2_FLAG_RW,
905 ctx->superblock, blocksize, io_ptr,
908 com_err(ctx->program_name, retval,
909 _("while trying to re-open %s"),
913 ctx->fs->priv_data = ctx;
914 ctx->fs->now = ctx->now;
915 ctx->fs->flags |= EXT2_FLAG_MASTER_SB_ONLY;
916 ctx->fs->super->s_kbytes_written += kbytes_written;
918 /* Set the superblock flags */
919 e2fsck_clear_recover(ctx, recover_retval);
922 * Do one last sanity check, and propagate journal->s_errno to
923 * the EXT2_ERROR_FS flag in the fs superblock if needed.
925 retval = e2fsck_check_ext3_journal(ctx);
926 return retval ? retval : recover_retval;
930 * This function will move the journal inode from a visible file in
931 * the filesystem directory hierarchy to the reserved inode if necessary.
933 static const char * const journal_names[] = {
934 ".journal", "journal", ".journal.dat", "journal.dat", 0 };
936 void e2fsck_move_ext3_journal(e2fsck_t ctx)
938 struct ext2_super_block *sb = ctx->fs->super;
939 struct problem_context pctx;
940 struct ext2_inode inode;
941 ext2_filsys fs = ctx->fs;
944 const char * const * cpp;
945 int group, mount_flags;
947 clear_problem_context(&pctx);
950 * If the filesystem is opened read-only, or there is no
951 * journal, then do nothing.
953 if ((ctx->options & E2F_OPT_READONLY) ||
954 (sb->s_journal_inum == 0) ||
955 !(sb->s_feature_compat & EXT3_FEATURE_COMPAT_HAS_JOURNAL))
959 * Read in the journal inode
961 if (ext2fs_read_inode(fs, sb->s_journal_inum, &inode) != 0)
965 * If it's necessary to backup the journal inode, do so.
967 if ((sb->s_jnl_backup_type == 0) ||
968 ((sb->s_jnl_backup_type == EXT3_JNL_BACKUP_BLOCKS) &&
969 memcmp(inode.i_block, sb->s_jnl_blocks, EXT2_N_BLOCKS*4))) {
970 if (fix_problem(ctx, PR_0_BACKUP_JNL, &pctx)) {
971 memcpy(sb->s_jnl_blocks, inode.i_block,
973 sb->s_jnl_blocks[15] = inode.i_size_high;
974 sb->s_jnl_blocks[16] = inode.i_size;
975 sb->s_jnl_backup_type = EXT3_JNL_BACKUP_BLOCKS;
976 ext2fs_mark_super_dirty(fs);
977 fs->flags &= ~EXT2_FLAG_MASTER_SB_ONLY;
982 * If the journal is already the hidden inode, then do nothing
984 if (sb->s_journal_inum == EXT2_JOURNAL_INO)
988 * The journal inode had better have only one link and not be readable.
990 if (inode.i_links_count != 1)
994 * If the filesystem is mounted, or we can't tell whether
995 * or not it's mounted, do nothing.
997 retval = ext2fs_check_if_mounted(ctx->filesystem_name, &mount_flags);
998 if (retval || (mount_flags & EXT2_MF_MOUNTED))
1002 * If we can't find the name of the journal inode, then do
1005 for (cpp = journal_names; *cpp; cpp++) {
1006 retval = ext2fs_lookup(fs, EXT2_ROOT_INO, *cpp,
1007 strlen(*cpp), 0, &ino);
1008 if ((retval == 0) && (ino == sb->s_journal_inum))
1014 /* We need the inode bitmap to be loaded */
1015 retval = ext2fs_read_bitmaps(fs);
1020 if (!fix_problem(ctx, PR_0_MOVE_JOURNAL, &pctx))
1024 * OK, we've done all the checks, let's actually move the
1025 * journal inode. Errors at this point mean we need to force
1026 * an ext2 filesystem check.
1028 if ((retval = ext2fs_unlink(fs, EXT2_ROOT_INO, *cpp, ino, 0)) != 0)
1030 if ((retval = ext2fs_write_inode(fs, EXT2_JOURNAL_INO, &inode)) != 0)
1032 sb->s_journal_inum = EXT2_JOURNAL_INO;
1033 ext2fs_mark_super_dirty(fs);
1034 fs->flags &= ~EXT2_FLAG_MASTER_SB_ONLY;
1035 inode.i_links_count = 0;
1036 inode.i_dtime = ctx->now;
1037 if ((retval = ext2fs_write_inode(fs, ino, &inode)) != 0)
1040 group = ext2fs_group_of_ino(fs, ino);
1041 ext2fs_unmark_inode_bitmap2(fs->inode_map, ino);
1042 ext2fs_mark_ib_dirty(fs);
1043 ext2fs_bg_free_inodes_count_set(fs, group, ext2fs_bg_free_inodes_count(fs, group) + 1);
1044 ext2fs_group_desc_csum_set(fs, group);
1045 fs->super->s_free_inodes_count++;
1049 pctx.errcode = retval;
1050 fix_problem(ctx, PR_0_ERR_MOVE_JOURNAL, &pctx);
1051 fs->super->s_state &= ~EXT2_VALID_FS;
1052 ext2fs_mark_super_dirty(fs);
1057 * This function makes sure the superblock hint for the external
1058 * journal is correct.
1060 int e2fsck_fix_ext3_journal_hint(e2fsck_t ctx)
1062 struct ext2_super_block *sb = ctx->fs->super;
1063 struct problem_context pctx;
1064 char uuid[37], *journal_name;
1067 if (!(sb->s_feature_compat & EXT3_FEATURE_COMPAT_HAS_JOURNAL) ||
1068 uuid_is_null(sb->s_journal_uuid))
1071 uuid_unparse(sb->s_journal_uuid, uuid);
1072 journal_name = blkid_get_devname(ctx->blkid, "UUID", uuid);
1076 if (stat(journal_name, &st) < 0)
1079 if (st.st_rdev != sb->s_journal_dev) {
1080 clear_problem_context(&pctx);
1081 pctx.num = st.st_rdev;
1082 if (fix_problem(ctx, PR_0_EXTERNAL_JOURNAL_HINT, &pctx)) {
1083 sb->s_journal_dev = st.st_rdev;
1084 ext2fs_mark_super_dirty(ctx->fs);