1 jbd2: fix FS corruption possibility in jbd2_journal_destroy() on umount path
3 From: OGAWA Hirofumi <hirofumi@mail.parknet.co.jp>
5 On umount path, jbd2_journal_destroy() writes latest transaction ID
6 (->j_tail_sequence) to be used at next mount.
8 The bug is that ->j_tail_sequence is not holding latest transaction ID
9 in some cases. So, at next mount, there is chance to conflict with
10 remaining (not overwritten yet) transactions.
13 write transaction (id=11)
14 write transaction (id=12)
15 umount (id=10) <= the bug doesn't write latest ID
18 write transaction (id=11)
24 transaction (id=12) <= valid transaction ID, but old commit
27 Like above, this bug become the cause of recovery failure, or FS
30 So why ->j_tail_sequence doesn't point latest ID?
32 Because if checkpoint transactions was reclaimed by memory pressure
33 (i.e. bdev_try_to_free_page()), then ->j_tail_sequence is not updated.
34 (And another case is, __jbd2_journal_clean_checkpoint_list() is called
35 with empty transaction.)
37 So in above cases, ->j_tail_sequence is not pointing latest
38 transaction ID at umount path. Plus, REQ_FLUSH for checkpoint is not
41 So, to fix this problem with minimum changes, this patch updates
42 ->j_tail_sequence, and issue REQ_FLUSH. (With more complex changes,
43 some optimizations would be possible to avoid unnecessary REQ_FLUSH
48 journal->j_tail_sequence =
49 ++journal->j_transaction_sequence;
51 Increment of ->j_transaction_sequence seems to be unnecessary, but
54 Signed-off-by: OGAWA Hirofumi <hirofumi@mail.parknet.co.jp>
55 Signed-off-by: Theodore Ts'o <tytso@mit.edu>
56 Cc: stable@vger.kernel.org
59 fs/jbd2/journal.c | 17 ++++++++++++-----
60 1 file changed, 12 insertions(+), 5 deletions(-)
62 diff -puN fs/jbd2/journal.c~ext4-umount-fix fs/jbd2/journal.c
63 --- linux/fs/jbd2/journal.c~ext4-umount-fix 2016-02-28 10:16:41.997093008 +0900
64 +++ linux-hirofumi/fs/jbd2/journal.c 2016-02-28 23:01:25.865647233 +0900
65 @@ -1408,11 +1408,12 @@ out:
67 * jbd2_mark_journal_empty() - Mark on disk journal as empty.
68 * @journal: The journal to update.
69 + * @write_op: With which operation should we write the journal sb
71 * Update a journal's dynamic superblock fields to show that journal is empty.
72 * Write updated superblock to disk waiting for IO to complete.
74 -static void jbd2_mark_journal_empty(journal_t *journal)
75 +static void jbd2_mark_journal_empty(journal_t *journal, int write_op)
77 journal_superblock_t *sb = journal->j_superblock;
79 @@ -1430,7 +1431,7 @@ static void jbd2_mark_journal_empty(jour
80 sb->s_start = cpu_to_be32(0);
81 read_unlock(&journal->j_state_lock);
83 - jbd2_write_superblock(journal, WRITE_FUA);
84 + jbd2_write_superblock(journal, write_op);
86 /* Log is no longer empty */
87 write_lock(&journal->j_state_lock);
88 @@ -1716,7 +1717,13 @@ int jbd2_journal_destroy(journal_t *jour
89 if (journal->j_sb_buffer) {
90 if (!is_journal_aborted(journal)) {
91 mutex_lock(&journal->j_checkpoint_mutex);
92 - jbd2_mark_journal_empty(journal);
94 + write_lock(&journal->j_state_lock);
95 + journal->j_tail_sequence =
96 + ++journal->j_transaction_sequence;
97 + write_unlock(&journal->j_state_lock);
99 + jbd2_mark_journal_empty(journal, WRITE_FLUSH_FUA);
100 mutex_unlock(&journal->j_checkpoint_mutex);
103 @@ -1975,7 +1982,7 @@ int jbd2_journal_flush(journal_t *journa
104 * the magic code for a fully-recovered superblock. Any future
105 * commits of data to the journal will restore the current
107 - jbd2_mark_journal_empty(journal);
108 + jbd2_mark_journal_empty(journal, WRITE_FUA);
109 mutex_unlock(&journal->j_checkpoint_mutex);
110 write_lock(&journal->j_state_lock);
111 J_ASSERT(!journal->j_running_transaction);
112 @@ -2021,7 +2028,7 @@ int jbd2_journal_wipe(journal_t *journal
114 /* Lock to make assertions happy... */
115 mutex_lock(&journal->j_checkpoint_mutex);
116 - jbd2_mark_journal_empty(journal);
117 + jbd2_mark_journal_empty(journal, WRITE_FUA);
118 mutex_unlock(&journal->j_checkpoint_mutex);