jbd2: fix hung processes in jbd2_journal_lock_updates()
Toshiyuki Okajima found out that when running
for ((i=0; i < 100000; i++)); do
if ((i%2 == 0)); then
chattr +j /mnt/file
else
chattr -j /mnt/file
fi
echo "0" >> /mnt/file
done
process sometimes hangs indefinitely in jbd2_journal_lock_updates().
Toshiyuki identified that the following race happens:
jbd2_journal_lock_updates() |jbd2_journal_stop()
---------------------------------------+---------------------------------------
write_lock(&journal->j_state_lock) | .
++journal->j_barrier_count | .
spin_lock(&tran->t_handle_lock) | .
atomic_read(&tran->t_updates) //not 0 |
| atomic_dec_and_test(&tran->t_updates)
| // t_updates = 0
| wake_up(&journal->j_wait_updates)
prepare_to_wait() | // no process is woken up.
spin_unlock(&tran->t_handle_lock) |
write_unlock(&journal->j_state_lock) |
schedule() // never return |
We fix the problem by first calling prepare_to_wait() and only after that
checking t_updates in jbd2_journal_lock_updates().
Reported-and-analyzed-by: Toshiyuki Okajima <toshi.okajima@jp.fujitsu.com>
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
diff --git a/fs/jbd2/transaction.c b/fs/jbd2/transaction.c
index a0e41a4..35ae096 100644
--- a/fs/jbd2/transaction.c
+++ b/fs/jbd2/transaction.c
@@ -517,12 +517,13 @@
break;
spin_lock(&transaction->t_handle_lock);
- if (!atomic_read(&transaction->t_updates)) {
- spin_unlock(&transaction->t_handle_lock);
- break;
- }
prepare_to_wait(&journal->j_wait_updates, &wait,
TASK_UNINTERRUPTIBLE);
+ if (!atomic_read(&transaction->t_updates)) {
+ spin_unlock(&transaction->t_handle_lock);
+ finish_wait(&journal->j_wait_updates, &wait);
+ break;
+ }
spin_unlock(&transaction->t_handle_lock);
write_unlock(&journal->j_state_lock);
schedule();