jbd2: Add sanity check for attempts to start handle during umount
authorTheodore Ts'o <tytso@mit.edu>
Thu, 28 Oct 2010 01:30:04 +0000 (21:30 -0400)
committerTheodore Ts'o <tytso@mit.edu>
Thu, 28 Oct 2010 01:30:04 +0000 (21:30 -0400)
An attempt to modify the file system during the call to
jbd2_destroy_journal() can lead to a system lockup.  So add some
checking to make it much more obvious when this happens to and to
determine where the offending code is located.

Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
fs/jbd2/checkpoint.c
fs/jbd2/transaction.c

index 5247e7f..524800d 100644 (file)
@@ -299,6 +299,16 @@ static int __process_buffer(journal_t *journal, struct journal_head *jh,
                transaction->t_chp_stats.cs_forced_to_close++;
                spin_unlock(&journal->j_list_lock);
                jbd_unlock_bh_state(bh);
+               if (unlikely(journal->j_flags & JBD2_UNMOUNT))
+                       /*
+                        * The journal thread is dead; so starting and
+                        * waiting for a commit to finish will cause
+                        * us to wait for a _very_ long time.
+                        */
+                       printk(KERN_ERR "JBD2: %s: "
+                              "Waiting for Godot: block %llu\n",
+                              journal->j_devname,
+                              (unsigned long long) bh->b_blocknr);
                jbd2_log_start_commit(journal, tid);
                jbd2_log_wait_commit(journal, tid);
                ret = 1;
index f3479d6..6bf0a24 100644 (file)
@@ -156,6 +156,7 @@ alloc_transaction:
         */
 repeat:
        read_lock(&journal->j_state_lock);
+       BUG_ON(journal->j_flags & JBD2_UNMOUNT);
        if (is_journal_aborted(journal) ||
            (journal->j_errno != 0 && !(journal->j_flags & JBD2_ACK_ERR))) {
                read_unlock(&journal->j_state_lock);