Re: [PATCH] jbd2: Fix a race between checkpointing code and journal_get_write_access()

[Jan Kara <jack@suse.cz>]

On Mon 13-07-09 10:44:47, Theodore Tso wrote:
> On Mon, Jul 13, 2009 at 03:29:58PM +0200, Jan Kara wrote:
> >   I've been doing some more tests here and there's one more path which emits
> > the warning when deleting symlinks (falsely this time). I'll send you the fix
> > when I verify that I cannot trigger the warning anymore.
> 
> Ah, that would explain why I'm stil seeing the warning when fsstress
> runs.  Was it doing it on long symlinks or short symlinks, out of
> curiosity?
  Long symlinks - the problem was that __dispose_buffer() called from
ext4_invalidatepage() did __jbd2_journal_unfile_buffer() (jbddirty bit was
moved to dirty bit) and then __jbd2_journal_file_buffer() which complained.
As we cleared the jbddirty bit immediately afterwards it was quite
straightforward to fix.  I've audited all other occurences of
jbd2_journal_unfile_buffer() and we discard the buffer head after all other
calls...

								Honza
-- 
Jan Kara <jack@suse.cz>
SUSE Labs, CR