On Mon 06-06-11 10:16:30, Ted Tso wrote:
> On Mon, May 30, 2011 at 05:12:58PM +0200, Jan Kara wrote:
> > /*
> > - * For the unlocked version of this call, also make sure that any
> > - * hanging journal_head is cleaned up if necessary.
> > + * For the unlocked version of this call, also drop buffer_head reference.
> > *
> > * __jbd2_journal_refile_buffer is usually called as part of a single locked
> ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
>
> Doesn't this paragraph refer to jbd2_journal_refile_buffer(), not
> __jbd2_journal_refile_buffer()? Or am I missing something?
Hmm, the comment seems to be wrong. The comment about buffer_head
reference does not apply anymore. I'll fix that.
> > void jbd2_journal_refile_buffer(journal_t *journal, struct journal_head *jh)
> > {
> > struct buffer_head *bh = jh2bh(jh);
> >
> > + /* Get reference so that buffer cannot be freed before we unlock it */
> > + get_bh(bh);
>
> OK, so we're adding a get_bh(bh) call to jbd2_journal_refile_buffer(),
> which we're not freeing later in the function. So this means every
> single place where we call jbd2_journal_refile_buffer(), we'd better
> add put_bh(bh) or bhrelse(bh) call, right?
Well, we are adding get_bh() but we are now also using
jbd2_journal_put_journal_head() instead of
jbd2_journal_remove_journal_head() and the former call does additional
__brelse().
Probably a better way to look at this is that
jbd2_journal_remove_journal_head() now gets a reference and puts it at the
end of that function. jbd2_journal_put_journal_head() cares about releasing
bh reference held by journal_head. So the logic ends up being simpler than
it used to be.
> So in fs/jbd2/commit.c, line 418, in jbd2_journal_commit_transaction(),
> I see a call to jbd2_journal_refile_buffer(), which the patch doesn't
> seem to adjust. Looks like this could cause a buffer leak?
>
> In your testing, have you checked to the slab cache to make sure there
> isn't any memory leakage going on with buffer heads?
Not really - I now ran fsxlinux and fsstress and both the number of
buffer_head and journal_head slabs seem to be under control.
Honza
--
Jan Kara <jack@xxxxxxx>
SUSE Labs, CR
--
To unsubscribe from this list: send the line "unsubscribe linux-ext4" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
