On Mon 09-05-11 13:39:07, Allison Henderson wrote:
> On 5/9/2011 7:42 AM, Jan Kara wrote:
> >On Mon 09-05-11 10:22:37, Ted Tso wrote:
> >>On Mon, May 09, 2011 at 04:05:37PM +0200, Jan Kara wrote:
> >>> Yes. ext4_append() can return ENOSPC and passed bh will get set to NULL
> >>>without being marked dirty.
> >>
> >>Ah, so the right fix then is to add to make the cleanup code like this:
> >>
> >> ext4_mark_inode_dirty(handle, dir);
> >> ext4_handle_dirty_metadata(handle, dir, frame->bh);
> >>+ ext4_handle_dirty_metadata(handle, dir, bh2);
> >>+ if (bh)
> >>+ ext4_handle_dirty_metadata(handle, dir, bh);
> >> dx_release(frames);
> >> return retval;
> >>
> >>Agreed?
> > Not quite. make_indexed_dir() does frame->bh = bh and bh = bh2 before
> >calling do_split(). So bh2 is not really carrying a valid buffer reference
> >at this point - even more so because do_split() does brelse() on the passed
> >bh so it need not be around when are at this point. The code is a real
> >mess. But for example attached patch will work because both callers of
> >do_split() do brelse() anyway.
> >
> > Honza
>
> Hi all,
>
> Oh, I understand the problem now. Sorry for the late response, I
> had to stop and dig around with this one for a bit. Would people
> prefer to add the new code before the do_split like this:
>
> + ext4_handle_dirty_metadata(handle, dir, frame->bh);
> + ext4_handle_dirty_metadata(handle, dir, bh);
> +
> de = do_split(handle,dir, &bh, frame, &hinfo, &retval);
> if (!de) {
> /*
> @@ -1421,8 +1425,6 @@ static int make_indexed_dir(handle_t *handle,
> struct dentry *dentry,
> * with corrupted filesystem.
> */
> ext4_mark_inode_dirty(handle, dir);
> - ext4_handle_dirty_metadata(handle, dir, frame->bh);
> - ext4_handle_dirty_metadata(handle, dir, bh);
> dx_release(frames);
> return retval;
> }
This would be fine with me as well. It has a slight overhead of marking
buffer dirty twice but I don't think it really matters.
> I've tested both patches and they both seem to resolve the null
> pointer. The only other solution that comes to mind would be to add
> flags to the do_split to skip the brelse or to do the mark dirty
> before the brelse as you suggest.
Yes, I don't mind which of them Ted chooses...
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
