On Mon 14-10-13 21:50:54, Ming Lei wrote:
> On Mon, Oct 14, 2013 at 8:58 PM, Jan Kara <jack@xxxxxxx> wrote:
> > Umm, I guess I see what are you pointing at. Thanks for catching that.
> > mpage_process_page_bufs() always adds a buffer to mpd even if nr_to_write
> > is already <= 0. But I would somewhat prefer not to call
> > mpage_prepare_extent_to_map() at all when nr_to_write <= 0. So a patch
> > like:
> > ret = mpage_prepare_extent_to_map(&mpd);
> > if (!ret) {
> > - if (mpd.map.m_len)
> > + if (mpd.map.m_len) {
> > ret = mpage_map_and_submit_extent(handle, &mpd,
> > &give_up_on_write);
> > - else {
> > + done = (wbc->nr_to_write <= 0);
> > + } else {
> >
> > Should also fix your problem, am I right?
>
> I am afraid it can't, because we need to stop scanning page cache
> if end of file is reached.
That should be OK. mpage_process_page_bufs() won't add a buffer beyond
EOF so we end the extent at EOF and next time we don't add anything to the
extent. My change wouldn't change anything in this.
> nr_to_write will become negative inside mpage_map_and_submit_extent(),
> that is why I fix it inside mpage_prepare_extent_to_map().
Yes, mpage_map_and_submit_extent() creates negative nr_to_write but only
because mpage_prepare_extent_to_map() asked for mapping too big extent. But
if I'm reading the code correctly we first ask for writing the extent of
just the right size (nr_to_write becomes 0) but then ext4_writepages() asks
mpage_prepare_extent_to_map() for another extent and it will create a single
page extent for mapping before it finds out nr_to_write <= 0 and
terminates. Am I understanding the problem correctly?
After thinking about it again, moving the condition in
mpage_prepare_extent_to_map() as you did is probably better that what I
suggested. Just move it more up in the loop - like after page->index > end
condition. So that we don't unnecessarily lock the page etc.
Honza
--
Jan Kara <jack@xxxxxxx>
SUSE Labs, CR
--
To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
