Re: [PATCH] nfs: fix NR_FILE_DIRTY underflow

[Andrew Morton <akpm@osdl.org>]

On Wed, 13 Dec 2006 17:29:21 -0800
Andrew Morton <akpm@osdl.org> wrote:

> a) we're now calling try_to_release_page() with a potentially-dirty
>    page, whereas it was previously clean.
> 
>    I wouldn't expect ->releasepage() implementations to go looking at
>    PG_Dirty, because that's not what they're suppoed to be interested in. 
>    But they might do, dunno.

Still an issue, probably minor.

> b) If invalidate_complete_page2() failed due to, say, dirty buffer_heads
>    then we now have a clean page with dirty buffers.  That is an illegal
>    state and the page will leak permanently.
>
>    I _think_ that's what the was_dirty logic is in there for: to
>    preserve the correct page-vs-buffers dirtiness coherency.  But I'd need
>    to do some 2.5.x changelog-dumpster-diving to be sure.

no, that's bs.  The patch looks OK from that POV: try_to_release_page()
will be able to clear clean buffers from a dirty page.

And in fact if it did that, it will then clean the page for us (see
test_clear_page_dirty() in try_to_free_buffers()).

But we still need the clear_page_dirty() in invalidate_complete_page2() in
case we didn't call try_to_release_page() at all.

> Trond, please define precisely and completely and without reference to
> the existing implementation: what behaviour does NFS want?

But this would be nice.