Re: [PATCH 2/4] AFS: Add a function to excise a rejected write from the pagecache

[Andrew Morton <akpm@linux-foundation.org>]

On Fri, 25 May 2007 00:08:43 +0100
David Howells <dhowells@redhat.com> wrote:

> Andrew Morton <akpm@linux-foundation.org> wrote:
> 
> > hm.  I don't see why that race window would be a problem in practice: the
> > page-exciser does a lock_page();wait_on_page_writeback() as normal, then
> > proceeds with its business?
> 
> No.  The page-exciser ends (cancels) PG_writeback, not waits for it (something
> has to clear the flag).  The problem is that the truncation routines may be
> sat there holding a lock on the page whilst waiting for PG_writeback to go
> away - so we have to clear PG_writeback before we can think about getting
> PG_lock:-(

But we already covered that?  Your exciser can do an unconditional
end_page_writeback(), because it is this thread of control which did the
set_page_writeback().  So we end up with:

	end_page_writeback(page);
	lock_page(page);
	wait_on_page_writeback(page);
	<now nail it>

> > But given that this doesn't work right for some reason, can we use PG_error
> > and then handle that appropriately in the filesystem's ->prepare_write() and
> > ->page_mkwrite()?
> 
> Possibly, though I'd rather they didn't see such a page.

Well someone needs to be taught all about this case.  Question is, should
it be the VFS, or should it just be the address_space(s) which brought
this state about, and which care about it?