Re: [patch] fs: improved handling of page and buffer IO errors

[Nick Piggin <npiggin@suse.de>]

On Thu, Oct 23, 2008 at 10:59:49AM +0100, steve@chygwyn.com wrote:
> > Btw, at least for the readpage case, a return of AOP_TRUNCATED_PAGE should
> > be checked for, which would indicate (along with !PageUptodate()) whether we
> > need to retry the read. page_mkwrite though, as you point out, is a
> > different story.
> > 	--Mark
> >
> Yes, and although I probably didn't make it clear I was thinking
> specifically of the page fault path there where both readpage and
> page_mkwrite hang out.
> 
> Also, I've looked through all the current GFS2 code and it seems to
> be correct in relation to Miklos' point on PageUptodate() vs
> page->mapping == NULL so I don't think any changes are required there,
> but obviously that needs to be taken into account in filemap_fault wrt
> to retrying in the lock demotion case. In other words we should be
> testing for page->mapping == NULL rather than !PageUptodate() in that
> case,

PageUptodate is OK for the filemap_fault check AFAIKS, because it does
a find_lock_page and runs the check under lock (so it can't be truncated
or invalidated), in order to prevent fault vs truncate / invalidate races.