Re: 2.6.16-rc6-mm2

[Trond Myklebust <trond.myklebust@fys.uio.no>]

On Mon, 2006-06-12 at 17:24 -0700, Christoph Lameter wrote:
> I guess this is correct. nfs_clear_page_writeback also ignores the page 
> if req->wb_page == NULL. Trond?

Sorry I'm late in responding. I'm currently on vacation in Iceland.

This is most probably an issue with nfs_cancel_requests(). We shouldn't
be calling nfs_clear_page_writeback() without _first_ decrementing
NR_UNSTABLE. I'll have a look, but I may be a bit unresponsive until I
get to a new place with an internet connection.

Cheers,
  Trond

> On Mon, 12 Jun 2006, Christoph Lameter wrote:
> 
> > On Mon, 12 Jun 2006, Cedric Le Goater wrote:
> > 
> > > Unable to handle kernel NULL pointer dereference at 0000000000000007 RIP:
> > >  [<ffffffff8025b017>] dec_zone_page_state+0x1/0x5b
> > 
> > Seems that req->wb_page may be NULL.
> > 
> > This patch may fix it but we may miss an unstable page then. We may 
> > have to move the decrement of NR_UNSTABLE to a different location when
> > wb_page is still valid.
> > 
> > Index: linux-2.6.17-rc6-cl/fs/nfs/write.c
> > ===================================================================
> > --- linux-2.6.17-rc6-cl.orig/fs/nfs/write.c	2006-06-12 13:37:47.321243148 -0700
> > +++ linux-2.6.17-rc6-cl/fs/nfs/write.c	2006-06-12 15:13:48.020908204 -0700
> > @@ -1419,7 +1419,8 @@ static void nfs_commit_done(struct rpc_t
> >  		nfs_mark_request_dirty(req);
> >  	next:
> >  		nfs_clear_page_writeback(req);
> > -		dec_zone_page_state(req->wb_page, NR_UNSTABLE);
> > +		if (req->wb_page)
> > +			dec_zone_page_state(req->wb_page, NR_UNSTABLE);
> >  	}
> >  }
> >  
> > 
> >