Bug#508866: linux-image-2.6.26-1-amd64: NFS going stale for stat() for renamed files like .Xauthority
On Sun, 28 Dec 2008, Moritz Muehlenhoff wrote:
> On Tue, Dec 16, 2008 at 03:03:39PM +1100, Tim Connors wrote:
> > Package: linux-image-2.6.26-1-amd64
> > Version: 2.6.26-10
> > Severity: normal
> >
> > Identical to ubuntu bug 269954:
> > https://bugs.launchpad.net/ubuntu/+source/linux/+bug/269954
> >
> > I've got a 2.6.26-1-amd64 box that is getting the occasional stale
> > return result from an stat() access for .Xauthority, which is a pain
> > because it means I can't open new X clients when this happens. It is
> > rather rare, happening from thrice a day to once per week, and I
> > suspect it depends on the pattern of usage of whether I am logging
> > into other machines often with ssh (which atomically renames
> > .Xauthority on the remote host, giving this bug the opportunity to
> > arise), or not.
> >
> > Opening an xterm locally on the 2.6.26-1 client does this:
> >
>
> [..]
>
> >
> > The ubuntu bug says that this started in 2.6.24, and still exists (in
> > somewhat mitigated form) in 2.6.27. I haven't tried mainline kernels
> > yet because this is a production box, although this thread doesn't
> > give me much encouragement that Trond has even been able to reproduce
> > the bug: http://marc.info/?t=121316971700004&r=1&w=2
>
> Given that this is a longstanding problem and subtle to catch I'd
> suggest to try to reproduce this with 2.6.28 and to open a bugreport
> at bugzilla.kernel.org if it can be reproduced with 2.6.28. This way
> all information on this issue is collected in a single place and easier
> to track for the NFS developers.
I have done only a tiny bit of testing so far (and I am on leave from
today to go to the LCA, so won't be at work in the environment where I was
triggering this bug, so won't be able to test this properly for another
couple of weeks), but a semi-reliable way of triggering the bug didn't
trigger the couple of times I tested it.
However, since bisecting where the alleged fix was introduced will take
some time, getting the fix into lenny might be an issue given that lenny
will ship with the buggy 2.6.26...
I do think this is a rather important fix to get into lenny though, since
it will be embarrassing if debian stable won't be fit to run in
environments with nfs hosted home directories - precisely the environment
where sensible people use debian stable instead of crappy distributions
like redhat :)
Anyone got time to search through the kernel changelogs for 2.6.28 to work
out where the fix (if it exists) was introduced? I've got to pack my
bags...
--
TimC
Octopuses don't like astro turf much. That's a *great* piece of trivia
to drop into conversations. I must remember it.
-- Lloyd Gilbert in AFAFDA
Reply to: