Red Hat Bugzilla – Bug 989044
gssproxy-0.2.3-6.fc19.x86_64 & kernel-3.10.3-300.fc19.x86_64: BUG: unable to handle kernel NULL pointer dereference at 0000000000000009
Last modified: 2013-08-15 05:59:23 EDT
Created attachment 778966 [details]
gssproxy-0.2.3-6.fc19.x86_64 causes a kernel Oops with kernel-3.10.3-300.fc19.x86_64. This doesn't happen with the previous kernel-3.9.9-302.fc19.x86_64.
It appears that recent upstream changes in gssproxy might resolve this issue:
Would it be possible to release an updated gssproxy to test?
I've attached a trace.
Nevermind pushing out a quick update, I've built one myself that updates gssproxy with all the changes in the master branch.
I still get the Oops when clients try to mount. I've attached the TAKE_2 trace.
Created attachment 778967 [details]
We haven't changed anything in the gssproxy protocol, witness the fact it works with an earlier kernel.
So I suspect breakage in newer kernels.
Bruce can you investigate please?
FWIW -- I cannot boot 3.10.3-300 at all (i7-2600K / dz68bc m.b.).
Kernel panic/No logs saved, but I probably can take picture of the screen
if this is of any interest.
Thanks I can reproduce this (or something similar) now, so should have a fix soon.
(In reply to J. Bruce Fields from comment #5)
> Thanks I can reproduce this (or something similar) now, so should have a fix
Thanks, Bruce. Where should I be looking for the fix? I'd be willing to test out a Koji scratch build if that would help you confirm.
Created attachment 781378 [details]
fix gss-proxy bugs
Apologies, this was completely my mistake: there were two bugs found during the Fedora backport that I somehow forgot to get upstream.
Also I found two more bugs in the upstream.
A combined diff that should be sufficient for Fedora is attached.
Tomorrow I'll do a more careful check of what needs to go upstream and what to Fedora.
Bruce, I see you posted "gss-proxy fixes for 3.11" to the linux-nfs list. The patches seem to relate to this issue. Will the Fedora 3.10 kernels get a backport of these patches? 3.11 seems a long way off.
(In reply to Anthony Messina from comment #8)
> Bruce, I see you posted "gss-proxy fixes for 3.11" to the linux-nfs list.
> The patches seem to relate to this issue. Will the Fedora 3.10 kernels get
> a backport of these patches? 3.11 seems a long way off.
The first three are marked for stable, which means as soon as Linus pulls them those three should get pulled for 3.10.x--no need to wait for the 3.11 release.
If we need the fix sooner, those three patches could also be backported to Fedora now.
My attempt at a Koji scratch build: http://koji.fedoraproject.org/koji/taskinfo?taskID=5694191
(In reply to J. Bruce Fields from comment #10)
> My attempt at a Koji scratch build:
Your scratch build does resolve the issue reported here. Thanks. I'd love to see the patches backported to Fedora's 3.10.x series.
Reassigning to the kernel component where it belongs.
Can we get a kernel maintainer to include this patches for the next Fedora kernel build ?
Is there a chance to get this fix backported to the 3.10.6 Fedora kernels being worked on today?
(In reply to Anthony Messina from comment #16)
> Is there a chance to get this fix backported to the 3.10.6 Fedora kernels
> being worked on today?
Duh! Nevermind. I see it should be included from upstream (https://www.kernel.org/pub/linux/kernel/v3.x/ChangeLog-3.10.6). Thanks Bruce.
This issue is resolved with 3.10.6-200.fc19.x86_64. Thanks to all of you.