[ovs-discuss] Kernel Crash with OVS 1.4.6 and Linux 3.5

Jesse Gross jesse at nicira.com
Wed Jan 8 01:34:27 UTC 2014


On Tue, Jan 7, 2014 at 11:07 AM, Chris J Arges
<chris.j.arges at canonical.com> wrote:
> On 01/06/2014 01:55 PM, Jesse Gross wrote:
>> On Mon, Jan 6, 2014 at 11:37 AM, Chris J Arges
>> <chris.j.arges at canonical.com> wrote:
>>> Hi,
>>> I've been trying to determine a fix for the 1.4.6 series of openvswitch
>>> due to a bug when using a 3.5 series linux kernel with the ovs 1.4.6
>>> dkms module.
>>>
>>> The details on how to reproduce and backtrace are here:
>>> https://bugs.launchpad.net/ubuntu/+source/openvswitch/+bug/1262692
>>>
>>> Commit 703133de331a7a7df47f31fb9de51dc6f68a9de8 (ip: generate unique IP
>>> identificator if local fragmentation is allowed) in the linux kernel if
>>> reverted fixes the issue. Looking through the openvswitch code, the only
>>> difference I suspect is if skb->local_df is changed.
>>
>> That patch changed the function signature of ip_select_ident() so when
>> it is called from OVS pointers get dereferenced as different types
>> than they actually are.
>>
>
> The following patch fixes the issue for me; however I'm unsure of how to
> ifdef the fix. The change appears in upstream around the 3.12-rc, but
> this change has been applied to many linux stable versions already
> (v3.2.52 for example).
>
> Any suggestions for a proper patch for ovs series that have the
> ip_select_ident call?

The easiest thing to do is probably just to backport the new version
and always use it on kernels < 3.12. There's no harm in using a
version bundled with OVS.

OVS 1.4 only supports up to Linux 3.2 and we don't generally add
support for newer kernels to stable release series. I guess if you say
that it affects stable versions of 3.2 then we could consider it
though.



More information about the discuss mailing list