[ovs-discuss] OvS 2.5 regression in VXLAN tunnels

[pravin shelar]

On Thu, Jul 21, 2016 at 4:06 AM, Jakub Libosvar <jlibosva at redhat.com> wrote:
> On 20/07/16 23:13, pravin shelar wrote:
>>
>> On Fri, Jul 15, 2016 at 6:01 AM, Jakub Libosvar <jlibosva at redhat.com>
>> wrote:
>>>
>>> Hi,
>>>
>>> in OpenStack Neutron project we use Open vSwitch on a single node to
>>> simulate tunneling between multiple nodes. After commit [1] was merged,
>>> we
>>> observe openflow output rule being applied on packets, that injects
>>> packets
>>> into tunnels but we don't see any packets on the other side of the
>>> tunnel.
>>>
>>> I wrote a simple script to build this environment in order to make it
>>> easy
>>> to reproduce [2]. I didn't attach it, I can if requested.
>>>
>>> Ways to reproduce:
>>>    - fetch the script (e.g. fullstack_env.sh)
>>>    - run as root: fullstack_env.sh setup
>>>    - try ping via tunnel: ip net e nsA ping 10.20.0.10
>>>
>>> In previous OvS 2.4 were icmp replied.
>>>
>> Thanks for detailed report.
>> I sent out a fix for master branch. can you try it?
>>
>> ref: http://openvswitch.org/pipermail/dev/2016-July/075768.html
>>
> Thanks for sending fix promptly. I tried the patch applied on master
> (compiled kmod and modprobed) but I still can't see packets delivered to the
> peer side of tunnel using [2] script.
>
> Does [2] work for you?
>
> Thank you,
> Jakub
>
>
I have pushed fix for vxlan issue both upstream and out of tree kernel
module. with these fixes the test case from the script work for me.


Thanks.