[ovs-discuss] pf_packet/af_packet socket with vlan tagging does not strip vlan tag

Jesse Gross jesse at nicira.com
Wed May 27 20:34:48 UTC 2015

I think all of the discussion from the original thread still applies.
I believe that this issue must affect a ton of other things in the
kernel (including various filtering ton with iptables) so an
OVS-specific solution is not likely to go over well. The only real
solution is to initialize skb->protocol for raw sockets in the
upstream kernel.

On Wed, May 27, 2015 at 8:39 AM, Ben Pfaff <blp at nicira.com> wrote:
> Jesse and Pravin, can you comment on the best way to deal with this
> long-standing bug?
> On Tue, May 26, 2015 at 09:18:30PM -0700, Jon Gainsley wrote:
>> Hi Ben,
>> Thanks for your response. Yeah, that's exactly it. I tracked it down to
>> pop_vlan(), which was not popping the vlan because skb->protocol was set to
>> ETH_P_ALL, rather than ETH_P_8021Q. The reason it works on the first few
>> packets is without a rule, the packet is upcalled, which clones skb and
>> properly assigns skb->protocol by parsing the packet.
>> If it's a kernel bug, it's not fixed in 3.16. I've modified
>> datapath/actions.c:pop_vlan to check the packet's eth proto and set
>> skb->protocol to ETH_P_8021Q if that's what's in the packet, so the vlan
>> gets popped correctly. I don't know if you have any comments on if that's a
>> valid workaround, but it's what I'm going to try for now and see how it
>> goes.
>> Thanks!
>> -Jon
>> On Tue, May 26, 2015 at 6:22 PM, Ben Pfaff <blp at nicira.com> wrote:
>> > On Wed, May 20, 2015 at 01:45:37PM -0700, Jon Gainsley wrote:
>> > > Hi,
>> > >
>> > > I'm wondering if anyone can help me with this. I am having trouble
>> > > inserting raw vlan tagged packets into the vswitch using an AF_PACKET
>> > > socket. I have the following simple bridge:
>> >
>> > This kind of sounds like this kernel bug described in the thread here:
>> >         http://openvswitch.org/pipermail/dev/2011-December/013948.html
>> >
>> > I don't know whether it's been fixed.
>> >

More information about the discuss mailing list