Re: [RFC net-next 0/2] net/sched: cls_flower, act_mirred: VXLAN redirect using TC

[Jiri Pirko <jiri@resnulli.us>]

Mon, Aug 15, 2016 at 11:08:04AM CEST, amir@vadai.me wrote:
>On Mon, Aug 15, 2016 at 11:17:40AM +0300, Amir Vadai wrote:
>> On Mon, Aug 15, 2016 at 09:11:22AM +0200, Jiri Pirko wrote:
>> > Sun, Aug 14, 2016 at 07:53:30PM CEST, xiyou.wangcong@gmail.com wrote:
>> > >On Sun, Aug 14, 2016 at 7:06 AM, Amir Vadai <amir@vadai.me> wrote:
>> > >> tc qdisc add dev $ETH ingress
>> > >>
>> > >> # ENCAP rule for ARP
>> > >> tc filter add dev $ETH protocol 0x806 parent ffff: prio 11 \
>> > >>                 flower \
>> > >>                 action mirred egress redirect dev $VXLAN enc_src_ip 11.11.0.1 enc_dst_ip 11.11.0.2 enc_key_id 11 enc_dst_port 4789
>> > >>
>> > >> # ENCAP rule for ICMP
>> > >> tc filter add dev $ETH protocol ip parent ffff: prio 10 \
>> > >>                 flower ip_proto 1 \
>> > >>                 action mirred egress redirect dev $VXLAN enc_src_ip 11.11.0.1 enc_dst_ip 11.11.0.2 enc_key_id 11 enc_dst_port 4789
>> > >>
>> > >
>> > >I don't like this. This makes mirred action unnecessarily
>> > >complex, it should really just mirror or redirect packets as
>> > >it is, why it should be aware of tunnel information?
>> > >
>> > >I think you probably need to introduce a new tc action
>> > >for these tunnel information and pipe it to mirred.
>> > 
>> > that is the first thing that I thinked of when I saw the patch. I think
>> > you can introduce act_vxlan similar to act_vlan.
>> 
>> introducing a new action was the first thing I thought of, but it felt
>> problematic because the actual encap is done by the redirection to the
>> vxlan device. This action is only responsible to supply the metadata and
>> work tightly with the mirred. It is not exactly like vlan that the
>> push/pop actions can live without mirroring/redirecting.
>> But still as all of you said, it makes mirred complex with stuff that
>> shouldn't be there. And between the two options it is better to
>> introduce a new action.
>> 
>> I will go in this direction.
>> 
>> Thanks,
>> Amir
>
>Any objection to the following?
>
># ENCAP rule
>tc filter add dev $ETH protocol ip parent ffff: prio 10 \
>		flower ip_proto 1 \
>		action set_tunnel_key src_ip 11.11.0.1 dst_ip 11.11.0.2 key_id 11 dst_port 4789 \

Looks fine to me.


>		action mirred egress redirect dev $VXLAN 
>
># DECAP rule
>tc filter add dev $VXLAN protocol ip parent ffff: prio 10 \
>		flower \
>			enc_src_ip 11.11.0.2 enc_dst_ip 11.11.0.1 enc_key_id 11 \
>			ip_proto 1 \
>		action mirred egress redirect dev $ETH