[ovs-discuss] Problems with MPLS header pushing

[Ben Pfaff]

On Tue, Jul 28, 2015 at 02:33:12PM +0200, thomas.morin at orange.com wrote:
> 2015-07-27, Ben Pfaff:
> >If MPLS is going to work well in Open vSwitch, we need a developer to
> >champion implementing and testing it.  None of the OVS developers right
> >now have a use case for MPLS, so we don't use it and don't test it, and
> >therefore don't find bugs, missing features, performance problems, etc.,
> >and no one has incentive to improve the situation.
> 
> There are many people/projects based on OVS and needing an MPLS dataplane,
> in particular for use cases where OVS has to interconnect with IP/MPLS
> routers for BGP VPNs:
> - the reference driver for networking-bgpvpn Neutron project
> - work in progress in ODL for BGP VPN interconnection
> - Nuage Networks solution
> 
> So I'm not pessimistic, I expect this to happen.

Well, that's good.

> >If you are able and willing to work on MPLS in OVS, please do!
> 
> Well I'll be humble here: without some support, my mastership of OVS in
> insufficient to allow me to dive in kernel/userspace interactions and fix
> things in an amount of time reasonable compared to how important MPLS is in
> the 2.3 branch.
> 
> In that particular case (trying to fix tat bug in the 2.3.x branch for MPLS)
> I could use some help: a hint on where to start to find why
> match->mask->key.ip.proto ends up being something else than 0xff in
> match_validate (datapath/flow_netlink.c).

Usually that would mean that userspace wanted to wildcard everything at
L4.