Re: [PATCH v2.57] datapath: Add basic MPLS support to kernel

On Tue, May 20, 2014 at 07:05:28PM -0700, Jesse Gross wrote:

On Tue, May 20, 2014 at 3:48 AM, Simon Horman [off-list ref] wrote:

quoted

On Mon, May 19, 2014 at 06:34:05PM -0700, Jesse Gross wrote:

quoted

On Thu, May 15, 2014 at 4:07 PM, Simon Horman [off-list ref] wrote:

quoted

diff --git a/datapath/actions.c b/datapath/actions.c
index 603c7cb..7c3ae0c 100644
--- a/datapath/actions.c
+++ b/datapath/actions.c

When we set the EtherType at the bottom of the function, I don't think
that it is correct in the presence of VLAN tags because it will set
the outer EtherType.

I believe this comes back to the problem we have with tag ordering. A
problem which after the most recent discussion of this topic I planned to
kick into the long grass by only allowing push MPLS actions on packets with
a well defined tag order.

Right, I subconsciously keep forgetting this even though I know it's there.

quoted

This is the purpose of the white list in ovs_nla_copy_actions__().  It is
supposed to only push MPLS actions for flows with an IPv4, IPv6, ARP, RARP
or MPLS ethtype.  However, I now think that the white list likely does not
work for VLAN packets as their flow's ethtype will be the inner-ethtype
(the one inside the VLAN tag) rather than a VLAN ethtype.

I'm unsure how you would like to handle this but one possibility would be
simply for push_mpls() to return an error if it is called on an skb with a
VLAN tag present or if the ethtype doesn't match a white list. Another is
to set the inner ethtype.

Can't we just look at/track flow->tci to see if there is a vlan?

Yes :)

I had been thinking in terms of OpenFlow flows and thought
this wouldn't work. But I have verified that it does with datapath flows.