[ovs-dev] [PATCH v4 ovn] Don't suppress localport traffic directed to external port

[Ihar Hrachyshka]

On Wed, Jul 14, 2021 at 8:07 PM Numan Siddique <numans at ovn.org> wrote:
>
> On Wed, Jul 14, 2021 at 4:59 PM Ihar Hrachyshka <ihrachys at redhat.com> wrote:
> >
> > Recently, we stopped leaking localport traffic through localnet ports
> > into fabric to avoid unnecessary flipping between chassis hosting the
> > same localport.
> >
> > Despite the type name, in some scenarios localports are supposed to
> > talk outside the hosting chassis. Specifically, in OpenStack [1]
> > metadata service for SR-IOV ports is implemented as a localport hosted
> > on another chassis that is exposed to the chassis owning the SR-IOV
> > port through an "external" port. In this case, "leaking" localport
> > traffic into fabric is desirable.
> >
> > This patch inserts a higher priority flow per external port on the
> > same datapath that avoids dropping localport traffic.
> >
> > Fixes: 96959e56d634 ("physical: do not forward traffic from localport
> > to a localnet one")
> >
> > [1] https://docs.openstack.org/neutron/latest/admin/ovn/sriov.html
> >
> > Signed-off-by: Ihar Hrachyshka <ihrachys at redhat.com>
>
> Thanks Ihar for v4.
>
> All the tests pass now.  The patch LGTM.  I applied to the main branch.
> I also added the "Reported-at:
> https://bugzilla.redhat.com/show_bug.cgi?id=1974062" tag in
> the commit message.
>
> Does this need a backport ?  Looks like to me.  I tried backporting
> but the added test case fails.
> On 21.06 we don't have the pflow_output and lflow_output separation.
> That could be the reason
> it is failing.  If the fix is required, can you please take a look and
> submit the patch for branch-21.06.
>

I've sent a backport to 21.06 that returns false from
binding_handle_port_binding_changes when external port deleted. This
fixes the test failure.

Let me know if it's too hacky.

Thanks.
Ihar