[ovs-dev] [PATCH] Allow to create tunnel ports with the same config

[Ihar Hrachyshka]

On Fri, Nov 20, 2020 at 1:16 PM Ben Pfaff <blp at ovn.org> wrote:
>
> On Fri, Nov 20, 2020 at 12:44:30PM -0500, Ihar Hrachyshka wrote:
> > On Thu, Nov 19, 2020 at 12:19 PM Ben Pfaff <blp at ovn.org> wrote:
> > >
> > > On Thu, Nov 19, 2020 at 08:31:44AM -0500, Ihar Hrachyshka wrote:
> > > > On Thu, Nov 19, 2020 at 12:51 AM Ben Pfaff <blp at ovn.org> wrote:
> > > > >
> > > > > On Wed, Nov 18, 2020 at 10:35:01PM -0500, Ihar Hrachyshka wrote:
> > > > > > It's a legal setup where tunnel ports with the same config are created
> > > > > > on different bridges served by Open vSwitch. Specifically, multiple
> > > > > > OVN controllers may emulate multiple chassis running on the same
> > > > > > physical host, in which case they may need to create separate tunnel
> > > > > > ports to connect to the same remote chassis on their respective
> > > > > > bridges.
> > > > > >
> > > > > > Signed-off-by: Ihar Hrachyshka <ihrachys at redhat.com>
> > > > >
> > > > > That makes sense for sending packets, but which one is supposed to
> > > > > receive a packet when one arrives for that tunnel?
> > > > >
> > > >
> > > > If we are talking in OVN context, each virtual ovn-controller chassis
> > > > on the same host has to have a different IP / port to distinguish
> > > > between chassis. I don't think it's a useful configuration to have two
> > > > two incoming tunnel ports with the same config for the same IP that
> > > > are not served by separate DST IP addresses.
> > >
> > > I think that this code rejects tunnel ports with exactly the same
> > > configuration, though.  If the two ports were configured with different
> > > local IP addresses, then this code would not flag a conflict.  Do I
> > > misunderstand?
> > >
> >
> > Local IP addresses are not part of tunnel interface options map. Only
> > remote_ip and dst_port are. Does it address your concern?
>
> I don't see where you're getting that.  Tunnel configuration allows both
> local and remote IPs to be specified.  The match structure has slots for
> both IP addresses, and both of them may be specified.
>
> If the remote (or local) IP addresses are different, then this warning
> won't come up.  And you said that the remote IP addresses are different:
> "I don't think it's a useful configuration to have two two incoming
> tunnel ports with the same config for the same IP that are not served by
> separate DST IP addresses."
>
> tnl_port_receive() looks up the tunnel that should receive a packet.  It
> calls into tnl_find().  This calls tnl_find_exact() for each of several
> lookup tables.  In the case where a warning would be given here, we know
> that tnl_find_exact() would find two matching elements in the map,
> because that's the function that we use to know to issue the warning.
> If we disable the warning and insert the second match anyway, there will
> be two elements in the map with exactly the same key, tnl_find_exact()
> will only ever return one of them, thus the other will never receive any
> packets.
>
> What am I missing?
>

Oh I missed that local_ip *may* be configured for a tunnel endpoint,
it's just not configured by OVN controller *right now*. We can deal
with it in OVN controller instead of OVS then. Sorry for the noise,
this patch can be dropped.

Ihar