[PATCH] Filter: Add MPLS label route attribute

[Ondrej Zajicek]

On Fri, May 07, 2021 at 03:54:40PM -0700, Trisha Biswas wrote:
> Hi all,
> 
> While BIRD 2.x has underlying support for MPLS (applying labels to static
> routes and syncing with kernel), there is no option to set labels via
> configuration. The attached patch adds support to set or read labels using
> filters. Currently this supports the addition of one label per route, and
> can be extended to stacked labels in future.
> 
> Please let me know if you have any questions.

Hi

The patch looks fine (there is a minor issue that reading the attribute
could check 'labels' field to see if there is any, and if it is not, it
should return implicit-null label (3)).

There are some conceptual issues with it (and that is also why we did not
implemented it already):

First, note that all nexthop attributes (gw, iface, ...) a kind of
broken, as they only allow to set the first nexthop and ignore ECMP.
This attribute behaves in a similar manner, so it is not a big issue,
just note that we plan to change/rework that in the future.

Second, MPLS allows to use whole stack of labels (we allow to set it for
static routes), but we cannot represent it in the filter language, so the
patch allows to access only the first one (so it is handled as simple
integer). You wrote that it can be extended to stacked labels in future,
but it seems to me that it would be more (backwards-incompatible) change
than extension (e.g. from int type to some int-list type).

Third, note that there are two kinds of MPLS labels for a route - local
label (label expected on received packets) and remote label of nexthop
(label or labels attached to forwarded packets - this is the label stack
in nexthop structure). I would prefer to reserve 'mpls' attribute name
for the first one, so perhaps we could call it 'gw_mpls', 'encap_mpls'
(like in ip-route tools), or something like that. 

It is true that having limited attribute is better than no attribute, so
i would prefer to merge it with some note in documentation that it is
experimental and likely will change in the future to handle multiple
labels and under 'gw_mpls' name. Is that acceptable?

-- 
Elen sila lumenn' omentielvo

Ondrej 'Santiago' Zajicek (email: santiago at crfreenet.org)
OpenPGP encrypted e-mails preferred (KeyID 0x11DEADC3, wwwkeys.pgp.net)
"To err is human -- to blame it on a computer is even more so."