Hi all,

as noted on mic a few minutes ago, I do not think this draft is a good
idea.  The two specific points I made on mic are these:

- LLDP is intentionally designed to bypass STP; you want to know what is
  on the other end of the link even if the port is blocked.  Can you
  discard the routes if the port is blocked?  Sure.  But now you have an
  interaction not only RIB<>LLDP, but also STP to get the port state and
  keep up to date.  Of course no sane person runs STP in a scenario you
  would use this LLDP feature in, but I have learned to never
  underestimate the "creativity" of operators in coming up with weird
  setups.

- MACsec is by default negotiated on the :03 special Ethernet multicast
  group.  LLDP by default runs on :0E.  These two groups don't have the
  same "reach": :03 transits more devices than :0E.  That means your
  "normal" MACsec SA won't be with the device you're talking LLDP with.
  Both the MACsec and LLDP specs allow configuring the respectively
  other multicast group, but - that is not default config and not
  commonly done.

Furthermore, no operator I know actually *wants* MACsec on LLDP.  They
want to know what's connected where while debugging their network,
possibly including MACsec.  That's a reasonable stance, IMHO.  But since
you've tied yourself to LLDP, you're now left without any way of
authenticating & securing these routes.  And this functionality is
really the most useful when crossing admin boundaries - worst place to
lack security on!

I believe the above points are just symptomatic of the layer break that
is happening here and you're "reaping the rewards".  Routes IMHO just
don't belong in LLDP.

And with all that said, I don't see what the point of sticking this in
LLDP is.  The protocol itself doesn't give you anything that any random
other multicast discovery protocol does.  And you're defining a new
feature; there's no compatibility argument.  The only functional
distinction I can see is the use of the limited scope 01:80:C2:00:00:0E
ethernet multicast group.  If that is your goal, we should look at
achieving that - without going through LLDP.

(And I don't think :0E [closest TPMR] is the correct choice; if anything
I think the function you're describing might go into :03 [closest
non-TPMR bridge, and also 802.1X/MACsec termination])

Regards,


equi


P.S.: yes, I know TPMRs are quite rare and you'll probably never see one
in a datacenter.  But the fact that these things just don't quite fit
together in their definitions doesn't go away just because the case is
rare.  OpenBSD has a TPMR implementation.

_______________________________________________
rtgwg mailing list -- [email protected]
To unsubscribe send an email to [email protected]

Reply via email to