Host Router Support for OSPFv2
draft-ietf-ospf-ospfv2-hbit-12

Thank you for the work put into this document. The short document is easy to read even if I wonder whether it is useful to spend time on IPv4-only protocol ;-)

The deployment issue (section 5) had raised a DISCUSS of mine and I appreciated your reply, so, I have cleared this DISCUSS.

Feel free to ignore my COMMENTs and NIT.

== COMMENTS ==

-- Generic --
Mentioning that the H-bit of OSPFv2 is the reverse of the R-bit of OSPFv3 would be useful.

-- Section 1 --
A description of "Closet Switches" would be welcome.

-- Section 3 --
Isn't the wording "host router" an oxymoron ?


== NITS ==

-- Section 8 --
I recommend reading and following the suggestions of RFC 8126 (how to write the IANA considerations)

I'm going to complain about some wording in Section 5 that Ben already called out, but I'll try to put in some specific suggestions for corrections here:

   In normal operation, there is no guarantee that the RI LSA will reach
   all routers in an area in a timely manner, which may result in
   forwarding loops in partial deployments.

This wording makes it sound exactly the opposite of what you mean, that if the RI LSA *does* reach all routers in a timely manner it can cause forwarding loops.  I suggest this:

NEW
   In normal operation, it is possible that the RI LSA will fail to
   reach all routers in an area in a timely manner.  That can result
   in forwarding loops in partial deployments.
END

   For example, if a new
   router joins an area, which previously had only H-bit capable routers
   with H-bit set then it may take some time for the RI to propagate to
   all routers.

First, change "area, which" to "area that" (no comma).  That fixes a usage problem.

But second, Ben and I are both unsure whether you mean that the new router does or doesn't support the H bit, or whether it matters.  Maybe the right approach here is to say a little more about what happens.  Something like this (adjust as needed to make it correct):

NEW
   For example, if a new
   router joins an area that previously had only H-bit capable routers
   with H-bit set then it may take some time for the RI to propagate to
   all routers.  While it is propagating, the area as a whole is unsure of
   the status of the new router, and that can cause <what problem?>
END

   o  All routers, with the H-bit set, MUST advertise all of the
      router's non-stub links with a metric equal to MaxLinkMetric

Both commas need to be removed here.

   o  All routers supporting H-Bit MUST check all the RI LSAs of nodes
      in the area before actively running the modified SPF to account
      for the H-bit in order to verify that all routers are in routing
      capability.

This is very awkwardly worded and is really hard to decipher.  I *think* you mean to say this:

NEW
   o  All routers supporting the H-Bit MUST check the RI LSAs of all
      nodes in the area to verify that all nodes support the H-Bit before
      actively using the H-Bit feature.
END

Did I get that right?

Abstract

   The Open Shortest Path First Version 2 (OSPFv2) does not have a
   mechanism for a node to repel transit traffic if it is on the
   shortest path.  This document defines a bit (Host-bit) that enables a

nit: I suggest to add "protocol" after "(OSPFv2)" to match the definite
article "The".

Section 1

   The OSPFv2 specifies a Shortest Path First (SPF) algorithm that

(same nit about adding "protocol")

   This functionality is particularly useful for a number of use cases:

nit: "this functionality" seems to refer to "the SPF algorithm that
identifies transit verticies based on their adjacencies", so I suggest
rewording to "such functionality would be useful" or "A mechanism to
move traffic away from the shortest path" or similar.

Section 4

I suggest noting that the (lettered) sub-procedures of step (2) remain
unchanged.

Section 5

   In normal operation, there is no guarantee that the RI LSA will reach
   all routers in an area in a timely manner, which may result in
   forwarding loops in partial deployments.  For example, if a new
   router joins an area, which previously had only H-bit capable routers
   with H-bit set then it may take some time for the RI to propagate to
   all routers.

It's currently only implicit that this new router does not support the
H-bit; shall we make it explicit?

   o  All routers supporting H-Bit MUST check all the RI LSAs of nodes
      in the area before actively running the modified SPF to account
      for the H-bit in order to verify that all routers are in routing
      capability.  If any router does not advertise the Host Router

nit: the grammar here is a little wonky, particularly for "all routers
are in routing capability" but perhaps also for "to account for the
H-bit".

Section 6

   When calculating the path to an OSPF AS-External-LSA or NSSA-LSA
   [RFC3101] with a Type-2 metric, [...]

nit: is this saying "calculating the path to [an LSA]"?  That's not a
usage I'm familiar with; can the AS-External-LSA or NSSA-LSA really
serve as a destination in this sense?

Section 7

Thank you for phrasing this as "this document requests the IANA to
assign", since until these specific values are officially assigned we
are technically "squatting" on them.  (The respective registration
policies of Standards Action and IETF Review give us pretty good control
that nothing else is going to swoop in on them, though.)

Three comments/questions:

1) This sentence in section 3:
"An OSPFv2 router advertising a router-LSA with the H-bit
   set indicates that it MUST NOT be used as a transit router (see
   Section 4) by other OSPFv2 routers in the area supporting the
   functionality."
Isn't the MUST here too restrictive? What if the router is the part of the only path to a certain host?
Or is the assumption that this host is some kind of endhost/deadend, but then it should never be on the shortest path anyway, or?

Later on you say
"When the H-bit is set, the OSPFv2 router is a Host (non-transit)
   router and is incapable of forwarding transit traffic."
However, there might also be routers that don't understand the H bit and therefore will route traffic over this host anyway, no?

2) Shouldn't this document update RFC2328, given section 4 and this sentence in section 2:
"If the H-bit is set then the calculation of the shortest-
   path tree for an area, as described in section 16.1 of [RFC2328], is
   modified by including a check to verify that transit vertices DO NOT
   have the H-bit set (see Section 4)."
(And why is DO NOT in upper letters?)

3) Is there an attack that by spoofing the H-bit an attacker could influence the routing such that traffic is router over a malicious node instead? I guess there are multiple ways to impact the routing that way and this might not be specific to the H bit but maybe still worth thinking about it once more...?

Nit:
Section 5: "The RI LSA MUST be area-scoped.  Bit:" -> I guess the "Bit:" should be removed.