[mpls] Rtgdir early review of draft-ietf-mpls-bfd-directed-07

[Carlos Pignataro <[e-mail]>]

Reviewer: Carlos Pignataro
Review result: Has Issues

Hello

I have been selected to do a routing directorate “early” review of this draft.
https://datatracker.ietf.org/doc/draft-ietf-mpls-bfd-directed/

The routing directorate will, on request from the working group chair, perform
an “early” review of a draft before it is submitted for publication to the
IESG. The early review can be performed at any time during the draft’s lifetime
as a working group document. The purpose of the early review depends on the
stage that the document has reached.

The MPLS chairs have requested an early review from the directorate with the
objective of improving document quality.  This document has had three
unsuccessful WG LCs.

For more information about the Routing Directorate, please see
​http://trac.tools.ietf.org/area/rtg/trac/wiki/RtgDir

Document: draft-ietf-mpls-bfd-directed-07.txt
Reviewer: Carlos Pignataro
Review Date: Early July, 2017
Intended Status: Standards Track

Summary:
I have significant concerns about this document.
I also recommend a BFD WG Chair or appointee to review this document.

Comments:

First, I have a general concern about the architectural approach in this
document.

This document is modeled after RFC 7110. RFC 7110 describes the specification
of a return path for MPLS LSP Ping. MPLS LSP Ping uses a request/reply
command/response paradigm, in which receipt of an Echo Request elicits the
generation of an Echo Reply.

BFD for MPLS, however, uses a different approach and paradigm (as per RFC
5884). An MPLS LSP Ping packet is used as a bootstrap, signaling discriminator
value for a persistent BFD session. After the MPLS LSP Ping signals the
Discriminator (via MPLS LSP Ping TLV) to use, then BFD control messages are
sent back and forth.

However, while the BFD session is UP and  BFD control messages and being sent
back and forth, and while no MPLS LSP Ping packets are sent after bootstrapping
-- what happens if the return path changes (e.g., the return LSP goes down,
gets unconfigured, etc.)?

In that case, not only this mechanism can actually make things worst, because
it results in a false negative, but also the document does not specify how the
system should recover. The I-D seems to assume complete topological
invariability for it to work long-term, since it does not specify any mechanism
to update or to deal with such a failure or change scenario.

On the other hand, there is already a BFD mechanism without the bootstrapping
setup and with a command/response like behavior, that is S-BFD, RFC 7881. That
one is notably missing from this draft.

Further, there seem to be a number of potentially erroneous assumptions made,
see below.

Additional Comments:

     Bidirectional Forwarding Detection (BFD) Directed Return Path

The title should include that this is *only* for MPLS BFD.

   When a BFD session monitors an explicitly routed unidirectional path
   there may be a need to direct egress BFD peer to use a specific path
   for the reverse direction of the BFD session.

Scope: is this solution targeted only for "explicitly routed unidirectional
path", and the solution to have the reply come back the exact reverse
direction? That does not seem to be the case and the solution.

   [RFC5880], [RFC5881], and [RFC5883] established the BFD protocol for
   IP networks.  [RFC5884] and [RFC7726] set rules of using BFD
   asynchronous mode over IP/MPLS LSPs.  These standards implicitly
   assume that the egress BFD peer will use the shortest path route
   regardless of route being used to send BFD control packets towards
   it.

Is "These standards" referring to the three former or the four latter?

   For the case where a LSP is explicitly routed it is likely that the
   shortest return path to the ingress BFD peer would not follow the
   same path as the LSP in the forward direction.  The fact that BFD
   control packets are not guaranteed to follow the same links and nodes
   in both forward and reverse directions is a significant factor in
   producing false positive defect notifications, i.e. false alarms, if
   used by the ingress BFD peer to deduce the state of the forward
   direction.

There may be an implicit mis-assumption in this text and overall approach: the
fact that traffic flows on one direction does not imply that the reverse
direction using the same interfaces and nodes would actually be consequently
properly programmed and working.

   This document defines the BFD Reverse Path TLV as an extension to LSP
   Ping [RFC8029] and proposes that it is to be used to instruct the
   egress BFD peer to use an explicit path for its BFD control packets
   associated with a particular BFD session.

This text assumes that the BFD return path is MPLS. However, my understanding
from RFC 5884 is that this is not necessarily the case, and the return can be
IP.

   When BFD is used to monitor unidirectional explicitly routed path,
   e.g.  MPLS-TE LSP, BFD control packets in forward direction would be
   in-band using the mechanism defined in [RFC5884] and [RFC5586].

Which BFD uses RFC 5586? RFC5586 says that is not needed:

   "Some of these functions can be supported using existing
   tools such as Virtual Circuit Connectivity Verification (VCCV)
   [RFC5085], Bidirectional Forwarding Detection for MPLS LSPs (BFD-
   MPLS) [BFD-MPLS], LSP-Ping [RFC4379], or BFD-VCCV [BFD-VCCV]."

And then:

   o  a failure detection by ingress node on the reverse path cannot be
      interpreted as bi-directional failure unambiguously and thus
      trigger, for example, protection switchover of the forward
      direction without possibility of being a false positive.

   To address this scenario the egress BFD peer would be instructed to
   use a specific path for BFD control packets.

But using a specific path for return cannot either imply "interpreted as
bi-directional failure unambiguously", so the scenario is not *addressed*.

   The BFD Reverse Path TLV carries information about the path onto
   which the egress BFD peer of the BFD session referenced by the BFD
   Discriminator TLV MUST transmit BFD control packets.  The format of
   the BFD Reverse Path TLV is as presented in Figure 1.

What does the remote endpoint do with that "MUST" if the return FEC goes away?

There also seem to be some self-contradiction. This document says:

   LSP ping, defined in [RFC8029], uses BFD Discriminator TLV [RFC5884]
   to bootstrap a BFD session over an MPLS LSP.  This document defines a
   new TLV, BFD Reverse Path TLV, that MUST contain a single sub-TLV
   that can be used to carry information about the reverse path for the
   BFD session that is specified by value in BFD Discriminator TLV.

And then says:

   Reverse Path field contains a sub-TLV.

But then says:

   None, one or more sub-TLVs MAY be included in the BFD Reverse
   Path TLV.  If none sub-TLVs found in the BFD Reverse Path TLV, the
   egress BFD peer MUST revert to using the default, i.e., over IP
   network, reverse path.

So is it only one, or none/one/multiple?

I believe it needs to be multiple since then a Tunnel can be specified. But the
document as-is seems self-contradicting.

Further, where has that "default" been defined as "over IP network"?

There's another contradiction here:

   If the egress LSR cannot find the path specified in the Reverse Path
   TLV it MUST send Echo Reply with the received Reverse Path TLV and
   set the Return Code to "Failed to establish the BFD session.  The
   specified reverse path was not found" Section 3.3.  The egress BFD
   peer MAY establish the BFD session over IP network as defined in
   [RFC5884].

So the response is "Failed to establish the BFD session." But then it MAY
establish the session? And, again, what if the path is found at bootstrap but
lost afterwards?

4.  Use Case Scenario

The fact that A-B-C-D-G-H works does not mean that the reverse, H-G-D-C-B-A,
will work.

6.  Security Considerations

   Security considerations discussed in [RFC5880], [RFC5884], [RFC7726],
   and [RFC8029], apply to this document.

There seem to be additional security considerations with returns taking
explicit paths, and should be expanded in here.

Net-net, I do have concerns about this document. I believe it is not ready to
advance, and could use more whiteboard time as well as a review by BFD experts.

Best,

Carlos Pignataro.