Path Computation Element Communication Protocol (PCEP) Procedures and Extensions for Using the PCE as a Central Controller (PCECC) of LSPs
RFC 9050

Note: This ballot was opened for revision 11 and is now closed.

Erik Kline (was Discuss) Yes

Alvaro Retana No Objection

Comment (2021-02-22 for -12)
(0) The fact that the procedures (§5) are presented before introducing the messages/objects (§6-7) makes reading this document harder and more complex than it has to be.  Consider changing the order or at least adding forward references in §5.

(1) §5.2:  Is there a reason for the messages from rfc8231 to be in parenthesis?

(2) §5.4:  

   The PCECC-CAPABILITY sub-TLV SHOULD NOT be used without the
   corresponding Path Setup Type being listed in the PATH-SETUP-TYPE-
   CAPABILITY TLV.  If it is present without the corresponding Path
   Setup Type listed in the PATH-SETUP-TYPE-CAPABILITY TLV, it MUST be
   ignored.

When is it ok to use the PCECC-CAPABILITY sub-TLV without the corresponding PST?  If the result is that it will be ignored, then I don't understand why the use of both is not required.

(3) §5.5.1/§5.5.4: "ingress MAY further choose to deploy a data plane check mechanism and report the status back to the PCE"  Is this (checking and reporting) specified somewhere?  Because you're using normative language, please add a reference.

A similar statement is made in §5.5.7: "ingress PCC MAY choose to apply any OAM mechanism to check the status of LSP in the Data plane and MAY further send its status in a PCRpt message to the PCE".

(4) §5.5.3: s/central controller instructions...is done/central controller instructions...are done

(5) §5.5.8: "The PCC SHOULD allocate the Label and SHOULD report to the PCE using the PCRpt message."   When is it ok for the PCC to not allocate and/or report?  IOW, why are these actions only recommended and not required?  Note that the very next paragraph requires the behavior.

(6) §7.3/§7.3.1:  In the out-label case, "it is mandatory to encode the next-hop information".  Should this information point at a directly connected IP address/interface, or can it point at a remote next-hop (which may be resolved through a routing protocol)?  What if the expected conditions are not met?

Benjamin Kaduk No Objection

Comment (2021-02-24 for -12)
I agree with Roman about the prospects for ensuring a solid security
baseline with what is essentially greenfield deployment.

(throughout) Is the TLV LSP-IDENTIFIER or LSP-IDENTIFIERS (with final
'S')?

Thanks to Yaron Sheffer for the secdir reviews, and to the authors for
updating in light of that review.

I note in a few places in the section-by-section comments that the
figures seem to indicate a 'D' flag in PCInitiate and PCUpd messages,
and the only D flag I see mentioned in the prose is the Delegate flag in
a PCRpt.  This seems particularly important to check and get right
(though I admit that I could just be missing something).

Section 1

   A PCE-based Central Controller (PCECC) can simplify the processing of
   a distributed control plane by blending it with elements of SDN and
   without necessarily completely replacing it.  Thus, the LSP can be
   calculated/setup/initiated and the label forwarding entries can also
   be downloaded through a centralized PCE server to each network
   devices along the path while leveraging the existing PCE technologies
   as much as possible.

nit: "each network device" singular.

Section 2

   The terminology used in this document is the same as that described
   in the draft [RFC8283].

"That RFC doesn't look like a draft to me"

Section 3

   This document also allows a case where the label space is maintained
   by the PCC itself, and the labels are allocated by the PCC, in this
   case, the PCE should request the allocation from PCC as described in
   Section 5.5.8.

nit: comma splice.

Section 4

   4.  PCEP procedures need to provide a mean to update (or clean up)
       the label-download entry to the PCC.

   5.  PCEP procedures need to provide a mean to synchronize the labels
       between the PCE and the PCC via PCEP messages.

nits: "provide a means" plural (twice); s/the label-download entry/label
entries downloaded/ (I think)

Section 5.4

   A legacy PCEP speaker (that does not recognize the PCECC Capability
   sub-TLV) will ignore the sub-TLV in accordance with [RFC8408] and
   [RFC5440].  As per [RFC8408], the legacy PCEP speaker (that does not
   support PST), it will:

Sending a PCErr for an unrecognized PST in the
PATH-SETUP-TYPE-CAPABILITY would break extensibility; the 21/1 error
type/value pair is only used in RFC 8408 when a PST is attempted to be
used in a PCRpt, PCInitiate, or PCUpd.  I think we should just say that
it ignores the PST in the PATH-SETUP-TYPE-CAPABILITY TLV.

Section 5.5.1

   An LSP-IDENTIFIER TLV MUST be included for PCECC LSPs, the tuple
   uniquely identifies the LSP in the network.  [...]

Which tuple?

Also, RFC 8231 says that the LSP-IDENTIFIERS TLV (note final 'S') must
be used for RSVP-signaled LSPs, but PCECC is not (to my knowledge)
requiring the use of RSVP.  Do we need to say anything to generalize
LSP-IDENTIFIERS for other use?

   The ingress PCC MUST also set D (Delegate) flag (see [RFC8231]) and C
   (Create) flag (see [RFC8281]) in the LSP object of the PCRpt message
   to the PCE in the initial exchange.  The PCC responds with a PCRpt
   message with the status set to "GOING-UP" and carrying the assigned
   PLSP-ID (see Figure 1).  [...]

nit: "responds" feels a bit out of place here, since the first sentence
has already talked about setting flags in the PCRpt.  Switching the
order of the sentences might help, but there still wouldn't be a very
clear connection in the prose between the PCRpt and triggering
PCInitiate.

                                                              Each PCC
   further responds with the PCRpt messages including the central
   controller instruction (CCI) and the LSP objects.  The PCC responds
   with a PCRpt message to acknowledge the central controller
   instruction.

Likewise, the second "responds" here feels out of place.

                |PCC    |                              |  PCE  |
                 |ingress|                              +-------+
          +------|       |                                  |
          | PCC  +-------+                                  |
          | transit| |                                      |
   +------|        | |<--PCInitiate,PLSP-ID=0,PST=TBD1,D=1--| PCECC LSP
   |PCC   +--------+ |                                      | Initiate

[sorry for truncation] In the PCInitiate line, what does D=1 refer to?
The only mention of a D flag I can find is that the PCC must set D=1 in
the initial PCRpt to delegate the LSP.

       |       |     |<---PCUpd,PLSP-ID=2,PST=TBD1,D=1------| PCECC LSP
       |       |     |      (UP)                            | Update

Likewise, what is the 'D=1' in PCUpd?

(Both remarks seem to apply to Figure 2 as well.)

       - The O bit is set as before (and thus not included)


            Figure 2: PCE-Initiated PCECC LSP (PCC allocation)

(It doesn't look like we currently indicate the O bit in Figure 1, so
this remark feels a little out of place.  We do indicate the O bit in
Figure 3, though.)

Section 5.5.3

   As per [RFC8281], following the removal of the Label forwarding
   instruction, the PCC MUST send a PCRpt message.  The SRP object in
   the PCRpt MUST include the SRP-ID-number from the PCInitiate message
   that triggered the removal.  The R flag in the SRP object MUST be
   set.

This text seems to indicate that the R flag is set in the SRP object in
the PCRpt, but this does not seem to be reflected in Figure 5.

Section 5.5.4

                                New CC-IDs are used to identify the
   updated instructions, the identifiers in the LSP object identify the
   existing LSP.  [...]

nit: comma splice.

       |       |     |<---PCUpd,PLSP-ID=1,PST=TBD1,D=1-----| PCECC
       |       |     |    SRP=S                            | LSP Update

(I remain unsure about the D flag in the PCUpd.)

Section 5.5.8

   the Label.  If the allocation is successful, the PCC MUST report via
   the PCRpt message with the CCI object.  Else, it MUST send a PCErr
   message with Error-Type = TBD5 ("PCECC failure") and Error Value =
   TBD9 ("Invalid CCI").  If the value of the Label in the CCI object is
   valid, but the PCC is unable to allocate it, it MUST send a PCErr
   message with Error-Type = TBD5 ("PCECC failure") and Error Value =
   TBD10 ("Unable to allocate the specified CCI").

I might be misreading, but this seems to say that if allocation failed
but the value of the label in the CCI object is valid, you have to send
*two* PCErr messages, one with TBD9 and one with TBD10 (there are two
MUST-level requirements that seem to both apply).  I mostly assume that
just one would suffice, so a bit of rewording might be in order.

   If the PCC wishes to withdraw or modify the previously assigned
   label, it MUST send a PCRpt message without any Label or with the
   Label containing the new value respectively in the CCI object.  The
   PCE would further trigger the removal of the central controller
   instruction as per this document.

This seems quite vague about which CCI is to be removed from where.

Section 6.1

   At most two instances of the CCI object can be included, in the case
   of transit LSR to encode both in-coming and out-going label
   forwarding instructions.  Other instances MUST be ignored for P2P
   LSP.  [...]

It's a little hard to square up "at most two instances" and "other
instances MUST be ignored", even if the former doesn't use normative
language.

Section 7.1.1

   o  L bit (Label): if set to 1 by a PCEP speaker, the L flag indicates
      that the PCEP speaker support and willing to handle the PCECC

nits: "supports and is willing"

Section 7.3

   CC-ID:  A PCEP-specific identifier for the CCI information.  A PCE
      creates a CC-ID for each instruction, the value is unique within
      the scope of the PCE and is constant for the lifetime of a PCEP
      session.  The values 0 and 0xFFFFFFFF are reserved and MUST NOT be
      used.

In the vein of draft-gont-numeric-ids-sec-considerations, please include
some discussion on whether the uniqueness property is the only one
needed (e.g., no ordering or gap detection), as well as some analysis of
what information is leaked (including side channels) if the CC-ID is
revealed to outside parties.

My preliminary instinct is that if the value is only in scope for a
single live PCEP session (i.e., two fixed peers) and the session is
protected by TLS, there is no harm in doing sequential allocation and
that makes ensuring uniqueness easier, but there are any number of ways
in which such a trivial analysis could be flawed.

Section 7.3.1

IANA already shows IPV4-ADDRESS and IPV6-ADDRESS TLVs allocated by RFC
8779, with what appears to be identical structure to these.  Why are new
TLV types necessary?

Section 9

I agree with the secdir reviewer that it would be worthwhile to discuss
authorization as well as authentication.  In a system with the
privileged central controller, confirming that a given authenticated
entity is authorized to act as the central controller is important to
the overall security of the system (so that a compromised PCC node
cannot claim to be a PCE to other, uncompromised, PCC nodes).  This
might be done, for example, via an attribute in the PCE's X.509
certificate used for PCEPS or a local policy with a specific accept-list
of X.509 certificate.

Section 9.1

I think we should reiterate the guidance that PCCs need to check that
labels provided by the PCE are in the proper range.

   general precaution, it is RECOMMENDED that this PCEP extension be
   activated on mutually-authenticated and encrypted sessions across
   PCEs and PCCs belonging to the same administrative authority, using
   TLS [RFC8253], as per the recommendations and best current practices
   in [RFC7525].

It's probably best to cite this as BCP 195, as there is likely to be an
updated version in the next couple years.

                                 [RFC8281] provides a mechanism to
   protect PCC by imposing a limit.  The same can be used for the PCECC
   operations as well.

nit: either "PCCs" or "the PCC".

Section 11.6

   IANA is requested to create a new sub-registry to manage the Flag
   field of the CCI object called "CCI Object 16-bits Flag Field".  New

Are these flags expected to be specific to the Object-Type 1 (MPLS
Label) CCI Object?  If so, perhaps the registry name should indicate
that.

Section 13.2

We generally see RFC 8126 listed as normative when used for the
registration procedures when defining a new registry.

Per
https://www.ietf.org/about/groups/iesg/statements/normative-informative-references/
it seems that the recommended behavior with RFCs 8253 and 7525 should
make them normative references.

Murray Kucherawy No Objection

Comment (2021-02-24 for -12)
No email
send info
In Section 2, RFC 8283 isn't a "draft".

In Section 5.5.1:

   Once the label operations are completed, the PCE SHOULD send a PCUpd
   message to the ingress PCC.

Why "SHOULD"?  Is there another option?   Why might an implementer do something else?

The SHOULDs elsewhere in Section 5 are probably worth a second look too.

Robert Wilton No Objection

Comment (2021-02-25 for -12)
Hi,

Thanks for the document, I have not had time to review in detail.

The length of the abstract does stand out to me, and it might be helpful if it could be shortened.  E.g. I suspect that the first two paragraphs are probably not required in the abstract and are best covered in the introduction.

On section 10:

   A PCE or PCC implementation SHOULD allow to configure to enable/
   disable PCECC capability as a global configuration.   => 
   A PCE or PCC implementation SHOULD allow the PCECC capability
   to be enabled/disabled as part of the global configuration.

Also probably change "is enabled on the PCEP session" to "is enabled on a PCEP session"?

Regards,
Rob

Roman Danyliw No Objection

Comment (2021-02-24 for -12)
Thank you to Yaron Sheffer for the SECDIR review and the updates made as a result to improve the Security Considerations.  I endorse the revised text that minimally RECOMMENDs the use of “mutually-authenticated and encrypted sessions.”  My question is why can’t we go even further and require (use MUST) this crucial provisioning channel always be protected.  When would we NOT want to use TLS?  I appreciate that mandating the use of security primitives in routing is challenging due to a long tail of legacy investment.  However, this extension seems like a near "green field" focused on a modern, SDN architecture which seems unlikely to have less capable legacy elements.

Éric Vyncke (was Discuss) No Objection

Comment (2021-03-02 for -13)
Thank you for the work put into this document. Thank you for addressing my previous DISCUSS points; I have cleared my previous DISCUSS points but nevertheless please upload yet-another-revised I-D before sending it to the RFC Editor.

Section 7.3.1 should use " IPv6 address: the 128-bit IPv6 link-local address of the interface." rather than " IPv6 address: A 128-bit IPv6 address of the node."

I hope that this helps to improve the document,

Regards,

-éric

== DISCUSS [kept for history] ==

-- Section 7.3.1 --
LINKLOCAL-IPV6-ID-ADDRESS TLV: I fail to understand why there are two addresses in this TLV while others have one one ? Also is 'local' and 'remote' really global addresses ?

== COMMENTS [kept for history] ==

A minor comment: the abstract is clear but probably a little too long for an abstract.

-- Section 7.3 --
Just wonder why  LINKLOCAL-IPV6-ID-ADDRES is not mentioned in this section but well in the next one ?

(Deborah Brungard; former steering group member) Yes

Yes ( for -11)
No email
send info

(Alissa Cooper; former steering group member) No Objection

No Objection ( for -12)
No email
send info

(Barry Leiba; former steering group member) No Objection

No Objection ( for -12)
No email
send info