Re: [mpls] Benjamin Kaduk's No Objection on draft-ietf-mpls-summary-frr-rsvpte-08: (with COMMENT)

[Tarek Saad <[e-mail]>]

Hi Benjamin,

Thank you much for reviewing the document and for the provided comments. We have uploaded version -09 which addresses those - diffs at: https://tools.ietf.org/rfcdiff?url2=draft-ietf-mpls-summary-frr-rsvpte-09.txt

Please see inline for further clarification.

On 2/19/20, 7:22 PM, "Benjamin Kaduk via Datatracker" <[e-mail]> wrote:

    Benjamin Kaduk has entered the following ballot position for
    draft-ietf-mpls-summary-frr-rsvpte-08: No Objection
    
    When responding, please keep the subject line intact and reply to all
    email addresses included in the To and CC lines. (Feel free to cut this
    introductory paragraph, however.)
    
    
    Please refer to https://urldefense.com/v3/__https://www.ietf.org/iesg/statement/discuss-criteria.html__;!!NEt6yMaO-gk!SGyRA2gLM4trHPJWqXsD2TAPzQ9xZ1xLq3IBuMDFFNYXAFNS-dAL8YO1QCvSbQ$ 
    for more information about IESG DISCUSS and COMMENT positions.
    
    
    The document, along with other ballot positions, can be found here:
    https://urldefense.com/v3/__https://datatracker.ietf.org/doc/draft-ietf-mpls-summary-frr-rsvpte/__;!!NEt6yMaO-gk!SGyRA2gLM4trHPJWqXsD2TAPzQ9xZ1xLq3IBuMDFFNYXAFNS-dAL8YNzdi70mQ$ 
    
    
    
    ----------------------------------------------------------------------
    COMMENT:
    ----------------------------------------------------------------------
    
    Nice and simple.  Thanks!
    
    RFC 2961 says that MESSAGE_ID objects are "always generated and
    processed over a single hop between RSVP neighbors", but IIRC the PLR
    and MP need not be immediate neighbors.  Has this restriction from RFC
    2961 already been lifted by some other document that we can reference
    (e.g., in Section 3.1.x where we say "a MESSAGE_ID object as defined by
    [RFC2961]")?
[TS]: the PLR and MP can be multiple hops away. However after FRR, the signaling from the PLR is tunneled over the bypass tunnel, so the PLR and MP will appear as 1-hop away so RFC2961 is applicable without changes.
    
    Is it generally understood that "Reserved for future use." means "set to
    zero on transmit and ignore on receipt, until further notice"?
[TS]: this was modified to:
NEW:
           Reserved for future use. MUST be set to zero when sending
           and ignored on receipt.
END
    
    Section 1
    
       similar number of LSPs that ingress on the same link.  In the event
       of the failure of the link or neighbor node, the RSVP-TE control
       plane of the node when acting as a PLR becomes busy rerouting
       protected LSPs signaling over the bypass tunnel(s) in one direction,
    
    nit: I think there's a singular/plural (possessive?) mismatch near
    "LSPs".
[TS]: this was revised to:
NEW:
   In the event of the failure of the link or neighbor node,
   the RSVP-TE control plane of the node (when acting as a PLR node)
   becomes busy rerouting protected LSPs over the bypass tunnel(s) in
   one direction . . .
END
    
    Section 3
    
    
       The PLR SHOULD assign the same Bypass_Group_Identifier to all
       protected LSPs that egress the same protected interface and are
       protected by the same bypass tunnel.  This minimizes the number of
       bypass tunnel SFRR groups, and optimizes the amount of signaling
       needed between the PLR and the MP after FRR.
       [...]
       The PLR SHOULD assign the same Bypass_Group_Identifier to all
       protected LSPs that share the egress link, and bypass tunnel as long
       as the protected LSP(s) have the common group attributes, including
       the modified tunnel sender address used for backup path
       identification as described in [RFC4090].
    
    Is one of these a superset of the other?
[TS]: this was revised to eliminate repetition.
NEW:
   A PLR node SHOULD assign the same Bypass_Group_Identifier to all
   protected LSPs provided that the protected LSPs:

   o  share the same outgoing protected interface,

   o  are protected by the same bypass tunnel, and

   o  are assigned the same tunnel sender address that is used for
      backup path identification after FRR as described in [RFC4090].
END
    
       The MP maintains the PLR group assignments learned via signaling, and
       acknowledges the group assignments via signaling.  Once the PLR
       receives the acknowledgment, FRR signaling can proceed as group
       based.
    
    nit: "group-based" is (1) hyphenated, and (2) an adjective, so we need a
    noun to hang it off of.
[TS]: revised to:
NEW:
   The MP maintains the PLR node group assignments learned from
   signaling, and acknowledges the group assignments to the PLR node via
   signaling.  Once the PLR node receives the group assignment
   acknowledgment from the MP, the FRR signaling can proceed based on
   Summary FRR procedures as described in this document.
END
    
       The PLR node that supports Summary FRR procedures adds an Extended
       ASSOCIATION object with B-SFRR-Ready Extended Association ID in the
    
    nit: I'd suggest s/The/A/ since there's probably more than one PLR node
    that meets these criteria, globally.
[TS]: addressed globally, as suggested.
    
    Section 3.1.2
    
       to [RFC2961].  The MESSAGE_ID object flags SHOULD be cleared when
       transmitted by the PLR and ignored when received at the MP.
    
    When might this SHOULD be ignored?  (Are there cases where a MP might
    assign semantics to a received flag that was not intentionally set by
    the PLR with intent to induce those semantics?)
[TS]: this was changed to MUST - see below:
NEW:
   according to [RFC2961].  The MESSAGE_ID object flags MUST be cleared
   when transmitted by the PLR node and ignored when received at the MP
   node.
END
    
       Resv message (with exception of the MESSAGE_ID).  If the fields do
       not match, or if B-SFRR-Ready Extended ASSOCIATION object is absent
       in a subsequent refresh, the PLR node MUST consider the protected LSP
       as not Summary FRR capable.
    
    This "absent in a subsequent refresh" makes me wonder about race
    conditions where the PLR tries to refresh and the MP removes the
    B-SFRR-Ready nature in its Resv, but the PLR attempts to engage the
    bypass before the Resv makes its way to the PLR -- the PLR thinks the
    bundled protection is active but the MP does not.  Is this just "normal
    operation" and not worth worrying about?
[TS]: thanks for highlighting it. This case will recover using normal protocol mechanisms. A new paragraph was added to clarify this case:
NEW:
   A race condition may arise for a previously Summary FRR capable
   protected LSP when the MP node triggers a refresh that does not
   contain the B-SFRR-Ready Extended ASSOCIATION object, while at the
   same time, the PLR triggers Summary FRR procedures due to a fault
   occurring concurrently.  In this case, it is possible that the PLR
   triggers Summary FRR procedurees on the protected LSP before it can
   receive and process the refresh from the MP node.  As a result, the
   MP will receive a Srefresh with a Message_Identifier that is not
   associated with any state.  As per [RFC2961], this results in the MP
   generating an Srefresh NACK for this Message_Identifier and sending
   it back to the PLR.  The PLR processes the Srefresh NACK and replays
   the full Path state associated with the Message_Identifier, and
   subsequently recovering from this condition.
END
    
    Section 3.2.x
    
          The Bypass_Group_Identifier that is previously signaled by the PLR
          using the Extended Association object.  One or more
          Bypass_Group_Identifiers MAY be included.
    
    nit: s/is/was/ (I think?)
[TS]: OK, adopted.
    
          Replacement TIME_VALUES object to be applied to all LSPs
          associated with each of the following Bypass_Group_Identifiers
          after receiving the B-SFRR-Active Extended ASSOCIATION Object.
    
    nit: s/following/preceding/ (I think?)
[TS]: OK, adopted.
    
    Section 3.3
    
       The facility backup method introduced in [RFC4090] takes advantage of
       MPLS label stacking (PLR imposing additional MPLS label post FRR) to
       allow rerouting of protected traffic over backup path.  The backup
    
    nit: s/over backup path/over the backup path/
[TS]: OK, adopted.
    
    Section 3.3.2
    
       Note, an MP may receive more than one RSVP Path message with the B-
       SFRR-Ready Extended ASSOCIATION object from different upstream PLR
       node(s).  In this case, the MP node is expected to save all the
       received MESSAGE_IDs from the different upstream PLR node(s).  After
       a failure, the MP node determines and activates the associated
       Summary Refresh ID to use once it receives and processes the RSVP
       Path message containing B-SFRR-Active Extended ASSOCIATION object
       that is signaled over the bypass tunnel from the PLR, as described
       Section 3.4
    
    What is a "Summary Refresh ID" and where is it defined?  I don't see it
    either here or in RFC 2961.
 [TS]: the correct term that RFC2961 uses is Srefresh Message_Identifier from the MESSAGE_ID object.
NEW:
   After a failure, the MP node determines and activates the
   state(s) associated with the Bypass_Group_Identifier(s) received in
   the RSVP Path message containing B-SFRR-Active Extended ASSOCIATION
   object that is signaled over the bypass tunnel from the PLR node, as
   described Section 3.4
END
    
       The PLR MUST signal non-Summary FRR capable LSPs over the bypass
       tunnel before signaling the Summary FRR capable LSPs.  This is needed
       to allow for the case where the PLR node recently changed a bypass
       assignment and the MP has not processed the change yet.
    
    Talking through this, there's two main cases for "changed a bypass
    assignment", right -- "added an LSP to a group" and "removed an LSP from
    a group"?  For the "removed from a group" case I see how this helps,
    since the PLR sends an explicit change and the MP can assume that the
    explicit change overrides any former group membership.  But I'm not sure
    how/whether this helps with the "added to a group" change.
[TS]: In case of adding a protected LSP to a group, the PLR only declares it Summary FRR capable after the MP acknowledges the B-SFRR back - as described in section 3:
"
   Once the PLR node receives the group assignment
   acknowledgment from the MP, the FRR signaling can proceed based on
   Summary FRR procedures as described in this document.
"

    
    Section 3.4.1
    
       The RSVP_HOP_Object field in the B-SFRR-Active Extended ASSOCIATION
       ID is set to the common RSVP_HOP that was used by the PLR in
       Section 3.3 of this document.
    
    I see something more plausible as a target for this reference in Section
    3.2(.x) than in Section 3.3(.x).
[TS]: this was revised as below:
NEW:
   The RSVP_HOP_Object field in the B-SFRR-Active Extended ASSOCIATION
   ID is set to a common object that will be applied to all LSPs
   associated with the Bypass_Group_Identifiers that are carried in the
   B-SFRR-Active Extended ASSOCIATION ID.
END    
    Section 3.4.2
    
       1.  The RSVP_HOP object is copied from the B-SFRR-Active Extended
           ASSOCIATION ID.
    
       2.  The TIME_VALUES object is copied from the TIMES_VALUE field in
           the B-SFRR-Active Extended ASSOCIATION ID.  The TIME_VALUES
    
    nit: I suggest using a parallel linguistic construction for all the
    steps (e.g., always or never include "from the <FOO> field in").
[TS]: the steps were normalized as suggested.
    
    Section 5
    
       When using procedures defined in this document, FRR (or the reroute
       of protected LSP(s) on to the bypass tunnel) can be activated on per
       group of protected LSP(s).  This allows an intruder to potentially
       impact and manipulate a set of protected LSP that are assigned to the
       same bypass tunnel group.
    
    I'd consider saying something about how "new attacks enabled by
    these mechanisms would also be possible without these mechanisms, just
    at a higher cost in signalling messages" (with the possible exception of
    the race condition I mentioned earlier).

[TS]: Indeed, added. Such attacks are possible even without the mechanisms proposed in this document at the higher cost resulting from the extra per LSP signaling.

Regards,
Tarek