Re: [mpls] George can yu look at this - Re: end of WGLC, RE: working group last call for draft-ietf-mpls-lsp-ping-reply-mode-simple-01

[Nobo Akiya <[e-mail]>]

Hi Adrian,

Many thanks for very helpful comments. I have carefully read your comments
and made modifications to section 3.2 in my private copy.

[OLD]

   1.  The Reply Mode Order TLV MAY be included in MPLS echo request.

   2.  The Reply Mode Order TLV MUST NOT be included in MPLS echo reply.

   3.  The Reply Mode field of an MPLS echo request MUST be set to a
       valid value even when supplying the Reply Mode Order TLV.  The
       initiator LSR SHOULD set the Reply Mode field of MPLS echo
       request to a value that corresponds to a return path which most
       likely to be available, in case the responder LSR does not
       understand the Reply Mode Order TLV.

   4.  If a responder LSR understands the Reply Mode Order TLV but the
       TLV is not valid (due to conditions described in the items 6, 8
       and 9 immediately below), then the responder LSR MUST only use
       the value described in the Reply Mode field of received MPLS echo
       request.

   5.  If a responder LSR understands the Reply Mode Order TLV and the
       TLV is valid, then the responder LSR MUST consider the Reply Mode
       values described in the TLV and MUST NOT use the value described
       in the Reply Mode field of received MPLS echo request.  In other
       words, a valid Reply Mode Order TLV overrides the value specified
       in the Reply Mode field of received MPLS echo request.

   6.  Reply Mode Order TLV MUST contain at least one Reply Mode value,
       and SHOULD contain at least two Reply Mode values.

   7.  A Reply Mode value, except for Reply Mode value 5 (Reply via
       Specified Path), MUST NOT be repeated (i.e., MUST NOT appear
       multiple times) in the Reply Mode Order TLV.

   8.  The Reply Mode value 5 (Reply via Specified Path) MAY be included
       more than once in the Reply Mode Order TLV.  However, in such
       case a Reply Path TLV MUST be included for all instances of the
       Reply Mode value 5 included in the Reply Mode Order TLV.  In
       other words, 3 instances of the Reply Mode value 5 in the Reply
       Mode Order TLV will require 3 instances of the Reply Path TLVs.

   9.  The Reply Mode value 1 (Do not reply) MUST NOT be used in the
       Reply Mode Order TLV.

   If a responder LSR receives a Reply Mode Order TLV which does not
   comply to the rules described above, then the responder LSR MUST
   ignore the Reply Mode Order TLV.


[NEW]

   1.  The Reply Mode Order TLV MUST NOT be included in MPLS echo reply.
       If the initiator LSR receives an MPLS echo reply with the Reply
       Mode Order TLV, the initiator LSR MUST ignore the whole Reply
       Mode Order TLV and MUST only use the value from the Reply Mode
       field of the received MPLS echo reply.  It may be beneficial for
       implementations to provide counters and/or loggings, with
       appropriate log dampening, to record this error case.

   2.  The Reply Mode Order TLV MAY be included in MPLS echo request.

   3.  The Reply Mode field of an MPLS echo request MUST be set to a
       valid value even when supplying the Reply Mode Order TLV.  The
       initiator LSR SHOULD set the Reply Mode field of MPLS echo
       request to a value that corresponds to a return path which most
       likely to be available, in case the responder LSR does not
       understand the Reply Mode Order TLV.

   4.  If a responder LSR understands the Reply Mode Order TLV but the
       TLV is not valid (due to conditions described in the items 6, 7,
       8 and 9 immediately below), then the responder LSR MUST ignore
       the whole Reply Mode Order TLV and MUST only use the value from
       the Reply Mode field of the received MPLS echo request.  It may
       be beneficial for implementations to provide counters and/or
       loggings, with appropriate log dampening, to record this error
       case.

   5.  If a responder LSR understands the Reply Mode Order TLV and the
       TLV is valid, then the responder LSR MUST consider the Reply Mode
       values described in the TLV and MUST NOT use the value described
       in the Reply Mode field of received MPLS echo request.  In other
       words, a valid Reply Mode Order TLV overrides the value specified
       in the Reply Mode field of received MPLS echo request.

   6.  Reply Mode Order TLV MUST contain at least one Reply Mode value.

   7.  A Reply Mode value, except for Reply Mode value 5 (Reply via
       Specified Path), MUST NOT be repeated (i.e., MUST NOT appear
       multiple times) in the Reply Mode Order TLV.

   8.  The Reply Mode value 5 (Reply via Specified Path) MAY be included
       more than once in the Reply Mode Order TLV.  However, in such
       case a Reply Path TLV MUST be included for all instances of the
       Reply Mode value 5 included in the Reply Mode Order TLV.  In
       other words, 3 instances of the Reply Mode value 5 in the Reply
       Mode Order TLV will require 3 instances of the Reply Path TLVs.

   9.  The Reply Mode value 1 (Do not reply) MUST NOT be used in the
       Reply Mode Order TLV.


Thanks!

-Nobo


On Sat, Apr 25, 2015 at 9:17 AM, Adrian Farrel <[e-mail]> wrote:

> Hi Nobo,
>
>
>
> Let's come back to my original review comments.
>
>
>
> > Section 3.2 gives clear instructions and guidance on forming the Reply
>
> > Mode Order TLV but not on what to do if a received TLV deviates from the
>
> > MUST and MUST NOT instructions. Options might include ignoring errors,
>
> > ignoring the TLV, ignoring the message. But presumably not sending an
>
> > error response (because how would you know how to send it?)
>
>
>
> You responded:
>
>
>
> | [NOBO] That’s a good point. What the document really state is that The
>
> | Reply Mode value 5 (Reply via Specified Path) MAY be repeated but all
>
> | other Reply Mode values MUST NOT be repeated. Will update the
>
> | document to make it clear.
>
>
>
> That's a fine thing to say, but doesn't answer my question about how an
> implementation is supposed to behave when a received Reply Mode Order TLV
> is "malformed".
>
>
>
> You posted draft-ietf-mpls-lsp-ping-reply-mode-simple-02. Section 3.2
> swapped the order of points 4 and 5, and enhanced the text to read...
>
>
>
>    4.  If a responder LSR understands the Reply Mode Order TLV but the
>
>        TLV is not valid (due to conditions described in the items 6, 8
>
>        and 9 immediately below), then the responder LSR MUST only use
>
>        the value described in the Reply Mode field of received MPLS echo
>
>        request.
>
>
>
> That seems to cover most bases. Good.
>
> It might also be helpful to say "...the responder LSR MUST silently ignore
> the whole Reply Mode Order TLV and MUST only use the value from the Reply
> Mode field of the received MPLS echo request."
>
> Saying this helps clarify the behavior.
>
>
>
> The only things I don't find covered are:
>
>
>
> a. how to handle a violation of the seventh point
>
>
>
>    7.  A Reply Mode value, except for Reply Mode value 5 (Reply via
>
>        Specified Path), MUST NOT be repeated (i.e., MUST NOT appear
>
>        multiple times) in the Reply Mode Order TLV.
>
>
>
>    I think you can address this by adding "7" to the list of conditions
> in point 4.
>
>
>
> b. how to handle a violation of the second point
>
>
>
>    2. The Reply Mode Order TLV MUST NOT be included in MPLS echo reply.
>
>
>
>    I think this will need additional text.
>
>
>
> I note that after the numbered bullets you also now have the following
> text...
>
>
>
>    If a responder LSR receives a Reply Mode Order TLV which does not
>
>    comply to the rules described above, then the responder LSR MUST
>
>    ignore the Reply Mode Order TLV.
>
>
>
> This is better than point 4 and covers everything.
>
> You might consider whether "ignore" means "silently ignore" or whether you
> want to give advice about logging. It seems probable that (because of the
> nature of ping) if you do suggest logging, you also want to describe some
> form of thresholding or damping.
>
>
>
>
>
> To address Loa's concern...
>
> The TLV is from the optional range. It can be ignored if it is not
> understood, and does not require to be included.
>
> Loa asserted that the document says that the TLV MUST be included, but I
> don't find this in -02.
>
> Therefore, I think this is all fine.
>
>
>
> Adrian
>
>
>
> *From:* Nobo Akiya [mailto:[e-mail]]
> *Sent:* 25 April 2015 05:50
> *To:* [e-mail]
> *Cc:* Loa Andersson; t.petch; Ross Callon; mpls;
> [e-mail];
> [e-mail]
> *Subject:* Re: [mpls] George can yu look at this - Re: end of WGLC, RE:
> working group last call for draft-ietf-mpls-lsp-ping-reply-mode-simple-01
>
>
>
> Hi Adrian, Tom, Loa,
>
>
>
> This extension is currently structured such that it allows for backwards
> compatibility (i.e., transit LSR not supporting this mechanism will not
> return "malformed request" ... because the TLV is optional).
>
>
>
> The result is that this mechanism becomes a best effort mechanism, and we
> will not get the full benefit until all LSRs along the LSP (and other LSRs
> which could falsely receive the echo request) implements this extension.
>
>
>
> One way to allow the initiator LSR to determine whether or not the
> responder LSR understood this TLV, and still keeping the backwards
> compatibility, is that we keep the Reply Mode Order TLV as an optional TLV,
> but require (i.e., MUST) the responder LSR understanding this TLV to
> include the Reply Mode Order TLV in the echo reply, potentially with result
> of parsing/handling that TLV.
>
>
>
> However, that's probably not the path we want to go, as all (or most)
> optional TLVs will have to do something similar (i.e., include the same TLV
> in the echo reply). This will quickly result in echo reply packet bloat ...
> we should prevent that as echo reply usually tends to include more
> information (ILS, DSMAP/DDMAP per nexthop, multipath Sub-TLVs per nexthop,
> etc).
>
>
>
> To me, the right way to solve this is to create a capability TLV that
> mandates the reponder LSR to return which features it supports in bitmaps
> or something very compact. But this can be done in a separate effort/draft.
>
>
>
> In short, my preference for this document is to go as is (after
> incorporating comments from Tom).
>
>
>
> Thanks!
>
>
>
> -Nobo
>
>
>
> On Thu, Apr 23, 2015 at 4:38 PM, Adrian Farrel <[e-mail]>
> wrote:
>
> Thanks Loa,
>
> You captured it. The edge cases need to be nailed down.
>
> Personally I have no particular preference for where it is nailed, but I
> don't
> want it flapping in the breeze.
>
> A
>
> > -----Original Message-----
> > From: mpls [mailto:[e-mail]] On Behalf Of Loa Andersson
> > Sent: 23 April 2015 13:59
> > To: t.petch; Nobo Akiya
> > Cc: Ross Callon; mpls; [e-mail];
> draft-ietf-mpls-lsp-ping-reply-
> > [e-mail]
> > Subject: [mpls] George can yu look at this - Re: end of WGLC, RE: working
> group
> > last call for draft-ietf-mpls-lsp-ping-reply-mode-simple-01
> >
> > Tom,
> >
> > The question you and Adrian asks is what to do if the Reply Mode
> > Order TLV is missing.
> >
> > There is something fishy here, I'd like George to look at this.
> >
> > The LSP Ping design says that TLVs from this range may be silentsly
> > dropped, my take is that we don't need to specify anything more
> > than that.
> >
> > Now, we say that it MUST be present, and after thinking around a bit
> > I wonder if the TLV should be assigned from the mandatory range instead?
> >
> > Or if the MUST be present means that the message will be malformed if
> > it is not there, and the message should be discarded.
> >
> > OK - now I'm confused.
> >
> > /Loa
> >
> > On 2015-04-23 13:02, t.petch wrote:
> > > ---- Original Message -----
> > > From: "Loa Andersson" <[e-mail]>
> > > Sent: Wednesday, April 22, 2015 2:15 PM
> > >
> > >> Tom,
> > >>
> > >> On 2015-04-20 00:07, Nobo Akiya wrote:
> > >>>      <tp>
> > >>>
> > >>>      Yes but ... I think that it is a change of meaning.  Is is
> > > enough just
> > >>>      to ignore the TLV or should the whole PDU be discarded?  I find
> > > it
> > >>>      difficult to know but don't feel strongly about that choice so
> > > will go
> > >>>      with what you suggest.
> > >>>
> > >>>      </tp>
> > >>
> > >> So I don't misunderstand what you are saying. It seems to me like the
> > >> comments made by Adrian and you actually requires a "change of
> > > meaning",
> > >> that is kind of essence of a "comment", right?
> > >>
> > >> As for what to do with if the TLV is not recognized, it is
> > >> intentionally requested from a space where it can be silently dropped
> > >> (i.e. "ignored").
> > >>
> > >>      The new TLV Type value should be assigned from the range
> > >>      (32768-49161) specified in [RFC4379] section 3 that allows the
> TLV
> > >>      type to be silently dropped if not recognized.
> > >>
> > >>        Type   Meaning                            Reference
> > >>        ----   -------                            ---------
> > >>        TBD1   Reply Mode Order TLV               this document
> > >>
> > >> What is it that I miss?
> > >
> > > Nothing serious.  My initial thought was to echo Adrian, that, at least
> > > in this context, there should be an indication what to do if a MUST or
> > > SHOULD was violated without just then having a clear sense of what it
> > > should be instead.
> > >
> > > The I-D did require (MUST) one entry in the TLV and wanted (SHOULD)
> > > more.  Adding what to do if that did not happen I was seeing as
> > > clarification.  I then read Nobo as proposing going a bit further
> saying
> > > requires (MUST) one or more.  Which might lead to boxes taking a
> > > simplistic approach and always putting in the new TLV with a single
> > > entry and ignoring the traditional TLV.  Not a problem just a change
> > > from what others might think that they have consented to.
> > >
> > > On the question of what to do when the rules are violated, again I did
> > > not initially think of what the action should be.  On reflection, I am
> > > still unsure.  I understand that the Reply Mode Order TLV  is optional
> > > and so can be ignored when not understood; that's fine.  But if it is
> > > understood and can be seen to be defective, should the box with that
> > > knowledge discard just that TLV and accept the remainder of the
> message?
> > > Or should it argue that if this TLV is defective, then likely the rest
> > > is as well and should be ignored?  I am unsure.
> > >
> > > If there is scope for a breach of security, or taking a hit in
> > > performance, then ignore is the right policy. If the requirement is to
> > > get as much data as possible from a failing network, then use it is the
> > > right policy.  As long as the I-D is clear, I am not too fussed which
> > > way it goes.  I am content with the changes that Nobo has proposed.
> > >
> > > Tom Petch
> > >
> > >> /Loa
> > >>
> > >> --
> > >>
> > >>
> > >> Loa Andersson                        email: [e-mail]
> > >> Senior MPLS Expert                          [e-mail]
> > >> Huawei Technologies (consultant)     phone: [phone]
> > >
> >
> > --
> >
> >
> > Loa Andersson                        email: [e-mail]
> > Senior MPLS Expert                          [e-mail]
> > Huawei Technologies (consultant)     phone: [phone]
> >
> > _______________________________________________
> > mpls mailing list
> > [e-mail]
> > https://www.ietf.org/mailman/listinfo/mpls
>
>
>