Re: [Idr] AD Review of draft-ietf-idr-bgp-ls-segment-routing-ext-11

[Alvaro Retana <[e-mail]>]

On March 8, 2019 at 10:41:58 PM, Ketan Talaulikar (ketant) ([e-mail])
wrote:

Ketan:

Hi!


...

(1) Consistency in the specification of the new TLVs. Most of the
descriptions point at other documents, but they don't do so in a consistent
manner: some at the start of the section, others when pointing at the
values, etc. Please be consistent! Given that §2.4/2.5 already have a
summary, I personally find the information in the individual sections
redundant. Instead of a statement at the start of a section, I prefer you
to be specific about the values -- see §2.1.2 for an example.
[KT] I have made some changes to address this and will wait for some
clarification from you on this point. Please check my responses on such
points further below.

It looks like we have different opinions. :-)

Proposal: for consistency, let’s use the approach from rfc7752, ok?

In rfc7752 I see a couple of different things:

(1) For TLVs that are formatted exactly as IGP TLVs: simple tables point at
the origin TLV, and (if needed) a short explanation.  For example, look for
TLV 258 (§3.2.2).  The SR Algorithm TLV is an example of this, where the
format is exactly the same as the corresponding IGP TLVs.  No need to even
include a figure for the TLV...

Note that others, like the SR Local Block TLV are in a similar situation:
 the format, with minor exceptions, come from equivalent IGP TLVs.  In this
case, pointing at the specific IGP TLV and mentioning the difference (no
Flags field in OSPF, for example) should be ok.

(2) For TLVs that have values that don’t map to a specific IGP TLV: the
fields are defined in function of where information comes from.  For
example, see the Multi-Topology ID TLV (§3.2.1.5).  I think that the Source
Router Identifier (Source Router-ID) TLV is an example.


Until we agree on the formatting I’m not considering the related comments
(yours or mine) below, with a couple of exceptions.  Maybe we can find some
time next week to talk about this.

...

...
200 Some of the TLVs defined in this document contain fields (e.g. flags)
201 whose semantics need to be interpreted accordingly to the respective
202 underlying IS-IS, OSPFv2 or OSPFv3 protocol. The receiver of the
203 BGP-LS update for any of the NLRIs MUST check the Protocol-ID of the
204 NLRI and refer to the underlying protocol specification in order to
205 parse such fields. The individual field descriptions in the sub-
206 sections below point to the relevant underlying protocol
207 specifications for such fields.

[major] "The receiver of the BGP-LS update for any of the NLRIs MUST check
the Protocol-ID of the NLRI and refer to the underlying protocol
specification in order to parse such fields." This text is a general
statement ("for any of the NLRIs") that seems to want to Update rfc7752,
where Protocol-ID is defined...but I don't think that is the intent, right?
Note that rfc7752 doesn't explicitly mandate that the "receiver...refer to
the underlying protocol specification" -- it just implies it when talking
about the Opaque TLVs.
[KT] Actually in this case, the protocol needs to be used by the consumer
of the information to interpret certain fields in the BGP-LS TLVs. E.g.
Adjacency SID in sec 2.2.1, the flags field is different in OSPF and ISIS
and based on the flags the SID/index/label field needs to be decoded.

IOW, this seems like a significant change as related to other BGP-LS specs.
[KT] I will agree that it is a deviation from other BGP-LS specs including
the base where we have tried to define IGP agnostic flags. I think the
placement of this sentence gives the wrong impression that it wants to
update rfc7752. It may be better placed if this was move (and may be
repeated) for only those TLV fields where this kind of protocol specific
decoding is necessary. I have updated the draft for this.

e.g. for sec 2.2.1, I would propose to add the following text at the end of
the TLV fields description:

"The Flags and, as an extension, the SID/Index/Label fields of this TLV
need to be interpreted accordingly to the respective underlying IS-IS,
OSPFv2 or OSPFv3 protocol. The consumer of the BGP-LS interested in this
TLV information MUST check the Protocol-ID of the BGP-LS Link NLRI and
refer to the underlying protocol specification in order to parse these
fields."

Note also that §5 (Manageability Considerations) has the following text,
which I think is in contradiction of the one above:
[KT] I think this section should be using the term "consumer of the BGP-LS
information" and not "receiver". Fixed this.

Because what the consumer does is outside the scope of this document, I
think that the best path forward is to simply delete the text above
completely.  Note that now (placed in the different sections) is, at best,
redundant (because the sections specify the fields anyway...


...

...

290 Length: Variable.

[major] Yes, it's variable, but it has to be at least 12. What should a
router do if the length is < 12? After that, there are only specific valid
[KT] I have updated for the minimum length for the TLVs. However, it is not
very helpful to try and list all other valid values (in some cases, cannot
even say "in multiple of X").

lengths: 12, 13, 19, 21, etc. What if the length is not one of those values?




I couldn't find guidance in rfc7752. The closest statement is from §6.2.2
(Fault Management), where one of the "checks for determining if a message
is malformed...Does any fixed-length TLV correspond to the TLV Length field
in this document?" This TLV is not fixed-length, but the possibilities are
finite.

Note that rfc7752 says that if the length is not the expected one, then the
whole BGP-LS attribute is discarded. For this specific case, it means that
the controller wouldn't have complete knowledge of the network, even if the
information derived from the IGP is correct...
[KT] I would take this in the generic RFC7752 context and not specific for
this draft.

Yes.  I agree with your comments and point of view.

About the specific lengths above: yes, it’s hard to characterize the valid
lengths.  Given the limitations in rfc7752, what I’m trying to do is to be
able to characterize these TLVs as having a fixed length because that is
the only case covered in rfc7752. ;-)


...

...
339 2.1.4. SR Local Block TLV
...
380 Flags: 1 octet of flags. None are defined at this stage.

[major] Only the corresponding ISIS sub-TLV has Flags defined -- and I
realize that the text above is the same as in
draft-ietf-isis-segment-routing-extensions. I think you really don't want
this field to evolve independently. IOW, please use a description like the
you used for the Flags in the SR-Capabilities TLV.
[KT] Actually, I would prefer the BGP-LS flags and encodings to evolve
independent of the IGPs (but mapping and encompassing their encodings) so
that consumers don't need to be exposed to IGP specifics (where possible)
for decoding/interpretation of values that can be modelled in an IGP
independent manner. We could not do this for some of the initial
TLVs/fields since there wasn't a consensus on this within
authors/contributors. I hope we can going forward unless the WG feels
differently.

I think that the hard/confusing part about this approach would be that some
bits in the flags may be important for the BGP-LS consumers…and others
not.  This means that if the Flags are not in sync, we could run into a
case where no more BGP-LS flags exist for a new (and interesting) IGP flag.

This specific TLV doesn’t bother me too much because there is a Reserved
field in it…so that could be used to diverge, if needed.


...
393 2.1.5. SRMS Preference TLV
...
427 The use of the SRMS Preference TLV is defined in
428 [I-D.ietf-isis-segment-routing-extensions],
429 [I-D.ietf-ospf-segment-routing-extensions] and
430 [I-D.ietf-ospf-ospfv3-segment-routing-extensions].

[major] This document only defines how to carry information in BGP-LS, not
what to do with it. IOW, I think that the last paragraph is not needed, it
is out of scope of this document...
[KT] This is true for almost everything in BGP-LS since it is only a
carrier of information. This info is meant for the consumer of the BGP-LS
info on how this is used. E.g. it helps pick mapping of one SRMS server
over the other - this way the consumer application would be able to get the
right label to prefix mapping as advertised by that server.

But by providing instructions/hints for the consumer we are going beyond
the scope of the document.

IOW, if we want the consumer operation, including semantic checks, etc. to
be in scope…then let’s make it in scope for everything.

In the specific case of this TLV, I think that knowing the source of the
information provides the other background needed and the text above is
still not needed.

...

...
749 2.3.4. Range TLV

751 The range TLV is used in order to advertise a range of prefix-to-SID
752 mappings as part of the Segment Routing Mapping Server functionality
753 [I-D.ietf-spring-segment-routing-ldp-interop], as defined in the
754 respective underlying IGP SR extensions
755 [I-D.ietf-ospf-segment-routing-extensions],
756 [I-D.ietf-ospf-ospfv3-segment-routing-extensions] and
757 [I-D.ietf-isis-segment-routing-extensions]. The Prefix-NLRI to which
758 the Range TLV is attached MUST be advertised as a non-routing prefix
759 where no IGP metric TLV (TLV 1095) is attached.

[major] What do you mean in the final sentence? It sounds as if the IGP
metric TLV and the Range TLV are mutually exclusive and should never be
advertised together. Is that it? If so, what should the receiver do if they
are?
[KT] I have rephrased this since they are not mutually exclusive. The point
about the IGP Metric TLV was that a consumer of a Prefix NLRI originated on
account of a SRMS mapping should not mistake this prefix to be like a
normal prefix reachability - it can detect this by the absence of the IGP
Metric TLV which is included for all "real" prefix reachability
advertisements.

The updated text says that a “consumer...MUST NOT mis-interpret a Prefix
NLRI, that been advertised with a Range TLV...as a normal routing prefix
(i.e. prefix reachability) unless there is also an IGP metric TLV (TLV
1095) attached to it.”

Here we’re giving instructions to the consumer…and they are not clear: what
does “mis-interpret” leads to?

Instead, maybe something like this:  ...a Prefix NLRI, that been advertised
with a Range TLV…is considered a normal routing prefix (i.e. prefix
reachability) only when there is also an IGP metric TLV (TLV 1095) attached
to it…    This way we still convey the purpose, but avoid trying to specify
what the consumer should do.


...

994 A consumer of the BGP-LS information is retrieving this information

995 from a BGP protocol component that is doing the signaling over a BGP-
996 LS session, via some APIs or a data model (refer Section 1 and 2 of
997 [RFC7752]). The handling of semantic or content errors by the
998 consumer would be dictated by the nature of its application usage and
999 hence is beyond the scope of this document. This document only
1000 introduces new Attribute TLVs and an error in them would result in
1001 only that specific attribute being discarded with an error log.

[nit] s/is retrieving this/retrieves this
[KT] ack - fixed

[minor] "...retrieving this information...over a BGP-LS session, via some
APIs or a data model (refer Section 1 and 2 of [RFC7752])." I think that
even mentioning that an API or data model can be used instead of BGP-LS is
a stretch -- that is not how I interpret the initial sections in rfc7752
(which are just background sections), and there are no formal API/data
model definition.
[KT] How else would Alto Server or a PCE component get access to the BGP-LS
information? There would be some APIs or a model. It may not be formal but
there needs to be one?

rfc7752 specifies how the consumer uses a BGP-LS session…not anything
else.  The full sentence is this: "A consumer of the BGP-LS information
retrieves this information from a BGP protocol component that is doing the
signaling over a BGP-LS session, via some APIs or a data model (refer
Section 1 and 2 of [RFC7752]).”

A consumer of BGP-LS uses BGP-LS….

This is a minor point anyway.  The important part is about semantic
checking being out of scope.


...

1009 6. Security Considerations

1011 The new protocol extensions introduced in this document augment the
1012 existing IGP topology information that was distributed via [RFC7752]

Hmmm… It looks like this was also truncated — I wonder why that is
happening.

I put the rest of the review below.

Thanks!

Alvaro.



1009 6.  Security Considerations


1011   The new protocol extensions introduced in this document augment the

1012   existing IGP topology information that was distributed via [RFC7752].

1013   The Security Considerations section of [RFC7752] also applies to

1014   these extensions.  The procedures and new TLVs defined in this

1015   document, by themselves, do not affect the BGP-LS security model

1016   discussed in [RFC7752].


[nit] s/that was distributed/that is distributed


1018   BGP-LS SR extensions enable traffic engineering use-cases within the

1019   Segment Routing domain.  SR operates within a trusted domain (refer

1020   Security Considerations section in [RFC8402] for more detail) and its

1021   security considerations also apply to BGP-LS sessions when carrying

1022   SR information.The SR traffic engineering policies using the SIDs

1023   advertised via BGP-LS are expected to be used entirely within this

1024   trusted SR domain (e.g. between multiple AS/domains within a single

1025   provider network).  Therefore, precaution is necessary to ensure that

1026   the SR information collected via BGP-LS is limited to specific

1027   controllers or applications in a secure manner within this SR domain.


[nit] s/trusted domain (refer Security Considerations section in [RFC8402]
for more detail)/trusted domain [RFC8402]


[nit] s/SR information.The SR traffic/SR information. The SR traffic


[major] "SR operates within a trusted domain...[RFC8402]...and its security
considerations also apply to BGP-LS sessions when carrying SR information."
 The Security Considerations in rfc8404 really only talk about the data
plane -- I don't see how they apply to the BGP-LS sessions.


[major] "...precaution is necessary to ensure that the SR information
collected via BGP-LS is limited to specific controllers or applications..."
 This sounds as if you're referring to information that (once collected)
can be shared between controllers -- I think that case is out of scope.  If
you are trying to talk about the BGP sessions, then I think the language
needs a little more work.  BTW, the paragraph below also talks about BGP
sessions; suggestion: keep the common topics together.


[major] The end of the last sentence says "...in a secure manner within
this SR domain".  Assuming that we're talking about BGP sessions, what does
that mean?  Does it mean anything beyond what BGP is normally specified to
do?  If not, then I would recommend taking that piece of text out to not
invite more questions than needed.


[minor] You talk about "controllers or applications", but in the rest of
the document you have mostly used "consumer" (which is a good thing because
it shows consistency with rfc8402).  Suggestion: use "consumers" here as
well.


1029   The isolation of BGP-LS peering sessions is also required to ensure

1030   that BGP-LS topology information (including the newly added SR

1031   information) is not advertised to an external BGP peering session

1032   outside an administrative domain.


[major] What does "isolation of BGP-LS peering sessions" mean?  Why is it
not Normatively REQUIRED?


[major] From prior experience (see draft-ietf-idr-te-pm-bgp), the
SecDir/Sec ADs asked for text related to why it is ok to transport the new
information in BGP.  This is the text that resulted from that discussion:


   The TLVs introduced in this document are used to propagate IGP

   defined information ([I-D.ietf-lsr-isis-rfc7810bis] and [RFC7471].)

   These TLVs represent the state and resource availability of the IGP

   link.  The IGP instances originating these TLVs are assumed to

   support all the required security and authentication mechanisms (as

   described in [I-D.ietf-lsr-isis-rfc7810bis] and [RFC7471]) in order

   to prevent any security issue when propagating the TLVs into BGP-LS.

   The advertisement of the link attribute information defined in this

   document presents no additional risk beyond that associated with the

   existing set of link attribute information already supported in

   [RFC7752].


Note that this text is complementary to what is already stated in the first
paragraph -- but goes into a more explicit explanation and brings in the
security considerations from the documents where the IGP extensions are
defined.  Consider adding something similar here.



...

1126 9.2.  Informative References

...

1159   [RFC8402]  Filsfils, C., Ed., Previdi, S., Ed., Ginsberg, L.,

1160              Decraene, B., Litkowski, S., and R. Shakir, "Segment

1161              Routing Architecture", RFC 8402, DOI 10.17487/RFC8402,

1162              July 2018, <https://www.rfc-editor.org/info/rfc8402>.


[major] I think this should be a Normative reference.