[Last-Call] Re: [Lsr] 答复: 答复: Re: 【Can you concatenate several pieces together without one "explicit key" to identify them belong to the same segment】Re: Rtgdir last call review of draft-ietf-lsr-multi-tlv-08

"Adrian Farrel" <adrian@xxxxxxxxxxxx> · Thu, 20 Feb 2025 12:05:05 -0000

Just jumping in on this one before I respond to your email to me…

I believe that the key MUST be present in MP-TLV only if it is already defined (and must be present) for the base TLV. 
Thus, this document does not change any key definitions and does not need to. It does not define any new keys or any requirement for new keys.

Why are keys used in base TLVs?
Because multiple TLVs of the same type may legitimately be present to describe different instances of an entity, so they must have a key to distinguish them, show what they are describing, and show they are not accidental duplicates.

MP-TLV introduces no additional need for keys.

Cheers,
Adrian

From: Aijun Wang <wangaijun@xxxxxxxxxxxxxxx> 
Sent: 20 February 2025 04:35
To: 'Joel Halpern' <jmh@xxxxxxxxxxxxxxx>; rtg-dir@xxxxxxxx
Cc: draft-ietf-lsr-multi-tlv.all@xxxxxxxx; last-call@xxxxxxxx; lsr@xxxxxxxx; 'The IESG' <iesg@xxxxxxxx>
Subject: [Lsr] 答复: [Last-Call] 答复: Re: 【Can you concatenate several pieces together without one "explicit key" to identify them belong to the same segment】Re: Rtgdir last call review of draft-ietf-lsr-multi-tlv-08

Hi, Joel:

We should notice the fact，or the difference between them---that----the “key“ information is not important or must-be part for the traditional IS-IS implementation, then there is no any definition in the current RFCs.
But, for MP-TLV, it is the MUST-BE present information. Then, it should be defined explicitly, or else, the interoperability can’t be assured from the implementation of different vendors.

The above is the main differences between the traditional IS-IS implementation and the implementation and deployment of MP-TLV in operator network.
We should exclude the deployment example that negotiated by the vendors offline.

Best Regards

Aijun Wang
China Telecom

发件人: Joel Halpern [mailto:jmh@xxxxxxxxxxxxxxx] 
发送时间: 2025年2月20日 12:06
收件人: Aijun Wang <wangaijun@xxxxxxxxxxxxxxx>; rtg-dir@xxxxxxxx
抄送: draft-ietf-lsr-multi-tlv.all@xxxxxxxx; last-call@xxxxxxxx; lsr@xxxxxxxx; 'The IESG' <iesg@xxxxxxxx>
主题: Re: [Last-Call] 答复: [Lsr] Re: 【Can you concatenate several pieces together without one "explicit key" to identify them belong to the same segment】Re: Rtgdir last call review of draft-ietf-lsr-multi-tlv-08

Aijun, may I suggest another way of phrasing what I am hearing people who disagree with your analysis are asserting?  (If I get it wrong, I am sure they will correct me.)
1) We have working IS-ID
2) In order to function, IS-IS implementations already must recognize when they receive two conlicting dvertisements
2') That is, for every advertisement that can possibly be duplicated with variation, there already is a well-defined notion common to all implementations of the key for those advertisements
3) The multi-tlv effort you are objecting to uses that same keying to allow for the case where the two variations that are received are to be combined, instead of merely selecting one.
4) Given that this necessary behavior has been operation for MANY years, the apparent conclusion is that it is already sufficiently specified.
Yours,
Joel
On 2/19/2025 10:56 PM, Aijun Wang wrote:
Hi, Adrian:

Thanks for your detail analysis and logical deduction! 
It help us to understand the arguments of each other. Let me give my thoughts to your analysis:

First, I must point out that the base of your assumptions is NOT CORRECT: 
There is no such “Fixed part” that can be used as the “key” information for slicing the large TLV, defined in the related RFCs, for every possible MP-TLVs that are listed in draft-ihttps://datatracker.ietf.org/doc/html/draft-ietf-lsr-multi-tlv-09#section-9

We should all know there are fixed fields defined in these TLVs/sub-TLVs, but that can’t be solely used as the “Fixed part” to concatenate the several pieces together.

We can take the TLV 22(Extended IS Reachability) as one example(also in this WGLC document). 
For clarity, I extracted the related content from the document directly(https://datatracker.ietf.org/doc/html/draft-ietf-lsr-multi-tlv-09#section-3.2.1):

As an example, consider the Extended IS Reachability TLV (type 22). A neighbor in this TLV is specified by:¶
7 octets of system ID and pseudonode number¶
3 octets of default metric¶
Optionally one or more of the following link identifiers:¶
IPv4 interface address and IPv4 neighbor address as specified in [RFC5305]¶
IPv6 interface address and IPv6 neighbor address as specified in [RFC6119]¶
Link Local/Remote Identifiers as specified in [RFC5307]¶
The key consists of the 7 octets of system ID and pseudonode number plus the set of link identifiers which are present.
We can check the original definition of RFC for TLV 22----RFC 5305----There is no any “key” definition information in this RFC----Verify what I have said above that there is no key definition for every possible MP-TLVs in current RFCs.

What constitutes the key of TLV 22 is only mentioned in this WGLC document------“The key consists of the 7 octets of system ID and ……”

Even we continuing our discussions based on such key definition information, can you tell me what is the “Fixed part” that you mentioned when slicing the TLV in your example?
I think you should know, according to the description of this document, “the set of link identifiers” is optionally. 

I can image if the sender/receiver are from one vendor, they can keep the “Fixed part” in different pieces same, but, how to ensure they are same if the devices from different vendors?

Comparing to this undefined and ambiguity of “Fixed part”, or “key” information for each possible MP-TLV, other issues that you raised(slicing not in the boundary of sub-TLV, temporarily duplicate of a TLV in two LSPs etc. can be easily mitigated).

I think you admitted the “key” must be represented in every of piece of the sliced TLV/sub-TLVs(for the “keyed TLV” that you mentioned). 
Then, where is “key” definition for each of such TLV/sub-TLVs?

Best Regards

Aijun Wang
China Telecom

发件人: forwardingalgorithm@xxxxxxxx [mailto:forwardingalgorithm@xxxxxxxx] 代表 Adrian Farrel
发送时间: 2025年2月19日 19:21
收件人: 'Aijun Wang' <wangaijun@xxxxxxxxxxxxxxx>; 'Les Ginsberg (ginsberg)' <ginsberg@xxxxxxxxx>; rtg-dir@xxxxxxxx
抄送: draft-ietf-lsr-multi-tlv.all@xxxxxxxx; last-call@xxxxxxxx; lsr@xxxxxxxx; 'The IESG' <iesg@xxxxxxxx>
主题: [Lsr] Re: 【Can you concatenate several pieces together without one "explicit key" to identify them belong to the same segment】Re: Rtgdir last call review of draft-ietf-lsr-multi-tlv-08

The correct formation of a TLV includes certain fields that must be present and other fields (such as sub-TLVs) that may be present and may recur. All instances of a TLV must be formatted correctly. That means that when a TLV is repeated as part of an MP-TLV each component TLV must be correctly formatted.

In other words, t is incorrect to read this document as simply saying that the “overflow” appears in a second TLV structure with the same Type, and with the data continuing. 

Consider (please enable non-proportional font)

| T=t | L | Fixed part | sub-TLV1 | sub-TLV2 | sub-TLV3 |
            1 ...                ... 255 ...

There is too much to fit into one TLV structure.
The incorrect MP-TLV would be…
| T=t | L | Fixed part | sub-TLV1 |
| T=t | L | sub-TLV2 | sub-TLV3 |

The correct MP-TLV would be
| T=t | L | Fixed part | sub-TLV1 |
| T=t | L | Fixed part | sub-TLV2 | sub-TLV3 |

There are two cases:
TLVs that have a key defined
TLVs that don’t have a key defined

If the TLV has a key, that is usually found in the fixed part.
Thus, the key is found in both component TLVs.

What are the concerns?
That for a keyed TLV, the component TLVs will not be correctly associated. 
Not a problem because the key is present in each component TLV.
That the component TLVs will not be concatenated in the right order
Not a problem because the sub-TLVs (i.e., the not fixed part) are allowed to be
present in any order.
That fragments of a sub-TLV will be concatenated in the wrong order.
Not a problem because partitioning must be at a sub-TLV boundary
That the fixed part will differ in component TLVs.
This is identified as an error.
How do I know which fields in the fixed part comprise the key?
If you have implemented support for a TLV, you must understand the key.
If you have not implemented support for the TLV, you don’t care because you 
ignore the TLV.
Your question:
Can you concatenate several pieces together without one “explicit key” to identify them belong to the same segment?
I say “yes”. Because, without a key you know that the TLV is “one of a kind”. 
The only legitimate duplication of a TLV is when it includes a key.
Note: There are two cases identified in the document where (without MP-TLV) an un-keyed TLV may be duplicated.
sender makes an error by including multiple versions of a TLV
the sender is transiting a TLV from one LSP to another (so it appears in both LSPs)
These are both handled by allowing concatenation to proceed, and the composed TLV to be discovered to be incorrectly formatted.
For the first case, this will result in the errored composed TLV being discarded. Good.
For the second case, there will be a transitory case where the errored composed TLV is discarded, but when the LSP is refreshed, everything will resolve. 

My question back to you would be to ask you to give an example where this goes wrong?

Thanks,
Adrian

From: Aijun Wang <wangaijun@xxxxxxxxxxxxxxx> 
Sent: 19 February 2025 02:40
To: adrian@xxxxxxxxxxxx; 'Les Ginsberg (ginsberg)' <ginsberg@xxxxxxxxx>; rtg-dir@xxxxxxxx
Cc: draft-ietf-lsr-multi-tlv.all@xxxxxxxx; last-call@xxxxxxxx; lsr@xxxxxxxx; 'The IESG' <iesg@xxxxxxxx>
Subject: 答复: [Lsr] 【Can you concatenate several pieces together without one "explicit key" to identify them belong to the same segment】Re: Rtgdir last call review of draft-ietf-lsr-multi-tlv-08

Hi, Adrian:

I have read your Rtgdir reviews and update suggestions on the current document. Appreciate for your efforts!
But, I don’t know why you ignore the issues that you have concerned previously also about the “explicit key” definition of each possible IS-IS TLV/sub-TLV.(Please refer to: https://mailarchive.ietf.org/arch/msg/lsr/Xga5YrD6ObbNDvFFK_nVDXlipJA/).

Let’s try to don’t loop the past arguments and make the life better.
Then, would you, and also other experts/reviewer that pass this document answer the following simple, straightforward question:

Can you concatenate several pieces together without one “explicit key” to identify them belong to the same segment?

If the answer is “can”, please tell me how?
If the answer is “can’t”, then, where is necessary “explicit key” in this document? 
And, if there is none of such “explicit key” information, what’s the value of this document?

Let’s be clear for the further discussion, the implicit negotiations solution to solve the interoperability is not STANDARD solution.

I copied this discussions also to the IESG mail list for further evaluations of the IESG experts.

Best Regards

Aijun Wang
China Telecom

发件人: forwardingalgorithm@xxxxxxxx [mailto:forwardingalgorithm@xxxxxxxx] 代表 Adrian Farrel
发送时间: 2025年2月14日 18:10
收件人: 'Les Ginsberg (ginsberg)' <ginsberg@xxxxxxxxx>; rtg-dir@xxxxxxxx
抄送: draft-ietf-lsr-multi-tlv.all@xxxxxxxx; last-call@xxxxxxxx; lsr@xxxxxxxx
主题: [Lsr] Re: Rtgdir last call review of draft-ietf-lsr-multi-tlv-08

Many thanks for your responsiveness, Les.

V9 addresses all of my nits adequately. I will update the status of my review in the Datatracker.

As to…

> That said, after posting V9, I thought maybe something like what I show below
> might be added to Section 4. I am not convinced it is needed – but let me know
> what you think.
> 
> “Each TLV that is part of an MP-TLV MUST be parsable independent of other
> TLVs in the MP-TLV. Breaking of a single sub-TLV or other data unit across TLVs
> MUST NOT be done.”

I agree with you that this is “not needed.” 
I do believe it would be somewhat helpful to avoid people falling into error.
I leave it to you and your co-authors to decide.

Cheers,
Adrian

-- 
last-call mailing list -- last-call@xxxxxxxx
To unsubscribe send an email to last-call-leave@xxxxxxxx