[tsvwg] On the proposal for 4895bis (SCTP-AUTH)

[Magnus Westerlund <[e-mail]>]

Hi,

In regards to today’s presentation in TSVWG for draft-ietf-tsvwg-rfc4895-bis-03<https://datatracker.ietf.org/doc/draft-ietf-tsvwg-rfc4895-bis/>

On Proposal number 1 on making an empty CHUNKS parameter have the meaning protect all chunks that can be protected. My concern which I tried to express is that as SCTP AUTH is dependent on multiple parameters to and that this proposal creates a dependency for CHUNKS on the HMAC Algo parameter. So the receiver needs to process HMAC-Algo before it can interpret CHUNKS parameter. That I think is a considerable  change to prior. However, I do see the mitigating factor in this case that SCTP-AUTH will process all the three parameters to execute so this non-backwards compatible change in general behavior can be made to work. So I have a no strong opinion on its usage.

A question in regards to this proposal.  To my understanding there are 21 registered chunks. Where four can’t be included. So the current saving is at most 17 bytes. Even if we add a couple of more chunks, we are still only saving some 25 bytes. Is it worth that change for something that is only transmitted once in each direction.

On Proposal 2:
I strongly support addition of replay protection in the AUTH chunk. However, two considerations.

First, it is not necessary to include all the 64-bit of the sequence number, instead it should be sufficient to include only the lower 32-bits and only include the full 64-bits extended sequence number in the authentication tag calculation.

Secondly, the replay window needs to be large enough so that it covers the time period that corresponds to any latenecy differencies (including queuing delay) between different paths. Otherwise the heart beats being sent to verify the availability of alternative paths when multihoming. I think one can consider if one needs to have path specific sequence numbers. However, as there are no formal path identifiers in SCTP, the sender would have to enumerate its source-destination pairs and include that ID in the SCTP-AUTH chunk and include that in the data covered by the authentication tag.

Cheers

Magnus