Re: [Ietf-behave] Conflicting ICMP payload translation reqs: MUSTvs. SHOULD
I'm just watching out for the poor Gen-ART reviewer who gets the last
document in this string...
- Is it the intention that the ICMP document UPDATEs the UDP document, in
the RFC Editor sense of the word?
- Is it the intention that the TCP document updates the ICMP document?
Given that the UDP draft is in the RFC Editor Queue, and the ICMP and TCP
documents are not even assigned to a shepherding AD in the tracker yet, your
best path forward really may be to issue UDP, UPDATE it with ICMP, and refer
to ICMP for the ICMP requirements supporting TCP... assuming that the TCP
draft doesn't pass the ICMP draft in the chute.
This does tend to make our archival document series denser and less
implementer-friendly, but sometimes "good enough" is the best we can do.
If I got this hand grenade over the review transom, I'd definitely be asking
this at IETF Last Call time, and probably at IESG Telechat time as well, so
thinking about this now is probably better than thinking about it some other
times...
Thanks,
Spencer
I full agree with Fernando. I do believe, all normative requirements
related to
ICMP should be in the ICMP document and not spread across multiple
documents.
Repeating in multiple drafts in different words is prone to language
interpretation problems. I also believe, a NAT MUST rewrite the payload.
As for the requirement in UDP draft (soon to be an RFC) - During the WG
last
call when this issue was brought up, folks wanted to leave the requirement
in
there so the publication is not delayed (I dont recall discussig the
SHOULD vs
MUST for the requirement, however.).
This exception for the UDP draft is understandable. However, the
requirement
shoudlnt have to be repeated in the TCP draft. TCP draft could simply
refer the
requirement in the ICMP draft rather than list as a distinct TCP
requirement.
My 2c.
regards,
suresh
--- Fernando Gont <fernando@xxxxxxxxxxx> wrote:
At 19:45 24/10/2006, Saikat Guha wrote:
>TCP currently aligns itself with ICMP/MUST because receiving the ICMP
>messages and being able to link them to the actual connection is
>essential for path MTU discovery.
>
> From my reading of the documents, the requirements in ICMP/TCP and UDP
>are in conflict.
>
>Is there any particular UDP has a SHOULD and not a MUST? If so, should
>the recommendation be narrowed only to UDP payloads inside ICMP packets?
As a meta comment, I'd suggest that all the ICMP-related stuff be
included only in the ICMP document, rather than spread it over three
documents (ICMP, TCP, and UDP).
As for whether to rewrite the ICMP payload is a MUST or SHOULD, I'd
vote for the "MUST".
Whether one likes it or not, ICMP is a signalling protocol that is
part of the TCP/IP suite. You cannot just NAT the transport protocol,
but kill the protocol that aids in signalling error conditions.
If anybody wants to get rid of ICMP, the one should get rid of it for
"non-NATed" traffic, and only then the NAT-related specs should follow.
Just my two cents,
