[tcpm] Re: draft-eggert-tcpm-tcp-abort-timeout-option-00.txt

[Wesley Eddy <[e-mail]>]

On Wed, Apr 21, 2004 at 05:41:56PM +0200, Lars Eggert wrote:
> 
> A negotiation always involves two segments. So in the example you give, 
> the initiator cannot currently continue the negotiation using the first 
> ACK. It made its offer in the SYN and got the response in the SYN-ACK.
>


Ah, that makes more sense.

 
> >Removing the ability to send an ATO on the first ACK would rid the draft
> >of some inconsistency (ie between figure 2 and the first paragraph of 2.1).
> 
> I see no inconsistency - but the wording may need some improvement. The 
> idea is that an INITIATOR can only use a segment with the SYN flag for 
> its ATO offer, i.e., the SYN and SYN-ACK. The RESPONDER will respond in 
> the very next segment, which obviously cannot be the initial SYN but 
> must be either the SYN-ACK or first ACK.
> 
> Does this clarify the intent?


Yes quite a bit.  Although I see the lack of sending an ATO option on the
initial SYN as indicating either "I don't care" or "I don't grok the ATO
option", and so in both cases, it won't really matter what the ATO in the
SYN-ACK is, because the other side will either be cool with it or not
understand it.

Keeping with the spirit of other TCP options, it might be best to put the
ATO option in the SYN if you understand it, and not put it at all in the
SYN-ACK if it wasn't in the SYN, on the goal of not breaking fragile stacks
with options they can't parse.  That would also remove the need to worry
about the option coming in on non-SYN options.  It's typical to use different
option parsing routines for syn-set and syn-unset segments (sort of slow and
fast paths through the option parsing), and so disallowing an ATO option on
the first ACK also simplifies implementation.

-Wes