Re: [tsvwg] incorrectly coalesce packets [was: Re: RDMA Support by UDP FRAG Option]

[Joseph Touch <[e-mail]>]

> On Jun 20, 2021, at 5:52 AM, Rodney W. Grimes <[e-mail]> wrote:
> 
> Joe,
> 	I caught this comment and wish for some eloboration so I
> can better understand what may of already been identified, where it
> is called out, and what can be done to make corrections.
> 
> ...
>> 
>> In TCPM we?ve identified a number of additional issues with TCP offload, notably regarding how they incorrectly coalesce packets with different TCP headers.  
>> 
>> If we?re not calling out these behaviors as the bugs they are, there?s little point in doing much of anything in the IETF.
>> 
>> Joe
> 
> I have a larger scope on the "different TCP headers" in that IMHO
> it should be clearly spelled out that even presently reserved bits
> should be considered when coalesce of packets occur, as if it is
> not considered potential future uses of these reserved bits may cause
> issues.  An example is that in our experimenting with SCE we found
> that some ACK thinning code out there combines ACK's that have
> different values in the former NS bit we used for feeding back
> ESCE (Echo Some Congestion Experienced).  Investigation of the
> code infact yielded that it ignored all reserved bits in its
> decision to combine packets.
> 
> Has there been any published RFC text that addresses these problems?

None that I am aware. We mention it in draft-ietf-tcpm-edo as a concern. In TCPM, there were proposals to require an EDO integrity check to detect such erroneous coalescing, but we decided there not to require it for similar reasons - i.e., that it’s a losing game to design protocols against deliberate incorrect implementation.

It would be useful to explore current offloading behavior regarding unknown options, though.

Joe