Re: [tsvwg] RDMA Support by UDP FRAG Option

[Joseph Touch <[e-mail]>]

> On Jun 21, 2021, at 7:34 AM, Tom Herbert <[e-mail]> wrote:
...
> IETF standards specify behavior of protocols,

They DEFINE protocols, including their behavior.

> they do not mandate how
> protocols are implemented.

Agreed, except that the implementation is faulty unless if follows the definition.

> An implementation that produces the
> specified behavior is correct and conformant with the standard.

If the ones you cite always produced correct answers, yes.

But your claim that “I’ve never seen an error” is like saying “but my [broken] watch is fine because I only look at it twice a day - at 8am each time”.

Just because you never experienced a corner case doesn’t mean your code is correct. It means the code hasn’t been sufficiently tested.

…
> In my opinion, the best chance for UDP options to be deployable and
> successful is to require that the surplus area always sums to zero.

When UDP CS==0, OCS provides that capability. When OCS isn’t zero, it doesn’t (and cannot) matter to the UDP layer. It should never matter to any other layer, because that layer is either entirely inside the UDP data or completely encloses the IP packet. There are no protocols that encapsulate UDP but not its outer IP; UDP (like all IETF transport protocols) is not even defined in the absence of an IP wrapper.

If you have a case you believe falls outside what I’ve just indicated, please provide it. I believe it will demonstrate an error in how protocols are encapsulated, not an error where OCS will be required when UDP CS == 0.

Joe