Re: [tsvwg] RDMA Support by UDP FRAG Option

[Tom Herbert <[e-mail]>]

On Sun, Jun 20, 2021 at 9:34 PM Joseph Touch <[e-mail]> wrote:
>
>
>
> On Jun 20, 2021, at 9:11 PM, Tom Herbert <[e-mail]> wrote:
>
>
>
> On Sun, Jun 20, 2021, 8:58 PM Joseph Touch <[e-mail]> wrote:
>>
>>
>>
>> On Jun 20, 2021, at 8:43 PM, Tom Herbert <[e-mail]> wrote:
>>
>>
>>
>> On Sun, Jun 20, 2021, 8:04 PM Joseph Touch <[e-mail]> wrote:
>>>
>>>
>>>
>>> On Jun 20, 2021, at 6:24 PM, Tom Herbert <[e-mail]> wrote:
>>>
>>> …
>>>
>>> Length  is the length  in octets  of this user datagram  including  this
>>> header  and the data.   (This  means  the minimum value of the length is
>>> eight.)
>>>
>>>
>>> Please stop repeating any claim to the contrary, except to note an error in an existing implementation (which should be corrected, not propagated).
>>
>>
>> I am describing how real implementation actually *works*.
>>
>>
>> I do not doubt that, but you *are* propagating an error.
>
>
> That is your opinion.
>
>
> It is explicit in the text cited above.
>
>
>>
>> There is no error because the correct checksum per RFC768 can be produced.
>>
>>
>> That logic is equivalent to “we do math in base 6” because you only ever tried to add 2+2=4.
>>
>> It’s both faulty as logic and incorrect.
>
>
> No, it's not. The algorithms I described are provably correct. One's complement addition is commutative and associative so:
>
> Sum (whole_packet_from_UDP_header) = Sum(UDP_packet) + Sum(Surplus_area)
>
>
> Agreed, where the first one is “whole ***IP** packet from UDP header”. The UDP packet has a length and that’s not it.
>
> So if Sum(Surplus_area == 0)
>
>
> That is not a requirement anywhere in UDP, but note that there are two zeroes in ones complement arithmetic.
>
> then Sum (whole_packet_from_UDP_header) = Sum(UDP_packet),
>
>
> Sure. And if Sum(Surplus_area == 1, then Sum(IP payload) == Sum(UDP packet) + 1.
>
> and therefore rather an offload performs the computation over the just the UDP packet or includes the surplus area they get the same correct result.
>
>
> Said more specifically, an offload that is implemented INCORRECTLY will still work.
>
IETF standards specify behavior of protocols, they do not mandate how
protocols are implemented. An implementation that produces the
specified behavior is correct and conformant with the standard.

> Yes.
>
> Again, we already support that when OCS is included for many versions of the draft now. We don’t support it when it isn’t, but that’s only allowed when UDP CS=0. These two are coupled for reasons other than making erroneous checksum work.
>
> What’s your point?

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.

>
> Joe
>
>