Re: [tsvwg] I-D Action: draft-ietf-tsvwg-udp-options-08.txt

[Joseph Touch <[e-mail]>]

I’ll check that - thanks for the pointer.

> On Feb 29, 2020, at 9:47 AM, raffaele <[e-mail]> wrote:
> 
> On 2019-12-31 15:11, Joe Touch wrote:
>> Hi, Mike,
>> Many thanks for these detailed comments. They’ll be incorporated in
>> the next rev.
>> Joe
>>> On Dec 28, 2019, at 2:53 PM, C. M. Heard <[e-mail]> wrote:
>>>> - fix the figure and description of OCS to use a 16-bit checksum
>>> The work on this section is incomplete, as it does not specify
>>> that a two-byte pseudo header containing the length of the surplus
>>> area be conceptually prefixed to the surplus area data. As noted in
>>> https://tools.ietf.org/html/draft-fairhurst-udp-options-cco-00,
>>> that pseudo-header is necessary to fix the most common middlebox
>>> traversal issue that has been found (i.e., calculating the UDP
>>> checksum based in the IP payload length instead of the UDP length).
>>> Since the result will not be an overall ones-complement sum of
>>> zero, that language needs to be dropped from the draft.
> 
> 
> Hello everyone,
> 
> I wanted to add one thing about how the pseudo-header (the Options length) is added to the checksum, just in case it hasn't been already considered.
> 
> In the current OCS design the potential misalignment between the start of the Options area and the OCS field is taken into account through a byte swap.
> 
> However, the misalignment between the start of UDP header and the OCS field should be also taken into account, when adding the Options length.
> If OCS field is aligned with UDP header (and thus with UDP Length), the Options length should be added as-is.
> If OCS is not aligned with UDP header, the Options length should be byte-swapped and then added.
> 
> 
> Sorry for unnecessary caveat if you have already considered this.
> 
> Best regards,
> Raffaele Zullo
> 
> 
> 
> 
> 
> 
>>> I would also recommend that the next version of the draft
>>> explicitly mention that for the purpose of the checksum
>>> calculation the surplus area needs to be conceptually
>>> padded by zero bytes that are not actually transmitted
>>> if its start and/or end are on odd boundaries relative
>>> to the start of the UDP data area.
>>>> - define OCS as required when UDP CS != 0
>>> Changes look good.
>>>> - change ACS from a 16-bit CRC to CRC32c
>>> Changes look good, but it should be explicitly stated whether it is
>>> REQUIRED that an option-aware receiver discard a packet with
>>> an incorrect ACS (which I believe is the intent of the draft) or
>>> whether it is at the receiver's discretion (presumably not, since
>>> ACS is listed as one of the options that an options-aware
>>> implementation is required to support).
>>>> There are more than a few other changes underway based on feedback
>>>> from
>>>> the last meeting. The concept behind those changes will be posted
>>>> for
>>>> feedback when developed further for WG feedback before being
>>>> included in
>>>> the next update.
>>>> I.e., please don't treat the fact that the rest hasn't changed as
>>>> anything beyond "TBD".
>>>> Comments on *these changes* welcome, of course.
>>> Editorial: section numbers for Echo (6.) and Experimental (6.1) are
>>> incorrect.
>>> Substantive comments later on the proposed path forward presented at
>>> IETF 106.
>>> Mike Heard