[tsvwg] summary of issues for draft-touch-tsvwg-udp-options-05.txt

[Joe Touch <[e-mail]>]

Hi, all,

Thanks for the lively discussion! I hope this bodes well for accepting
this as a WG doc. Again, I'd prefer if I had deployable running code,
but we do have multiple implementations that worked, just not ready for
prime time.

Here's a quick summary of the issues just raised, both for me (for the
next edit) and in case any new issues are of interest.

Please do let me know if I missed or incorrectly summarized anything
below...

Thanks!

Joe

-----------------------------------------------------

Editorial:

- fix Identification field length in Fig 15v s. text (should be 32 bits)

-----------------------------------------------------

Use cases:

- add DNSSEC impact on segment size and IP fragment problems with
network traversal

    add ref to draft-muks-dns-message-fragments app-layer solution

    note that FRAG works well when request is small because it can
indicate FRAG support for large response

- discuss impact to UDP API

    who sets these? OS or app?

    which ones are transparent? which are NOT?

-----------------------------------------------------

Option design:

FRAG:

    - should FRAG have a reassembly checksum?

    - should FRAG have a post-reassembly length verification?

    - use LITE to support backwards transparent FRAG (thanks, C. M. 
Heard!) (if so, need final reassembly checksum!)

    - what happens if LITE and FRAG are present with non-trivial
covered/not-covered areas?

    - rename "segmentation" to avoid confusion with IP layer mechanism?
(or is SAR too overloaded with ATM?)

ACS:

    - address ACS interaction with LITE and FRAG (ACS covers UDP
payload, but NOT IP or UDP header except len?)

    - clarify ACS covers each received UDP segment or fragment, NOT
reassembled result

    - indicate which CRC-16 polynomial

OCS:

    - clarify OCS covers header within each received UDP segment or
fragment, NOT reassembled result

    - should OCS be mandatory, and if so, make it just the UDP checksum
on the option area?

LITE

    - revise Sec 7 vs. UDP Lite and omit EOL discussion (not relevant
given new swap mechanism)

EOL

    - explain possible uses (given removal from Section 7)

    - if removed, consider changing NOP to 0

Another security option?

    - HMAC for IPv6 SR option(see Tom Herbert's GUE variant)

ALL (general):

    - revisit MUST IMPLEMENT list (add MSS, FRAG, TIME?)

    - create a list of required ordering

        OCS first (always)

        AE next (if it appears)

        TIME, MSS

        FRAG (last? otherwise security/safety mechanisms wouldn't
protect the reassembly process?)

-----------------------------------------------------