Re: [tsvwg] Fwd: New Version Notification for draft-touch-tsvwg-udp-options-05.txt

[Joe Touch <[e-mail]>]

Hi, Derek,


On 2/28/2017 1:46 PM, Derek Fawcus wrote:
> On Tue, Feb 28, 2017 at 01:17:16p.m. -0800, Joe Touch wrote:
>> Hi, Tom,
>>
>> OK - so I think you and Derek are basically arguing for an Internet
>> checksum to verify reassembly, rather than using a new or stronger one.
> Actually,  I made a mistake.  I attributed to Tom what was I believe
> was actually suggested by C. M. Heard - namely the suggestion to
> combine FRAG+LITE as a way of allowing existing hosts to discard fragments.
>
> In that he was suggesting adding an overall reassembled checksum,
> which would be necessary as otherwise the repurposed reassembled
> Lite payloads would not have a checksum.  I was simply pointing
> out that in that scenario,  the same could be achieved with a
> single CHECKSUM option, carried in only one of the FRAG packets.
It matters whether the options are considered to apply before or after
reassembly.

The trick is what is being reassembled - is it the UDP payload only or
something larger?

Again, I need to think about this a little more. I'd really like to
stick to something simple.

> So there would be no need to change the FRAG header definiton to
> add a checksum there.
>
> In the normal case of reassembling FRAG'ed payloads,  I don't see
> that an overall checksum gains us anything.
>
> Possibly an overall CRC would have some value?  That would seem to
> require a second type of ACS-like option, i.e REASSEMBLE-ACS?
> Again it would only need to be sent with only one of the FRAG
> containing packets.  Is such an addition worth it?

The two options are:

- add a checksum inside the FRAG option that applies the reassembled whole
- define the options in such a way that some apply before reassembly
(e.g., before each FRAG option in each received UDP packet) and some
apply after (we'd have to be careful in defining what is created as a
result.

The second variant is cleaner if there's a way to do it. However we have
some possibly conflicting issues:

    - OCS clearly ought to apply for each fragment (otherwise you might
err in reassembly and wait too long to catch it)
    - AE should apply to each fragment (same rule as OCS - you have
enough info to prevent an attack on reassembly)
    - ACS would usefully apply after reassembly
    - FRAG shouldn't appear more than once
    - TIME is useful only before FRAG
    - MSS is useful only before FRAG (and we might want two different
MSS values - the conventional MSS and a UDP reassembly MSS!)
 
Again, this deserves a bit more digestion...

Joe
    -