Re: [dccp] Soliciting input on UDP encapsulation for DCCP

[Michael Welzl]

Hi!

but I think that the partial checksum extension header should also
include
the UDP header, because it is applied when the UDP header

checksum is zero - which means that there is no other ckecksum applied

on the UDP header... DCCP could, and probably should, make up for

that.

[Tom P.] This is a good point (and the first time it's been brought up
to my knowledge).  There are problems however.  The UDP header (ports,
at least) may have been changed by a NAPT in the path, which would
invalidate the partial checksum in the DCCP header (if it included the
UDP header).

A possible workaround to this would be to include only the UDP length
field in the DCCP partial checksum (we know that the checksum field is
zero so there's no need to include that).  This would give at least

protection for the UDP fields that shouldn't be changed in the network.

Does that work for you?

yes... without port numbers that doesn't help much though, i guess,
but it's probably better than nothing

[Tom P.] So are you suggesting that we also define DCCP encapsulated in UDP-Lite? I'm not opposed to that, but I'm not sure I see much benefit for DCCP, as UDP-Lite has the same NAT traversal problems that DCCP has.

yes, i was suggesting that. it should be easy to do, but you got me
convinced, i think it's probably pointless

cheers
michael