[dccp] Re: RFC - DCCP and NAT

[Ian McDonald <imcdnzl at gmail.com>]

> WRT the discussion: DCCP has several provisions for checksum coverage, even
> having an feature negotiation about it, but all require that at least the
> DCCP header be included in the checksum, so barring tricks like Ian
> suggested we better implement NAT support right now, that this protocol
> isn't yet an RFC and spread the word like was done for ECN.
> 

I was thinking about this as I biked up to university today and had
another thought on how we can tackle this.

Why don't we change the checksum to exclude IP source and destination
for IPv4? IPv4 already does the checksum for source and header so the
data is already covered. We are not at risk of data corruption as it
is already covered. This would remove the need for the NAT layer, and
remove the need for my original solution. I have discussed this with
others here and they think that it sounds good too (they didn't like
my original solution either).

For IP v6 we can leave as is. The reason for this is two fold:
- IPv6 does not checksum the header (unlike IPv4)
- IPv6 doesn't use NAT (correct me if I am wrong)

In some ways I don't like IPv4 and IPv6 being different but they are
already different in this protocol and others already (e.g. IPsec).
The checksum code is in separate routines today so it presents no
issue there.

To me it seems like I have just untied the Gordian knot on this one
but I would certainly appreciate comments on it.

Regards,

Ian