[rtcweb] DTLS, DTLS-SRTP, and 5-tuples

[Simon Perreault <[e-mail]>]

This is a follow-up to this thread on discuss-webrtc:
https://groups.google.com/forum/#!topic/discuss-webrtc/i67xq-FWtoM

I see two distinct problems:

1. It must be possible for a DTLS connection to jump from one ICE 
candidate to another. That is, a DTLS connection must not be bound to a 
single 5-tuple. This can be deduced from a good understanding of ICE, 
but it could certainly be stated more clearly.

2. SRTP keying material extracted from a DTLS connection must also not 
be bound to a single 5-tuple, in case media jumps from one ICE candidate 
to another. That's more problematic because RFC 5764 states very clearly 
that keying material is to be bound to the DTLS 5-tuple, modulo forking.

If I'm not mistaken, the right thing to do would be to bind keying 
material to all local ICE candidates (multiple 3-tuples), and treat any 
incoming media flow as indicated in RFC 5764 page 15, with one mapping 
table per 3-tuple. For outgoing flows, I guess it would make most sense 
to bind keying material to the 5-tuples for which the sender has 
obtained consent.

Is this correct? If so, would it be useful to add text explaining this 
in detail?

Simon