[rtcweb] Encryption consideration: SDES and TLS-SRP

["Fabio Pietrosanti (naif)" <[e-mail]>]

Hi all,

i wanted to ask the WG opinion on the encryption framework for the RTCWEB.

In particular as we know today there is a strong need to have an
encryption-by-default on most new protocols.

Today:
- Email provider encryption by default traffic
- Mobile operators encrypt by default traffic
- Almost all web transaction use encryption

So it's realistic and practical to think that VoIP future is
encrypted-by-default to at least basically protect against the casual
wiretapper. (Can we evaluate this assumption?)

In that case we may have different threat scenario:
- Eavesdropping by third party with access to user's communication line
- Eavesdropping by who is providing the telephony service

A simpler protection scenario would at least provide protection against
third party with access to user's communication line.
A stronger protection scenario would also provide protection against who
is providing the service.

Here we should consider the security of:
- Signaling
- Media

Signaling has to be protected with TLS.
Media has to be protected with SRTP that rely on DTLS or on the new
proposal to use SDES [1] (idea that i like).

So basically in both cases there is the need to rely on a TLS-secured
channel and here we should evaluate the possible weak points.

As you know the TLS based trust model based on "Browser/OS Pre-loaded
Trusted CA" revealed serious risks posed to internet trust model [2] .

So, because we are defining a stack that should look forward the future,
we cannot not-consider seriously the serious issues related with TLS and
way to manage it.

An idea that came to my mind is that the specification, to face a
future-proof approach to the internet trust issues, could mandate to
support the implementation of TLS-SRP.

TLS-SRP is a new standard that let the TLS key exchange be authenticated
by a login/password with no need to trust a specific CA.
Status of implementation of RFC5054 http://tools.ietf.org/html/rfc5054
is available on http://trustedhttp.org/wiki/Main_Page .

If we embrace SDES key exchange and TLS-SRP authentication the overall
protocol stack would be simpler than having to deal with DTLS and
relying only on Trusted-CA architectures.

[1]
http://www.ietf.org/proceedings/interim/2012/01/31/rtcweb/slides/rtcweb-7.pdf
[2] http://ritter.vg/p/2012-TLS-Survey.pdf

p.s. I'm quite new to the list, so i may repeat some discussion
elements. In that case i kindly ask to leave to archive pointer to
already made discussion.