Re: [rtcweb] SDP Security Descriptions (RFC 4568) and RTCWeb

[Martin Thomson <[e-mail]>]

On 25 April 2013 16:19, Eric Rescorla <[e-mail]> wrote:
> I think we've discussed most of the topics Martin raises already and I'm
> not sure I have much to add (which isn't to say I necessarily agree).

I just wanted them on the record.

> However, I think the below is incorrect and I'm not sure I have made
> that point before.
>
>> Practical Matters of Security - You aren't necessarily more secure
>> because you are using DTLS-SRTP.  The complete security story requires
>> not just DTLS-SRTP, but also that both peers use identity assertions.
>> Otherwise its security properties are basically the same as security
>> descriptions.
>>
>> The default mode of operation for getUserMedia is to return media that
>> is accessible to the web site.  The same for RTCPeerConnection.  That
>> means that the site can see and modify your media in your browser,
>> even if it can't tamper with it on the network.
>
> It's certainly true that the site has access to the media with DTLS if you
> don't use identity assertions/isolated streams. However, what it doesn't
> have is *invisible* access. I.e., it must do something that is user visible,
> which allows for the detection of cheating by the site. By contrast, if SDES
> is used
> then the site can simply passively monitor all your traffic, or at least
> any that goes through its network and you can't detect it.

That's a fair point, and one I missed.  My bad.  But it's only
relevant to the extent that it is possible to observe that the
provided certificate is a phony.  It's not like the server is
obligated to provide a particular CN, so it's mostly down to just
looking at fingerprints.

It's also hard to retroactively audit due to the way (current)
implementations mint new certificates for every interaction.  I'm
fairly sure that (current) implementations don't make this
particularly easy to examine in browser chrome, but even if they did
it's hard to imagine a good usability story for accessing the
information.