Re: [rtcweb] Asking TLS for help with media isolation

[Paul Kyzivat <[e-mail]>]

On 4/4/14 4:50 PM, Martin Thomson wrote:
> Fair questions...
>
> On 4 April 2014 13:42, Paul Kyzivat <[e-mail]> wrote:
>> - IIUC this isolation is a browser thing. If the other end doesn't
>>    have a browser then what does it mean?
>
> Yes, this is specific to a browser.  A non-browser peer would probably
> respect this isolation property without any effort at all.  After all,
> few non-browser applications permit the running of arbitrary,
> remotely-source code in the way that a browser does.

But ISTM there would need to be a different set of limitations for 
non-browsers. For instance, to not echo the stream back to the source. 
And not to store the stream in a place that the sender can retrieve.

Maybe it is possible to define a set of rules, but I don't have a vision 
of what that would be.

>> - Isn't the isolation supposed to be on a per-stream basis?
>>    (Or is it really on a peer-connection basis.) If it is per-stream
>>    then how does signaling it on a per-TLS basis work?
>
> The question of scope is an important one.  The isolation property is,
> at least at the most abstract layer, applicable to a MediaStreamTrack.
>   I believe that the right approach is to broaden this scope to
> RTCPeerConnection, where the identity controls exist.  This means that
> you won't be able to mix isolated and non-isolated streams in the same
> RTCPeerConnection.

I'll wait to see what others have to say about this.
I haven't been able to wrap my head around it.

>> - The use of DTLS to do SRTP keying is a sort of hack, since it
>>    isn't about the DTLS payload at all. I would think that whatever
>>    new you introduce into TLS for this would most naturally apply
>>    to the DTLS payload (the data channels).
>
> This is true only in the strictest academic or abstract sense.  We are
> already taking things from the DTLS handshake and using those things
> to make assertions about the properties of the SRTP.  Certificates.
>
> Since you raise data channels, I'll note that we have no way to use
> data channels in an isolated fashion.  Thus, I conclude that if we
> intend to make this property consistent, we should not be using data
> channels in the same RTCPeerConnection as media streams.  Of course,
> I'm open to taking the pragmatists approach too, unless we think that
> there is a use case for - say, file transfer - where it makes sense
> for data to be isolated in the same way I'm proposing that media be.

Again I'm interested to hear what others have to say about this.

	Thanks,
	Paul