Re: [rtcweb] draft-ietf-rtcweb-rtp-usage-12 Client-to-Mixer Audio Level

[tim panton <[e-mail]>]

On 5 Mar 2014, at 14:05, Cullen Jennings (fluffy) <[e-mail]> wrote:

> 
> Well if people don’t want to allow the application to choose to encrypt this header or not, I will be arguing for MUST NOT encrypt the RTP headers - all of them. 
> 
> A recurring theme at IETF is that people trying to protect E2E privacy, do things that result in the real world having the only way they can build applications is to build middle boxes that pretend to to be an end to the things on both sides. This means that the middle boxes end up with 100% ability to change everything and we loose all the protection we hoped to get with E2E security. 

To me this smacks of special pleading, it would be very useful (for support reasons) if the javascript (optionally)
got access to the names and characteristics of the input devices. We decided as a group that a) the risks of fingerprinting out weighed the benefits b) that we didn’t need a control surface to expose an option for this.

I feel this is an equivalent case, but we seem to be looking at making a different decision.

I’m confident that the desired result could be achieved securely with a combination of webAudio sending appropriate volume data down a data channel set up to the middle box for this purpose.
I realise that would require some extensive rework in the middle box, but thats what happens when you connect a middle box to an ostensibly peer-to-peer system.

Tim.



> 
> 
> On Mar 5, 2014, at 1:59 PM, tim panton <[e-mail]> wrote:
> 
>> 
>> On 5 Mar 2014, at 12:32, Cullen Jennings (fluffy) <[e-mail]> wrote:
>> 
>>> 
>>> I am opposed to mandating that Client-to-Mixer Audio Level RTP header is REQUIRED to be encrypted. It means that we can not really build conferencing system where the mixers do not have access to decrypt the media.
>> 
>> I have to ask:
>> Does that matter enough to weaken user security for everyone else? 
>> 
>> 
>>> 
>>> I would like to propose that the JS application can control  if the header is encrypted or not.  
>> 
>> I am deeply uncomfortable about that. I am not a good enough cryptographer to know, but my instinct
>> is that exposing the fact that (for example) most of the top bits of the ulaw data are zeros in this packet
>> must be a risk.
>> 
>> I’m pretty confident that one could detect the pitch of the speaker and probably fingerprint that down
>> to individuals or languages with enough data.
>> 
>>> 
>>> I am aware of the paper that suggests these audio levels can reveal the contents of the conversation. There is some element of truth to this. There is also some element of truth to the point that if this was true, speech recognition system would work better than they do. Encrypting this or not is a risk that the application using the header extension needs to evaluate, and WebRTC system needs to provide the applications with a control surfaces that allows them to use this in the way the applications desires.
>> 
>> I am also uncomfortable about the number of little ad-hoc control surfaces that we seem to be mandating
>> without an overall design.
>> 
>> My take is that we should push this out to 2.0 when hopefully we will have a designed control surface in place.
>> 
>>> 
>>> I will note that an normal application that used this header could decide if it was encrypted or not, what is different in RTCWeb is that we are building a platform and my view is that this platform should allow the applications build on the platform to decide the tradeoffs of risk for encrypting this or not. 
>> 
>> That isn’t the line we have taken in any other encryption discussion.
>> 
>> Tim.
>> 
>