Re: [Perc] Double questions

[Nils Ohlmeier <[e-mail]>]

> On Mar 21, 2017, at 15:58, Eric Rescorla <[e-mail]> wrote:
> - Normal header
> - Non-modifiable Extensions
> - OHB
> - Modifiable Extensions
> - Encrypted Payload
> 
> And now you encrypt that. As I understand it, the OHB, if present,
> must replicate the inner header:
> 
>       The OHB MUST replicate the information found in the
>       RTP header following the application of the inner cryptographic
>       transform.
> 
> Does that mean it must be of 3-byte length? If so, why, because the
> outer transform shouldn't be changing any of these values. I.e.,
> at the point where the packet gets to the media relay, PT and SEQ
> should match between the header and OHB.

Yes that is my current reading of the draft that if the originating
client adds the OHB it needs to duplicate PT and SEQ even though
it has obviously not modified these.
I guess the question here is if it is worth to have a shorter OHB
header for the scenarios where PT and SEQ are the same (which could
be used by originating client as well as relays which don't change
any of this)?

> 
> DECRYPTION
> Finally, upon decryption, you strip the outer transform, and get:
> 
> - Normal header
> - Non-modifiable Extensions
> - OHB'
> - Modifiable Extensions'
> - Encrypted Payload
> 
> You then remove the modifiable extensions, stomp the stuff in the
> normal header with the OHB, and then strip the inner transform.

To be precise I think you need to take OHB and modifiable extensions out
to make the decryption work. After the decryption you re-insert the
modifiable extensions into the packet, because they might be of interest
for your RTP stack.

> I note that S 5.3. says:
> 
>    o  The PT from the outer SRTP packet is used for normal matching to
>       SDP and codec selection.
> 
>    o  The sequence number from the outer SRTP packet is used for normal
>       RTP ordering.
> 
> So, does that mean that the original PT/SEQ (as reconstructed from
> OHB) are not used for anything other than to make the decryption work?
> They're totally discarded?

That is my understanding, yes.

  Nils