Re: [rtcweb] JSEP: Why always offer actpass?

[Stefan Håkansson LK <[e-mail]>]

On 30/11/14 14:51, Makaraju, Maridi Raju (Raju) wrote:
>> Hi,
>>
>>>> RFC 7345 (UDPTL-DTLS) says the following:
>>>>
>>>> "Once an offer/answer exchange has been completed, either
>>>> endpoint
>> MAY
>>>> send a new offer in order to modify the session.  The
>>>> endpoints
>> can
>>>> reuse the existing DTLS association if the key fingerprint
>>>> values
>> and
>>>> transport parameters indicated by each endpoint are unchanged.
>>>> Otherwise, following the rules for the initial offer/answer
>> exchange,
>>>> the endpoints can negotiate and create a new DTLS association
>>>> and, once created, delete the previous DTLS association,
>>>> following the same rules for the initial offer/answer exchange.
>>>> Each endpoint needs to be prepared to receive data on both the
>>>> new and old DTLS associations as long as both are alive."
>>>>
>>>> So, I guess that can be read in a way that the setup attribute
>>>> as such
>> does not impact previously
>>>> negotiated TLS roles - unless the key fingerprint values and/or
>>>> transport
>> parameters are modified.
>>>>
>>>> The SCTP-SDP draft currently says that a subsequent offer must
>>>> not change
>> the previously negotiated roles. But, I guess
>>>> we could say something similar as in RFC 7345.
>>>
>>> I have struggled with this language quite a bit from the
>>> implementation
>> perspective. I think we need to explicitly state
>>> that endpoints MUST reuse the existing association if the key
>>> fingerprint
>> values and transport parameters indicated
>>> by each endpoint are unchanged.
>>
>> We could take such general approach.
>>
>> However, for ‘SCTP/DTLS’ (DTLS over SCTP), I assume that the DTLS
>> connection will also be re-established if the underlying SCTP
>> association is re- established - even if no transport parameters
>> etc are changed.
>>
>> Also, RFC 6083 says:
>>
>> “Each DTLS connection MUST be established and terminated within the
>> same SCTP association.”
>>
>>
>>> The way this language reads to me is that endpoints can reuse the
>>> existing
>> association if they want to, but they can create a new one if they
>> don't. Also, when this new
>>> association is created, it is unclear if old setup roles MUST be
>>> preserved
>> or if they MUST be selected based on the current offer/answer
>> exchange. It seems the current
>>> specification language is not strong or clear enough on this
>>> topic.
>>
>> In my opinion, IF a new DTLS connection needs to be established
>> (for whatever reasons), the current roles should be used.
>
> <Raju> When ICE is NOT used, how does the offerer know that the
> answerer does not change the fingerprint and/or transport parameters?
> I guess it does not know. So, offerer has to be prepared for new DTLS
> association by offering actpass. When ICE is used, the answerer can't
> change transport parameter unless offerer does ICE restart (which
> changes offerer transport parameters); Ref [1] is very clear on this
> indicating "DTLS handshake procedure is repeated". However, even when
> ICE is used, I do not find any restriction about answerer not
> changing fingerprint. So, even without ICE restart answerer can
> trigger a DTLS renegotiation by changing its fingerprint.
>
> To conclude all this, IMO whether ICE is used or not, sending actpass
> for all new offers may be cover all possible scenarios.

That is also my conclusion based on the discussion so far.

I also think the JSEP draft as far as detailed out is correct, but info 
about how the implementation should behave for Subsequent answers is 
missing. Text saying that the role must be maintained (unless there is 
an ICE restart) should be put in there.