Re: [rtcweb] Data channel: Handling of packets on unknown channels

[Harald Alvestrand <[e-mail]>]

Den 28. juni 2018 16:35, skrev Michael Tuexen:
>> On 28. Jun 2018, at 16:22, Harald Alvestrand <[e-mail]> wrote:
>>
>> Den 28. juni 2018 14:51, skrev Michael Tuexen:
>>>> On 28. Jun 2018, at 13:30, Harald Alvestrand <[e-mail]> wrote:
>>>>
>>>> In considering the datachannel API, we encountered one interesting race
>>>> condition:
>>>>
>>>> A: <configure for datachannel>
>>>>
>>>> A: CreateOffer(), SetLocalDescription(), send SDP
>>>>
>>>> B: SetRemoteDescription, CreateAnswer, SetLocalDescription, send SDP
>>>>
>>>> B: Configure an externally defined data channel, with #3249
>>>>
>>>> B: Send a message on #3249
>>>>
>>>> A: SetRemoteDescription
>>>>
>>>> A: Wait a while (THE PAUSE)
>>>>
>>>> A: Configure #3249
>>>>
>>>> Now, if a message comes in to A on #3249 during THE PAUSE, what is the
>>>> implementation to do?
>>> Isn't that some kind or error condition?
>>>
>>> If that it true, one could apply:
>>>
>>>   If a message with an unsupported PPID is received or some error
>>>   condition related to the received message is detected by the receiver
>>>   (for example, illegal ordering), the receiver SHOULD close the
>>>   corresponding data channel.  This implies in particular that
>>>   extensions using additional PPIDs can't be used without prior
>>>   negotiation.
>>
>>
>> The receiver can't close the datachannel if the datachannel doesn't
>> exist yet, so this doesn't work for that case.
> I was assuming that the SCTP receives a user message on a stream. When
> this message is delivered to its upper layer, doesn't this layer know
> that there is no data channel? I would assume that this layer triggers
> the stream reset procedure. I'm not saying that the user (for example
> via a JS API) is involved... I'm more talking about implementing this
> iside the browser..


Exactly, it could close the stream, but it can't close the data channel
since it doesn't exist.

I think closing the stream would be a mistake, since that would make the
outcome about whether you end up with the datachannel or not racy;
discarding data will give you a working datachannel once A gets around
to configuring it.