[Serusers] NAT traversal solution by reinvite?

[Greger V. Teigre]

Federico Giannici wrote:
> Greger V. Teigre wrote:
>> Are you trying to solve NATing for devices behind the same NAT or
>> behind different NATs?
>> If it's the second, you must remember that the NAT problem includes
>> being able to traverse the NAT with packets from the outside.
>> Normally, a symmetric meda stream is based on incoming packets
>> coming on the same port as outgoing and the outgoing must have been
>> initiated first. Couple this with the fact that restricted and
>> port-restricted NATs will stop incoming from other IPs than the
>> source of the outgoing packet and that symmetric will not work
>> anyway, you may be able to solve full cone NATs.  However, this is
>> the type of NAT that most STUN clients discover correctly anyway.
>
> I agree with you for the case of symmetric NATs, but they wouldn't
> work with STUN neither...

That's correct.

> For port-restricted NATs I don't understand the problem. If a
> re-invite is sent to a different IP, then the NAT will be opened for
> that IP when the UA will start the stream. Don't he?

Yes, if the port stays the same and the UA initiates the stream right away.

> A problem could be if, for the re-invite, the UA use a different port
> for RTP then the one used in the previous invite!

Exactly, but this is not the UA, it's the NAT allocating the port (you 
didn't answer whether you are solving the problem of two UAs behind same NAT 
or behind different NATs).

> Is this permitted in the RFC?

The UA will probably use the same port, but a reINVITE may even change IP... 
The problem is the NAT allocating another port on the public interface.
g-)

> Is this common in UAs implementation?
>
> Thanks.
>
>
>
>> Federico Giannici wrote:
>>
>>> I'd like to ask to somebody with more knowledge of me if a possible
>>> solution to NAT traversal is really feasible.
>>>
>>> For various reasons, we DON'T want to use an RTP proxy.
>>> We'd like to avoid the use of STUN because: 1) creates hairpin
>>> problems; 2) many UAC have a bad STUN client code implementation; 3)
>>> it requires additional configuration by the final user.
>>>
>>> It seems to me that with the nathelper's message rewriting functions
>>> it is possible to solve every problem for the SIP protocol.
>>>
>>> Moreover, as we have the REAL IP of the UA (in the original SIP
>>> messages) we could also avoid haipin problems: it is sufficient to
>>> use the original IP/Port of the two UAs if both have the same
>>> natted IP (ie they are behind the same NAT). This doesn't work when
>>> the UAs are behind multiple NATs, but this is a relatively uncommon
>>> case. So, the unresolved problem is with the RTP data, because we don't
>>> know what will be the NATted port so we cannot correctly mangle the
>>> SDP data in the INVITE message.
>>>
>>> Am I correct up to this point?
>>>
>>> Now, I'm asking myself if it is feasible to use a "MINI RTP Proxy"
>>> that receives the initial INVITEs, discovering the NATted RTP ports,
>>> and then IMMEDIATLY RE-INVITE the two UAs to connect directly each
>>> other. So only the first RTP packet is actually proxed, all
>>> subsequent traffic will be directly between the two UAs.
>>> I think that something similar is done by Asterisk.
>>>
>>> Is this feasible?
>>>
>>> If it is, then we could have a good solution to NAT Traversal:
>>> 1) No Hairpin problems (for one NAT cases)
>>> 2) No problems of the normal RTP proxy (waste of bandwidth, longer
>>> delays, bad scalability).
>>> 3) Will work with all type of NATs except for symmetric ones (the
>>> same that work with STUN).
>>> 4) Simpler UAC configuraton: only username, password and sip server.
>>>
>>>
>>> Thanks.