[OpenSIPS-Users] [OpenSIPS-Devel] Why the best response is 408 instead of 486 when parallel forking?

[Dan Pascu]

On Monday 27 October 2008, Iñaki Baz Castillo wrote:
> 2008/10/27 Iñaki Baz Castillo <ibc at aliax.net>:
> >> I do not think that 6xx should be considered before 4xx or 3xx. 6xx
> >> means global response and if you get a 4xx and a 6xx at the same
> >> time, it is obvious that a device took a global decision that
> >> another device doesn't agree upon. 6xx should only be sent when a
> >> device knows _for sure_ that no other device can answer a call and
> >> it can give a final answer in the name of all devices (which is
> >> practically almost never when you have parallel forking).
>
> I understand what you say. You are talking about devices that decide
> to reply a 6XX. An endpoint should never sent a 6XX (since it could
> also break sequential forwarding to a voicemail server for example),
> but the fact is that, in terms of RFC 3261, a 6XX must break parallel
> and serial forking and must be chosen as winner reply.

It doesn't matter who sends the 6xx (device, proxy, gateway, ...). Once 
you use parallel forking, 6xx should never be used IMO. 6xx means that 
some device/proxy/gateway _knows_ that _no other device_ can answer the 
call for whatever reason. Now if you used parallel forking and you also 
receive a 4xx/3xx from another device, than it is obvious that the other 
device didn't agree with the assumption the first device made. Which 
means that the 6xx reply was a bad assumption. Even more, at some point 
you cannot know if an upstream device did parallel forking or not. So 
even if you are sure you do not do parallel forking and you can assume 
you know there is a final global 6xx answer, you have no idea if that 
won't clash with a 3xx/4xx somewhere upstream.

For that reason, I think that 6xx is unusable, unless used in strictly 
controlled environments that can guarantee no parallel forking end to end 
or if all devices involved in routing a message end to end can 
communicate between them and agree that one can give a final 6xx reply.

So are you saying that we should base the selection logic in the proxy on 
the assumption that devices are well behaved and won't send a 6xx? What 
if a proxy sends a 6xx because a clueless admin wrote a script where he 
used 6xx because he thought they are better? Will you contact all the 
proxies/devices/gateways out there and ask them nicely to fix their 
behavior because your proxy cannot work properly? Don't you see someone's 
ability to cause DOS using this?

-- 
Dan