[SR-Users] rtpproxy and CANCEL in serial forked scenarios

[Jan Klepal]

Hello,

On Friday 10 of February 2012 14:02:08 Daniel-Constantin Mierla wrote:
> Hello,
> 
> On 2/9/12 5:13 PM, Andreas Granig wrote:
> > Hi,
> > 
> > I ran into a scenario with couple of serial forks where kamailio loops
> > to itself and, due to the looping, the INVITE to a new branch happens
> > before the CANCEL to an old branch. What I do at the moment is force
> > rtpproxy in branch route, and stop rtpproxy in the failure route.
> > 
> > The problem with this scenario is that in rtpproxy_offer and
> > unforce_rtpproxy, the rtpproxy module only passes the call-id and
> > from-tag to rtpproxy (because there is no to-tag yet), which then in
> > unforce_rtpproxy for a CANCEL deletes all calls related to it (because
> > it can only match on from-tag and call-id, obviously). This means that
> > when I do a subsequent rtpproxy_answer for my new branch, rtpproxy
> > doesn't find the session anymore, since it has been removed with the
> > CANCEL.
> > 
> > A fix I can think of is to also pass the branch of the top-most via to
> > rtpproxy in order to perform a more fine-grained matching. Are there
> > solutions to that out there somewhere, or is it something I should just
> > introduce, eg. as a new param to offer/answer/unforce functions?
> > Objections? Other approaches?
> 
> if you do rtpproxy handling on an instance that received a parallel
> forked call, then it is better to add the condition on top via branch, I
> don't see other solution for now and I am not aware of an implementation
> for it. I would add a new flag to rtpproxy parameters to send the branch
> value, that will allow to use old rptproxy versions when the flag is not
> given.
> 

I've experienced similar problem with no-answer forwarding. In my scenario 
call was torn down from rtpproxy in onreply_route when "487 Request 
terminated" was received from B phone (upon CANCEL). I am using Kamailio 3.2 
default configuration with rtpproxy_manage() and workaround is to check if 
transaction is canceled for negative replies in onreply_route and only then 
call rtpproxy_manage().

Can anybody think of scenario when this workaround will not clean up RTP proxy 
session?

I checked rtpproxy_manage function and patch should be really simple:

--- a/modules/rtpproxy/rtpproxy.c
+++ b/modules/rtpproxy/rtpproxy.c
@@ -1736,8 +1736,12 @@ rtpproxy_manage(struct sip_msg *msg, char *flags, char 
*ip)
                                        (ip!=NULL)?1:0);
                }
        } else if(msg->first_line.type == SIP_REPLY) {
-               if(msg->first_line.u.reply.statuscode>=300)
+               if(msg->first_line.u.reply.statuscode>=300) {
+                       if(tmb.t_gett!=NULL && tmb.t_gett()!=NULL
+                               && !(tmb.t_gett()->flags & T_CANCELED))
+                               return -1;
                        return unforce_rtp_proxy_f(msg, 0, 0);
+               }
                if(nosdp==0) {
                        if(tmb.t_gett==NULL || tmb.t_gett()==NULL
                                        || tmb.t_gett()==T_UNDEFINED)

Cheers,
Jan