[sr-dev] How to change the reply msg in the onreply_route

[Min Wang]

Hi Andrei

Thanks so much for the detailed information!

I will try the easy way as you suggested. :)


Kind Regards
 
Min Wang

> -----Original Message-----
> From: Andrei Pelinescu-Onciul [mailto:andrei at iptel.org]
> Sent: Monday, February 22, 2010 5:21 AM
> To: Min Wang
> Cc: sr-dev at lists.sip-router.org
> Subject: Re: [sr-dev] How to change the reply msg in the onreply_route
> 
> On Feb 19, 2010 at 18:24, Min Wang <wang at basis-audionet.com> wrote:
> >
> > Hi Andrei:
> >
> > Thanks for the quick response.
> >
> > > >
> > > > The pseudo codes is like:
> > > >
> > > > t=tmb.t_gett();
> > > > send_request(method_ACK);
> > > > send_request(method_BYE);
> > > > reason_code = 488
> > > > tmb.t_reply(t->uas.request,reason_code,reason_text);
> > > > (Not sure if t_reply can be called here or not.)
> > >
> > > >
> > > > The ACK/Bye is sent to U2, 488 is sent to U1,
> > > >
> > > > But the issue is it seems the original 200 OK is forwarded to U1
as
> > > > well.
> > > >
> > > > How to avoid it?
> > >
> > > You need either to DROP the reply (add drop in the script after
your
> > > function), or use the generic onreply_route ( [0] or unnamed).
> >
> > So:
> > onreply_route[1]
> > {
> >       if (my_function_check("")) {
> >              release_call_onreply("");
> > 			Drop;
> >       }
> >
> > }
> >
> > >
> > > Anyway you would need to modify t_reply, or it won't work from the
> > > onreply_route (at least not in sr 3.*).
> >
> > I guess what you mean is : even with drop, this function seems need
to
> > be modified ( since the reply is 200 OK.)
> 
> Yes, you're right.
> >
> > In the tm/t_reply.c->reply_received:
> >                 run_top_route(onreply_rt.rlist[t->on_reply], p_msg,
> > &ctx);
> >                 if ((ctx.run_flags&DROP_R_F)  && (msg_status<200)) {
> >                         if (unlikely(replies_locked)) {
> >                                 replies_locked = 0;
> >                                 UNLOCK_REPLIES( t );
> >                         }
> >                         goto done;
> >                 }
> >
> 
> Sorry, I forgot about this.
> 
> >
> > What is the good reason for checking msg_status<200 there?
> 
> A final reply cannot be dropped. The timers are already stopped and if
> this is the only active branch, dropping the reply might leave it
> forever in memory (the code expects that either relay_reply() or
> local_reply() is always called for a final reply).
> Dropping 200 would also not make much sense for a sip proxy.
> 
> You could try changing reply_received:
> - replace stop_rb_timer(&uac->request) in  if (msg_status >= 200) with
>   stop_rb_retr(&uac->request)
> - add if (msg_status>=200) stop_rb_timer(&uac_request) after the
>   execution of the reply route (if (t->on_reply) {...} <here> ).
> 
> Note however that you might run into other problems and you would need
a
> lot of testing.
> 
> An easier way is to use the on_reply[0] route. This is executed by the
> core prior to tm's reply_received and tm on_reply routes. If you drop
a
> reply from here, tm won't see it at all. You still need to modify
> w_t_reply_wrp (w_t_reply()) which right now can be executed only from
> request or on failure routes. If you are going to execute it from the
main
> onreply_route (onreply_route[0]{} or onreply_route{}), you would need
to
> call t_reply() like for a request route and not t_reply_unsafe().
> This is different from the case where you would want to execute
> t_reply/w_t_reply() from the onreply_route[ != 0] (in this case you
> would need t_reply_unsafe(), since the transaction is already locked).
> 
> The script will look like:
> 
> onreply_route {
>       if (my_function_check("")) {
>               release_call_onreply("");
>               Drop;
>        }
> }
> 
> and release_call_onreply(""):
> 
>  if (tmb.t_check(msg, &branch)>0){
>     send_request(method_ACK);
>     send_request(method_BYE);
>     reason_code = 488
>     tmb.t_reply(t->uas.request,reason_code,reason_text); # where
>     # tmb.t_reply is the modified version that calls t_reply()
>     # internally
>  }
> 
> 
> Andrei