Hi Daniel,
Taking in considerations your input I did some changes to the previous proposal:
1. I used the SEND_PR_BUFFER wrapper for sending out all messages
2. I moved the declaration of "struct ip_addr ip" variable towards the start of the relay_reply() function and removed the other style infringing variable send_res ;
3. I safeguarded the run_onsend() call with checking first that p_msg is neither NULL nor set to FAKED_REPLY.
As far as I understand, FAKED_REPLY should be used when automatically locally generating a reply, right? Additionally, locally generated replies which are triggered from the config via the t_reply() function are not processed here. However, only the code which passes through the t_reply() is served by the event_route[tm:local-request], right?
Would you say that the current patch addresses all the statefully relayed replies cases?
Apart from this, without some more knowledgeable input, I don't know any other unifying solution for this patch (e.g. call onsend route for both forwarded and local generated messages as you suggested).
Thank you,
Lucian
On 12/04/2014 12:58 AM, Daniel-Constantin Mierla wrote:Hello,
On 03/12/14 18:14, Lucian Balaceanu wrote:
Hi Daniel,
Indeed, locally generated replies provoke a segmentation fault in my code because of the missing msg pointer in such cases.
I am currently testing a version where I generate the msg from the buf to be sent by using the parse_msg() function.
Conceptually, do you see any problems with this approach?
there are other attributes that might not be set in this case, such as source ip/port, local ip/port...
Cheers,
Daniel
Thank you,
Lucian
On 11/27/2014 10:37 AM, Daniel-Constantin Mierla wrote:Hi Lucian,
the patch is still in radar, just that last month was crazy busy from various aspects and didn't get the time to handle it.
On 27/11/14 09:09, Lucian Balaceanu wrote:
Hi Daniel,
I must admit this run_onsend() patch for stateful replies creation was not quite a success story. However, I think it serves a practical purpose, for example in Homer tracing and could be useful for the community. Again, I propose my past solution, with some questions:
1. I am unsure if the place I introduced the run_onsend call is appropriate since the buf used for msg_send is constructed
build_res_buf_from_sip_req and build_res_buf_from_sip_res calls.
These functions are to build the reply from request (local generated reply) or from the incoming reply (forwarding reply).
I am wondering what would take to call onsend route for both forwarded and local generated messages (no matter is request or reply) -- might become simpler by having the code in the wrapper function sending to the network from core, on the other hand could break scripting as no msg structure is available for local generated messages. Otherwise the benefit can be also in coherency.
2. Also, we can maybe unite these if call branches I created:
send_res = msg_send(&uas_rb->dst, buf, res_len);
send_res = SEND_PR_BUFFER( uas_rb, buf, res_len );
SEND_PR_BUFFER seems to be a wrapper around msg_send() for safety and debugging processes, so it can be used.
3. Do you think a get_send_socket snippet as follows should be inserted before the if (onsend_route_enabled(SIP_REPLY)){ :
if(dst.send_sock == NULL) {
dst.send_sock=get_send_socket(msg, &dst.to, dst.proto);
if (dst.send_sock==0){
LM_ERR("cannot forward reply\n");
}
}
Is the above referring to the patch attached? If the send_sock is needed elsewhere after the IF block, then should be inserted before.
One remark, new variables should be declared at the beginning of the blocks, you mixed the style with two new vars you added -- this is more a suggestions as we try to stay compatible with old c standards -- I know it is not everywhere in the code, but we should aim at it.
Cheers,
Daniel
Thank you,
Lucian
On 10/29/2014 02:15 PM, Daniel-Constantin Mierla wrote:Hello Lucian,
I applied your patch with some fixes.
I haven't checked with stateful replies, at some point a function from core should be used. You can go ahead and see if it works, if not, let me know and I can look into it as well. You can follow the callbacks for TMCB_RESPONSE_OUT or TMCB_RESPONSE_FWDED inside tm code, they should lead to the place where a sip response is going to be sent out.
Cheers,
Daniel
On 27/10/14 12:51, Lucian Balaceanu wrote:
Hello Daniel,
I must admit I only saw your mail last Friday. Until the 10th of October I was also on vacation. I know that you actually committed some of the changes together with your comments on the 12th this month.
I don't know if we can consider the topic of the patch closed. As far as I understand, the state-full replies have not been addressed, right? (There should be a change in the t_reply.c) I followed the code to the relay_reply but I did not yet come to find the send function. Should I pursue further?
Thank you,
Lucian Balaceanu
Hi Lucian,
somehow I forgot to follow up on this. But we need to get sorted out soon, before we release, so it works as expected with the new version. See more comments inline.
On 17/09/14 18:09, Lucian Balaceanu wrote:
Hi Daniel,
Please forgive me for my delay in responding to your mail.
Please find attached a second version of the onsend_route_reply patch (which again has some problems). As per your previous indications I did the following:
Issue1
From performances point of view, there can be added a config parameter to enable running of onsend_route for replies:
onsend_route_reply = 0|1
Following http://www.asipto.com/pub/kamailio-devel-guide/#c08add_parameters I have tried to add onsend_route_reply parameter. The code compiles, but when trying to start kamailio with this parameter inside, the parsing fails with syntax errors signaling:
0(1321) :<core> [cfg.y:3423]: yyerror_at(): parse error in config file kamailio-basic.cfg.4.1, from line 107, column 1 to line 108, column 0: syntax error
0(1321) : <core> [cfg.y:3423]: yyerror_at(): parse error in config file kamailio-basic.cfg.4.1, from line 107, column 1 to line 108, column 0:
ERROR: bad config file (2 errors)
The issue is:
+<INITIAL>{ONSEND_RT_REPLY} { yylval.intval=atoi(yytext); + yy_number_str=yytext; return NUMBER; } It should be: +<INITIAL>{ONSEND_RT_REPLY} { yylval.intval=atoi(yytext); + yy_number_str=yytext; return ONSEND_RT_REPLY; }
Issue2
#define onsend_enabled(rtype) (onsend_rt.rlist[DEFAULT_RT]?((rtype==SIP_REPLY)?onsend_route_reply:1):0)That is to say you see it best to take the chek for onsend_rt.list[DEFAULT_RT] from inside run_onsend() function and call this onsend_enabled(...) before the run_onsend()?
This is to detect whether the onsend_route should be executed for SIP replies. The condition being:
- if is a sip reply and onsend_route is set and the onsend_route_reply parameter is 1
Stateful processing for replies is mainly done in t_reply.c from tm module. At some point there should be a send buffer function call.
Issue3
On the other hand, is onsend_route also executed for local requests? I had in mind it is only for received requests that are forwarded ... Iirc, on onsend_route, the sip message is the one received, the outgoing content being accessible via $snd(buf).I agree with you with taking out the locally generated requests and only left the run_onsend call in do_forward_reply function (inside forward.c).
Could you point me to the reply relaying function that is called for state-full processing?
Cheers,
Daniel
Thank you and sorry again for my late answer,
Lucian
-- Daniel-Constantin Mierla http://twitter.com/#!/miconda - http://www.linkedin.com/in/miconda
-- Daniel-Constantin Mierla http://twitter.com/#!/miconda - http://www.linkedin.com/in/miconda
-- Daniel-Constantin Mierla http://twitter.com/#!/miconda - http://www.linkedin.com/in/miconda
-- Daniel-Constantin Mierla http://twitter.com/#!/miconda - http://www.linkedin.com/in/miconda
-- Daniel-Constantin Mierla http://twitter.com/#!/miconda - http://www.linkedin.com/in/miconda Book: SIP Routing With Kamailio - http://www.asipto.com Kamailio Advanced Training, Sep 28-30, 2015, in Berlin - http://asipto.com/u/kat