answer_machine.pl dies with Invalid argument at lib/Net/SIP/Simple/RTP.pm line 408.

[pali]

Test scenario:

Run:

$ perl -Ilib bin/answer_machine.pl -d 100 -W /tmp/test.pcmu -D /tmp/save sip:test@localhost:2000

And then via linphonec application create a SIP call to sip:test@localhost:2000. When Net::SIP::Simple::RTP tries to send packet it dies and exit whole answer_machine.pl script.

Reason is that linphonec uses localhost <--> localhost connection for SIP packets, but in SDP IN IP4 announces either public address or NATed address. Not the localhost address. And when Net::SIP::Simple::RTP::_send_rtp() calls send() function it fail because UDP packet from 127.0.0.1 to public IPv4 address is not possible to send.

So this is not a problem in answer_machine.pl, but in Net::SIP::Simple::RTP module. It should not die as this is not an error of Net::SIP, but error of other side (linphonec). Instead Net::SIP should exit active call.

To debug this problem I applied following patch:

diff --git a/lib/Net/SIP/Simple/RTP.pm b/lib/Net/SIP/Simple/RTP.pm
index d6cf418..7e3418a 100644
--- a/lib/Net/SIP/Simple/RTP.pm
+++ b/lib/Net/SIP/Simple/RTP.pm
@@ -384,13 +384,13 @@ sub _send_rtp {
     }
 
     die $! if ! defined $buf or $buf eq '';
-    if (0) {
+#    if (0) {
 	DEBUG(50, "%s -> %s seq=%d ts=%x",
-	    ip_sockaddr2string(getsockname($sock)),
-	    ip_sockaddr2string($addr),
+	    Net::SIP::Util::ip_sockaddr2string(getsockname($sock)),
+	    Net::SIP::Util::ip_sockaddr2string($addr),
 	    $seq, $timestamp
 	);
-    }
+#    }
 
     # add RTP header
     $rtp_event = 0 if ! defined $rtp_event;

which shown me real problem in debug log:

1529101979.2434 DEBUG:<50> Net::SIP::Dispatcher::Eventloop::loop[172]: trigger timer(rtpsend) 1529101979.24203 repeat=0.02
1529101979.2435 DEBUG:<50> Net::SIP::Simple::RTP::_send_rtp[388]: 127.0.0.1:7616 -> X.X.X.X:7078 seq=1 ts=a0
1529101979.2436 DEBUG:<100> Net::SIP::Simple::RTP::_send_rtp[407]: send 160 bytes to RTP
Invalid argument at lib/Net/SIP/Simple/RTP.pm line 408.
1529101979.2438 DEBUG:<100> Net::SIP::Simple::Call::DESTROY[131]: done

[noxxi]

I don't think I fully understand the problem. Could you please add more debugging and a trace of the SIP messages exchanged (perl -MNet::SIP::Debug=99 ..) and ideally also a packet capture which includes any RTP traffic?

[pali]

I tried to send you an email with whole terminal debug output, but your email server does not want to accept it... I'm periodically getting SMTP 451:

Trusted TLS connection established to mx1.genua.de[80.154.94.10]:25: TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)
C7B631685: host mx1.genua.de[80.154.94.10] said: 451 Please try again later. (in reply to RCPT TO command)
Trusted TLS connection established to mx2.genua.de[80.154.94.20]:25: TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)
C7B631685: to=<Steffen_Ullrich@genua.de>, relay=mx2.genua.de[80.154.94.20]:25, delay=160, delays=159/0.01/0.86/0.07, dsn=4.0.0, status=deferred (host mx2.genua.de[80.154.94.20] said: 451 Please try again later. (in reply to RCPT TO command))

[pali]

After more failed attempts, finally mx1.genua.de accepted email with "250 Message accepted for delivery.". So full debug output should be in your mailbox now.

[noxxi]

The problems you had with mail delivery are due to Greylisting, i.e. an anti-spam measure. As for the problem with the answer machine: Net::SIP::Simple expects that the RTP connections are done from the same network interface as SIP. Since you give no registrar this interface is taken from the SIP URL, i.e. localhost in your case which means 127.0.0.1. It looks like that linphone is taking a different interface instead on the same machine, i.e. 192.168.200.102. To properly work within this setup you should use a domain which maps to this interface or use the IP address directly, i.e. sip:test@192.168.200.102:2000 instead of sip:test@localhost:2000.

[pali]

Yes, that is a problem. But I would expect that Net::SIP does not crash (exit whole perl process with exit code 22) when connection is from different network. Rather I would expect that Net::SIP close SIP call (with optional some error reason).

[noxxi]

I've changed the code in edeb43e to ignore the error in send, i.e. it will simply not succeed in sending anything. That's about the behavior you get from other software like linphone which also did not complain that the address given in the SDP by the answer machine is not reachable.

[pali]

Great, thank you! Seems that I could be able to replace asterisk with Net::SIP which just depends on perl and core modules.

(Probably the better solution would be to invoke some callback when such error happen, so application itself could write own logic how to handle error - e.g. ignore it, wait some time or hangup call immediately; just suggestion)