Using UDP-SIP with Exchange UM and Lync 2010

Attachment: https://bennettadelson.wordpress.com/2012/06/04/using-udp-sip-with-exchange-um-and-lync-2010/kamailio-cfg/ (remember to change extension)

Attachment: asterisk.tar.gz (remember to change extension)

I am working on and off with a client that is deploying Exchange 2010 Unified Messaging and Lync 2010 in their environment. They want to use Exchange UM with a hosted SIP-based VoIP system from a provider that I will refer to as “PhoneCo” for the sake of discussion. Furthermore, they want their Lync environment to work with the Exchange voicemail, and by the way, think it would be nice if they could experiment with Enterprise Voice functionality. Luckily, PhoneCo offers SIP trunks, and will trunk from the hosted VoIP environment to Exchange UM. So all is good, right?

The Problem Statement

Ha ha, of course I am joking. Because although Microsoft talks SIP, and PhoneCo talks SIP, we hit upon a long-standing issue. Microsoft refuses to support UDP SIP (they have their reasons, I won’t debate the point here) while PhoneCo refuses to support TCP SIP. Thus, we have an impasse.

Solution Overview

The official, standard answer to this is to use a Session Border Controller (SBC), which is essentially a SIP middleman box that can do UDP on one end and TCP on the other. A typical SBC also includes firewalling intelligence to prevent denial-of-service and other such nasty behavior. As a result, they generally start at thousands and quickly get into tens of thousands of dollars. In this customer’s case, the SIP trunk is going to be over a private MPLS connection directly between the hosted PBX and the on-premises Microsoft tools, so the customer didn’t want to pay for a lot of security they didn’t need just to deal with this issue.

The customer found a commercial product named Brekeke SIP Server that appears to be $500 to start. This is nice in that (1) it is commercial and (2) it can run on Windows, although it is Java-based so it’s a little messy and gives you one more thing to deal with patching every day or two.

We wanted to see if there was an open-source way to solve this problem. We found a way, and this post documented what we came up with. I have replicated the scenario in a lab, and have since actually simplified things a bit. I have also corrected something we had done to work around an Asterisk “bug” (in quotes because the bug states it’s not really an Asterisk bug) that came up while we were simulating the PhoneCo setup.

So first, here’s the list of VMs that are in the UC Lab:

Hostname	IP	Description
dc.uclab.local	172.30.1.10	Domain Controller
exchange.uclab.local	172.30.1.12	Exchange 2010
freepbx.uclab.local	172.30.1.11	PhoneCo stand-in
lync.uclab.local	172.30.1.13	Lync 2010
siprouter.uclab.local	172.30.1.14	SIP middleman
tmg.uclab.local	172.30.1.1	TMG 2010
internalclient.uclab.local	172.30.1.100	Test Lync/SIP client

The PhoneCo stand-in is a FreePBX installation using the FreePBX Linux distribution. I am not going to go into details on installing that into a VM because there are plenty of guides on getting that to work. For the purposes of this post I’m going to pretend Asterisk can’t do TCP SIP because that’s what we are looking at with PhoneCo. This also means ignoring all the online info about getting Asterisk to talk to Lync and Exchange using TCP SIP. (Note: Some of these guides assuming port 5065 for talking to Exchange, which is a partial solution. I’ll get into why that’s wrong later on.)

The SIP middleman – SIP router – is a Linux-based CentOS machine running the Kamailio open-source SIP router package. Kamailio is a mature, solid package that is quite amazing in some of what it can do, but I’m ignoring about 99% of it, I think. We may end up needing some of the NAT support eventually at the client, which I’m not getting into here and don’t need for the lab, but otherwise a lot of functionality is actually not in play here.

Preparing the CentOS Machine

So let’s get to it.

1. I began with a basic minimal CentOS 6.2 installation. Note that I’ve had repeated issues with the Hyper-V Integration Components on this OS so far, so I didn’t bother with them – for a lab it’s not critical. For production you’d care a lot more – the customer uses VMware so this particular issue did not come up.
2. Next, I logged in as root via SSH (PuTTY is your friend here) and accepted the key when prompted:
   
   
3. I ran yum updateto get all of the current updates for the OS, and rebooted to get the updated kernel loaded.
4. Using vi, I created /etc/yum.repos.d/kamailio.repowith:

   [kamailio]
   name=Kamailio
   baseurl=http://download.opensuse.org/repositories/home:/kamailio:/telephony/CentOS_CentOS-6/
   enabled=1
   gpgcheck=0

   This looks like this:

   

5. I confirmed that the new repository was visible with yum repolist:
6. I then confirmed that there was a package I could install in that repository with yum list kamailio:
   
7. After confirming the package, I installed it with yum install kamailio:
   
   
8. So now I need to configure the beast. Kamailio comes with a very long sample configuration file. Most of it is noise for my use. I tried to trim it down as safely as possible, as well as better fit what I wanted. So using the following commands I saved the shipped file:

   cd /etc/kamailio
   mv kamailio.cfg kamailio.cfg.original
   vi /etc/kamailio.cfg

   And then made mine, which I will explain later after finishing the build instructions:

   #!KAMAILIO
   
   # Remote Hosts
   #!subst "/SIP_UDP_HOST/172.30.1.11/"
   #!subst "/EXCHANGE_UM/172.30.1.12/"
   #!subst "/LYNC_MEDIATION/172.30.1.13/"
   
   listen=172.30.1.14:5060
   listen=172.30.1.14:5065
   listen=172.30.1.14:5067
   
   ####### Global Parameters #########
   
   memdbg=5
   memlog=5
   
   debug=2
   
   log_facility=LOG_LOCAL0
   
   fork=yes
   children=4
   
   disable_tcp=no
   
   auto_aliases=no
   
   /* uncomment and configure the following line if you want Kamailio to
      bind on a specific interface/port/proto (default bind on all available) */
   #listen=udp:10.0.0.10:5060
   
   # life time of TCP connection when there is no traffic
   # - a bit higher than registration expires to cope with UA behind NAT
   tcp_connection_lifetime=3605
   
   ####### Modules Section ########
   
   mpath="/usr/lib/kamailio/modules_k/:/usr/lib/kamailio/modules/"
   
   loadmodule "kex.so"
   loadmodule "tm.so"
   loadmodule "tmx.so"
   loadmodule "sl.so"
   loadmodule "pv.so"
   loadmodule "maxfwd.so"
   loadmodule "usrloc.so"
   loadmodule "textops.so"
   loadmodule "siputils.so"
   loadmodule "xlog.so"
   loadmodule "sanity.so"
   loadmodule "ctl.so"
   loadmodule "cfg_rpc.so"
   loadmodule "mi_rpc.so"
   
   # ----- tm params -----
   # auto-discard branches from previous serial forking leg
   modparam("tm", "failure_reply_mode", 3)
   # default retransmission timeout: 30sec
   modparam("tm", "fr_timer", 30000)
   # default invite retransmission timeout after 1xx: 120sec
   modparam("tm", "fr_inv_timer", 120000)
   
   server_header="Server: PhoneCo Intransigence Coping Solution (PICS) 2.0";
   
   ####### Routing Logic ########
   route {
           if(is_method("OPTIONS")) {
                   xlog("L_INFO","OPTIONS from $si");
                   sl_send_reply("200", "Yes, Microsoft, I am alive");
                   exit();
           }
   
           xlog("L_INFO", "*** M=$rm RURI=$ru F=$fu T=$tu IP=$si ID=$ci");
   
           # Route Exchange extensions
           if((to_uri=~"sip:5992") || (to_uri=~"sip:5999")) {
                   xlog("L_NOTICE", "EXCHANGE UM call, $proto port $op, $ru, $fU");
                   t_on_reply("1");
   
                   # https://issues.asterisk.org/jira/browse/ASTERISK-16862
                   # http://imaucblog.com/archive/2009/10/03/part-1-how-to-integrate-exchange-2010-or-2007-with-trixbox-2-8/
                   replace("Diversion: <sip:5999@SIP_UDP_HOST>;reason=unconditional","MCB-Stripped-Header: Diversion");
   
                   switch($op) {
                           case 5060:
                                   xlog("L_NOTICE", "Redirecting to TCP 5060");
                                   t_relay_to("tcp:EXCHANGE_UM:5060");
                                   exit();
                                   break;
                           case 5065:
                                   xlog("L_NOTICE", "Redirecting to TCP 5065");
                                   t_relay_to("tcp:EXCHANGE_UM:5065");
                                   exit();
                                   break;
                           case 5067:
                                   xlog("L_NOTICE", "Redirecting to TCP 5067");
                                   t_relay_to("tcp:EXCHANGE_UM:5067");
                                   exit();
                                   break;
                   }
           }
   
           # Route Lync extensions
           if(to_uri=~"sip:5...") {
                   replace("To: <sip:", "To: <sip:+");
                   xlog("L_NOTICE", "LYNC call to $tu");
                   t_relay_to("tcp:LYNC_MEDIATION:5068");
                   exit();
           }
   
           # Route the rest to Asterisk
           xlog("L_NOTICE", "Asterisk call to $tu");
           forward_udp("SIP_UDP_HOST", 5060);
   }
   
   onreply_route[1] {
           xlog("L_NOTICE", "Handling reply from Exchange relay, status $rs");
           switch($rs) {
                   case 302:
                           xlog("L_NOTICE", "Saw 302 Redirect response, checking details...");
                           if(search(";transport=Tcp")) {
                                   xlog("L_NOTICE", "Saw TCP redirection, changing redirection to UDP");
                                   replace(";transport=Tcp", ";transport=Udp");
                           } else {
                                   xlog("L_NOTICE", "302 was not matched (!)");
                           }
                           exit();
                           break;
                   case 100:
                           xlog("L_NOTICE", "Saw 100, leaving alone...");
                           exit();
                           break;
           }
   
   }

    

9. I stared the daemon (read: service) with /etc/rc.d/init.d/kamailio start and confirmed that it started  with /etc/rc.d/init.d/kamailio status:
10. I confirmed it was listening (netstat –an | grep 506):
11. I then opened up the firewall to allow those ports in (okay, thats a lie – I floundered a bit before remembering I had to do this) by editing /etc/sysconfig/iptables and adding after the --dport 22line:

    		-A INPUT -p tcp -m state --state NEW -m tcp --dport 5060 -j ACCEPT
    		-A INPUT -p tcp -m state --state NEW -m tcp --dport 5065 -j ACCEPT
    		-A INPUT -p tcp -m state --state NEW -m tcp --dport 5067 -j ACCEPT
    		-A INPUT -p udp -m state --state NEW -m udp --dport 5060 -j ACCEPT
    		-A INPUT -p udp -m state --state NEW -m udp --dport 5065 -j ACCEPT
    		-A INPUT -p udp -m state --state NEW -m udp --dport 5067 -j ACCEPT

    This looks like this when it’s done:
    

12. I then made this kick in by restarting the firewall with /etc/rc.d/init.d/iptables restart.
13. I next added system logger support for the configured log source by editing /etc/rsyslog.confand adding:

    local0.*                                                 /var/log/kamailio.log

    

14. I then made this kick in by reloading the logger configuration with /etc/rc.d/init.d/syslog reload.
    
15. I don’t want this log to grow uncontrollably so I configured the logrotate daemon to make a new log every day and save seven of them by creating /etc/logrotate.d/kamailiowith:

    /var/log/kamailio.log {
    	rotate 7
    	missingok
    	daily
    }

    

Preparing Exchange 2010 and Lync 2010

This is normal Exchange and Lync SIP configuration so I’m not going to get into great detail here. The following are the key points:

• Make sure Lync has a TCP listener on port 5068 on the mediation server of your choice. There’s no high availability here so pick one and go. As quick hints of where this is done in Topology Builder:
  
  
  After publishing and running Bootstrapper (Lync Setup) on the Mediation Server as instructed by Topology Builder I ran into (what I consider to be) a bug in Lync shown via the event log – there were LS Mediation Server messages 25075 and 25031 indicating no TCP port is enabled, then that the TCP port was requested but ignored. Restarting the Mediation Server service sorted it out. The Kamailio log will show this working (e. g. tail /var/log/kamailio.log):
  
• For Exchange, make sure you have TCP enabled on the UM server (requires a service restart to kick in) and that you have an appropriate IP gateway and unsecured telephone extension dial plan configured against that gateway:
  
  

And that’s it!

So What Does the Configuration Mean?

OK, so what the heck does the configuration I gave you above mean?  Let’s go through it:

#!KAMAILIO

This is a signature for the configuration file.

# Remote Hosts

#!subst "/SIP_UDP_HOST/172.30.1.11/"
#!subst "/EXCHANGE_UM/172.30.1.12/"
#!subst "/LYNC_MEDIATION/172.30.1.13/" 

listen=172.30.1.14:5060
listen=172.30.1.14:5065
listen=172.30.1.14:5067

This is the super important customization part. The three subst lines replace all references to those text strings with the appropriate IP addresses, while the listen lines allow the router to accept traffic on its IP on three ports – 5060, 5065, and 5067. The latter two are because Exchange – for reasons known to Microsoft but not me – takes UM connections on port 5060 but then redirects them to 5065 or 5067. Remember how above I said that some sites use 5065 and that’s wrong?  That’s because they are assuming all redirects are to 5065, but Exchange might want 5067.

Anyway, the next lines are some configuration stuff that is from the default that I left alone mainly because either the settings were fine (e. g. the syslog facility used) or because I didn’t know the implications in changing them (e. g. the children process count); there’s also the enabling of TCP (normally disabled):

####### Global Parameters ######### 
memdbg=5

memlog=5 
debug=2
log_facility=LOG_LOCAL0 
fork=yes

children=4 
disable_tcp=no 
auto_aliases=no 

# life time of TCP connection when there is no traffic
# - a bit higher than registration expires to cope with UA behind NAT
tcp_connection_lifetime=3605

Next are the modules that I am loading. I know I need some of these for sure – there are others I don’t know about so I left well-enough alone and kept them there:

####### Modules Section ######## 
mpath="/usr/lib/kamailio/modules_k/:/usr/lib/kamailio/modules/" 
loadmodule "kex.so"
loadmodule "tm.so"
loadmodule "tmx.so"
loadmodule "sl.so"
loadmodule "pv.so"
loadmodule "maxfwd.so"
loadmodule "usrloc.so"
loadmodule "textops.so"
loadmodule "siputils.so"
loadmodule "xlog.so"
loadmodule "sanity.so"
loadmodule "ctl.so"
loadmodule "cfg_rpc.so"
loadmodule "mi_rpc.so" 

# ----- tm params -----
# auto-discard branches from previous serial forking leg
modparam("tm", "failure_reply_mode", 3)
# default retransmission timeout: 30sec
modparam("tm", "fr_timer", 30000)
# default invite retransmission timeout after 1xx: 120sec
modparam("tm", "fr_inv_timer", 120000)

The next line sets a server header seen in the SIP headers. It is a fun way to point out that PhoneCo was annoying me as well as to hide the actual software being used:

server_header="Server: PhoneCo Intransigence Coping Solution (PICS) 2.0"

Now comes the real meat. It starts the routing logic for incoming SIP calls looking for the OPTIONS call that Lync and Exchange make every nanosecond (approximately) to check to see if their SIP peers are alive. Hence the status text – the code is all that really matters:

####### Routing Logic ########

route {
        if(is_method("OPTIONS")) {
                xlog("L_INFO","OPTIONS from $si");
                sl_send_reply("200", "Yes, Microsoft, I am alive");
                exit();
        }

The next line just acts as a debugging log showing what came in:

        xlog("L_INFO", "*** M=$rm RURI=$ru F=$fu T=$tu IP=$si ID=$ci");

The dollar-sign pseudo-variables are documented here, should you care: http://www.kamailio.org/wiki/cookbooks/3.2.x/pseudovariables

Anyway, moving on, we have the Exchange routing. Looking at this now, I probably want the two extensions (one for the auto-attendant and one for subscriber access) to be substituted variables, but that will be 2.1 I guess:

# Route Exchange extensions
        if((to_uri=~"sip:5992") || (to_uri=~"sip:5999")) {
                xlog("L_NOTICE", "EXCHANGE UM call, $proto port $op, $ru, $fU");
                t_on_reply("1");

This basically says “if a SIP call is made to extension 5992 or extension 5999, then do this…” and starts by indicating that we are going to do a transactional SIP redirect that, when we see a reply, should go to reply handler “1“, which will come later. After that, we have:

        # https://issues.asterisk.org/jira/browse/ASTERISK-16862
        # http://imaucblog.com/archive/2009/10/03/part-1-how-to-integrate-exchange-2010-or-2007-with-trixbox-2-8/
        replace("Diversion: <sip:5999@SIP_UDP_HOST>;reason=unconditional","MCB-Stripped-Header: Diversion");

Why is this here? Basically, Asterisk does something we don’t want it to do on the Exchange redirect – adds an extra SIP Diversion header – and we want that extra header to go away. I need to replace it with something though, so I just made up a vendor header and used that. This is safe as SIP agents – like HTTP server and clients – ignore headers that they don’t know. Next, we take the UDP session and do a transactional redirect to TCP:

        switch($op) {
                case 5060:
                        xlog("L_NOTICE", "Redirecting to TCP 5060");
                        t_relay_to("tcp:EXCHANGE_UM:5060");
                        exit();
                        break;
                case 5065:
                        xlog("L_NOTICE", "Redirecting to TCP 5065");
                        t_relay_to("tcp:EXCHANGE_UM:5065");
                        exit();
                        break;
                case 5067:
                        xlog("L_NOTICE", "Redirecting to TCP 5067");
                        t_relay_to("tcp:EXCHANGE_UM:5067");
                        exit();
                        break;
                }
        }

I couldn’t come up with a “smart” way to do this better; this is a little wordy but it is clear what is happening. I next route the Lync calls (adding the E.164 “+” sign along the way) based on extension pattern (all other 5xxx extensions besides the two special case ones above), with all others going to the Asterisk side:

        # Route Lync extensions
        if(to_uri=~"sip:5...") {
                replace("To: <sip:", "To: <sip:+");
                xlog("L_NOTICE", "LYNC call to $tu");
                t_relay_to("tcp:LYNC_MEDIATION:5068");
                exit();
        }

        # Route the rest to Asterisk
        xlog("L_NOTICE", "Asterisk call to $tu");
        forward_udp("SIP_UDP_HOST", 5060);
}

Notice that I do forward_udpinstead of t_relay_to because I don’t care about maintaining transactional state in the case of going back to Asterisk, so there’s no reason to waste resources on it. I just tell Kamailio to throw it over the wall and forget about it.

Finally, I handle the reply from Exchange. This is why I made the Exchange piece transactional:

onreply_route[1] {
        xlog("L_NOTICE", "Handling reply from Exchange relay, status $rs");
        switch($rs) {
                case 302:
                        xlog("L_NOTICE", "Saw 302 Redirect response, checking details...");
                        if(search(";transport=Tcp")) {
                                xlog("L_NOTICE", "Saw TCP redirection, changing redirection to UDP");
                                replace(";transport=Tcp", ";transport=Udp");
                        } else {
                                xlog("L_NOTICE", "302 was not matched (!)");
                        }
                        exit();
                        break;
                case 100:
                        xlog("L_NOTICE", "Saw 100, leaving alone...");
                        exit();
                        break;
        }

Notice if I get a redirect from Exchange (which I will for port 5060) I change that from a Tcp redirect to a Udp redirect, then send it on its way.

So, this is what is in the lab right now. I think this works – until PhoneCo gets the line in place we won’t know 100% but I think this is close if it isn’t completely right. We’ll see.

Hope this helps you in your integration scenarios!

— Michael C. Bazarewsky
Principal Consultant, Windows Server and Security