Hello Phillipp,
the problem exposed by you here is somehow similar to the topic discussed on devel@openser.org in the mail thread "Processing REGISTER requests".
Mainly is how to attach a resource to a contact address and identify it properly (in this case is the tcp connection). Since we have all over NATs, we cannot rely on the contact address. The latest version of openser stores in the usrloc database the associated NAT ip and port along with contact address. This is not enough when you have multiple levels of NATs (e.g., two different networks with same private IPs being visible from outside behind same nat - this gets a conflict when same user has phones in both networks).
It is clear that in your case you have to keep the tcp connection open for the whole call. The problem is how to identify the connection from the contact address. We have to address this soon and find the proper solution for this case as well as for usrloc.
Cheers, Daniel
On 10/26/05 12:51, Alexander Ph. Lintenhofer wrote:
Hi All,
I just wanted to ask you once again about the TCP-alias riddle. I found out, that there is a problem with the combination of fix_nated_contact(), force_tcp_alias() and NAT:
Imagine following situation:
Alice behind NAT: socket 172.16.0.6:2421 Nat-Box translates this to 192.168.0.13:6007
The Outbound-Proxy of Alice is 192.168.0.1
Bob is registered with 192.168.1.1
1.) Her INVITE: ==================================================================== INVITE sip:bob@192.168.1.1:2331;transport=tcp;line=wxqurd1s SIP/2.0 [...] Via: SIP/2.0/TCP 172.16.0.6:2421;received=192.168.0.13; branch=z9hG4bK-wm9jcstcboys;rport=6007 From: "Alice" sip:alice@atlanta.com;tag=sufzmxi0us To: "Bob" sip:bob@biloxi.com;tag=c9550czwtn [...] Contact: sip:alice@172.16.0.6:2421;transport=tcp;line=fyyuh6tl [...] ====================================================================
2.) fix_nated_contact() doesn't work with TCP (look at nathelper.c). force_tcp_alias() now creates following tuple as TCP-alias: 192.168.0.13:6007 to 192.168.0.13:2421
Reason: The TCP-alias is not built solely from the Via-header as suggested in the draft. The portnumber is taken from the Via-header and the IP-address is taken from the source of the incoming datagram. I read it in the sourcecode and assured it by contacting Andrei!
3.) So as a result of the notfixed Contact-header of Alice's INVITE the BYE of Bob is addressed to 172.16.0.6:2421. But no TCP-alias exists for this socket :-(
4.) I made following test by rewriting the Contact-header.... ==================================================================== if (method=="INVITE") { replace("Contact: <sip:alice@172.16.0.6:2421;transport=tcp;", "Contact: <sip:alice@192.168.0.13:2421;transport=tcp;"); } ==================================================================== ...with success. Now TCP-alias works as you can see on my Ethereal-trace below! Compare the destination port of the packet to the destination port of the RURI! ==================================================================== [...] Transmission Control Protocol, Src Port: 5060 (5060), Dst Port: 6007 (6007), ... Session Initiation Protocol Request-Line: BYE sip:alice@192.168.0.13:2421;transport=tcp;line=fyyuh6tl SIP/2.0 Message Header [...] From: "Bob" sip:bob@biloxi.com;tag=kcsveifugd To: "Alice" sip:alice@atlanta.com;tag=ricaq5cy15 Contact: sip:bob@192.168.1.1:2331;transport=tcp;line=wxqurd1s [...] ====================================================================
Another solution: Comment the lines in nathelper.c which force the return in case of TCP or TLS. Now all works well!
But why??????????????
regards, Philipp
Users mailing list Users@openser.org http://openser.org/cgi-bin/mailman/listinfo/users