On Thu, Apr 12, 2012 at 4:23 AM, Matt DiMeo <mattdimeo at yahoo.com> wrote: > I'd like to give pjsip the ability to act on ICMP Destination Unreachable > messages, which gets sent by a host when it receives a UDP packet on a port > nobody's listening to. It looks like I could just modify > tsx_timer_callback in sip_transaction.c to terminate a transaction if a > retransmit timeout happens on a transaction after receiving a matching > ICMP. This would let me detect that asterisk is shut down on a pbx quicker > than waiting for all the retransmits to time out. > > That's a neat idea, as it doesn't introduce too many new flows to the transaction state machine. > Is it safe to assume that a transaction is always a communication with one > specific IP address? I'm a little unclear on whether the transaction would > cycle through ip addresses for a server that resolves to multiple hosts, or > whether those would be different transactions handled at a higher level in > the stack. > > Request can be retried due to various reasons, e.g. multiple DNS SRV entries, failed auto-switch to TCP ( http://trac.pjsip.org/repos/wiki/Using_SIP_TCP#switch), 401/407 auth, and 3xx redirection. The rule of thumb is the same transaction instance will be used if the new request uses the same branch value. And in this case, the DNS SRV retries and TCP switch do use the same branch (although in this case I guess you don't need to worry about the TCP switch since it uses TCP). Also one UDP transport can be shared by many transactions that may send to different destinations so in this case it could be hard to know which transaction the ICMP packet should be processed by. -Benny -m@ > > -------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.pjsip.org/pipermail/pjsip_lists.pjsip.org/attachments/20120414/587896a7/attachment.html>
