On 29/06/15 02:11 AM, Juha Heinanen wrote:

as i reported on sr-users list, i noticed that sometimes rtpengine requests from kamailio set ip header don't fragment flag that causes icmp error message from a router on the path to rtpengine. as result, rtpengine modules things that the rtpengine is dead.

enclosed is an example of such an ip packet.

after i made this kind of patch to rtpengine.c, the flag is not set anymore:

*** /usr/src/orig/kamailio/modules/rtpengine/rtpengine.c 2015-06-16 12:33:56.497242900 +0300 --- rtpengine.c 2015-06-29 09:03:26.959479313 +0300

---

*** 1059,1064 **** --- 1059,1065 ---- struct addrinfo hints, *res; struct rtpp_set *rtpp_list; struct rtpp_node *pnode;

• int ip_mtu_discover = IP_PMTUDISC_DONT;

  if(rtpp_set_list==NULL ) return 0;

---

*** 1121,1126 **** --- 1122,1131 ---- return -1; }

• 	setsockopt(rtpp_socks[pnode->idx], IPPROTO_IP,
  

• 		   IP_MTU_DISCOVER, &ip_mtu_discover,
  

• 		   sizeof(ip_mtu_discover));
  

• 	if (bind_force_send_ip(pnode->idx) == -1) {
  		LM_ERR("can't bind socket\n");
  		close(rtpp_socks[pnode->idx]);
  
  

any comments on this?

Good idea. I'm not sure how well this is supported on other (non-Linux) Unixes, so it might be advisable to enclose this in #ifdefs.

Cheers