Hi all & hi Daniel ,
We had a crash with version 4.0.5 today . I looks like it happens due to a race condition between a provisional response (183) and the generation of a 408 .
We send the INVITE, get a `100 Trying` from upstream , then we get a `183 Session Progress` from upstream , then after exactly 1 minute , we get another `183 Session Progress` from upstream. At this moment kamailio crashed.
Some tm params we have : modparam("tm", "fr_timer", 5000)
modparam("tm", "fr_inv_timer", 60000) <- our final reply timer is 1 minute . modparam("tm", "restart_fr_on_each_reply", 1)
GDB output and logs here:
kamailio 4.0.5 crash - in fake_req() - race cond - Pastebin.com
kamailio 4.0.5 crash - in fake_req() - race cond - Paste... (gdb) bt #0 0x00007f4d93129a32 in memcpy () from /lib64/libc.so.6 View on pastebin.com Preview by Yahoo
In fake_req() from t_reply.c , the pointer shmem_msg->new_uri.s which is passed to memcpy becomes invalid.
memcpy( faked_req->new_uri.s, shmem_msg->new_uri.s,faked_req->new_uri.len); (gdb) p shmem_msg->new_uri $3 = {s = 0x0, len = 0}
If this bug is not supposed to be fixed in 4.2.0 , we can help with debugging .
Regards, Dragos
Hello,
I expect to be due to an issue fixed by commit f32354f9aae3afadca98e5a1cf254d9286a3a985 .
Apparently it was not backported to 4.0. I am considering to release 4.0.7 this Friday, where to backport appropriate fixes from 4.1 branch.
Cheers, Daniel
On 04/11/14 18:18, Dragos Oancea wrote:
-
Daniel-Constantin Mierla -
Dragos Oancea