[OpenSIPS-Devel] [ opensips-Patches-3463527 ] nat_traversal TM callback registration

[SourceForge.net]

Patches item #3463527, was opened at 2011-12-21 14:44
Message generated for change (Tracker Item Submitted) made by aerringer
You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=1086412&aid=3463527&group_id=232389

Please note that this message will contain a full copy of the comment thread,
including the initial issue submission, for this request,
not just the latest update.
Category: modules
Group: 1.7.x
Status: Open
Resolution: None
Priority: 5
Private: No
Submitted By: Alan Erringer (aerringer)
Assigned to: Nobody/Anonymous (nobody)
Summary: nat_traversal TM callback registration

Initial Comment:
We noticed that with 1.7.0 and still with 1.7.1 that we saw increased CPU utilization over time.

We also encountered a seg fault which led us to look closely at the TM REQIN callback list:

Core was generated by `/usr/local/sbin/opensips -P /var/run/opensips.pid'.
Program terminated with signal 11, Segmentation fault.
#0  0x00002b67e330a640 in ?? ()
(gdb) bt
#0  0x00002b67e330a640 in ?? ()
#1  0x00002b67d7f0820f in run_reqin_callbacks (trans=0x2b67e7759520, req=0x8c3248,
    code=<value optimized out>) at t_hooks.c:248
#2  0x00002b67d7ef6265 in build_cell (p_msg=0x8c3248) at h_table.c:289
#3  0x00002b67d7f0fe91 in new_t (p_msg=0x8c3248) at t_lookup.c:973
#4  t_newtran (p_msg=0x8c3248) at t_lookup.c:1081
#5  0x00002b67d7f02e4b in t_relay_to (p_msg=0x2b67e7759520, proxy=0x2,
    flags=-1059770624) at t_funcs.c:199
#6  0x00002b67d7f14261 in w_t_relay (p_msg=0x8c3248, proxy=0x0, flags=0x0) at tm.c:1129
#7  0x0000000000410970 in do_action (a=0x7f77d8, msg=0x8c3248) at action.c:1280
#8  0x0000000000414906 in run_action_list (a=<value optimized out>, msg=0x8c3248)
... 

We didn't conclusively find the seg fault, but we did find the source of the CPU increase.

It appears that the nat_traversal module registers for TM reqin callbacks once per message rather than once during mod_init().

This results in the callback list becoming huge (1.7M elements!) over time.

We moved the registration into mod_init(), and it seems to perform as it should.

I'm including the modified module/nat_traversal.c file from 1.7.1.

Let me know if you have further questions, or if our fix doesn't seem correct.

Thanks much,

Alan Erringer, alan at pinger.com


----------------------------------------------------------------------

You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=1086412&aid=3463527&group_id=232389