[asterisk-dev] [Code Review] 4608: res_pjsip_phoneprov_provider: Fix reference leak on unload

[George Joseph]

> On April 9, 2015, 1:49 p.m., Corey Farrell wrote:
> > branches/13/res/res_pjsip_phoneprov_provider.c, lines 358-363
> > <https://reviewboard.asterisk.org/r/4608/diff/1/?file=73836#file73836line358>
> >
> >     To record our brief discussion about this change:
> >     <coreyfarrell> gtjoseph: so one thing I don't see how using an ao2_iterator instead of ao2_callback in users_apply_handler makes it more straight forward..
> >     <coreyfarrell> I don't want to hold up the fix, but that change doesn't make much sense to me
> >     <gtjoseph> there's no worry about the callback return value, OBJ_NODATA, etc.  Get the container and iterate over it.   It also eliminates a lot of overhead in the ao2_traversal code.
> >     <gtjoseph> unless the callback is controlling the traversal with it's return code, there's no point.

Actually, I think more people should use an ao2_iterator instead of ao2_callback when they can.  That internal_ao2_traverse code is ugly. :)


- George


-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviewboard.asterisk.org/r/4608/#review15164
-----------------------------------------------------------


On April 9, 2015, 1:57 p.m., George Joseph wrote:
> 
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> https://reviewboard.asterisk.org/r/4608/
> -----------------------------------------------------------
> 
> (Updated April 9, 2015, 1:57 p.m.)
> 
> 
> Review request for Asterisk Developers and Corey Farrell.
> 
> 
> Bugs: ASTERISK-24935
>     https://issues.asterisk.org/jira/browse/ASTERISK-24935
> 
> 
> Repository: Asterisk
> 
> 
> Description
> -------
> 
> res_pjsip_phoneprov_provider was leaking references to phoneprov objects due to a missing OBJ_NODATA in an ao2_callback in load_users().  Rather than adding the OBJ_NODATA, I changed load_users to use a more straightforward ao2_iterator.  This plugged the leak but exposed an unload order issue between res_pjsip_phoneprov_provider, res_phoneprov and res_pjsip.
> 
> res_pjsip_phoneprov_provider unloads first, then res_phoneprov, then res_pjsip.  Since res_pjsip_phoneprov_provider uses res_pjsip's sorcery instance, when it unloads, it's objects are still in the sorcery instance.  When res_pjsip unloads, it destroys all its objects including res_pjsip_phoneprov_provider's.  The phoneprov destructor then attempts to unregister the extension from res_phoneprov but because res_phoneprov is already cleaned up, its users container is gone and we get a FRACK.
> 
> Simple solution, check for the NULL users container before attempting to remove the entry. Duh.
> 
> 
> Diffs
> -----
> 
>   branches/13/res/res_pjsip_phoneprov_provider.c 434447 
>   branches/13/res/res_phoneprov.c 434447 
> 
> Diff: https://reviewboard.asterisk.org/r/4608/diff/
> 
> 
> Testing
> -------
> 
> Ran tests/res_phoneprov/res_phoneprov_provider.  No leaks in res_pjsip_phoneprov_provider and no FRACKs.
> 
> 
> Thanks,
> 
> George Joseph
> 
>

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.digium.com/pipermail/asterisk-dev/attachments/20150409/e44b3fd3/attachment-0001.html>