[asterisk-bugs] [Asterisk 0016840]: Deadlock in chan_local when obtaining locks on local_pvt->lock

[Asterisk Bug Tracker]

A NOTE has been added to this issue. 
====================================================================== 
https://issues.asterisk.org/view.php?id=16840 
====================================================================== 
Reported By:                bzing2
Assigned To:                
====================================================================== 
Project:                    Asterisk
Issue ID:                   16840
Category:                   Channels/chan_local
Reproducibility:            random
Severity:                   block
Priority:                   normal
Status:                     new
Asterisk Version:           1.6.2.2 
JIRA:                        
Regression:                 No 
Reviewboard Link:            
SVN Branch (only for SVN checkouts, not tarball releases): N/A 
SVN Revision (number only!):  
Request Review:              
====================================================================== 
Date Submitted:             2010-02-16 11:40 CST
Last Modified:              2010-02-16 11:45 CST
====================================================================== 
Summary:                    Deadlock in chan_local when obtaining locks on
local_pvt->lock
Description: 
Deadlock condition in chan_local when two brideged threads attempt to
obtain
locks to eachothers local_pvt structures in the opposite order.  All
processing of SIP traffic halts when an attempt is made to hangup the
call.

Thread A obtains a lock to its local_pvt->lock when in local_write.  When
in check_bridge a call is made to ast_bridged_channel, this in turn calls
local_bridgedchannel which attempts to obtain a lock on the bridged
channels
local_pvt->lock.

Thread B follows a similar path.  It holds its own lock, and during a call
to
ast_channel_masquerade again local_bridgedchannel is called.  This
attempts to
obtain a lock to the bridged channel.

At this point either side has deadlocked.  Then a user attempts to hangup
the
call and again local_bridgedchannel is called, and blocks while attempting
to
get the lock.  Unfortunalty sip netlock is held and everything grinds to
a
halt.

====================================================================== 

---------------------------------------------------------------------- 
 (0118105) bzing2 (reporter) - 2010-02-16 11:45
 https://issues.asterisk.org/view.php?id=16840#c118105 
---------------------------------------------------------------------- 
First attempt at a fix is to remove the check
        p->chan->_bridge != ast_bridged_channel(p->chan)
from the beginning of check_bridge.  If I understood correctly this can
only be true
when LOCAL_ALREADY_MASQED is also true, which is tested first. 
Unfortunatly I
know we cant trust the flag as we were already in the process of setting
up
the masqurade when it deadlocked, what effect that has on
local_queue_frame I
do not (as yet) know.

I belive this will work mainly because one call to check_bridge has
isoutbound
true and the other is false.  In the second case case no action is
preformed
whatever the result of ast_bridged_channel. 

Issue History 
Date Modified    Username       Field                    Change               
====================================================================== 
2010-02-16 11:45 bzing2         Note Added: 0118105                          
======================================================================