[Home]

Summary:ASTERISK-20537: Asterisk deadlocks between looking up extension from process_sdp and bridge execution from pbx_realtime
Reporter:Yaroslav Radilov (yarikrad)Labels:
Date Opened:2012-10-09 00:32:42Date Closed:2013-11-10 20:31:39.000-0600
Priority:MajorRegression?
Status:Closed/CompleteComponents:Channels/chan_sip/General PBX/pbx_realtime
Versions:1.8.16.0 Frequency of
Occurrence
Related
Issues:
duplicatesASTERISK-21040 Deadlock involving chan_sip.c, pbx.c and autoservice.c, locking on chan and &conclock
is duplicated byASTERISK-21228 Deadlock in pbx_find_extension when attempting an autoservice stop due to holding the context lock
Environment:Linux CentOsAttachments:( 0) backtrace-threads.txt
( 1) core-show-locks.txt
Description:I have got lock on asterisk 1.8.16.0. Lock happens once week on incoming call. It happens in different points of dialplan. Files with backtrace and core show locks included.

[mjordan edit]:

Deadlock appears to be between {{ast_write}} called from bridge initiated from {{pbx_realtime}} and attempting to stop the autoservice on the channel (called from {{ast_exists_extension}} from {{process_sdp}}).  The problem here is that the autoservice waits indefinitely until the channel comes out of autoservice, which it can't do since the channel is already locked by {{chan_sip}}.

The channel cannot be locked when calling {{ast_exists_extension}}.

Quoting from the source header:

{code}
* \note It is possible for autoservice to be started and stopped on c during this
* function call, it is important that c is not locked prior to calling this. Otherwise
* a deadlock may occur
{code}
Comments:By: Yaroslav Radilov (yarikrad) 2012-10-09 00:36:01.782-0500

files

By: Matt Jordan (mjordan) 2012-10-29 13:57:33.853-0500

Thank you for reporting this issue.  {{pbx_realtime}} is an extended support module, and as such, developer support for it comes primarily from the Asterisk developer community.  Response times may reflect that.

By: Matt Jordan (mjordan) 2013-11-10 20:31:39.212-0600

Closing out as a duplicate of ASTERISK-21228. Since that issue has received more traffic, this issue will be tracked there. Thanks!