[asterisk-bugs] [Asterisk 0018310]: [patch] hint state changes deadlock/race
Asterisk Bug Tracker
noreply at bugs.digium.com
Wed Jan 5 16:37:17 UTC 2011
A NOTE has been added to this issue.
======================================================================
https://issues.asterisk.org/view.php?id=18310
======================================================================
Reported By: one47
Assigned To:
======================================================================
Project: Asterisk
Issue ID: 18310
Category: Core/PBX
Reproducibility: always
Severity: major
Priority: normal
Status: ready for testing
Asterisk Version: 1.6.2.14
JIRA: SWP-2541
Regression: No
Reviewboard Link:
SVN Branch (only for SVN checkouts, not tarball releases): N/A
SVN Revision (number only!):
Request Review:
======================================================================
Date Submitted: 2010-11-15 12:19 CST
Last Modified: 2011-01-05 10:37 CST
======================================================================
Summary: [patch] hint state changes deadlock/race
Description:
VERY similar to https://issues.asterisk.org/view.php?id=18165, but this is a
different deadlock path, so I have
raised a separate report.
Thread 1: taskprocessor -> handle_statechange
Lock order = conlock, hints, hint, pvt
Thread 2: chan_sip.c: handle_request_do -> handle_response_notify
Lock order = pvt, conlock
Thread2 only gets into the conlock if STATECHANGEQUEUE is true, ie. there
are rapid state changes happening on a subscribed hint.
======================================================================
Relationships ID Summary
----------------------------------------------------------------------
related to 0018165 [patch] hint state changes deadlock pro...
has duplicate 0018535 app_queue deadlocks if weight is set
======================================================================
----------------------------------------------------------------------
(0130208) one47 (reporter) - 2011-01-05 10:37
https://issues.asterisk.org/view.php?id=18310#c130208
----------------------------------------------------------------------
It seems that this patch is okay, but that it does not go far enough...
1) Performance - This patch is no worse performance wise that the
unpatched code - The deadlock avoidance is called so rarely that it will
have an effect tending towards zero.
2) Additional issue. In chan_sip.c:cb_extensionstate(), there is a
potential call to chan_sip.c:transmit_state_notify()
Within transmit_state_notify(), if "notifycid" is enabled, then there is a
call into channel.c:ast_channel_search_locked(), which is a deadlock
waiting to happen, particularly if called with 4 other locks held as in
this case.
ast_channel_search_locked() has no deadlock avoidance code, so things lock
hard.
The only way I have found to cause the above to be a problem is to drop
call-files which generate a Local/ channel - The nature of the Local/
channel means that it grabs a channel lock, and then shortly afterwards
needs the context-lock - This is the reverse of what handle_statechange
does, so deadlocks...
3) Way forward - Did anyone (stefan?) manage to patch handle_statechange()
so that it no longer needs the context-lock? That certainly looks like a
promising way forward.
Issue History
Date Modified Username Field Change
======================================================================
2011-01-05 10:37 one47 Note Added: 0130208
======================================================================
More information about the asterisk-bugs
mailing list