From mboxrd@z Thu Jan 1 00:00:00 1970 From: Roger Heflin Subject: Re: Bug#557262: 2.6.31+2.6.31.4: XFS - All I/O locks up to D-state after 24-48 hours (sysrq-t+w available) - root cause found = asterisk Date: Sat, 21 Nov 2009 08:29:18 -0600 Message-ID: <4B07F93E.10502@gmail.com> References: <20091019030456.GS9464@discord.disaster> <20091020003358.GW9464@discord.disaster> <4B0729D8.3000105@debian.org> Mime-Version: 1.0 Content-Type: text/plain; charset=UTF-8; format=flowed Content-Transfer-Encoding: 7bit Return-path: In-Reply-To: <4B0729D8.3000105@debian.org> Sender: linux-raid-owner@vger.kernel.org To: Faidon Liambotis Cc: Justin Piszcz , 557262@bugs.debian.org, Dave Chinner , submit@bugs.debian.org, linux-kernel@vger.kernel.org, xfs@oss.sgi.com, linux-raid@vger.kernel.org, asterisk-users@lists.digium.com, Alan Piszcz List-Id: linux-raid.ids Faidon Liambotis wrote: > Justin Piszcz wrote: > > Found root cause-- root cause is asterisk PBX software. I use an > SPA3102. >> When someone called me, they accidentally dropped the connection, I called >> them back in a short period. It is during this time (and the last time) >> this happened that the box froze under multiple(!) kernels, always when >> someone was calling. > >> I don't know what asterisk is doing but top did run before the crash >> and asterisk was using 100% CPU and as I noted before all other processes >> were in D-state. >> >> When this bug occurs, it freezes I/O to all devices and the only way to >> recover >> is to reboot the system. > That's obviously *not* the root cause. > > It's not normal for an application that isn't even privileged to hang > all I/O and, subsequently everything on a system. > > This is almost probably a kernel issue and asterisk just does something > that triggers this bug. > > Regards, > Faidon I had an application in 2.6.5 (SLES9)...that would hang XFS. The underlying application was multi-threaded and both threads were doing full disks syncs every so often, and sometimes when doing the full disk sync the XFS subsystem would deadlock, it appeared to me tha one sync had a lock and was waiting for another, and the other process had the second lock and was waiting for the first... We were able to disable the full disk sync from the application and the deadlock went away. All non-xfs filesytems still worked and could still be accessed. I did report the bug with some traces but I don't believe anyone ever determined where the underlying issues was.