xfs
[Top] [All Lists]

Re: [PATCH] xfs: fix race while discarding buffers [V4]

To: xfs@xxxxxxxxxxx
Subject: Re: [PATCH] xfs: fix race while discarding buffers [V4]
From: Dave Chinner <david@xxxxxxxxxxxxx>
Date: Tue, 21 Aug 2012 09:21:10 +1000
In-reply-to: <20120820224751.GA22096@xxxxxxxxxxxxxxxxxxxxxxxx>
References: <1344621711-8049-1-git-send-email-cmaiolino@xxxxxxxxxx> <20120820205134.GB29979@xxxxxxx> <20120820224751.GA22096@xxxxxxxxxxxxxxxxxxxxxxxx>
User-agent: Mutt/1.5.21 (2010-09-15)
On Mon, Aug 20, 2012 at 07:47:51PM -0300, Carlos Maiolino wrote:
> Hi Ben,
> > 
> > 
> Thanks for the comments.
> > 
> > What was the symptom that led to the discovery of this problem?
> > 
> > Reviewed-by: Ben Myers <bpm@xxxxxxx>
> > 
> 
> It started with the messages like the example below being logged by syslog:
> 
> shrink_slab: xfs_buftarg_shrink+0x0/0x160 [xfs] negative objects to delete 
> nr=-61993820
> shrink_slab: xfs_buftarg_shrink+0x0/0x160 [xfs] negative objects to delete 
> nr=-146
> shrink_slab: xfs_buftarg_shrink+0x0/0x160 [xfs] negative objects to delete 
> nr=-240601220
> shrink_slab: xfs_buftarg_shrink+0x0/0x160 [xfs] negative objects to delete 
> nr=-152
> shrink_slab: xfs_buftarg_shrink+0x0/0x160 [xfs] negative objects to delete 
> nr=-2921236993
> 
> These messages came from shrink_slab().

Worth noting is that this warning came from a RHEL kernel, not a
mainline kernel. The mainline kernels don't screw up the nr_to_scan
calculations when a negative object count is returned to them.
Hence mainline kernels are not impacted by the accounting bug at
all...

Cheers,

Dave.
-- 
Dave Chinner
david@xxxxxxxxxxxxx

<Prev in Thread] Current Thread [Next in Thread>