xfs
[Top] [All Lists]

Re: [PATCH 0/3] serialize unmount against new buffer I/O

To: xfs@xxxxxxxxxxx
Subject: Re: [PATCH 0/3] serialize unmount against new buffer I/O
From: Brian Foster <bfoster@xxxxxxxxxx>
Date: Thu, 14 Jul 2016 13:29:25 -0400
Delivered-to: xfs@xxxxxxxxxxx
In-reply-to: <1468426595-35032-1-git-send-email-bfoster@xxxxxxxxxx>
References: <1468426595-35032-1-git-send-email-bfoster@xxxxxxxxxx>
User-agent: Mutt/1.6.1 (2016-04-27)
On Wed, Jul 13, 2016 at 12:16:32PM -0400, Brian Foster wrote:
> This is a rework of the directory readahead panic fix based on a buffer
> I/O count mechanism. See [1] for the original patch and subsequent
> discussion leading to the solution as constituted.
> 
> This is split into multiple patches simply to break out some refactoring
> and special buffer exclusion from the accounting mechanism. This
> prevents the original problem (xfs/311) as well as passes an initial
> xfstests run.
> 

FYI, I've hit an unmount hang with this while running generic/270. It
takes a few iterations to hit. I'm not sure if it's a race issue or more
of an explicit accounting issue, but I'll hold off on v2 until I can
track it down...

Brian

> Brian
> 
> [1] http://oss.sgi.com/pipermail/xfs/2016-June/049946.html
> 
> Brian Foster (3):
>   xfs: helper to set flags on uncached buffer reads
>   xfs: exclude never-released buffers from buftarg I/O accounting
>   xfs: track and serialize in-flight async buffers against unmount
> 
>  fs/xfs/xfs_buf.c   | 177 
> ++++++++++++++++++++++++++++++++++++++++-------------
>  fs/xfs/xfs_buf.h   |   8 +++
>  fs/xfs/xfs_log.c   |   4 +-
>  fs/xfs/xfs_mount.c |  12 ++--
>  4 files changed, 151 insertions(+), 50 deletions(-)
> 
> -- 
> 2.5.5
> 
> _______________________________________________
> xfs mailing list
> xfs@xxxxxxxxxxx
> http://oss.sgi.com/mailman/listinfo/xfs

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