[Top] [All Lists]

Re: kernel bug in xfs_lrw.c (centos v5.5, directio, aio)

To: Nohez <nohez@xxxxxxxx>
Subject: Re: kernel bug in xfs_lrw.c (centos v5.5, directio, aio)
From: Dave Chinner <david@xxxxxxxxxxxxx>
Date: Thu, 19 Aug 2010 01:55:34 +1000
Cc: xfs@xxxxxxxxxxx
In-reply-to: <alpine.LNX.2.00.1008181715380.12207@xxxxxxxxxxxxxxxxxxxx>
References: <alpine.LNX.2.00.1008171906220.21398@xxxxxxxxxxxxxxxxxx> <20100818114305.GR10429@dastard> <alpine.LNX.2.00.1008181715380.12207@xxxxxxxxxxxxxxxxxxxx>
User-agent: Mutt/1.5.20 (2009-06-14)
On Wed, Aug 18, 2010 at 08:21:31PM +0530, Nohez wrote:
> Thank you for your reply.
> I checked the archives and found a bug report. The function names in
> the call trace report are different from the ones I have reported.
> Since it's an application bug, it does not matter which filesystem
> is being used. Can I assume that the data corruption can happen on
> ext3 too?

It depends on the way the filesystem handles parallelism when doing
direct IO. In general, though, you cannot rely on the filesystem
maintaining data coherence if you mix concurrent buffered/mmap IO
and direct IO to the same file. Filesystems do make efforts to
minimise the possibility of coherency problems, but they cannot be
prevented entirely. Hence the delegation of responsibility to the
application issuing the direct IO.


Dave Chinner

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