[Top] [All Lists]

Re: XFS and AMD64

To: evilninja <evilninja@xxxxxxx>
Subject: Re: XFS and AMD64
From: David Chinner <dgc@xxxxxxx>
Date: Thu, 11 Aug 2005 09:10:55 +1000
Cc: linux-xfs@xxxxxxxxxxx
In-reply-to: <42FA784A.2050301@xxxxxxx>; from evilninja@xxxxxxx on Wed, Aug 10, 2005 at 11:57:30PM +0200
References: <20050808025536.GA2029@xxxxxxxxxxxxxxx> <p73irye3oyz.fsf@xxxxxxxxxxxxx> <42FA784A.2050301@xxxxxxx>
Sender: linux-xfs-bounce@xxxxxxxxxxx
User-agent: Mutt/
On Wed, Aug 10, 2005 at 11:57:30PM +0200, evilninja wrote:
> Andi Kleen schrieb:
> > 
> > There is a problem when a unclean log from the 32bit kernel is replayed
> > on the 64bit kernel (or the other way round). Just make sure 
> > you always have clean unmounts when you switch between 32bit and 64bit
> > kernels.
> hm, but how is the OP supposed to get his fs checked?

I don't think was the question the OP asked. It was a 32bit
userspace/64bit kernel interop question.

> as i understand it
> he has successfully run a 8TB xfs in 32bit. now it's somehow corrupt and
> the log was not replayed yet.

Different thread.

Please keep in mind that log replay is completely separate to
repairing the filesystem and doesn't run in userspace nor does it
consume much memory. You should always attempt to do log replay on
the same system and kernel or, in the case of hardware failures, on
a machine of the same architecture and endianness.

> but now he's forced to use a 64bit machine
> to make xfs_repair work (which takes up > 4GB of memory).
> so he *has* to switch from 32bit to 64bit with an unclean fs :-\

No, you only run xfs_repair after you've replayed the log.  Once
you've replayed the log you can move the filesystem to any machine
you like to repair it.


Dave Chinner
R&D Software Enginner
SGI Australian Software Group

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