xfs
[Top] [All Lists]

Re: null character filesystem corruption with 2.4.6

To: Colin Walters <walters@xxxxxxxxxxxxxxxxxx>
Subject: Re: null character filesystem corruption with 2.4.6
From: Seth Mos <knuffie@xxxxxxxxx>
Date: Sat, 14 Jul 2001 22:40:06 +0200 (CEST)
Cc: linux-xfs@xxxxxxxxxxx
In-reply-to: <87lmlrbf0w.church.of.emacs@cis.ohio-state.edu>
Sender: owner-linux-xfs@xxxxxxxxxxx
On Sat, 14 Jul 2001, Colin Walters wrote:

> Seth Mos <knuffie@xxxxxxxxx> writes:
> 
> > I have had this corruption using bad memory in the 2.4.0-test
> > releases.  After replacing it I have not been able to replicate
> > this.
> 
> Well, I've dropped findnull into my /etc/cron.daily, and I'll see if
> all the null corruption was just from my last kernel crash, or whether
> it is continually happening.

I believe it only happens after recovery from a dirty fs.

> > I have seen this behaviour on the list with kernels that have been
> > compiled with gcc-2.96-rh. What compiler did you use.?
> 
> Reading specs from /usr/lib/gcc-lib/powerpc-linux/2.95.4/specs
> gcc version 2.95.4 20010703 (Debian prerelease)

That could be a hint but it will be a difficult on to trace. I don't know
if this also happens with gcc-2.91-66 which I believe it does not.

> > Have you tried running an xfs_repair -n on th filesystem to see what
> > it thinks of it.
> 
> I didn't get any error messages from that (how does it differ from
> xfs_check?).

xfs_repair can check or repair a filesystem. I'm not sure what the
differences are. The filesystem could actually be fine eventhough you see
null characters in your file.

> > I believe that the original report on PPC kernels were linus based.
> 
> Hm, more often than not, the linus tree doesn't even compile on ppc,
> at least last time I checked.

I was afraid that was the state of the linus tree yes. The big dived of
linuxPPC.

> > The extended attribute of the file is empty. That's why it is also
> > giving back NULLS. This happens during recovery IIRC. So it is not
> > really corrupt in the sense of the checker.
> > 
> > Did you see this after XFS recovery of the filesystem?
> 
> Yes, the files are still filled with nulls, past reboots.

The file keeps consistent during reboots, with or without reboots.
It's the recovery process AFAIK that put's them there after a unclean
reboot.

> > I don't know if there is something we can do about this except
> > making EA work. It would probably be moot point then. I am not sure
> > if I am coorect on this issue though.
> 
> EA?

Extended Attributes. The file got written or touched and to disk but after
recovery it was given a number of empty extended attributes.
This is what you are seeing.

Cheers
Seth


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