Date: Tue, 13 May 2003 11:38:30 +0200
From: Axel Thimm <Axel.Thimm@physik.fu-berlin.de>
To: Steve Lord <lord@sgi.com>
Cc: linux-xfs@oss.sgi.com
Subject: Re: Zero filled files
Message-ID: <20030513093830.GF19979@puariko.nirvana>
References: <20030512102338.GA3268@puariko.nirvana> <1052742135.1173.1.camel@laptop.americas.sgi.com>
Mime-Version: 1.0
Content-Type: multipart/signed; micalg=pgp-sha1;
	protocol="application/pgp-signature"; boundary="GLp9dJVi+aaipsRk"
Content-Disposition: inline
In-Reply-To: <1052742135.1173.1.camel@laptop.americas.sgi.com>
User-Agent: Mutt/1.4.1i
Sender: linux-xfs-bounce@oss.sgi.com
Errors-to: linux-xfs-bounce@oss.sgi.com
Precedence: bulk
Content-Length: 1978
Lines: 59

--GLp9dJVi+aaipsRk
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
Content-Transfer-Encoding: quoted-printable

On Mon, May 12, 2003 at 07:22:13AM -0500, Steve Lord wrote:
> On Mon, 2003-05-12 at 05:23, Axel Thimm wrote:
> > I know about XFS's zeroing of files for such files that were still mark=
ed as
> > dirty when the crash occured. I wonder about two things:
> >=20
> > o Why are 15-20 minutes old files also wiped out? Are the write-backs t=
hat
> >   much delayed (even if there is not other FS/CPU load)?
> >=20
> > o What is the best way to find these files (why aren't their names simp=
ly
> >   dumped to the kernel logs)? Something like
> >=20
> >   find . -xdev -type f \! -empty | xargs <grep for non-zero character h=
ere>
> >=20
> >   What's the best expression for the last part?
> >=20
> > (Red Hat 8.0 & XFS 1.2)
> > --=20
> > Axel.Thimm@physik.fu-berlin.de
>=20
> You need to try the current cvs kernel, there was a major rework of xfs
> sync recently which should have fixed this.

Thanks. Are zero filled files now completely avoidable, or simply
happen less often? The nicest scenario would be for files not
completly rewritten to disk at fs crash to revert to the previous
version.

Failing that it would be good to have them pinned down in some
way. Currently I have extremely large downtimes at fs crashes due to
the checks I have to run over the fs to detect those files to get them
from backup. Maybe altering something in the inode that could be used
to flag damaged files? Or dumping the files out to the kernel logs
(the latter does not help much of course if your /var/log is hosed).
--=20
Axel.Thimm@physik.fu-berlin.de

--GLp9dJVi+aaipsRk
Content-Type: application/pgp-signature
Content-Disposition: inline

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.2 (GNU/Linux)

iD8DBQE+wL0WQBVS1GOamfERApSdAJ95QSKd7kDV1m1vbEd1TFzDLT5vxACfdAuq
kJ+6crrssczu4Q3LOCkmBCw=
=Vxqn
-----END PGP SIGNATURE-----

--GLp9dJVi+aaipsRk--