Skip site navigation (1)Skip section navigation (2)
Date:      Mon, 26 Apr 2004 16:23:20 -0700
From:      Kris Kennaway <kris@obsecurity.org>
To:        Dan Nelson <dnelson@allantgroup.com>
Cc:        FreeBSD-Questions <freebsd-questions@freebsd.org>
Subject:   Re: NFS occassionally gives "permission" denied in the middle of a large transfer
Message-ID:  <20040426232319.GA22217@xor.obsecurity.org>
In-Reply-To: <20040426230855.GB22344@dan.emsphone.com>
References:  <20040426182547.GF92049@seekingfire.com> <20040426183717.GF2771@dan.emsphone.com> <20040426184324.GH92049@seekingfire.com> <20040426184710.GA22344@dan.emsphone.com> <20040426224633.GN92049@seekingfire.com> <20040426230855.GB22344@dan.emsphone.com>

next in thread | previous in thread | raw e-mail | index | archive | help

--MGYHOYXEY6WxJCY8
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
Content-Transfer-Encoding: quoted-printable

On Mon, Apr 26, 2004 at 06:08:56PM -0500, Dan Nelson wrote:
> In the last episode (Apr 26), Tillman Hodgson said:
> > On Mon, Apr 26, 2004 at 01:47:10PM -0500, Dan Nelson wrote:
> > > In the last episode (Apr 26), Tillman Hodgson said:
> > > > On Mon, Apr 26, 2004 at 01:37:17PM -0500, Dan Nelson wrote:
> > > > > The only time I've seen incorrect permission denied messages is
> > > > > when mountd is refreshing the exports list.  It's not atomic,
> > > > > so there's a small window where the old exports have been
> > > > > deleted but the new ones aren't in place yet.
> > > >=20
> > > > Is there anything in the default weekly cron jobs that would do
> > > > something like that on the file server?
> > >=20
> > > I don't think so.  Mounting or dismounting local (not NFS)
> > > filesystems might do it, but I'm not sure.
> >=20
> > Including local filesystems that aren't exported?
> >=20
> > If so, that's interesting because I do that in my weekly periodic
> > cron job. It's a simple script that umounts a backup partition,
> > newfs's it, dumps a local copy of /home over to it, and then
> > re-mounts it (for easy user-accessible near-line backup storage that
> > doesn't require going to tape for the "real" backup).
>=20
> That's probably it, then.  /sbin/mount has code that sends SIGHUP to
> mountd on any mount operation.  Which implies that any manual mount
> request, including NFS mounts would, cause the problem.  Amd calls the
> mount syscall directly, bypassing /sbin/mount.
>=20
> Ideally, mountd would be able to compare the current and new export
> settings and only update the ones that changed, or have a way to create
> a new mountlist and ask the kernel to replace the old one in a single
> atomic operation.

Very interesting..you might have identified the problem that has
plagued me on one of my package build clusters (NFS was unusable
because of the frequency of EACCESS errors).  A difference between
these systems and others is that the NFS server is also used to build
packages, and does a lot of unmounting and remounting of local
filesystems.

I can't test this right now because all of the client machines are
dead, but I should be able to soon.

Thanks for the insight!

Kris


--MGYHOYXEY6WxJCY8
Content-Type: application/pgp-signature
Content-Disposition: inline

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.4 (FreeBSD)

iD8DBQFAjZnnWry0BWjoQKURAnwkAKDVCMJz1ThdxgGOxgvXZQ6ON1iIUQCfYNLE
J1+AzRXAz4UtcuzKCRB99bI=
=T21D
-----END PGP SIGNATURE-----

--MGYHOYXEY6WxJCY8--



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20040426232319.GA22217>