From owner-freebsd-current  Tue Jan  2 12:36: 4 2001
From owner-freebsd-current@FreeBSD.ORG  Tue Jan  2 12:36:01 2001
Return-Path: <owner-freebsd-current@FreeBSD.ORG>
Delivered-To: freebsd-current@freebsd.org
Received: from math.uic.edu (seifert.math.uic.edu [131.193.178.249])
	by hub.freebsd.org (Postfix) with SMTP id 8363337B400
	for <current@freebsd.org>; Tue,  2 Jan 2001 12:36:01 -0800 (PST)
Received: (qmail 69247 invoked by uid 30007); 2 Jan 2001 20:34:24 -0000
Date: Tue, 2 Jan 2001 14:34:24 -0600
From: Marc Culler <culler@math.uic.edu>
To: dfr@nlsystems.com, mike@smith.net.au, paulz@trantor.stuyts.nl,
	current@freebsd.org
Cc: consult@math.uic.edu
Subject: Weird NFS error using Solaris 8 client
Message-ID: <20010102143424.A69160@math.uic.edu>
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
X-Mailer: Mutt 1.0.1i
Sender: owner-freebsd-current@FreeBSD.ORG
Precedence: bulk
X-Loop: FreeBSD.ORG

In December 1998 a message from Paul van der Zwan with a similar
subject
( http://lists.openresources.com/FreeBSD/freebsd-current/msg00018.html )
started a thread dealing with a problem of bad timestamps for NFS
files.  By the end of the thread Doug Rabson had apparently solved the
problem for FreeBSD NFS clients and a Solaris 7 NFS server.
Unfortunately, the problem does not seem to have been solved yet for a
FreeBSD NFS server with Solaris 8 NFS clients.

Our FreeBSD NFS server is running 4.1.1-stable.  Our Solaris clients are
running Solaris 5.8.  I can produce our problem with the same test program
that was used in that message from 1998:

#include <fcntl.h>
main()
{
int rv;

rv=open("testfile1",O_CREAT|O_RDWR|O_EXCL,0666);
if ( rv < 0 )
  perror("testfile1");
rv=open("testfile2",O_CREAT|O_RDWR,0666);
if ( rv < 0 )
  perror("testfile2");
}

When I run the above program on a FreeBSD client using a FreeBSD NFS
server there are no errors reported.  I can list the files:

[culler@seifert culler]$ ls -l testfile*
-rw-------  1 culler  30007  0 Jan  2 14:07 testfile1
-rw-------  1 culler  30007  0 Jan  2 14:07 testfile2

But, if I now login to a Solaris 8 client which is using the same FreeBSD
NFS server I am not able to stat testfile1:

[culler@neumann culler]$ ls -l testfile*
ls: testfile1: Value too large for defined data type
-rw-------   1 culler   30007           0 Jan  2 14:07 testfile2

If I go back to the FreeBSD client and touch testfile1, the error
disappears:

[culler@seifert culler]$ touch testfile1

[culler@neumann culler]$ ls -l testfile1
-rw-------   1 culler   30007           0 Jan  2 14:11 testfile1

I presume that the problem still has to do with bad timestamps, but
under some slightly different situation than in December 1998.  Can
anyone help us?  Sadly, pine seems to trigger this bug, so we are
getting a fair number of these bad files. Thanks.

Marc Culler
Department of Mathematics
University of Illinois at Chicago

PS Here are the revision numbers for nfs_vnops.c and nfs_vfsops.c on our
FreeBSD test machines.
FreeBSD client:
 $FreeBSD: src/sys/nfs/nfs_vnops.c,v 1.150 2000/01/05 00:32:18 dillon Exp $
 $FreeBSD: src/sys/nfs/nfs_vfsops.c,v 1.91 2000/01/05 05:11:37 dillon Exp $
FreeBSD server:
 $FreeBSD: src/sys/nfs/nfs_vnops.c,v 1.150 2000/01/05 00:32:18 dillon Exp $
 $FreeBSD: src/sys/nfs/nfs_vfsops.c,v 1.91.2.1 2000/09/10 01:45:36 ps Exp $


To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-current" in the body of the message