From: Ulrich Gemkow <ulrich.gemkow@ikr.uni-stuttgart.de>
To: "J. Bruce Fields" <bfields@fieldses.org>
Subject: Re: Linux 3.7 + Sun solaris 10: Problems when reading dir from application
Date: Sat, 6 Jul 2013 17:00:38 +0200
Cc: Harald Dunkel <harald.dunkel@aixigo.de>, linux-nfs@vger.kernel.org
References: <201302041112.03587.ulrich.gemkow@ikr.uni-stuttgart.de> <20130705082740.3df4245a@dpcl082.ac.aixigo.de> <20130705192749.GB8288@fieldses.org>
In-Reply-To: <20130705192749.GB8288@fieldses.org>
MIME-Version: 1.0
Content-Type: Text/Plain;
  charset="iso-8859-1"
Message-Id: <201307061700.40451.ulrich.gemkow@ikr.uni-stuttgart.de>
Sender: linux-nfs-owner@vger.kernel.org

Hello,

answering as the original poster of the problem:

On Friday 05 July 2013 21:27:49 J. Bruce Fields wrote:
> On Fri, Jul 05, 2013 at 08:27:40AM +0200, Harald Dunkel wrote:
> > Hi Ulrich,
> > 
> > On Mon, 4 Feb 2013 21:24:08 +0100
> > Ulrich Gemkow <ulrich.gemkow@ikr.uni-stuttgart.de> wrote:
> > 
> > > Hello,
> > > 
> > > please allow a followup of my own on this:
> > > 
> > > On Monday 04 February 2013 11:12:03 Ulrich Gemkow wrote:
> > > > Hello,
> > > > 
> > > > we upgraded our fileserver from Linux 3.2 to linux 3.7.6 and now have
> > > > problems when accessing our nfs-mounted user homes from some sun-
> > > > applications (i.e. Adobe Framemaker):
> > > > 
> > > > In the applications file open box, no files are displayed. When entering
> > > > the filename by path, the file can be opened. So it seems some kind of
> > > > dir enumeration which is used by the sun applications is broken.
> > > > 
> > > > Other programs on the sun like ls work as before and show all files.
> > > > 
> > > > We are using NFSv3 (and cannot switch to v4). Our sun is a very old
> > > > machine running Sun Solaris 10.
> > > 
> > > When mounting with vers=2 on the sun (using NFSv2) the files
> > > "reappear", so this is a clear regression in NFSv3 between
> > > Linux 3.2 and Linux 3.7.
> 
> And the *only* thing you change is the kernel version, not nfs-utils or
> anything else in userspace?

Yes, only the kernel changed.

In the time since my original post in February I found the reason
for the problem:

The problem is caused by the patch "nfsd: vfs_llseek() with 32
or 64 bit offsets (hashes)". A discussion of regressions caused
by this patch can be found in the thread http://comments.gmane.org/gmane.comp.file-systems.ext4/37022
(started by you :-).

I solved the problem by disabling the 64-bit hash-extension.
I have not checked whether a better solution was found since
then.

> > Have you considered upgrading your Solaris version? I had tons 
> > of problems with NFS on Solaris10u6 and 10u8, including unresponsive
> > mount points, problems with delegations (esp. in the users' .ssh
> > directories and .Xauthority files) and strange "permission 
> > denied" error messages for some ACL feature I never configured on 
> > the server. 
> > 
> > NFS in Solaris 10u10 works much better together with Linux. I
> > haven't tried Solaris 11.
> > 
> > My servers run Squeeze and the Linux kernel from the squeeze-
> > backports repository (3.2.0-0.bpo.4-amd64).
> > 
> > > Maybe this can be fixed. I will be happy to give more info
> > > if someone is interested.
> 
> Most interesting would probably be packet captures in both the "good"
> and "bad" cases; so, something like:
> 
> 	tcpdump -s0 -wtmp.pcap
> 
> then reproduce the problem, then kill tcpdump and send tmp.pcap.
> 
> (And/or take a look at it yourself with "wireshark tmp.pcap", and there
> may be something obvious that jumps out even to a non-expert.)
> 
> --b.

I think this makes the proposed dump obsolete?

Thank you and best regards

-Ulrich


-- 
|-----------------------------------------------------------------------
| Ulrich Gemkow
| University of Stuttgart
| Institute of Communication Networks and Computer Engineering (IKR)
|-----------------------------------------------------------------------