[OpenAFS] PAGS and kernels

[Kirk Patton]

Hello,

I am having similar issues under Linux kernel 2.6.14.  I am running
a SGE(Sun Grid Engine) cluster and I am using OpenAFS 1.4.0.  I tried
upgrading the OpenAFS binaries on one of the farm boxes to 1.4.1, but I 
get the same issue.  Reverting back to 2.6.9, and everything is fine 
again.

The issue I am seeing is;
SGE is configured to use AFS via a home-grown program that klogs on 
behalf of the use.  SGE has two parameters it uses to get work with
AFS.  One is the path to a program that can get tokens granted for 
the user submitting jobs, and the other is the path to pagsh.

If I set the path to pagsh to '/bin/sh', the problem of missing
tokens under 2.6.14 goes away. But, then, I have a very insecure 
environment.

Does anyone have any suggestions on how to correct this?

Thanks,
Kirk



On Mon, Sep 11, 2006 at 09:16:46PM -0700, Mike Polek wrote:
> Oops... openafs 1.4.1, compiled from source rpm, built 5/24/2006.
> The kernel module was compiled in the last few weeks from the
> same original source RPM.... openafs-1.4.1-fc4.1.src.rpm.
> 
> Thx,
> Mike
> 
> Jeffrey Altman wrote:
> >It would help if you said what version of OpenAFS you were using.
> >
> >Mike Polek wrote:
> >
> >>Hi, all,
> >> I downloaded kernel 2.6.17.7 from kernel.org and built it as
> >>usual for my environment. I needed the newer kernel because
> >>I have some machines with recent NIC and SCSI interfaces
> >>that don't get picked up under the old 2.6.13-1.1532_FC4 kernel
> >>I've been using for my FC4 machines. Using pagsh, PAGS *appear*
> >>to work normally. However, my FTP server exhibits strange
> >>behaviors. It looks like it's running outside a PAG, even though
> >>the PAM library claims it's setting up the PAG properly when
> >>I turn debugging on.
> >> When I used the old kernel on a machine without the new
> >>hardware, but everything else on the O/S exactly the same
> >>(It's a diskless booting config, so the O/S is on the network),
> >>the FTP server seems to work properly. My conclusion is that
> >>the kernel isn't supporting PAGs properly. (Not unusual these
> >>days.)
> >> Is there a good way to figure out the extent of brokenness
> >>for PAG support for a given kernel? I.e. is there some
> >>set of tests I can perform to figure out where the problem
> >>has likely cropped up? And has anyone else experienced similar
> >>problems between 2.6.13 an 2.6.17, so maybe there is someplace
> >>in between that has both the drivers I need and correct PAG
> >>support? Or alternatively, is there a patch available for
> >>the 2.6.17 kernels?
> >>
> >>Thanks in advance,
> >>Mike Polek
> >>Pictage, Inc.
> >>mike@pictage.com

-- 
Kirk Patton
Unix Administrator
Transmeta Inc.
Tel. 408 919-3055