[lustre-discuss] [EXTERNAL] Re: RHEL 9.8 and 10.2 support

Fokke Dijkstra f.dijkstra at rug.nl
Mon Jun 29 11:40:57 UTC 2026


I managed to create a patched version of the lustre-client-dkms 2.15.8 rpm,
which fixes the issue that the kernel modules cannot be compiled on our
Alma Linux 9.8 client nodes.
Since thousands of users can login on or submit jobs to these clients we
absolutely need to be able to quickly fix privilege escalation exploits on
these nodes.

Here is a summary of the way I was able to rebuild the lustre-client-dkms
rpm:

The patches for the new kernels can be found via:
https://jira.whamcloud.com/browse/LU-20071
where LU-20071 is the label for the new Red Hat 9.8 release. I assume that
a similar procedure can be followed in the future for Red Hat 9.9 and its
derivatives.
Note that in our case we wanted to fix the LTS release 2.15.8. So look for
the branch b2_15 on the LU-20071 page. This specific patch can be found at:
https://review.whamcloud.com/c/fs/lustre-release/+/65797
Figuring this out took some time, as I did not know the structure of the
build system, and I first tried to use the patch for the latest release,
which would not apply.

After this I used the following procedure to build an updated
lustre-client-dkms rpm:

1. Download the Lustre source rpm from
https://downloads.whamcloud.com/public/lustre/latest-release/el9.7/client/SRPMS/lustre-client-dkms-2.15.8-1.el9.src.rpm
2. Install the rpm using:
  rpm -i lustre-client-dkms-2.15.8-1.el9.src.rpm
3. Go the the ~/rpmbuild/SOURCES directory and unpack the source file
  tar xzf lustre-2.15.8.tar.gz
4. Download the patch from
https://review.whamcloud.com/c/fs/lustre-release/+/65797, which can be
downloaded using the "DOWNLOAD" link. I selected c131e65.diff.zip
5. Unzip the patch into c131e65.diff
6. Apply the patch using something like:
  cd lustre-2.15.8
  patch -p1 < ../c131e65.diff
  cd ..
7. Create an updated source tar file:
  tar czf lustre-2.15.8.tar.gz lustre-2.15.8
8. Fix the hardcoded buildid in the spec file, by changing it from 1 to 2,
so the line reads:
%define buildid 2
9. Build the rpm (client without zfs, but with ldiskfs for our use case):
rpmbuild -ba rpmbuild/SPECS/lustre-dkms.spec --without servers --without
zfs --with ldiskfs

This results in
rpmbuild/RPMS/noarch/lustre-client-dkms-2.15.8-2.el9.noarch.rpm

Installing this rpm, together with lustre-client-2.15.8-1.el9.x86_64.rpm
resulted in a working client on Alma Linux 9.8. I've tested some nodes
using the IOR benchmark.

I hope these instructions are helpful to others. Please let me know if you
have suggestions to improve the procedure.

Kind regards,

Fokke Dijkstra

Op di 9 jun 2026 om 19:07 schreef Mohr, Rick <mohrrf at ornl.gov>:

> Looking through my scripts, the configure command was run with the
> following options (with paths to dirs substituted for variables) after the
> patch had been applied to the source code:
>
>   configure --with-linux=$KERN_DIR --with-linux-obj=$KERN_OBJ_DIR
> --with-o2ib=$OFED_DIR --disable-gss --disable-gss-keyring --disable-server
> --disable-tests --disable-iokit --with-mds-max-threads=8192
> --enable-mpitests=yes
>
> Then I just ran "make rpms" to create the packages.  Everything was pretty
> straightforward, although not all the options I used would necessarily be
> needed at another site.
>
> --Rick
>
>
> On 6/4/26, 12:35 PM, "Mohr, Rick" <mohrrf at ornl.gov <mailto:
> mohrrf at ornl.gov>> wrote:
>
>
> Well, I actually built the client using a wrapper script that James
> Simmons wrote to manage git checkouts, patch application, etc. so the steps
> were hidden by that process. I'll see if I can dig out the underlying
> commands and post them here.
>
>
> --Rick
>
>
>
>
>

-- 
Fokke Dijkstra <f.dijkstra at rug.nl> <f.dijkstra at rug.nl>
Team High Performance Computing
Center for Information Technology, University of Groningen
Postbus 11044, 9700 CA  Groningen, The Netherlands
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.lustre.org/pipermail/lustre-discuss_lists.lustre.org/attachments/20260629/44faea52/attachment.html>


More information about the lustre-discuss mailing list