[SRU] module drbd8 update kernel from 2.6.32-16 to 2.6.32-20

Can you verify if this is still present? I just did:

$ sudo apt-get install drbd8-utils linux-headers-generic
that succeeded in building drbd.ko

Then:
$ sudo modprobe drbd

I'm currently booted into -19-generic and have -21-generic installed.

It seems strange that you got the module built but it wont load, rather than failing to build.

If i did apt-get install linux-headers-2.6.32-20-generic linux-image-2.6.32-20-generic from a version with the 2.6.32-16 kernel and the drbd module, I have the same error :

Invalid module format when i restarted on the last kernel

The syslog tail the same thing, just :
drbd8 (8.3.7): Installing module on kernel 2.6.32-20-generic.

The status of dkms :
# dkms status
drbd8, 8.3.7, 2.6.32-16-generic, i686: installed
drbd8, 8.3.7, 2.6.32-20-generic, i686: installed

ah. so it would seem that dkms is not finding the correct version of headers if there are multiple present.
Does that sound right?

No, i don't think
Because, when i did this commands :

# dkms remove -m drbd8 -v 8.3.7 -a 2.6.32-20-generic
# dkms build -m drbd8 -v 8.3.7 -a 2.6.32-20-generic
# dkms install -m drbd8 -v 8.3.7 -a 2.6.32-20-generic

Without any other command, the module is successfully build and modprobe don't return any error.
drbd8 run perfectly

On 16.04.2010 21:38, Gaëtan Ferez wrote:
> No, i don't think
> Because, when i did this commands :
>
> # dkms remove -m drbd8 -v 8.3.7 -a 2.6.32-20-generic
> # dkms build -m drbd8 -v 8.3.7 -a 2.6.32-20-generic
> # dkms install -m drbd8 -v 8.3.7 -a 2.6.32-20-generic
>
> Without any other command, the module is successfully build and modprobe don't return any error.
> drbd8 run perfectly

So, when you tell dkms explicitly which headers to use and what module
version to build, everything is fine. If you leave autodetection to
dkms, it fails.

Hm... This sounds like dkms issue.

On 18.04.2010 21:15, Ante Karamatić wrote:

> Hm... This sounds like dkms issue.

I managed to reproduce the issue. I've booted 2.6.32-16 kernel and
installed drbd8-utils. drbd module loaded fine and worked. But when
rebooted in 2.6.32-21, it failed with the same symptoms:

drbd: disagrees about version of symbol module_layout

Purging and reinstalling drbd8-utils (and thus rebuilding drbd module),
drbd module was loadable and usable in 2.6.32-21.

So, I'm quite sure this is dkms issue with headers...

Hi,

I search a reason of this problem, and I found something, when dkms launch the drbd module compilation with make command, the Makefile of drbd take the kernel source dir of 2.6.32-16-generic :

DKMS make.log for drbd8-8.3.7 for kernel 2.6.32-21-generic (i686)
lundi 19 avril 2010, 11:04:01 (UTC+0200)
make: entrant dans le répertoire « /var/lib/dkms/drbd8/8.3.7/build/drbd »

    Calling toplevel makefile of kernel source tree, which I believe is in
    KDIR=/lib/modules/2.6.32-16-generic/build

test -f ../scripts/adjust_drbd_config_h.sh && \
  KDIR=/lib/modules/2.6.32-16-generic/build O= /bin/bash ../scripts/adjust_drbd_config_h.sh

but dkms launch this command :
make KERNELRELEASE=2.6.32-21-generic -C drbd KERNELDIR=/lib/modules/2.6.32-21-generic/build

I looked in the Makefile of drbd module.We have this line with research the KDIR :

  # for some reason some of the commands below only work correctly in bash,
  # and not in e.g. dash. I'm too lazy to fix it to be compatible.
  SHELL=/bin/bash

  DRBDSRC := $(shell pwd)
  export DRBDSRC

  # to be overridden on command line:
  PREFIX := /
  ifneq ($(wildcard ../build-for-uml),)
    #### for Philipp's convenience :)
    ARCH_UM := "ARCH=um"
    KDIR := /usr/src/linux-um
  else
    ifeq ($(wildcard /lib/modules/$(shell uname -r)/source),)
      KDIR := /lib/modules/$(shell uname -r)/build
    else
      KDIR := /lib/modules/$(shell uname -r)/source
      ifneq ("$(origin KDIR)", "command line")
        ifneq ($(wildcard /lib/modules/$(shell uname -r)/build),)
          O := /lib/modules/$(shell uname -r)/build
        endif
      endif
    endif
  endif

Great, thank you. I'm applying the patch for this issue and subscribing main sponsors.

Obviously, attaching, not applying :)

Proposed fix for lucid-proposed.

The fix does not work on my freshly kickstarted ubuntu lucid server. The result is the same, dkms complains that no kernel sources are installed despite both generic and server headers are present.

The reason is that dkms tries to use the build link in /lib/modules/2.6.32-21-generic (which does not exist) instead of the one in /lib/modules/2.6.32-21-server (which does and correctly points to the server headers).

The key to the whole issue is the use of uname -r (results in ...-generic) by dkms to determine where the kernel sources are. There either needs to be a fix that creates the correct ...-server reference out of 'uname -r' or a the kernel needs to be renamed ...-server so the build reference does not have to be changed.

Adding a link /lib/modules/2.6.32-21-generic/build -> /usr/src/2.6.32-21-server can serve as a workaround but is not a sustainable solution.

The fix as provided does not work.

Hi,

I also tried to test via a kernel 2.6.32-21-server to 2.6.32-22-server but I had no problem passing dkms is properly implemented with the patch on deb package

# apt-get install linux-headers-2.6.32-22-server linux-image-2.6.32-22-server
root@ubuntu:/var/lib/dkms/drbd8# dkms status
drbd8, 8.3.7, 2.6.32-22-server, x86_64: installed
drbd8, 8.3.7, 2.6.32-21-server, x86_64: installed

Statement of Impact: DRBD in lucid had a bug that will build a module for all kernels with headers of a running kernel. This bug will be fixed in maverick as soon as it is reproduce there.

The patch attached to this bug will fix the issue in lucid.

TEST CASE:

1. Boot old kernel
2. Install server kernel
3. Install linux-headers-server
4. Install drbd8-source
5. Reboot to the new kernel, drbd8 module will load.

I dont expect any regressions with this patch. If you have any questions please let me know.

Regards
chuck

crh did you install pached drbd8-source or the one from archives? The one in archives isn't fixed yet.

Accepted drbd8 into lucid-proposed, the package will build now and be available in a few hours. Please test and give feedback here. See https://wiki.ubuntu.com/Testing/EnableProposed for documentation how to enable and use -proposed. Thank you in advance!

Hi,

I tested the package in lucid-proposed, and the problem disappear.
drbd module kernel build correctly with the new linux kernel

drbd8, 8.3.7, 2.6.32-21-server, x86_64: installed
drbd8, 8.3.7, 2.6.32-22-server, x86_64: installed

Thanks

This bug was fixed in the package drbd8 - 2:8.3.7-1ubuntu2.1

---------------
drbd8 (2:8.3.7-1ubuntu2.1) lucid-proposed; urgency=low

  * Use KDIR instead of KERNELDIR variable for headers path
    - thanks Gaëtan Ferez for debugging
    - (LP: #562832)
 -- Ante Karamatic <email address hidden> Mon, 19 May 2010 09:22:17 +0200

copied to maverick, too.