[Bug 2077929] Re: Failed to insert openafs module on ARM platform: Exec format error

Mauricio Faria de Oliveira 2077929 at bugs.launchpad.net
Wed Sep 4 22:29:14 UTC 2024


Hi Chengen,

Thanks for the clarifications!

For Jammy, I guess it might be a kernel config difference between
gcp/azure, perhaps (as we both tested 6.5.0-1025-flavor).

Uploaded, based on your reproduction logs in Jammy.

For Focal, I tried reproducing with gcc-10 (the latest supported gcc in
Focal) and in GCP (where you could reproduce the problem), but the issue
still doesn't reproduce; so let's not modify the package.

Marked Focal as Won't Fix.

Thanks again!
Mauricio

...

$ lsb_release -cs
focal

$ uname -m
aarch64

$ uname -rv
5.15.0-1067-gcp #75~20.04.1-Ubuntu SMP Thu Aug 8 00:30:06 UTC 2024

mfo at mfo-focal-arm:~$ sudo dmesg | grep DMI:
[    0.053624] DMI: Google Google Compute Engine/Google Compute Engine, BIOS Google 08/06/2024

$ apt-cache show dkms | grep Depends: | grep c-compiler | head -n1
Depends: ..., gcc | c-compiler, ...

$ apt-cache show gcc-10 | grep Provides: | grep c-compiler | head -n1
Provides: c-compiler

$ sudo apt install -y gcc-10
$ sudo update-alternatives \
  --install /usr/bin/gcc gcc /usr/bin/gcc-10 100 \
  --slave /usr/bin/cpp cpp /usr/bin/cpp-10 

$ sudo apt install openafs-client

$ objdump -s -j .comment --no-show-raw-insn /lib/modules/5.15.0-1067-gcp/updates/dkms/openafs.ko | grep -m1 -A2 'GCC:'
 0000 00474343 3a202855 62756e74 75203130  .GCC: (Ubuntu 10
 0010 2e352e30 2d317562 756e7475 317e3230  .5.0-1ubuntu1~20
 0020 2e303429 2031302e 352e3000 00474343  .04) 10.5.0..GCC

$ sudo modprobe openafs
$ sudo dmesg | grep openafs
[  754.935849] openafs: loading out-of-tree module taints kernel.
[  754.935855] openafs: module license 'http://www.openafs.org/dl/license10.html' taints kernel.
[  754.937239] openafs: module verification failed: signature and/or required key missing - tainting kernel

-- 
You received this bug notification because you are a member of Ubuntu
Sponsors, which is subscribed to the bug report.
https://bugs.launchpad.net/bugs/2077929

Title:
  Failed to insert openafs module on ARM platform: Exec format error

Status in openafs package in Ubuntu:
  Fix Released
Status in openafs source package in Focal:
  Won't Fix
Status in openafs source package in Jammy:
  In Progress
Status in openafs source package in Noble:
  In Progress
Status in openafs source package in Oracular:
  Fix Released

Bug description:
  [Impact]
  The openafs module cannot be installed on the ARM platform, resulting in the following error:
  insmod /lib/modules/6.5.0-1025-gcp/updates/dkms/openafs.ko 
  modprobe: ERROR: could not insert 'openafs': Exec format error

  [Fix]
  There is an upstream commit that addresses this issue:

  commit 2f260825b96b76556aba82fc364c841357a23a7d
  Author:     Cheyenne Wills <cwills at sinenomine.net>
  Date: Thu Mar 28 12:51:46 2024 -0400

      Linux: Remove weak attribute for key_type_keyring

      The reference to the Linux symbol key_type_keyring uses a weak
      attribute to assist in determining if the symbol is exported.  This
      method was introduced in commit: 'keyring-updated-20080310' (b7fb842b1c)
      as a way to detect if the symbol was exported by the Linux kernel.  A
      later commit: 'linux-keyring-export-check-20090701' (4ca66112c2)
      introduced an autoconf test that eliminated the need for using the weak
      attribute, but the commit did not remove the attribute.

      On a Debian-12 system on ARM64 the use of the weak attribute in this
      case causes an error when loading the kernel module:
        "ERROR: could not insert 'openafs': Exec format error".
      The error is due to the Linux module loader not supporting the
      relocation entry type in the openafs kernel module for the
      key_type_keyring.  Further investigation showed that this problem could
      occur with gcc-12 on a ARM64 system (see below for additional
      information).

      The code in osi_groups.c uses a pointer, __key_type_keyring, to the
      key_type_keyring.  When this symbol is exported by the Linux kernel, the
      loader would resolve the address and __key_type_keyring would contain
      the address.  If the key_type_keyring symbol was not exported,
      __key_type_keyring would be set to NULL (due to the weak attribute
      associated with key_type_keyring).

      Remove the weak attribute for key_type_keyring when the configure
      test, introduced in the 4ca66112c2 commit, determines that the symbol
      is exported (EXPORTED_KEY_TYPE_KEYRING is defined).  When the symbol
      isn't exported (pre Linux 2.6.22), just set the pointer,
      __key_type_keyring to NULL.

      NOTE:

      The load error was reported on the openafs IRC channel by "clarkb"

      The problem of loading the openafs kernel module was also observed
      on a NixOS system as well, and is described in the NixOS ticket

        https://github.com/NixOS/nixpkgs/issues/284501

      A Debian-11 system did not have a problem loading the openafs kernel
      module. Investigation of the differences between Debian-11 and Debian-12
      showed that the problem is due to a series of fixes in the toolchain
      (in particular gcc-12) to address 'Cortex-A53 erratum #843419', which
      deals with how the ARM64 ADRP instruction is used. With gcc-12, the
      code generated uses a R_AARCH64_ADR_GOT for this particular case (which
      isn't supported by the Linux kernel module loader).  Gcc-11 created a
      R_AARCH64_ABS64 relocation entry type for the symbol.

  [Test Plan]
  1. Install the OpenAFS client using the command `sudo apt install openafs-client`.
  2. Load the kernel module by running `sudo modprobe --verbose openafs`.
  3. Confirm that the module loads successfully without any errors.

  [Where problems could occur]
  The patch removed the weak attribute from key_type_keyring.
  Previously, commit #4ca66112c2 (linux-keyring-export-check-20090701) introduced an autoconf test that made the weak attribute unnecessary, but the attribute was not removed at that time.
  If the patch causes a regression, it might prevent the openafs module from being installed successfully.

To manage notifications about this bug go to:
https://bugs.launchpad.net/ubuntu/+source/openafs/+bug/2077929/+subscriptions




More information about the Ubuntu-sponsors mailing list