Ansible playbook for CVE-2026-31431 mitigation

cve-2026-31431.yml

# Mitigation for CVE-2026-31431 ("Copy Fail") — algif_aead LPE
# https://xint.io/blog/copy-fail-linux-distributions
# Apply: ansible-playbook playbooks/cve-2026-31431.yml
#
# One mitigation per OS family, plus a cleanup pass for hosts that received
# the (now-abandoned) systemd seccomp drop-ins in earlier runs.
#
# Tags:
#   cve-kernel          RHEL/Alma 9, 10: add initcall_blacklist=algif_aead_init
#                       to GRUB. No reboot — the arg becomes active on the
#                       next boot.
#   cve-kernel-reboot   Superset of cve-kernel: writes the arg AND reboots
#                       (idempotent — skipped if already in /proc/cmdline).
#   cve-rmmod           Debian family: install /etc/modprobe.d/disable-algif.conf
#                       and rmmod algif_aead so the module is gone now and
#                       cannot be reloaded.
#   cve-systemd-remove  All hosts: remove sshd.service.d/ and [email protected]/
#                       cve-2026-31431.conf drop-ins, daemon-reload, restart
#                       sshd. [email protected] is intentionally not restarted
#                       (would kill rootless pods).
#
# Default (no --tags): every block runs, gated by ansible_os_family /
# ansible_distribution_major_version. RHEL hosts get kernel + reboot +
# systemd-remove; Debian hosts get rmmod + systemd-remove.
#
# Cleanup once the vendor kernel is patched:
#   RHEL/Alma:  grubby --update-kernel=ALL --remove-args=initcall_blacklist=algif_aead_init && reboot
#   Debian:     rm /etc/modprobe.d/disable-algif.conf

---
- hosts: all
  gather_facts: true
  become: yes
  serial: 1

  tasks:
    # --- kernel initcall blacklist (RHEL/Alma 9 and 10 only) ---
    # Prevents algif_aead_init from running at boot so the AEAD handler is
    # never registered with af_alg; bind(AF_ALG, ..., "aead") returns ENOENT
    # system-wide. algif_aead is built into vmlinux on these releases, so a
    # modprobe blacklist would do nothing.
    - name: Apply kernel initcall blacklist mitigation (RHEL/Alma 9 and 10)
      tags: [cve-kernel-reboot]
      when:
        - ansible_os_family == 'RedHat'
        - ansible_distribution_major_version in ['9', '10']
      block:
        - name: Read current GRUB kernel args
          ansible.builtin.command:
            cmd: grubby --info=ALL
          register: grubby_info
          changed_when: false
          tags: [cve-kernel, cve-kernel-reboot]

        - name: Add initcall_blacklist=algif_aead_init to all kernels
          ansible.builtin.command:
            cmd: grubby --update-kernel=ALL --args=initcall_blacklist=algif_aead_init
          when: "'initcall_blacklist=algif_aead_init' not in grubby_info.stdout"
          tags: [cve-kernel, cve-kernel-reboot]

        - name: Check whether running kernel has the mitigation active
          ansible.builtin.command:
            cmd: grep -qw initcall_blacklist=algif_aead_init /proc/cmdline
          register: cmdline_check
          changed_when: false
          failed_when: false

        - name: Reboot to activate the kernel mitigation
          when: cmdline_check.rc != 0
          block:
            - name: Trigger reboot
              ansible.builtin.shell: sleep 3; reboot
              ignore_errors: true
              changed_when: false
              async: 1
              poll: 0

            - name: Wait for server to come back after reboot
              ansible.builtin.wait_for_connection:
                timeout: 900
                delay: 40
              register: reboot_result

            - name: Reboot time
              ansible.builtin.debug:
                msg: "The system rebooted in {{ reboot_result.elapsed }} seconds."

            - name: Verify running kernel now includes the mitigation
              ansible.builtin.command:
                cmd: grep -qw initcall_blacklist=algif_aead_init /proc/cmdline
              changed_when: false

    # --- Debian family modprobe blacklist + unload ---
    # algif_aead is a loadable module on Debian/Ubuntu, so an install rule
    # plus rmmod removes the attack surface immediately and prevents
    # reloads (including auto-load via AF_ALG bind requests).
    - name: Apply algif_aead modprobe blacklist + unload (Debian family)
      tags: [cve-rmmod]
      when: ansible_os_family == 'Debian'
      block:
        - name: Write modprobe blacklist for algif_aead
          ansible.builtin.copy:
            dest: /etc/modprobe.d/disable-algif.conf
            owner: root
            group: root
            mode: '0644'
            content: |
              # CVE-2026-31431 mitigation. Prevents algif_aead from being loaded.
              install algif_aead /bin/false

        - name: Unload algif_aead if currently loaded
          ansible.builtin.command: rmmod algif_aead
          register: rmmod_result
          failed_when:
            - rmmod_result.rc != 0
            - "'not currently loaded' not in (rmmod_result.stderr | default(''))"
            - "'is not currently loaded' not in (rmmod_result.stderr | default(''))"
          changed_when: rmmod_result.rc == 0

        - name: Verify algif_aead is not loaded
          ansible.builtin.shell: "! lsmod | awk '{print $1}' | grep -qx algif_aead"
          changed_when: false

    # --- remove systemd AF_ALG drop-ins ---
    # Earlier iterations of this playbook installed RestrictAddressFamilies=~AF_ALG
    # drop-ins on sshd.service and [email protected]. The seccomp filter caused
    # issues for legitimate workloads and has been abandoned. Removing the
    # drop-ins is safe to run on hosts that never had them — file: state=absent
    # is a no-op there.
    - name: Remove systemd AF_ALG drop-ins
      tags: [cve-systemd-remove]
      block:
        - name: Remove sshd.service AF_ALG drop-in
          ansible.builtin.file:
            path: /etc/systemd/system/sshd.service.d/cve-2026-31431.conf
            state: absent
          register: sshd_dropin_removed

        - name: Remove [email protected] AF_ALG drop-in
          ansible.builtin.file:
            path: /etc/systemd/system/[email protected]/cve-2026-31431.conf
            state: absent
          register: user_dropin_removed

        - name: Reload systemd if drop-ins were removed
          ansible.builtin.systemd:
            daemon_reload: yes
          when: sshd_dropin_removed.changed or user_dropin_removed.changed

        # Restart sshd so the running daemon drops the now-removed
        # seccomp filter. Existing SSH sessions are not killed — only
        # the listener restarts. user@<uid>.service is intentionally
        # NOT restarted: that would kill rootless pods. The redundant
        # filter on running per-user instances clears naturally on the
        # next user-instance restart or host reboot.
        - name: Restart sshd to drop the seccomp filter
          ansible.builtin.systemd:
            name: sshd
            state: restarted
          when: sshd_dropin_removed.changed

Thanks a lot for sharing this. I wanted to add a comment for anyone who may be coming here to apply this remediation to HPC clusters running the Slurm job scheduler.

The remediation needs to be applied to slurmd as well on the compute nodes. Apply the second task in this playbook ("Deny AF_ALG for sshd-spawned processes") as a drop-in under /etc/systemd/system/slurmd.service.d/

And don't forget to restart your slurmd service after adding the drop-in.

Updated to add a second mitigation for RHEL 9/10 as mentioned here: https://news.ycombinator.com/item?id=47956504

Updated to remove the systemd workaround, since it was causing issues with containers and the initcall_blacklist is cleaner. Also added Debian. For Debian no reboot is needed. For RHEL it needs a reboot because the module is built in.

It seems that RHEL 8 also use built-in module, at least v8.10.

I can confirm that rhel 8.10 works with the playbook above. @antoine-mesobfc