github.com/siderolabs/talos
Affected
- introduced
0, fixed1.12.7(semver)
Fixed versions
1.12.7
Fix
- Fixed version
1.12.7go get github.com/siderolabs/[email protected]
Advisory
Also known as: None
Talos Linux has a local privilege escalation from untrusted workloads
Summary
A vulnerability in the Linux kernel's algif_aead subsystem (CVE-2026-31431, "copy.fail") allows an unprivileged container workload to corrupt arbitrary file page-cache pages via the AF_ALG crypto interface and splice(). On Talos Linux, this vulnerability can be chained into a complete node compromise: an attacker who can schedule a pod on a worker node can, without any elevated Kubernetes permissions, achieve arbitrary code execution as root on the host (by poisoning a binary inside a privileged pod, or poisoning a binary which runs with elevated privileges like a CNI binary), access host filesystem, including node secrets.
The exploit does not require kernel debugging, race conditions, or any prior privileges beyond the ability to create a pod.
Impact
An attacker with the ability to deploy a Kubernetes pod on an affected node can:
Patches
Upgrade to Talos v1.13.0 or Talos v1.12.7 which ships Linux kernel 6.18.25. The kernel fix for CVE-2026-31431 (algif_aead in-place optimization revert) was committed upstream in Linux 6.18.22 and is included in all Talos releases from v1.13.0 and Talos 1.12.7 onwards.
Workarounds
There are multiple workarounds available based on the situation, but we really recommend to upgrade.
Option 1 - Change kernel arguments
Add a kernel argument with initcall_blacklist=algif_aead_init by upgrading Talos to the same version.
> Note: this either requires setting machine.kernel.extraKernelArgs if using BIOS based boot or upgrading with a new image from factory/imager generated image by setting the extra kernel args. See Boot Assets
Option 2 - Deploy all workload pods with a seccomp profile denying creating AF_ALG socket creation
patch.yaml
machine:
seccompProfiles:
- name: copy-fail-block.json
value:
defaultAction: SCMP_ACT_ALLOW
syscalls:
- names:
- socket
action: SCMP_ACT_ERRNO
args:
- index: 0
value: 38
op: SCMP_CMP_EQ
Apply this patch to all machines in the cluster and set this for all the pod spec:
...
spec:
securityContext:
seccompProfile:
type: Localhost
localhostProfile: profiles/copy-fail-block.json
Option 3 - Block the syscall in runtime with a eBPF program
See copy-fail-blocker, this can be applied to a running system without a reboot, but it has to run before any other workloads are scheduled after a reboot.
References
High Basis: The source's own band: HIGH
0, fixed 1.12.7 (semver)1.12.7
1.12.7go get github.com/siderolabs/[email protected]Fix commands are the export's own generic commands. Ojuju never runs them. Check them against your setup first.
Check your own computer: guard packages check
Source: GitHub Advisory Database (reviewed advisories) · CC BY 4.0