Summary
A time-of-check to time-of-use (TOCTOU) vulnerability was found in containerd v2.1.0. While unpacking an image during an image pull, specially crafted container images could arbitrarily modify the host filesystem.
Impact
Pulling a malicious image is enough to corrupt or modify arbitrary host files on a node running the affected version of containerd — no container needs to start. In a Kubernetes context, that means a single Pod referencing an attacker-controlled image can compromise the node before the workload ever runs. Only containerd v2.1.0 is affected; earlier and later releases are not.
Detection
Identify nodes running containerd v2.1.0 with kubectl get nodes -o wide and inspect the CONTAINER-RUNTIME column. On the node, containerd --version confirms the running version. There is no specific runtime log signature exposed in the advisory — focus on the version inventory and on auditing recent image pulls from untrusted registries.
Mitigation
Upgrade containerd to v2.1.1 or later. v2.1.0 is the only affected version; rolling forward to the patched point release is the only fix.
Workaround until patched: ensure that only trusted images are used and that only trusted users have permissions to import images. Apply admission policy that pins images by digest, restricts allowed registries, and enforces image-signature verification (Cosign / Notation).