containerdに不適切な権限管理の脆弱性（CVE-2026-95837）

### Impact A vulnerability in containerd's CRI implementation allows a container restored from an untrusted checkpoint via the `CreateContainer` API to bypass the destination security context and execute with elevated privileges. When restoring a container from a checkpoint archive or annotated OCI image, CRIU restores process credentials, Linux capabilities, `no_new_privs`, and seccomp state directly from checkpoint data rather than enforcing the destination CRI `ContainerConfig`. An attacker who can run a container with a crafted checkpoint image can execute processes as root with full capabilities and no enforced seccomp filters despite restrictive security policies requested by the orchestrator. Additionally, containerd's CRI status reporting reflects the requested configuration rather than the actual restored process state, masking the privilege discrepancy from orchestrators. This issue affects Linux systems running containerd with CRI checkpoint restore enabled; users not utilizing CRI checkpoint restore are not affected. ### Patches This bug has been addressed in containerd 2.2.7 and 2.3.4 by disabling checkpoint restore via `CreateContainer` by default. The feature has been removed entirely in containerd 2.4.0. Users should update to these versions to resolve the issue. Running containers that were restored from untrusted checkpoints should be stopped, deleted, and recreated. Users who rely on restore via the `CreateContainer` API can re-enable the feature by setting the new `enable_experimental_restore_via_create` configuration option. However, enabling restore via `CreateContainer` leaves this vulnerability present, as containerd cannot enforce destination security policy during CRIU process restoration. ### Workarounds On unpatched versions, containerd does not provide a configuration option to disable checkpoint restore via `CreateContainer`. Deployments should restrict container creation permissions to trusted users, validate container image registries, or use admission controllers to reject images containing checkpoint data. ### References * [CRIU image security](https://checkpoint-restore.github.io/criu/security/image-security.html) ### Credits The containerd project would like to thank @l2yyd5 and @berkpolatCE for responsibly disclosing this issue in accordance with the [containerd security policy](https://github.com/containerd/project/blob/main/SECURITY.md). ### For more information If you have any questions or comments about this advisory: * Open an issue in [containerd](https://github.com/containerd/containerd/issues/new/choose) * Email us at [security@containerd.io](mailto:security@containerd.io) To report a security issue in containerd: * [Report a new vulnerability](https://github.com/containerd/containerd/security/advisories/new) * Email us at [security@containerd.io](mailto:security@containerd.io)

containerd is an open-source container runtime. From 2.1.0 until 2.2.7 and 2.3.4, containerd's CRI CreateContainer checkpoint restore is vulnerable because CRIU restores process security state from checkpoint data instead of enforcing the destination ContainerConfig. An attacker who can create a container from a crafted checkpoint can supply restored credentials, capabilities, no_new_privs, and seccomp state. The restored process can run as root with full capabilities and no enforced seccomp filters while CRI status reports the requested configuration. Systems that do not use CRI checkpoint restore are not affected. This issue is fixed in versions 2.2.7 and 2.3.4.