CRI-O's pods can break out of resource confinement on cgroupv2

Impact

What kind of vulnerability is it? Who is impacted?
All versions of CRI-O running on cgroupv2 nodes.
Unchecked access to an experimental annotation allows a container to be unconfined. Back in 2021, support was added to support an experimental annotation that allows a user to request special resources in cgroupv2. It was supposed to be gated by an experimental annotation: io.kubernetes.cri-o.UnifiedCgroup, which was supposed to be filtered from the list of allowed annotations . However, there is a bug in this code which allows any user to specify this annotation, regardless of whether it’s enabled on the node. The consequences of this are a pod can specify any amount of memory/cpu and get it, circumventing the kubernetes scheduler, and potentially be able to DOS a node.

Patches

Has the problem been patched? What versions should users upgrade to?
1.29.1, 1.28.3, 1.27.3

Workarounds

Is there a way for users to fix or remediate the vulnerability without upgrading?
use cgroupv1

References

Are there any links users can visit to find out more?