CRI-O sandbox state poisoning fixed
criticalAn attacker who could create a Pod with arbitrary annotations could poison persisted sandbox state. After a CRI-O restart or node reboot, containers recreated in the affected sandbox could mount attacker-controlled host paths, potentially exposing the CRI-O socket and enabling container escape.
Show release-note lines
A user who can create a Pod with arbitrary annotations can poison persisted CRI-O sandbox state. After a CRI-O restart or node reboot, containers recreated in the affected sandbox may mount attacker-controlled host paths, potentially exposing the CRI-O socket and enabling container escape. The users must wipe all running containers on the nodes when upgrading CRI-O Admission webhook rejecting Pod annotations and labels with the io.kubernetes.cri-o. prefix. For example, a validating webhook that denies any Pod whose metadata.annotations or metadata.labels contain keys starting with io.kubernetes.cri-o.github.com/cri-o/cri-o/security/advisories/GHSA-6wmc-5877-hgc9
CRI-O sandbox state poisoning after restart can expose crio.sockgithub.com/cri-o/cri-o/security/advisories/GHSA-6wmc-5877-hgc9