Rancher user retains access to clusters despite Global Role removal

Impact

A vulnerability has been identified within Rancher Manager, where after removing a custom GlobalRole that gives administrative access or the corresponding binding, the user still retains access to clusters. This only affects custom Global Roles that:

• Have a * on * in * rule for resources
• Have a * on * rule for non-resource URLs

For example

apiVersion: management.cattle.io/v3
kind: GlobalRole
metadata:
  name: custom-admin
rules:
  - apiGroups:
      - '*'
    resources:
      - '*'
    verbs:
      - '*'
  - nonResourceURLs:
      - '*'
    verbs:
      - '*'

Specifically:

• When a user is bound to a custom admin GlobalRole, a corresponding ClusterRoleBinding is created on all clusters that binds them to the cluster-admin ClusterRole.
• When such a GlobalRole or the GlobalRoleBinding (e.g., when the user is unassigned from this role in UI) is deleted, the ClusterRoleBinding that binds them to the cluster-admin ClusterRole stays behind.

This issue allows a user to continue having access to clusters after they have been unassigned from the custom admin global role or the role has been deleted.

Please consult the associated MITRE ATT&CK - Technique - Account Access Removal for further information about this category of attack.

Patches

This vulnerability is addressed by removing the corresponding ClusterRoleBindings whenever the admin GlobalRole or its GlobalRoleBindings are deleted. Previously orphaned ClusterRoleBindings are marked with the annotation authz.cluster.cattle.io/admin-globalrole-missing=true and should be deleted manually.

Orphaned ClusterRoleBindings can be listed with:

kubectl get clusterrolebinding -o jsonpath='{range .items[?(@.metadata.annotations.authz\.cluster\.cattle\.io/admin-globalrole-missing=="true")]}{.metadata.name}{"\n"}{end}'

Patched versions of Rancher include releases v2.12.3, v2.11.7.

Complications with the restricted admin functionality prevented the patches from being included in v2.10 and v2.9.

Workarounds

If the deployment can't be upgraded to a fixed version, users are advised to manually identify the orphaned and remove them.