Skip to content

Probe ServiceAccount can patch every node's key of the fleet-role-probe ConfigMap #270

Closed as not planned

Description

@Mearman

Every fleet-role-probe pod runs as the same ServiceAccount, whose Role grants get and patch on the shared fleet-role-probe ConfigMap. So any node's probe token can patch any node's key: a faulty or hostile node can write a false summary for another node, which the controller then judges, or enlarge the object for every node's write. The probe refuses to publish a summary over probe_entry_limit_bytes (#267), but that check runs in the probe and cannot bind a token used from outside it.

RBAC cannot narrow this, since resourceNames cannot name "the caller's node", and a per-node Lease, Node annotation or per-node ConfigMap would each need wider grants than the current one named object.

The fix is an admission check on the ConfigMap: a ValidatingAdmissionPolicy, installed by the ARC role with the probe's other objects, that allows a probe ServiceAccount's update only when every changed data key is the node named in the request's bound token (the authentication.kubernetes.io/node-name extra) and the key's value is within the entry limit. It should be tested in the mesh scenarios with a write to another node's key that the API refuses.

Refs #266

Activity

  1. Mearman commented on Nov 20, 2026

    @Mearman
    MemberAuthor

    Duplicate of #268, which now carries the token node-name mechanism and the mesh test described here, and also covers the autoscaler-status object.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions