CVE-2024-52280: github.com/rancher/steve's users can issue watch commands for arbitrary resources

A vulnerability has been discovered in Steve API (Kubernetes API Translator) in which users can watch resources they are not allowed to access, when they have at least some generic permissions on the type. For example, a user who can get a single secret in a single namespace can get all secrets in every namespace.

During a watch request for a single ID, the following occurs:

• In the case of a watch request for a single resource, Steve API will return a partition with the requested resource in it. In other cases, it will check the user’s access when constructing partitions.
• When a watch request for a single resource is issued, instead of using a client which impersonates the user making the request, Steve API will use the admin client, which can read all resources.

This allows any requester to see the contents of any object such as secret keys, signing certificates, API tokens.

Please consult the associated MITRE ATT&CK - Technique - Valid Accounts and MITRE ATT&CK - Technique - Container and Resource Discovery for further information about this category of attack.