Incomplete tracking in PostgreSQL of changes to role membership, role attributes, and database ownership allows a query to continue using cached row-level security policies after those changes require a different policy, via plan reuse. Stale policies continue until some other event invalidates the cache or connection termination ends the session. This permits a user to complete reads and modifications that were recently permitted but now forbidden. An attacker must tailor an attack to a particular applications pattern of privilege removal and role-specific row security policies. Versions before PostgreSQL 18.5, 17.11, 16.15, 15.19, and 14.24 are affected.
Weakness
The product has or supports multiple distributed components or sub-systems that are each required to keep their own local copy of shared data - such as state or cache - but the product does not ensure that all local copies remain consistent with each other.
Extended Description
In highly distributed environments, or on systems with distinct physical components that operate independently, there is often a need for each component to store and update its own local copy of key data such as state or cache, so that all components have the same “view” of the overall system and operate in a coordinated fashion. For example, users of a social media service or a massively multiplayer online game might be using their own personal computers while also interacting with different physical hosts in a globally distributed service, but all participants must be able to have the same “view” of the world. Alternately, a processor’s Memory Management Unit (MMU) might have “shadow” MMUs to distribute its workload, and all shadow MMUs are expected to have the same accessible ranges of memory. In such environments, it becomes critical for the product to ensure that this “shared state” is consistently modified across all distributed systems. If state is not consistently maintained across all systems, then critical transactions might take place out of order, or some users might not get the same data as other users. When this inconsistency affects correctness of operations, it can introduce vulnerabilities in mechanisms that depend on consistent state.