Summary
Fission's builder manager previously accepted Package resources whose referenced environment lived in a different Kubernetes namespace than the Package itself.
This mismatch could let a tenant bind to or influence environments outside their own namespace, undermining isolation and multi-tenant boundaries in affected clusters.
Affected context
The vulnerability appears in Fission deployments where multiple namespaces are used for logical or tenant isolation.
It affects the buildermgr controller's handling of Package Custom Resources when it does not enforce that Package.spec.environment.namespace is the same as Package.metadata.namespace.
The risk is most pronounced in multi-tenant or shared clusters where different namespaces are expected to be mutually isolated.
Exploit preconditions
- Attack surface: Kubernetes API server, via Fission Package Custom Resources
- Required access: Ability to create or modify Fission Package CRDs within some namespace
- Trigger point: Submission of a Package whose
spec.environment.namespacepoints to a different namespace than the Package's own metadata namespace - Why exploitation is possible: The buildermgr controller trusted the namespace specified in
spec.environmentwithout cross-checking it againstmetadata.namespace, breaking the intended namespace isolation model
Why this happens
This issue stems from missing authorization and validation logic around namespace relationships in custom resource handling.
Instead of deriving or enforcing the environment namespace from the Package's own namespace, the buildermgr controller relied on a user-controlled field (Package.spec.environment.namespace).
This maps to CWE-441: Unintended Proxy or Intermediary and CWE-862: Missing Authorization, as the controller effectively acted as an intermediary performing actions in other namespaces without proper checks, exposing resources across isolation boundaries.
Potential impact
- Cross-namespace association of Packages with environments in another namespace
- Potential privilege escalation if a less-privileged tenant can bind to more-privileged or trusted environments
- Violation of multi-tenant isolation assumptions within the Kubernetes cluster
- Increased blast radius for misconfigurations or compromised tenants due to unintended cross-namespace interactions
General mitigation guidance
Mitigation typically focuses on enforcing strict namespace alignment and minimizing tenant control over cross-namespace references.
Recommended defensive patterns include:
- Validating that all environment references in Package specifications are constrained to the Package's own namespace or to an explicitly allowed set of namespaces
- Treating any user-controlled namespace or resource reference fields as untrusted and subject to server-side validation
- Designing controllers to derive security-relevant context (such as namespace) from trusted metadata rather than from user-supplied spec fields
- Applying least-privilege RBAC so that controller service accounts and tenants cannot perform unnecessary cross-namespace operations
How teams detect issues like this
Detection can focus on both configuration state and controller behavior:
- Static or policy-based checks (e.g., admission controllers, OPA/Gatekeeper, Kyverno) that flag or block Package CRDs where
spec.environment.namespacediffers frommetadata.namespace - Code review patterns that look for controllers or operators trusting spec-level namespace fields instead of using the object's metadata namespace
- Cluster audits to identify existing Package resources whose environment references cross namespace boundaries
- Use of automated scanning tools that detect improper access control or cross-namespace references in Kubernetes custom resources

