Summary
A privilege misconfiguration in Fission's PodSpec safety validation allows tenant-created functions or environments to run with the CAP_SYS_TIME Linux capability.
Under multi-tenant Kubernetes deployments where tenants can define Function or Environment CRDs, this can let attacker-controlled code adjust system time in its container or node, undermining security controls.
Affected context
The issue arises in Fission-based, Kubernetes-native serverless platforms where:
- Tenants are allowed to create or modify Environment and Function Custom Resources
- PodSpec safety validation relies on a denylist-based capability filter
- Runtime containers inherit Linux capabilities directly from tenant-specified securityContext
In these contexts, omitted capabilities like CAP_SYS_TIME can slip through validation, granting broader privileges than intended.
Exploit preconditions
- Attack surface: Fission's tenant-facing Environment and Function CRDs exposed via Kubernetes API
- Required access: Ability for a tenant or attacker to create or edit Function/Environment CRDs in a target namespace
- Trigger point: Submission of a PodSpec (or container securityContext) that adds CAP_SYS_TIME to capabilities
- Why exploitation is possible: The admission webhook and merge-layer sanitization use a fixed capabilities denylist, failing to block CAP_SYS_TIME and potentially other powerful capabilities, so they are propagated into the final function container.
Why this happens
This vulnerability stems from a privilege enforcement design that uses an explicit denylist of "disallowed" Linux capabilities instead of defining a minimal allowed set.
Because CAP_SYS_TIME and other capabilities were not included in the denylist, tenant-specified securityContext.capabilities.add could request them and still pass the ValidatePodSpecSafety / ValidateContainerSafety admission checks.
This aligns with CWE-269 (Improper Privilege Management) and CWE-732 (Incorrect Permission Assignment for Critical Resource), as the system grants more privileges than necessary to untrusted tenant workloads.
Potential impact
- Tenant functions may adjust system time inside their container, potentially impacting time-sensitive logic or logging.
- On some environments, CAP_SYS_TIME may influence node-level time settings, disrupting cluster services and time-based security controls.
- Attackers could manipulate time-based authentication tokens, session expiration, or audit trails, complicating incident response.
- Misconfigured or overly permissive nodes could experience broader integrity and reliability issues due to time skew.
General mitigation guidance
Effective mitigation focuses on tightening capability management and reducing trust in tenant-supplied PodSpecs:
- Prefer allowlists of safe Linux capabilities rather than denylists of dangerous ones.
- Enforce strict, centralized policies for container securityContext, ensuring that tenants cannot escalate capabilities beyond a minimal baseline.
- Use admission controls and policy engines (e.g., PodSecurity standards, PSP replacements, or policy-as-code) to constrain securityContext and capabilities for all tenant workloads.
- Adopt least-privilege principles for serverless runtimes and avoid granting time, network, or kernel-level capabilities unless there is a clearly justified need.
How teams detect issues like this
Detection can combine policy enforcement, configuration review, and runtime monitoring:
- Static analysis and policy-as-code tools can flag Kubernetes manifests or CRDs that request privileged capabilities like CAP_SYS_TIME.
- Code review guidelines can require inspection of securityContext and capabilities in tenant-facing controllers and admission webhooks.
- Cluster-wide policies (e.g., OPA/Gatekeeper, Kyverno) can log or deny pods that specify disallowed capabilities, providing an auditable trail.
- Runtime security tools can monitor for containers running with unexpected capabilities or performing sensitive system time operations.

