GHSA-qf5v-m7p4-95rp (CVE-2026-50570)

Details

Fission v1.24.0 added PodSpec safety validation for tenant-facing Environment and Function CRDs (ValidatePodSpecSafety / ValidateContainerSafety admission webhook + sanitizeContainerSecurityContext executor merge layer), but the capability check was implemented as a fixed denylist of six Linux capabilities (SYS_ADMIN, NET_ADMIN, SYS_PTRACE, SYS_MODULE, DAC_READ_SEARCH, DAC_OVERRIDE). The denylist omitted CAP_SYS_TIME, among others. As a result, a tenant who could create a Function or Environment CRD could request securityContext.capabilities.add: ["SYS_TIME"], pass Fission's admission validation and merge-layer sanitization, and run attacker-controlled code with CAP_SYS_TIME in the resulting function or runtime container.

Demonstrated consequence: cross-tenant node integrity damage via CAP_SYS_TIME. The Linux real-time clock is not namespaced — time namespaces virtualize only MONOTONIC and BOOTTIME, never REALTIME — so a tenant container holding CAP_SYS_TIME could call clock_settime(CLOCK_REALTIME) and rewrite the shared node wall clock. That corrupts TLS / certificate validity windows, Kubernetes lease renewal, token expiry, scheduling, and time-series for every workload on the node.

The denylist also omitted SYS_RAWIO, BPF, SYS_RESOURCE, and MAC_ADMIN. Those are documented as evidence that the denylist is structurally incomplete (their practical impact is kernel-, LSM-, or device-cgroup-dependent and is not exercised in this report).

The deeper structural problem: a denylist on capabilities.add cannot constrain capabilities the OCI runtime grants by default — DAC_OVERRIDE is in the OCI default cap set and reaches the container regardless of any add check, partially mooting the denylist for its own entries. A capability allowlist (with drop:["ALL"] to remove the default set) is the only model that addresses both problems.