{ "id": "cks-mcq-sandboxing", "title": "Container Runtime Sandboxing", "category": "System Hardening", "difficulty": "Hard", "type": "mcq", "weight": 3, "description": "## Container Runtime Sandboxing\n\nA security team wants to run untrusted workloads with stronger isolation than the standard `runc` runtime — specifically, they want each container's system calls to be intercepted by a user-space kernel rather than reaching the host kernel directly.\n\nWhich solution achieves this, and how is it configured in Kubernetes?", "options": [ { "id": "a", "text": "Seccomp with `RuntimeDefault` — restricts which syscalls are allowed via a BPF filter, but syscalls still reach the host kernel" }, { "id": "b", "text": "AppArmor with a `deny` profile — enforces path-based MAC policies but does not intercept syscalls" }, { "id": "c", "text": "gVisor (`runsc`) via a `RuntimeClass` — intercepts all syscalls in user-space with its own kernel implementation, isolating the host kernel from the container" }, { "id": "d", "text": "A `PodDisruptionBudget` — prevents pod disruptions and provides workload isolation" } ], "correct_option": "c", "explanation": "**gVisor** (`runsc`) is a user-space kernel from Google that intercepts all container system calls before they reach the host kernel, providing strong isolation. In Kubernetes it is configured via a **RuntimeClass** resource and referenced in a pod with `spec.runtimeClassName`. Kata Containers is a similar alternative using lightweight VMs. Seccomp filters syscalls but they still hit the host kernel. AppArmor enforces file-access policies, not syscall interception.", "hints": [ { "title": "RuntimeClass and runtimeClassName", "body": "Create a `RuntimeClass` that maps a name to the container runtime handler (e.g., `handler: runsc`). Reference it in a pod with `spec.runtimeClassName: `.", "command": "kubectl explain runtimeclass\nkubectl explain pod.spec.runtimeClassName" } ], "setup_commands": [], "default_namespace": "default", "teardown_commands": [] }