K01: Insecure Workload Configurations

📋 What Is It?

Insecure Workload Configurations is about the manifest, not the app code. Pods and Deployments admitted with permissive securityContext fields - privileged: true, host namespaces, hostPath mounts, or dangerous Linux capabilities - turn the container boundary into a formality an attacker steps over in seconds. Because these fields default to the permissive side for backward compatibility, a workload that never sets a securityContext is already weaker than it should be.

K01 OWASP Rank

Manifest Focus Area

Easy To Prevent

⚠️ Common Exploits

Privileged Escape: privileged: true mounts a host disk device and writes to the node
Runtime Socket Abuse: A mounted docker.sock launches a new privileged container that owns the node
hostPath Node Access: A hostPath: / mount reads kubelet creds and drops static-pod manifests
Host PID Inspection: hostPID: true reads other processes' /proc/<pid>/environ for secrets
Host Network + Metadata: hostNetwork: true reaches the kubelet and cloud IAM metadata, bypassing NetworkPolicy
Token Theft: The automounted default ServiceAccount token is a ready-made cluster credential
Capability Abuse: Added SYS_ADMIN enables the cgroup release_agent escape without full privilege

🔴 Attack Flow

1. Foothold: code execution in a container
↓
2. Enumerate the sandbox (id, capsh, mounts, token)
↓
3. Escape via privileged / hostPath / docker.sock
↓
4. Loot node: kubelet creds, other Pods' secrets, metadata
↓
5. IMPACT: Stolen token drives cluster-wide takeover!

❌ Vulnerable Code

# Bad: every escape primitive in one manifest
apiVersion: v1
kind: Pod
metadata:
  name: legacy-agent
spec:
  hostPID: true              # sees all host processes
  hostIPC: true              # host shared memory
  hostNetwork: true          # host NICs + metadata reach
  containers:
    - name: agent
      image: agent:latest       # mutable tag, unknown provenance
      securityContext:
        privileged: true               # ~= root on the node
        runAsUser: 0                    # root in container
        allowPrivilegeEscalation: true
        readOnlyRootFilesystem: false
        capabilities:
          add: ["SYS_ADMIN", "NET_ADMIN"]  # escape primitives
        seccompProfile:
          type: Unconfined              # full kernel syscall surface
      volumeMounts:
        - name: host-root
          mountPath: /host                # entire node filesystem
        - name: docker-sock
          mountPath: /var/run/docker.sock   # runtime takeover
      # no resources.limits -> noisy-neighbour DoS
  volumes:
    - name: host-root
      hostPath: { path: / }
    - name: docker-sock
      hostPath: { path: /var/run/docker.sock }
  # default ServiceAccount token automounted -> cluster credential
                

✅ Secure Code

# Good: least-privilege, hardened securityContext
apiVersion: v1
kind: Pod
metadata:
  name: agent
spec:
  hostPID: false
  hostIPC: false
  hostNetwork: false
  automountServiceAccountToken: false   # app doesn't call the API
  securityContext:                       # Pod-level
    runAsNonRoot: true
    runAsUser: 10001
    runAsGroup: 10001
    fsGroup: 10001
    seccompProfile:
      type: RuntimeDefault              # syscall filtering on
  containers:
    - name: agent
      image: registry.example.com/agent@sha256:<digest>  # pinned by digest
      securityContext:                   # container-level wins
        privileged: false
        allowPrivilegeEscalation: false
        readOnlyRootFilesystem: true
        runAsNonRoot: true
        runAsUser: 10001
        capabilities:
          drop: ["ALL"]                  # add back nothing
      resources:
        requests: { cpu: "100m", memory: "128Mi" }
        limits:   { cpu: "500m", memory: "256Mi", ephemeral-storage: "1Gi" }
      volumeMounts:
        - name: tmp
          mountPath: /tmp                  # writable scratch, not the host
  volumes:
    - name: tmp
      emptyDir: {}
  # no hostPath, no docker.sock, no host namespaces
                

✓ Prevention Checklist

Set runAsNonRoot: true with an explicit non-zero runAsUser
Set allowPrivilegeEscalation: false and privileged: false
Drop ALL capabilities, add back only the one strictly needed
Set readOnlyRootFilesystem: true; mount writable dirs explicitly
Use seccompProfile: RuntimeDefault, never Unconfined
Forbid hostPID, hostIPC, hostNetwork and all hostPath mounts
Never mount the container runtime socket; use a rootless builder (Kaniko)
Set automountServiceAccountToken: false where the API isn't used
Set CPU, memory, and ephemeral-storage limits on every container
Enforce Pod Security Admission (restricted) plus Kyverno/Gatekeeper
Scan manifests in CI (Trivy, kubescape, kube-linter) before they reach the cluster

🔍 Detection & Tools

Scanning & Policy:

kube-bench kubeaudit Trivy kubescape Polaris kube-linter Checkov

Admission & Runtime:

Pod Security Admission Kyverno OPA/Gatekeeper Falco

How to Test:

Find privileged Pods: kubectl get pods -A -o json | jq ...privileged==true
List Pods using hostPID/hostIPC/hostNetwork
Flag hostPath volumes, especially / and docker.sock
Dry-run restricted PSA: kubectl label --dry-run=server ns ...
Run trivy config ./k8s and kubescape scan framework nsa ./k8s

🌍 Real-World

Privileged-container escape to the node: A build/CI/monitoring agent granted privileged or SYS_ADMIN "for convenience" becomes a node compromise once an app flaw gives code execution inside it.
Exposed runtime socket (docker.sock): The docker-in-docker build pattern mounts the runtime socket; any code in the Pod launches a privileged host-mounting container - node takeover with no kernel exploit.
Tesla cryptojacking (2018, widely reported): An administrative Kubernetes dashboard was reachable without authentication; attackers scheduled their own cryptomining Pods and reached cloud resources - a management surface plus permissive workloads.
Note: Treat each as a class of failure. Privileged escape, runtime-socket exposure, and cryptojacking of open/permissive workloads are repeatedly observed; the takeaway is the mechanism, not an exact statistic.

📌 Quick Tips

DO NOT set privileged: true
DO NOT mount hostPath: / or the runtime socket
DO NOT enable host namespaces
DO run as non-root and drop ALL capabilities
DO disable token automount when unused
DO enforce restricted Pod Security Admission
DO scan manifests in CI

📜 Compliance

Related Standards:

CIS Kubernetes Benchmark - Pod Security
NSA/CISA Kubernetes Hardening Guidance
NIST 800-190 Application Container Security
PCI-DSS Requirement 2.2 - Secure Configuration
Pod Security Standards restricted profile

📦 Insecure Workload Configurations