Skip to content
Hands-on IT IT Drills

Day 9: Open Services Without Authentication

Goal: You read a cache without a password – from inside and from outside – and lock it down three ways. · Duration: approx. 70 min · Prerequisite: Day 8

Briefing

“It’s only internal” is not security. A database or cache service without a password is open to every Pod in the cluster – and if it is published as a NodePort or LoadBalancer, it is also open to everyone on the network in front of it. Today you deploy a Redis without auth as a NodePort (one of the most common misconfigurations of all), grab its data, and lock it down with three measures: a password, no NodePort, Default Deny.

Drill 1: The misconfiguration – cache without a password, open to the outside

kubectl create namespace exposed
kubectl -n exposed create deployment cache --image=redis:7-alpine
kubectl -n exposed rollout status deploy/cache --timeout=120s
kubectl -n exposed expose deployment cache --type=NodePort --port=6379
kubectl -n exposed get svc cache

Expected (NodePort number varies):

NAME    TYPE       CLUSTER-IP    EXTERNAL-IP   PORT(S)          AGE
cache   NodePort   10.43.50.41   <none>        6379:31196/TCP   3s

Save your progress? Sign up for free

Drill 2: Exploit – from inside

The attacker Pod has no redis-cli, but the Redis protocol is text over TCP – bash and /dev/tcp are enough:

kubectl -n attacker exec shell -- bash -c '
exec 3<>/dev/tcp/cache.exposed.svc/6379
printf "PING\r\nSET loot pwned\r\nGET loot\r\n" >&3
timeout 2 cat <&3'

Expected:

+PONG
+OK
$5
pwned

You have read and written – without any login at all. Every Pod in the cluster can tamper with this data.

Save your progress? Sign up for free

Drill 3: Exploit – from outside

The NodePort makes the cache reachable on a port of every node, even from the network in front of the cluster. From the host (with the NodePort number from Drill 1):

NP=31196
exec 3<>/dev/tcp/127.0.0.1/$NP
printf "PING\r\nGET loot\r\n" >&3
timeout 2 cat <&3

Expected:

+PONG
$5
pwned

You now read the data written inside the cluster in Drill 2 from outside. A NodePort without auth is a wide-open barn door.

Save your progress? Sign up for free

Drill 4: Fix – password, ClusterIP, Default Deny

First, enforce a password:

kubectl -n exposed patch deploy cache --type=json \
  -p='[{"op":"add","path":"/spec/template/spec/containers/0/args",
        "value":["--requirepass","lab-redis-pw"]}]'
kubectl -n exposed rollout status deploy/cache --timeout=90s

Then take back the publication – internally, ClusterIP is enough:

kubectl -n exposed patch svc cache -p '{"spec":{"type":"ClusterIP"}}'

And as defense in depth, Default Deny (like Day 8):

kubectl -n exposed apply -f - <<'EOF'
apiVersion: networking.k8s.io/v1
kind: NetworkPolicy
metadata: {name: default-deny-ingress, namespace: exposed}
spec: {podSelector: {}, policyTypes: ["Ingress"]}
EOF

Save your progress? Sign up for free

Drill 5: Proof

Without a password, Redis now refuses (network briefly left open for testing):

kubectl -n attacker exec shell -- bash -c '
exec 3<>/dev/tcp/cache.exposed.svc/6379
printf "GET loot\r\n" >&3
timeout 2 cat <&3'

Expected:

-NOAUTH Authentication required.

The old NodePort is no longer reachable from the host:

(exec 3<>/dev/tcp/127.0.0.1/31196) || echo "Port geschlossen"

Expected:

bash: connect: Connection refused
Port geschlossen

And with Default Deny, the attacker never even reaches the port:

kubectl -n attacker exec shell -- nc -zv -w4 cache.exposed.svc 6379; echo "rc=$?"

Expected:

command terminated with exit code 1

Clean up:

kubectl delete namespace exposed

Save your progress? Sign up for free

Checkpoint

Each of the three layers works. You find services exposed as NodePort cluster-wide like this:

kubectl get svc -A --field-selector spec.type=NodePort

Expected: ideally none – every hit needs to be checked (does it really need a NodePort, and is the service behind it authenticated?).

Challenge

NodePorts often appear unnoticed because an expose or a Helm value sets them. Write a one-liner that lists all NodePort and LoadBalancer Services in the cluster with namespace and port – your ongoing inventory.

Hint: kubectl get svc -A -o json plus a filter on .spec.type.

Solution

kubectl get svc -A -o json | python3 -c '
import sys,json
for s in json.load(sys.stdin)["items"]:
    t=s["spec"].get("type")
    if t in ("NodePort","LoadBalancer"):
        ports=[p.get("nodePort") or p.get("port") for p in s["spec"]["ports"]]
        print(t, s["metadata"]["namespace"]+"/"+s["metadata"]["name"], ports)'

Expected: In a fresh RKE2, only the intended ones (e.g. the Ingress controller). Everything else is a finding.

Save your progress? Sign up for free

Enterprise Note

  • Internal services as ClusterIP; external access only via an Ingress/Gateway with authentication and TLS, not via NodePort.
  • Every data service (DB, cache, message broker) needs authentication – even “internal only” ones. Forbid default passwords and “no password” via admission (Day 12).
  • Default Deny as the third layer: even an unauthenticated service is then only reachable by the explicitly allowed sources.
  • Never publish admin UIs (Kubernetes Dashboard, database UIs) without auth or via NodePort; use a dedicated, tightly bound token for the Kubernetes Dashboard.