Disallowing Unsafe sysctls

Use Policy Controller to limit pods to safe sysctls

This guide demonstrates how to use the Sigstore Policy Controller to only allow pods that use sysctls to modify kernel behaviour to run with the safe set of parameters. You will create a ClusterImagePolicy that uses the CUE language to examine a pod spec that uses sysctls, and only allow admission into a cluster if the pod is running a safe set parameters.


To follow along with this guide outside of the terminal that is embedded on this page, you will need the following:

If you are using the terminal that is embedded on this page, then all the prerequsites are installed for you. Note that it may take a minute or two for the Kubernetes cluster to finish provisioning. If you receive any errors while running commands, retry them after waiting a few seconds.

Once you have everything in place you can continue to the first step and confirm that the Policy Controller is working as expected.

Step 1 - Checking the Policy Controller is Denying Admission

Before creating a ClusterImagePolicy, check that the Policy Controller is deployed and that your default namespace is labeled correctly. Run the following to check that the deployment is complete:

kubectl -n cosign-system wait --for=condition=Available deployment/policy-controller-webhook && \
kubectl -n cosign-system wait --for=condition=Available deployment/policy-controller-policy-webhook

When both deployments are finished, verify the default namespace is using the Policy Controller:

kubectl get ns -l policy.sigstore.dev/include=true

You should receive output like the following:

default   Active   24s

Once you are sure that the Policy Controller is deployed and your default namespace is configured to use it, run a pod to make sure admission requests are handled and denied by default:

kubectl run --image docker.io/ubuntu ubuntu

Since there is no ClusterImagePolicy defined yet, the Policy Controller will deny the admission request with a message like the following:

Error from server (BadRequest): admission webhook "policy.sigstore.dev" denied the request: validation failed: no matching policies: spec.containers[0].image

In the next step, you will define a ClusterImagePolicy that verifies a pod spec is using safe sysctl parameters.

Step 2 — Creating a ClusterImagePolicy

Now that you have the Policy Controller running in your cluster, and have the default namespace configured to use it, you can now define a ClusterImagePolicy to admit images.

Open a new file with nano or your preferred editor:

nano /tmp/cip.yaml

Copy the following policy to the /tmp/cip.yaml file:

apiVersion: policy.sigstore.dev/v1beta1
kind: ClusterImagePolicy
  name: unsafe-sysctls-mask-cue
  - version: "v1"
    resource: "pods"
  images: [glob: '**']
  authorities: [static: {action: pass}]
  mode: enforce
    includeSpec: true
    type: "cue"
    data: |
      spec: {
          sysctls: [...{
            name: "kernel.shm_rmid_forced" |
                  "net.ipv4.ip_local_port_range" |
                  "net.ipv4.ip_unprivileged_port_start" |
                  "net.ipv4.tcp_syncookies" |

This policy will ensure that any pod that has a sysctl defined in its spec will only be admitted if it matches a parameter from the list.

Save the file and then apply the policy:

kubectl apply -f /tmp/cip.yaml

You will receive output showing the policy is created:


Next, you will test the policy with a failing pod spec. Once you have confirmed that the admission controller is rejecting pods using unsafe sysctls, you’ll create a pod with a safe parameter and admit it into your cluster.

Step 3 — Testing the ClusterImagePolicy

Now that you have a policy defined, you can test that it successfully rejects or accepts admission requests.

Use nano or your preferred editor to create a new file /tmp/pod.yaml and copy in the following pod spec that uses an unsafe sysctl:

apiVersion: v1
kind: Pod
  name: yolo
    - name: kernel.msgmax
      value: "65536"
  - name: "app"
    image: docker.io/ubuntu

Apply the pod spec and check for the Policy Controller admission denied message:

kubectl apply -f /tmp/pod.yaml
Error from server (BadRequest): error when creating "/tmp/pod.yaml": admission webhook "policy.sigstore.dev" denied the request: validation failed: failed policy: unsafe-sysctls-mask-cue: spec.containers[0].image
index.docker.io/library/ubuntu@sha256:854037bf6521e9c321c101c269272f756e481fb5f167ae032cb53da08aebcd5a failed evaluating cue policy for ClusterImagePolicy: failed to evaluate the policy with error: spec.securityContext.sysctls.0.name: 5 errors in empty disjunction: (and 5 more errors)

The first line shows the error message and the failing ClusterImagePolicy name. The second line contains the image ID, along with the specific CUE error message showing the policy violation.

Edit the /tmp/pod.yaml file and change the sysctls section to use the following safe parameter:

- name: net.ipv4.tcp_syncookies
  value: "1"
- name: net.ipv4.tcp_syncookies
  value: "1"

Save and apply the spec:

kubectl apply -f /tmp/pod.yaml

The pod will be admitted into the cluster with the following message:

pod/yolo created

Since the `net.ipv4.tcp_syncookies` sysctl is considered safe and only runs in specific Kubernetes namespaces, the Policy Controller evaluates the pod spec against the CUE policy and admits the pod into the cluster.

Delete the pod once you're done experimenting with it:

kubectl delete pod yolo

Last updated: 2024-05-10 13:11