This article is more than one year old. Older articles may contain outdated content. Check that the information in the page has not become incorrect since its publication.

Kubernetes Validating Admission Policies: A Practical Example

Admission control is an important part of the Kubernetes control plane, with several internal features depending on the ability to approve or change an API object as it is submitted to the server. It is also useful for an administrator to be able to define business logic, or policies, regarding what objects can be admitted into a cluster. To better support that use case, Kubernetes introduced external admission control in v1.7.

In addition to countless custom, internal implementations, many open source projects and commercial solutions implement admission controllers with user-specified policy, including Kyverno and Open Policy Agent’s Gatekeeper.

While admission controllers for policy have seen adoption, there are blockers for their widespread use. Webhook infrastructure must be maintained as a production service, with all that entails. The failure case of an admission control webhook must either be closed, reducing the availability of the cluster; or open, negating the use of the feature for policy enforcement. The network hop and evaluation time makes admission control a notable component of latency when dealing with, for example, pods being spun up to respond to a network request in a "serverless" environment.

Validating admission policies and the Common Expression Language

Version 1.26 of Kubernetes introduced, in alpha, a compromise solution. Validating admission policies are a declarative, in-process alternative to admission webhooks. They use the Common Expression Language (CEL) to declare validation rules.

CEL was developed by Google for security and policy use cases, based on learnings from the Firebase real-time database. Its design allows it to be safely embedded into applications and executed in microseconds, with limited compute and memory impact. Validation rules for CRDs introduced CEL to the Kubernetes ecosystem in v1.23, and at the time it was noted that the language would suit a more generic implementation of validation by admission control.

Giving CEL a roll - a practical example

Kubescape is a CNCF project which has become one of the most popular ways for users to improve the security posture of a Kubernetes cluster and validate its compliance. Its controls — groups of tests against API objects — are built in Rego, the policy language of Open Policy Agent.

Rego has a reputation for complexity, based largely on the fact that it is a declarative query language (like SQL). It was considered for use in Kubernetes, but it does not offer the same sandbox constraints as CEL.

A common feature request for the project is to be able to implement policies based on Kubescape’s findings and output. For example, after scanning pods for known paths to cloud credential files, users would like the ability to enforce policy that these pods should not be admitted at all. The Kubescape team thought this would be the perfect opportunity to try and port our existing controls to CEL and apply them as admission policies.

Show me the policy

It did not take us long to convert many of our controls and build a library of validating admission policies. Let’s look at one as an example.

Kubescape’s control C-0017 covers the requirement for containers to have an immutable (read-only) root filesystem. This is a best practice according to the NSA Kubernetes hardening guidelines, but is not currently required as a part of any of the pod security standards.

Here's how we implemented it in CEL:

kind: ValidatingAdmissionPolicy
  name: "kubescape-c-0017-deny-resources-with-mutable-container-filesystem"
  failurePolicy: Fail
    - apiGroups:   [""]
      apiVersions: ["v1"]
      operations:  ["CREATE", "UPDATE"]
      resources:   ["pods"]
    - apiGroups:   ["apps"]
      apiVersions: ["v1"]
      operations:  ["CREATE", "UPDATE"]
      resources:   ["deployments","replicasets","daemonsets","statefulsets"]
    - apiGroups:   ["batch"]
      apiVersions: ["v1"]
      operations:  ["CREATE", "UPDATE"]
      resources:   ["jobs","cronjobs"]
    - expression: "object.kind != 'Pod' || object.spec.containers.all(container, has(container.securityContext) && has(container.securityContext.readOnlyRootFilesystem) &&  container.securityContext.readOnlyRootFilesystem == true)"
      message: "Pods having containers with mutable filesystem not allowed! (see more at"
    - expression: "['Deployment','ReplicaSet','DaemonSet','StatefulSet','Job'].all(kind, object.kind != kind) || object.spec.template.spec.containers.all(container, has(container.securityContext) && has(container.securityContext.readOnlyRootFilesystem) &&  container.securityContext.readOnlyRootFilesystem == true)"
      message: "Workloads having containers with mutable filesystem not allowed! (see more at"
    - expression: "object.kind != 'CronJob' || object.spec.jobTemplate.spec.template.spec.containers.all(container, has(container.securityContext) && has(container.securityContext.readOnlyRootFilesystem) &&  container.securityContext.readOnlyRootFilesystem == true)"
      message: "CronJob having containers with mutable filesystem not allowed! (see more at"

Match constraints are provided for three possible API groups: the core/v1 group for Pods, the apps/v1 workload controllers, and the batch/v1 job controllers.

The validations include the CEL rules for the objects. There are three different expressions, catering for the fact that a Pod spec can be at the root of the object (a naked pod), under template (a workload controller or a Job), or under jobTemplate (a CronJob).

In the event that any spec does not have readOnlyRootFilesystem set to true, the object will not be admitted.

Using the CEL library in your cluster

Policies are provided as Kubernetes objects, which are then bound to certain resources by a selector.

Minikube is a quick and easy way to install and configure a Kubernetes cluster for testing. To install Kubernetes v1.26 with the ValidatingAdmissionPolicy feature gate enabled:

minikube start --kubernetes-version=1.26.1  --feature-gates='ValidatingAdmissionPolicy=true'

To install the policies in your cluster:

# Install configuration CRD
kubectl apply -f
# Install basic configuration
kubectl apply -f
# Install policies
kubectl apply -f

To apply policies to objects, create a ValidatingAdmissionPolicyBinding resource. Let’s apply the above Kubescape C-0017 control to any namespace with the label policy=enforced:

# Create a binding
kubectl apply -f - <<EOT
kind: ValidatingAdmissionPolicyBinding
  name: c0017-binding
  policyName: kubescape-c-0017-deny-mutable-container-filesystem
        policy: enforced

# Create a namespace for running the example
kubectl create namespace policy-example
kubectl label namespace policy-example 'policy=enforced'

Now, if you attempt to create an object without specifying a readOnlyRootFilesystem, it will not be created.

# The next line should fail
kubectl -n policy-example run nginx --image=nginx --restart=Never

The output shows our error:

The pods "nginx" is invalid: : ValidatingAdmissionPolicy 'kubescape-c-0017-deny-mutable-container-filesystem' with binding 'c0017-binding' denied request: Pods having containers with mutable filesystem not allowed! (see more at


Policy objects can include configuration, which is provided in a different object. Many of the Kubescape controls require a configuration: which labels to require, which capabilities to allow or deny, which registries to allow containers to be deployed from, etc. Default values for those controls are defined in the ControlConfiguration object.

To use this configuration object, or your own object in the same format, add a value to your binding object:

kind: ValidatingAdmissionPolicyBinding
  name: c0001-binding
  policyName: kubescape-c-0001-deny-forbidden-container-registries
    name: basic-control-configuration
        policy: enforced


Converting our controls to CEL was simple, in most cases. We cannot port the whole Kubescape library, as some controls check for things outside a Kubernetes cluster, and some require data that is not available in the admission request object. Overall, we are happy to contribute this library to the Kubernetes community and will continue to develop it for Kubescape and Kubernetes users alike. We hope it becomes useful, either as something you use yourself, or as examples for you to write your own policies.

As for the validating admission policy feature itself, we are very excited to see this native functionality introduced to Kubernetes. We look forward to watching it move to Beta and then GA, hopefully by the end of the year. It is important to note this feature is currently in Alpha, which means this is the perfect opportunity to play around with it in environments like Minikube and give a test drive. However, it is not yet considered production-ready and stable, and will not be enabled on most managed Kubernetes environments. We will not recommend Kubescape users use these policies in production until the underlying functionality becomes stable. Keep an eye on the KEP, and of course this blog, for an eventual release announcement.