Application Security Checklist

This checklist aims on providing basic guidelines on securing applications running in Kubernetes from a developer perspective. This list is not meant to be exhaustive and is meant to evolve over time.

On how to read and use this document:

• The order of topics does not reflect an order of priority.
• Some checklist items are detailed in the paragraph below the list of each section.
• This checklist assumes a developer is a Kubernetes cluster user that interacts with namespaced scope objects.

Base security hardening

The following checklist provide a base security hardening recommendations that would apply to most applications deploying to Kubernetes.

Application design

• Following the right security principles when designing applicaitons.
• Application configured with appropriate QoS class through resource request and limits
  • Memory limit is set for the workloads with a limit equal or inferior to the request.
  • CPU limit might be set on sensitive workloads.

Service account

• Avoid using the default ServiceAccount. Create ServiceAccounts for each workload or microservice.
• automountServiceAccountToken should be set to false unless the pod specifically requires access to the Kubernetes API to operate.

Pod-level securityContext recommendations

• Set runAsNonRoot: true.
• Configure the container to execute as a less privileged user (for example, using runAsUser and runAsGroup), and configure appropriate permissions on files or directories inside the container image.
• Optionally add a supplementary group with fsGroup to access persistent volumes.
• The application deploys into a namespace that enforces an appropriate Pod security standard. If you cannot control this enforcement for the cluster(s) where the application is deployed, take this into account either through documentation or additional defense in depth.

Container-level securityContext recommendations

• Disable privilege escalations using allowPrivilegeEscalation: false.
• Configure the root filesystem to be read-only with readOnlyRootFilesystem: true.
• Avoid running privileged containers (set privileged: false).
• Drop all capabilities from the containers, and add back only specific ones that are needed for operation of the container.

Role Based Access Control (RBAC)

• Permissions such as create, patch, update and delete should be only granted if necessary.
• Avoid creating RBAC permissions to create or update roles which can lead to privilege escalation.
• Review bindings for the system:unauthenticated group and remove them where possible, as this gives access to anyone who can contact the API server at a network level.

The create, update and delete verbs should be permitted judiciously. The patch verb if allowed on a Namespace can allow users to update labels on the namespace or deployments which can increase the attack surface.

For sensitive workloads, consider providing a recommended ValidatingAdmissionPolicy that further restricts the permitted write actions.

Image security

• Using a image scanning tool the scan an image before deploying containers in the Kubernetes cluster.
• Use container signing to validate the container image signature before deploying to the Kubernetes cluster.

Network Policies

• Configure NetworkPolicies to only allow expected ingress and egress traffic from the pods.

Make sure that your cluster provides and enforces NetworkPolicy. If you are writing an application that people will deploy to different clusters, consider whether you can assume that NetworkPolicy is available and enforced.

Advanced security hardening

This section of this guide covers some advanced security hardening points which might be valuable based on different Kubernetes environment setup

Linux container security

Configure Security Context for the pod-container.

• Set appropriate Seccomp profiles.
• Configure appropriate AppArmor policies.
• Assign appropriate SELinux labels to the contianers.

Runtime Classes

• Configure appropriate runtime classes for containers.

Some containers may require a different isolation level from what is provided by the default runtime of the cluster. runtimeClassname can be used in a podspec to define a differnt runtime class.

For sensitive workloads consider using kernel emulation tools like gVisor, or virtualized isolation using a mechanism such as kata-containers.

In high trust environments, consider using confidential virtual machines to improve cluster security even further.