Skip to main content
Version: v2.9

K3s Hardening Guides

This document provides prescriptive guidance for how to harden a K3s cluster intended for production, before provisioning it with Rancher. It outlines the configurations and controls required for Center for Information Security (CIS) Kubernetes benchmark controls.

note

This hardening guide describes how to secure the nodes in your cluster. We recommended that you follow this guide before you install Kubernetes.

This hardening guide is intended to be used for K3s clusters and is associated with the following versions of the CIS Kubernetes Benchmark, Kubernetes, and Rancher:

Rancher VersionCIS Benchmark VersionKubernetes Version
Rancher v2.7Benchmark v1.23Kubernetes v1.23
Rancher v2.7Benchmark v1.24Kubernetes v1.24
Rancher v2.7Benchmark v1.7Kubernetes v1.25 up to v1.26
note

In Benchmark v1.7, the --protect-kernel-defaults (4.2.6) parameter isn't required anymore, and was removed by CIS.

For more details on how to evaluate a hardened K3s cluster against the official CIS benchmark, refer to the K3s self-assessment guides for specific Kubernetes and CIS benchmark versions.

K3s passes a number of the Kubernetes CIS controls without modification, as it applies several security mitigations by default. There are some notable exceptions to this that require manual intervention to fully comply with the CIS Benchmark:

  1. K3s does not modify the host operating system. Any host-level modifications need to be done manually.
  2. Certain CIS policy controls for NetworkPolicies and PodSecurityStandards (PodSecurityPolicies on v1.24 and older) restrict cluster functionality. You must opt into having K3s configure these policies. Add the appropriate options to your command-line flags or configuration file (enable admission plugins), and manually apply the appropriate policies. See further for more details.

The first section (1.1) of the CIS Benchmark primarily focuses on pod manifest permissions and ownership. Since everything in the distribution is packaged in a single binary, this section does not apply to the core components of K3s.

Host-level Requirements​

Ensure protect-kernel-defaults is set​

The protect-kernel-defaults is no longer required since CIS benchmark 1.7.

Set kernel parameters​

The following sysctl configuration is recommended for all nodes type in the cluster. Set the following parameters in /etc/sysctl.d/90-kubelet.conf:

vm.panic_on_oom=0
vm.overcommit_memory=1
kernel.panic=10
kernel.panic_on_oops=1

Run sudo sysctl -p /etc/sysctl.d/90-kubelet.conf to enable the settings.

This configuration needs to be done before setting the kubelet flag, otherwise K3s will fail to start.

Kubernetes Runtime Requirements​

The CIS Benchmark runtime requirements center around pod security (via PSP or PSA), network policies and API Server auditing logs.

By default, K3s does not include any pod security or network policies. However, K3s ships with a controller that enforces any network policies you create. By default, K3s enables both the PodSecurity and NodeRestriction admission controllers, among others.

Pod Security​

K3s v1.25 and newer support Pod Security admission (PSA) for controlling pod security.

You can specify the PSA configuration by setting the defaultPodSecurityAdmissionConfigurationTemplateName field in the cluster configuration in Rancher:

spec:
  defaultPodSecurityAdmissionConfigurationTemplateName: rancher-restricted

The rancher-restricted template is provided by Rancher to enforce the highly-restrictive Kubernetes upstream Restricted profile with best practices for pod hardening.

Network Policies​

CIS requires that all namespaces apply a network policy that reasonably limits traffic into namespaces and pods.

note

This is a manual check in the CIS Benchmark. The CIS scan flags the result as a warning, because manual inspection is necessary by the cluster operator.

The network policies can be placed in the policy.yaml file in /var/lib/rancher/k3s/server/manifests directory. If the directory was not created as part of the PSP (as described above), it must be created first.

sudo mkdir -p -m 700 /var/lib/rancher/k3s/server/manifests

Here is an example of a compliant network policy:

---
kind: NetworkPolicy
apiVersion: networking.k8s.io/v1
metadata:
  name: intra-namespace
  namespace: kube-system
spec:
  podSelector: {}
  ingress:
    - from:
      - namespaceSelector:
          matchLabels:
            name: kube-system
---
kind: NetworkPolicy
apiVersion: networking.k8s.io/v1
metadata:
  name: intra-namespace
  namespace: default
spec:
  podSelector: {}
  ingress:
    - from:
      - namespaceSelector:
          matchLabels:
            name: default
---
kind: NetworkPolicy
apiVersion: networking.k8s.io/v1
metadata:
  name: intra-namespace
  namespace: kube-public
spec:
  podSelector: {}
  ingress:
    - from:
      - namespaceSelector:
          matchLabels:
            name: kube-public

The active restrictions block DNS unless purposely allowed. Below is a network policy that allows DNS-related traffic:

---
apiVersion: networking.k8s.io/v1
kind: NetworkPolicy
metadata:
  name: default-network-dns-policy
  namespace: <NAMESPACE>
spec:
  ingress:
  - ports:
    - port: 53
      protocol: TCP
    - port: 53
      protocol: UDP
  podSelector:
    matchLabels:
      k8s-app: kube-dns
  policyTypes:
  - Ingress

The metrics-server and Traefik ingress controller are blocked by default if network policies are not created to allow access.

---
apiVersion: networking.k8s.io/v1
kind: NetworkPolicy
metadata:
  name: allow-all-metrics-server
  namespace: kube-system
spec:
  podSelector:
    matchLabels:
      k8s-app: metrics-server
  ingress:
  - {}
  policyTypes:
  - Ingress
---
apiVersion: networking.k8s.io/v1
kind: NetworkPolicy
metadata:
  name: allow-all-svclbtraefik-ingress
  namespace: kube-system
spec:
  podSelector:
    matchLabels:
      svccontroller.k3s.cattle.io/svcname: traefik
  ingress:
  - {}
  policyTypes:
  - Ingress
---
apiVersion: networking.k8s.io/v1
kind: NetworkPolicy
metadata:
  name: allow-all-traefik-v121-ingress
  namespace: kube-system
spec:
  podSelector:
    matchLabels:
      app.kubernetes.io/name: traefik
  ingress:
  - {}
  policyTypes:
  - Ingress
note

You must manage network policies as normal for any additional namespaces you create.

API Server audit configuration​

CIS requirements 1.2.22 to 1.2.25 are related to configuring audit logs for the API Server. K3s does not create by default the log directory and audit policy, as auditing requirements are specific to each user's policies and environment.

If you need a log directory, it must be created before you start K3s. We recommend a restrictive access permission to avoid leaking sensitive information.

sudo mkdir -p -m 700 /var/lib/rancher/k3s/server/logs

The following is a starter audit policy to log request metadata. This policy should be written to a file named audit.yaml in the /var/lib/rancher/k3s/server directory. Detailed information about policy configuration for the API server can be found in the official Kubernetes documentation.

---
apiVersion: audit.k8s.io/v1
kind: Policy
rules:
- level: Metadata

Further configurations are also needed to pass CIS checks. These are not configured by default in K3s, because they vary based on your environment and needs:

  • Ensure that the --audit-log-path argument is set.
  • Ensure that the --audit-log-maxage argument is set to 30 or as appropriate.
  • Ensure that the --audit-log-maxbackup argument is set to 10 or as appropriate.
  • Ensure that the --audit-log-maxsize argument is set to 100 or as appropriate.

Combined, to enable and configure audit logs, add the following lines to the K3s cluster configuration file in Rancher:

spec:
  rkeConfig:
    machineGlobalConfig:
      kube-apiserver-arg:
        - audit-policy-file=/var/lib/rancher/k3s/server/audit.yaml    # CIS 3.2.1
        - audit-log-path=/var/lib/rancher/k3s/server/logs/audit.log   # CIS 1.2.18
        - audit-log-maxage=30                                         # CIS 1.2.19
        - audit-log-maxbackup=10                                      # CIS 1.2.20
        - audit-log-maxsize=100                                       # CIS 1.2.21

Controller Manager Requirements​

CIS requirement 1.3.1 checks for garbage collection settings in the Controller Manager. Garbage collection is important to ensure sufficient resource availability and avoid degraded performance and availability. Based on your system resources and tests, choose an appropriate threshold value to activate garbage collection.

This can be remediated by setting the following configuration in the K3s cluster file in Rancher. The value below is only an example. The appropriate threshold value is specific to each user's environment.

spec:
  rkeConfig:
    machineGlobalConfig:
      kube-controller-manager-arg:
        - terminated-pod-gc-threshold=10                              # CIS 1.3.1

Configure default Service Account​

Kubernetes provides a default service account which is used by cluster workloads where no specific service account is assigned to the pod. Where access to the Kubernetes API from a pod is required, a specific service account should be created for that pod, and rights granted to that service account.

For CIS requirement 5.1.5 the default service account should be configured such that it does not provide a service account token and does not have any explicit rights assignments.

This can be remediated by updating the automountServiceAccountToken field to false for the default service account in each namespace.

For default service accounts in the built-in namespaces (kube-system, kube-public, kube-node-lease, and default), K3s does not automatically do this.

Save the following configuration to a file called account_update.yaml.

---
apiVersion: v1
kind: ServiceAccount
metadata:
  name: default
automountServiceAccountToken: false

Create a bash script file called account_update.sh. Be sure to chmod +x account_update.sh so the script has execute permissions.

#!/bin/bash -e

for namespace in $(kubectl get namespaces -A -o=jsonpath="{.items[*]['metadata.name']}"); do
  kubectl patch serviceaccount default -n ${namespace} -p "$(cat account_update.yaml)"
done

Run the script every time a new service account is added to your cluster.

Reference Hardened K3s Template Configuration​

The following reference template configuration is used in Rancher to create a hardened K3s custom cluster based on each CIS control in this guide. This reference does not include other required cluster configuration directives, which vary based on your environment.

apiVersion: provisioning.cattle.io/v1
kind: Cluster
metadata:
  name: # Define cluster name
spec:
  defaultPodSecurityAdmissionConfigurationTemplateName: rancher-restricted
  enableNetworkPolicy: true
  kubernetesVersion: # Define K3s version
  rkeConfig:
    machineGlobalConfig:
      kube-apiserver-arg:
        - enable-admission-plugins=NodeRestriction,ServiceAccount     # CIS 1.2.15, 1.2.13
        - audit-policy-file=/var/lib/rancher/k3s/server/audit.yaml    # CIS 3.2.1
        - audit-log-path=/var/lib/rancher/k3s/server/logs/audit.log   # CIS 1.2.18
        - audit-log-maxage=30                                         # CIS 1.2.19
        - audit-log-maxbackup=10                                      # CIS 1.2.20
        - audit-log-maxsize=100                                       # CIS 1.2.21
        - request-timeout=300s                                        # CIS 1.2.22
        - service-account-lookup=true                                 # CIS 1.2.24
      kube-controller-manager-arg:
        - terminated-pod-gc-threshold=10                              # CIS 1.3.1
      secrets-encryption: true
    machineSelectorConfig:
      - config:
          kubelet-arg:
            - make-iptables-util-chains=true                          # CIS 4.2.7

Conclusion​

If you have followed this guide, your K3s custom cluster provisioned by Rancher will be configured to pass the CIS Kubernetes Benchmark. You can review our K3s self-assessment guides to understand how we verified each of the benchmarks and how you can do the same on your cluster.