|
This is unreleased documentation for Admission Controller 1.36-dev. |
Security hardening
SUSE Security Admission Controller strives to be secure with little configuration. In this section and its subpages, you can find hardening tips (with their trade-offs) to secure Admission Controller itself.
Please refer to the threat model for more information.
kubewarden-defaults Helm chart
Operators can obtain a secure deployment by installing all the Admission Controller Helm
charts. It’s recommended to install the kubewarden-defaults Helm chart and
enable its recommended policies with:
helm install --wait -n kubewarden kubewarden-defaults kubewarden/kubewarden-defaults \
--set recommendedPolicies.enabled=True \
--set recommendedPolicies.defaultPolicyMode=protect
This provides a default PolicyServer and default policies, in protect mode, to ensure the Admission Controller stack is safe from other workloads.
Verifying Admission Controller artifacts
Refer to the Verifying Admission Controller tutorial.
RBAC
Admission Controller describes RBAC configurations in different Explanations sections. Users can fine-tune the needed permissions for the Audit Scanner feature, as well as per Policy Server Service Account for the context-aware feature.
To view all Roles:
kubectl get clusterroles,roles -A | grep kubewarden
Per-policy permissions
For context-aware policies, operators specify fine-grained permissions per
policy under its spec.contextAwareResources, and those work in conjunction
with the Service Account configured for the Policy Server where the policy
runs.
Workload coverage
By default, Admission Controller excludes specific Namespaces from Admission Controller coverage. This is done to simplify first-time use and interoperability with other workloads.
Security-conscious operators can tune these Namespaces list via the
.global.skipNamespaces value for both the kubewarden-controller and
kubewarden-defaults Helm charts.
Pod Security Admission
From 1.23, Admission Controller’s stack is able to run in a Namespace where the
restricted`
Pod Security Standards are in place, with current Pod hardening best
practices.
To do that, you need to add the pod-security.kubernetes.io/enforce:
restricted label to the Admission Controller deployment Namespace.
kubectl label namespace kubewarden pod-security.kubernetes.io/enforce=restricted --overwrite
See the official documentation of Kubernetes' Pod Security Admission for more details.
SecurityContexts
The kubewarden-controller Helm chart configures the SecurityContexts and
exposes them in its values.yaml.
The kubewarden-defaults Helm chart allows for configuring the default Policy
Server .spec.securityContexts under .Values.policyServer.securityContexts.
For Policy Servers managed by operators, you can configure them via their
spec.securityContexts.
Namespaced policy host capabilities
Namespaced policies (AdmissionPolicy and AdmissionPolicyGroup) cannot
access Kubernetes resources via contextAwareResources, but they can exercise
other host capabilities: OCI registry queries, Kubernetes can_i checks, DNS
lookups, and certificate verification. These capabilities could be exploited by
a low-privileged user for information disclosure or reconnaissance (see our
threat model).
Each PolicyServer exposes a spec.namespacedPoliciesCapabilities field that
lists which host capability API call paths are available to its namespaced
policies. By default, all PolicyServers allow all host capability calls for
namespaced policies. This includes the PolicyServer default (installed with
the kubewarden-defaults Helm chart), and custom PolicyServers.
Cluster operators need to configure each PolicyServer’s
spec.namespacedPoliciesCapabilities by setting it to [] to deny all, or
provide an explicit allowlist.
For a full walkthrough, including an example on secure self-service Namespaces and policies useful for teams, see Controlling host capabilities for namespaced policies.