A new release of Runtime Enforcer just landed: v0.11.0 🥳
This release includes several fixes and a new CLI utility. It also has breaking changes in the Helm chart and in the telemetry export. Before you upgrade, read the upgrade notes.
wpgen: draft and check policies without a cluster
Before this release, you got a WorkloadPolicy from the learn phase in a cluster,
or you wrote it by hand. This release adds wpgen, a new standalone command line tool.
wpgen reads the executables in container images and writes a draft of a WorkloadPolicy.
It does not need a cluster, so you can run it on a workstation or in a CI
pipeline.
Draft a policy without running the container
wpgen drafts a WorkloadPolicy from one image, or from a workload file such
as the output of helm template:
wpgen image docker.io/library/nginx:1.27
helm template my-release ./chart | wpgen workload -
The draft starts in monitor mode. It allows every program and script of the
image, so you must trim it and test it before you use protect mode.
Find images with too many executables
wpgen reports how many executables each image contains. If a container
allows more than 50 executables, wpgen prints a warning. For example,
nginx:1.27 contains 504 executables, and a container usually runs only a few
of them.
An image with hundreds of executables has an attack surface that is too large.
wpgen raises awareness of this problem. Its report shows the number of
executables for each image. A CI pipeline can use this number to find these
images, block them, and notify their authors.
Find out if a policy still matches a new release
You have a WorkloadPolicy for version 1.0.0 of your application. Then
version 1.1.0 comes out, and you want to know if the policy is still valid.
wpgen check answers this question before you deploy the new version:
wpgen check policy.yaml --image ghcr.io/example/my-app:1.1.0 --list-not-allowed
wpgen check compares the paths that the policy allows with the executables
of the new image. The policy can come from the learn phase or from wpgen.
The report shows two kinds of difference:
- Stale paths: the policy allows them, but the new image does not contain them. The new release removed or renamed these executables.
- Executables of the new image that the policy does not allow. A new executable that the application might need is in this list.
If the report has a stale path, wpgen check exits with code 1. As a result,
a CI job can run it at each image update.
Get wpgen
The wpgen binaries are part of the
GitHub release.
They are available for Linux and macOS, on amd64 and arm64. The
CLI reference
lists all the commands and flags.
Helm chart: new layers for the values
The values of the chart have three layers: global, common, and the block
of each workload (controller, agent, debugger).
The chart merges maps, and it adds the lists of named items together. If two layers set the same key, the block of the workload wins.
The chart has these new keys:
global.imageRegistryextraEnvandextraEnvFromextraVolumesandextraVolumeMountstopologySpreadConstraintspriorityClassName- Labels and annotations for each workload
These values moved or changed:
| v0.10 | v0.11 |
|---|---|
Top-level imagePullSecrets | global.imagePullSecrets |
controller.pdb | controller.podDisruptionBudget |
env in a workload block | extraEnv |
controller.args and agent.args | Removed |
image.repository with the registry | image.registry (for example ghcr.io) and image.repository (for example kubewarden/runtime-enforcer/agent) |
The debugger also has a podDisruptionBudget block.
The chart no longer passes --metrics-bind-address to the controller, so the
controller no longer serves a metrics endpoint on port 8443. The chart also no
longer creates the metrics-auth and metrics-reader ClusterRoles.
The documentation explains the layers of the chart values.
Telemetry: use your own OpenTelemetry collector
Before, the chart deployed an OpenTelemetry collector by default. This release
removes the bundled collector and the telemetry block of the values. You
deploy the collector, and you give its URL to the chart:
global:
otel:
endpoint: https://otel-collector.observability.svc.cluster.local:4317
The export is off until you set global.otel.endpoint. Runtime Enforcer still
shows violations without telemetry, in the status of each WorkloadPolicy.
The rules for the endpoint are stricter:
- The endpoint must be a URL that starts with
http://orhttps://. With a value such ascollector:4317, the controller and the agent do not start. - An
https://URL selects TLS. Without a CA Secret, the controller and the agent trust only public certificate authorities (CAs). In v0.10, a missing CA Secret turned off TLS verification. global.otel.caSecretNamenames the Secret of a private CA.global.otel.clientCertificateSecretNamenames the Secret of a client certificate, for mutual TLS.- The controller and the agent no longer have the
--otlp-*flags. They read the standardOTEL_EXPORTER_OTLP_*environment variables through the OpenTelemetry SDK.
The telemetry page has the full procedure.
Provenance and signatures for the release artifacts
This release changes how we build and sign the release artifacts:
- Container images: the shared Kubewarden reusable workflow builds the images,
with SLSA Build L3 provenance (a signed record of how a build ran). The
identity of the signer changed, so the
cosigncommands for the images are different. kubectl-runtime_enforcerandwpgenbinaries: each binary has SLSA Build L3 provenance, an SPDX SBOM (a list of the components of the binary), and a keylesscosignsignature.- Base images: the controller and the agent moved from
scratchto distroless images. The controller uses thenonrootvariant.
The Verifying Runtime Enforcer tutorial has the commands for the images, the binaries, and the Helm chart.
Other improvements and bug fixes
- Runtime Enforcer no longer reports, blocks, or learns the executions that an
OCI runtime does at the start of a container, such as
runc init. This removes false positives. - The delete webhook ignores Pods in the
SucceededorFailedphase. As a result, the Pods of finished Jobs and CronJobs no longer block the deletion of aWorkloadPolicy. - The
podLabelsvalue can no longer replace the labels that the chart owns. Before, this broke the selectors. - The
values.schema.jsonfile of the chart references the upstream Kubernetes schemas. It accepts valid fields such ascapabilities.addandresources.limits.ephemeral-storage. - The CRD reference shows the optional fields correctly.
- The Rancher version annotation of the chart is 2.16.
- The project uses Go 1.27.4, with updates of the Kubernetes libraries, the OpenTelemetry SDK, gRPC, and NRI.
- The gRPC code generation uses
buf. - We accepted RFC 0009, about a configurable behavior for a missing
WorkloadPolicy. This release does not contain the implementation.
Upgrade notes
Runtime Enforcer supports the upgrade from 0.10.x to 0.11.0. It does not
support an upgrade across more than one minor version.
CAUTION: do not use helm upgrade --reuse-values. With this option, Helm
keeps the old image tags, and the Pods run the v0.10 images.
Before you upgrade, update your Helm values file. The chart ignores a value that it does not know, and it gives no error.
If you want to keep the telemetry export, deploy your own OpenTelemetry collector. The upgrade removes the bundled collector.
Set
global.otel.endpointto the URL of your collector.If a workload in
protectmode runs files through/proc/<pid>/root, setALLOW_FOREIGN_ROOT_EXECin its policy. Without this option, Runtime Enforcer blocks these executions.Give your updated values file to Helm with the
-foption:helm upgrade runtime-enforcer kubewarden/runtime-enforcer \ --namespace runtime-enforcer \ -f values.yaml \ --wait
The upgrade path page lists every change between v0.10 and v0.11.
Getting in touch
To ask a question or to report a problem, join us on Slack or open a thread in GitHub discussions.