[](https://microbadger.com/images/aquasec/kube-bench "Get your own image badge on microbadger.com")
kube-bench is a Go application that checks whether Kubernetes is deployed securely by running the checks documented in the [CIS Kubernetes Benchmark](https://www.cisecurity.org/benchmark/kubernetes/).
Note that it is impossible to inspect the master nodes of managed clusters, e.g. GKE, EKS and AKS, using kube-bench as one does not have access to such nodes, although it is still possible to use kube-bench to check worker node configuration in these environments.
There is also preliminary support for Red Hat's Openshift Hardening Guide for 3.10 and 3.11. Please note that kube-bench does not automatically detect Openshift - see below.
You can avoid installing kube-bench on the host by running it inside a container using the host PID namespace and mounting the `/etc` and `/var` directories where the configuration and other files are located on the host, so that kube-bench can check their existence and permissions.
> Note: the tests require either the kubelet or kubectl binary in the path in order to auto-detect the Kubernetes version. You can pass `-v $(which kubectl):/usr/bin/kubectl` to the above invocations to resolve this.
You can run kube-bench inside a pod, but it will need access to the host's PID namespace in order to check the running processes, as well as access to some directories on the host where config files and other files are stored.
The detection is done by verifying that mandatory components for master, as defined in the config files, are running (see [Configuration](#configuration)).
The default labels applied to master nodes has changed since Kubernetes 1.11, so if you are using an older version you may need to modify the nodeSelector and tolerations to run the job on the master node.
There is a `job-eks.yaml` file for running the kube-bench node checks on an EKS cluster. **Note that you must update the image reference in `job-eks.yaml`.** Typically you will push the container image for kube-bench to ECR and refer to it there in the YAML file.
There are two significant differences on EKS:
* It uses [config files in JSON format](https://kubernetes.io/docs/tasks/administer-cluster/kubelet-config-file/)
* It's not possible to schedule jobs onto the master node, so master checks can't be performed
If Go is installed on the target machines, you can simply clone this repository and run as follows (assuming your [$GOPATH is set](https://github.com/golang/go/wiki/GOPATH)):
kube-bench includes a set of test files for Red Hat's OpenShift hardening guide for OCP 3.10 and 3.11. To run this you will need to specify `--version ocp-3.10` when you run the `kube-bench` command (either directly or through YAML). This config version is valid for OCP 3.10 and 3.11.
Kubernetes config and binary file locations and names can vary from installation to installation, so these are configurable in the `cfg/config.yaml` file.
For each type of node (*master*, *node* or *federated*) there is a list of components, and for each component there is a set of binaries (*bins*) and config files (*confs*) that kube-bench will look for (in the order they are listed). If your installation uses a different binary name or config file location for a Kubernetes component, you can add it to `cfg/config.yaml`.
* **bins** - If there is a *bins* list for a component, at least one of these binaries must be running. The tests will consider the parameters for the first binary in the list found to be running.
* **podspecs** - From version 1.2.0 of the benchmark (tests for Kubernetes 1.8), the remediation instructions were updated to assume that the configuration for several kubernetes components is defined in a pod YAML file, and podspec settings define where to look for that configuration.
* **confs** - If one of the listed config files is found, this will be considered for the test. Tests can continue even if no config file is found. If no file is found at any of the listed locations, and a *defaultconf* location is given for the component, the test will give remediation advice using the *defaultconf* location.
* **unitfiles** - From version 1.2.0 of the benchmark (tests for Kubernetes 1.8), the remediation instructions were updated to assume that kubelet configuration is defined in a service file, and this setting defines where to look for that configuration.
remediation: "Edit the /etc/kubernetes/config file on the master node and set the KUBE_ALLOW_PRIV parameter to '--allow-privileged=false'"
scored: true
```
Recommendations (called `checks` in this document) can run on Kubernetes Master, Node or Federated API Servers.
Checks are organized into `groups` which share similar controls (things to check for) and are grouped together in the section of the CIS Kubernetes document.
These groups are further organized under `controls` which can be of the type `master`, `node` or `federated apiserver` to reflect the various Kubernetes node types.
If you decide that a recommendation is not appropriate for your environment, you can choose to omit it by editing the test YAML file to give it the check type `skip` as in this example:
```yaml
checks:
- id: 2.1.1
text: "Ensure that the --allow-privileged argument is set to false (Scored)"
type: "skip"
scored: true
```
No tests will be run for this check and the output will be marked [INFO].
Tests are the items we actually look for to determine if a check is successful or not. Checks can have multiple tests, which must all be successful for the check to pass.
When defining regular expressions in YAML it is generally easier to wrap them in single quotes, for example `'^[abc]$'`, to avoid issues with string escaping.
Going forward we plan to release updates to kube-bench to add support for new releases of the Benchmark, which in turn we can anticipate being made for each new Kubernetes release.
Our makefile contains targets to test your current version of kube-bench inside a [Kind](https://kind.sigs.k8s.io/) cluster. This can be very handy if you don't want to run a real kubernetes cluster for development purpose.
First you'll need to create the cluster using `make kind-test-cluster` this will create a new cluster if it cannot be found on your machine. By default the cluster is named `kube-bench` but you can change the name by using the environment variable `KIND_PROFILE`.
*If kind cannot be found on your system the target will try to install it using `go get`*
Next you'll have to build the kube-bench docker image using `make build-docker`, then we will be able to push the docker image to the cluster using `make kind-push`.
Finally we can use the `make kind-run` target to run the current version of kube-bench in the cluster and follow the logs of pods created. (Ctrl+C to exit)
Everytime you want to test a change, you'll need to rebuild the docker image and push it to cluster before running it again. ( `make build-docker kind-push kind-run` )
