A webhook token authentication plugin implementation backed by LDAP.
k8s-ldap-auth is released as a binary containing both client and server.
The server part provides two routes:
/auth
for the actual authentication from the CLI tool/token
for the token validation from the kube-apiserver.
The user created from the TokenReview will contain both uid and groups from the LDAP user so you can use both for role binding.
The same k8s-ldap-auth server can be used to authenticate with multiple kubernetes cluster since the ExecCredential it provides contains a signed token that will eventually be used by a kube-apiserver in a TokenReview that will be sent back.
I actually use this setup on quite a few clusters with a growing userbase.
Access rights to clusters and resources will not be implemented in this authentication hook, kubernetes RBAC will do that for you.
KUBERNETES_EXEC_INFO
is currently disregarded but might be used in future versions.
You can see the commands and their options with:
k8s-ldap-auth --help
# or
k8s-ldap-auth [command] --help
Pretty much all options can be set using environment variables and a few also read their values from files.
Create the password file for the bind-dn:
echo -n "bind_P@ssw0rd" > /etc/k8s-ldap-auth/ldap/password
The server can then be started with:
k8s-ldap-auth serve \
--ldap-host="ldaps://ldap.company.local" \
--bind-dn="uid=k8s-ldap-auth,ou=services,ou=company,ou=local" \
--search-base="ou=people,ou=company,ou=local"
Note that if the server do not know of any key pair it will create one at launch but will not persist it. If you want your jwt tokens to be valid across server instances, after restarts or behind a load-balancer, you should provide a key pair.
Key pair can be created with openssl:
openssl genrsa -out key.pem 4096
openssl rsa -in key.pem -outform PEM -pubout -out public.pem
Then, the server can be started with:
k8s-ldap-auth serve \
--ldap-host="ldaps://ldap.company.local" \
--bind-dn="uid=k8s-ldap-auth,ou=services,ou=company,ou=local" \
--search-base="ou=people,ou=company,ou=local" \
--private-key-file="path/to/key.pem"
--public-key-file="path/to/public.pem"
Now for the cluster configuration.
In the following example, I use the api version client.authentication.k8s.io/v1beta1
. Feel free to put another better suited for your need.
The following authentication token webhook config file will have to exist on every control-plane. In the following configuration it's located at /etc/kubernetes/webhook-auth-config.yml
:
---
apiVersion: v1
kind: Config
clusters:
- name: authentication-server
cluster:
server: https://<server address>/token
users:
- name: kube-apiserver
contexts:
- context:
cluster: authentication-server
user: kube-apiserver
name: kube-apiserver@authentication-server
current-context: kube-apiserver@authentication-server
If you're creating a new cluster with kubeadm, you can add the following to your init configuration file:
---
apiVersion: kubeadm.k8s.io/v1beta2
kind: ClusterConfiguration
apiServer:
extraArgs:
authentication-token-webhook-config-file: "/etc/ldap-auth-webhook/config.yml"
authentication-token-webhook-version: client.authentication.k8s.io/v1beta1
extraVolumes:
- name: "webhook-config"
hostPath: "/etc/kubernetes/webhook-auth-config.yml"
mountPath: "/etc/ldap-auth-webhook/config.yml"
readOnly: true
pathType: File
If the cluster was created with kubeadm, edit the kubeadm configuration stored in the namespace kube-system
to add the configuration from above: kubectl --namespace kube-system edit configmaps kubeadm-config
Editing this configuration does not actually update your api-server. It will however be used if you need to add a new control-plane with kubeadm join
.
On every control plane, edit the manifest found at /etc/kubernetes/manifests/kube-apiserver.yaml
:
spec:
containers:
- name: kube-apiserver
command:
- kube-apiserver
# ...
- --authentication-token-webhook-config-file=/etc/ldap-auth-webhook/config.yml
- --authentication-token-webhook-version=v1beta1
# ...
volumeMounts:
- mountPath: /etc/ldap-auth-webhook/config.yml
name: webhook-config
readOnly: true
# ...
volumes:
- hostPath:
path: /etc/kubernetes/webhook-auth-config.yml
type: File
name: webhook-config
Even though it's not specified anywhere, the --password
option and the equivalent $PASSWORD
environment variable as well as the configfile containing a password were added for convenience’s sake, e.g. when running in an automated fashion, etc. If not provided, it will be asked at runtime and, if available, saved into the client OS credential manager. The same can be said for the --user
options and $USER
environment variables.
Authentication can be achieved with the following command you can execute to test your installation:
k8s-ldap-auth auth --endpoint="https://<server address>/auth"
You can now configure kubectl
to use k8s-ldap-auth
to authenticate to clusters by editing your kube config file and adding the following user:
users:
- name: my-user
user:
exec:
# In the following, we assume a binary called `k8s-ldap-auth` is
# available in the path. You can instead put the full path to the binary.
# Windows paths do work with kubectl so the following would also work:
# `C:\users\foo\Documents\k8s-ldap-auth.exe`.
command: k8s-ldap-auth
# This field is used by kubectl to fill a template TokenReview in
# `$KUBERNETES_EXEC_INFO` environment variable. Not currently used, it
# might be in the future.
apiVersion: client.authentication.k8s.io/v1beta1
env:
# This environment variable is used within `k8s-ldap-auth` to create
# an ExecCredential. Future version of this authenticator might not
# need it but you'll have to provide it for now.
- name: AUTH_API_VERSION
value: client.authentication.k8s.io/v1beta1
# You can fill a USER environment variable to your username if you
# want to overwrite the USER from your system or to an empty one if you
# want the authenticator to ask for one at runtime.
- name: USER
value: ""
args:
- authenticate
# This is the endpoint to authenticate against. Basically, the server
# started with `k8s-ldap-auth server` plus the `/auth` route, used for
# authentication.
- --endpoint=https://k8s-ldap-auth/auth
installHint: |
k8s-ldap-auth is required to authenticate to the current context.
It can be installed from https://github.com/vbouchaud/k8s-ldap-auth.
# This parameter, when true, tells `kubectl` to fill the TokenReview in
# the `$KUBERNETES_EXEC_INFO` environment variable with extra config
# from the definition of the specific cluster currently targeted.
# This is not used today but might be in the future to allow for custom
# rules on a per-cluster basis.
provideClusterInfo: false
This user can be used by setting the --user
attribute for kubectl
:
kubectl --user my-user get nodes
You can also create contexts with it:
contexts:
- name: context1
context:
cluster: cluster1
user: my-user
- name: context2
context:
cluster: cluster2
user: my-user
current-context: context1
And then:
kubectl --context context2 get nodes
kubectl get nodes
Before you can actually get some result, you will have to upload some rolebindings to the cluster. As stated before, k8s-ldap-auth
provides the apiserver with an ExecCredential containing both LDAP username and groups so both can be used in ClusterRoleBindings and RoleBindings.
Beware: group DNs, username and user id are all set to lowercase in the TokenReview.
Given the following ldap users:
# User Alice
dn: uid=alice,ou=people,ou=company,ou=local
ismemberof: cn=somegroup,ou=groups,ou=company,ou=local
# User Bob
dn: uid=bob,ou=people,ou=company,ou=local
ismemberof: cn=somegroup,ou=groups,ou=company,ou=local
# User Carol
dn: uid=carol,ou=people,ou=company,ou=local
If I want to bind cluster-admin
ClusterRole to the user carol
, I can create a ClusterRoleBinding as following:
apiVersion: rbac.authorization.k8s.io/v1
kind: ClusterRoleBinding
metadata:
name: custom-cluster-admininistrators
roleRef:
apiGroup: rbac.authorization.k8s.io
kind: ClusterRole
name: cluster-admin
subjects:
- apiGroup: rbac.authorization.k8s.io
kind: User
name: carol
Let's say I want to bind the view
ClusterRole so that all user in the group somegroup
will have view access to a given namespace, I can create a RoleBinding such as:
apiVersion: rbac.authorization.k8s.io/v1
kind: RoleBinding
metadata:
name: namespace-users
namespace: somenamespace
roleRef:
apiGroup: rbac.authorization.k8s.io
kind: ClusterRole
name: view
subjects:
- apiGroup: rbac.authorization.k8s.io
kind: Group
name: cn=somegroup,ou=groups,ou=company,ou=local
Note: Kubernetes comes with some basic predefined roles for you to use.
A stripped binary can be built with:
make k8s-ldap-auth
A stripped and compressed binary can be build with:
make release
Docker release multi-arch image can be built and pushed with:
PLATFORM="linux/arm/v7,linux/amd64" make docker
PLATFORM
defaults to linux/arm/v7,linux/arm64/v8,linux/amd64
Docker images of this project are available for arm/v7, arm64/v8 and amd64 at vbouchaud/k8s-ldap-auth on docker hub and on quay.io at vbouchaud/k8s-ldap-auth.
Binaries for the following OS and architectures are available on the release page:
- linux/arm64
- linux/arm
- linux/amd64
- darwin/arm64
- darwin/amd64
- windows/amd64
k8s-ldap-auth.rb
is not in the official repository, you can install it from my repository with the following commands:
brew install vbouchaud/tap/k8s-ldap-auth
Or brew tap vbouchaud/tap
and then brew install k8s-ldap-auth
.
A Chart is hosted at vbouchaud/chartrepo. Please see its readme for more information on how to install it.
I originally started this project after reading Daniel Weibel's article "Implementing LDAP authentication for Kubernetes" (https://learnk8s.io/kubernetes-custom-authentication or https://itnext.io/implementing-ldap-authentication-for-kubernetes-732178ec2155).
- Group search for ldap not supporting
memberof
attribute ;