Switch branches/tags
Find file Copy path
360 lines (293 sloc) 19 KB

AWS IAM Authenticator for Kubernetes

A tool to use AWS IAM credentials to authenticate to a Kubernetes cluster. The initial work on this tool was driven by Heptio. The project recieves contributions from multiple community engineers and is currently maintained by Heptio and Amazon EKS OSS Engineers.

Why do I want this?

If you are an administrator running a Kubernetes cluster on AWS, you already need to manage AWS IAM credentials to provision and update the cluster. By using AWS IAM Authenticator for Kubernetes, you avoid having to manage a separate credential for Kubernetes access. AWS IAM also provides a number of nice properties such as an out of band audit trail (via CloudTrail) and 2FA/MFA enforcement.

If you are building a Kubernetes installer on AWS, AWS IAM Authenticator for Kubernetes can simplify your bootstrap process. You won't need to somehow smuggle your initial admin credential securely out of your newly installed cluster. Instead, you can create a dedicated KubernetesAdmin role at cluster provisioning time and set up Authenticator to allow cluster administrator logins.

How do I use it?

Assuming you have a cluster running in AWS and you want to add AWS IAM Authenticator for Kubernetes support, you need to:

  1. Create an IAM role you'll use to identify users.
  2. Run the Authenticator server as a DaemonSet.
  3. Configure your API server to talk to Authenticator.
  4. Set up kubectl to use Authenticator tokens.

1. Create an IAM role

First, you must create one or more IAM roles that will be mapped to users/groups inside your Kubernetes cluster. The easiest way to do this is to log into the AWS Console:

  • Choose the "Role for cross-account access" / "Provide access between AWS accounts you own" option.
  • Paste in your AWS account ID number (available in the top right in the console).
  • Your role does not need any additional policies attached.

This will create an IAM role with no permissions that can be assumed by authorized users/roles in your account. Note the Amazon Resource Name (ARN) of your role, which you will need below.

You can also do this in a single step using the AWS CLI instead of the AWS Console:

# get your account ID
ACCOUNT_ID=$(aws sts get-caller-identity --output text --query 'Account')

# define a role trust policy that opens the role to users in your account (limited by IAM policy)
POLICY=$(echo -n '{"Version":"2012-10-17","Statement":[{"Effect":"Allow","Principal":{"AWS":"arn:aws:iam::'; echo -n "$ACCOUNT_ID"; echo -n ':root"},"Action":"sts:AssumeRole","Condition":{}}]}')

# create a role named KubernetesAdmin (will print the new role's ARN)
aws iam create-role \
  --role-name KubernetesAdmin \
  --description "Kubernetes administrator role (for AWS IAM Authenticator for Kubernetes)." \
  --assume-role-policy-document "$POLICY" \
  --output text \
  --query 'Role.Arn'

You can also skip this step and use:

  • An existing role (such as a cross-account access role).
  • An IAM user (see mapUsers below).
  • An EC2 instance or a federated role (see mapRoles below).

2. Run the server

The server is meant to run on each of your master nodes as a DaemonSet with host networking so it can expose a localhost port.

For a sample ConfigMap and DaemonSet configuration, see example.yaml.

(Optional) Pre-generate a certificate, key, and kubeconfig

If you're building an automated installer, you can also pre-generate the certificate, key, and webhook kubeconfig files easily using aws-iam-authenticator init. This command will generate files and place them in the configured output directories.

You can run this on each master node prior to starting the API server. You could also generate them before provisioning master nodes and install them in the appropriate host paths.

If you do not pre-generate files, aws-iam-authenticator server will generate them on demand. This works but requires that you restart your Kubernetes API server after installation.

3. Configure your API server to talk to the server

The Kubernetes API integrates with AWS IAM Authenticator for Kubernetes using a token authentication webhook. When you run aws-iam-authenticator server, it will generate a webhook configuration file and save it onto the host filesystem. You'll need to add a single additional flag to your API server configuration:


On many clusters, the API server runs as a static pod. You can add the flag to /etc/kubernetes/manifests/kube-apiserver.yaml. Make sure the host directory /etc/kubernetes/aws-iam-authenticator/ is mounted into your API server pod. You may also need to restart the kubelet daemon on your master node to pick up the updated static pod definition:

systemctl restart kubelet.service

4. Set up kubectl to use authentication tokens provided by AWS IAM Authenticator for Kubernetes

This requires a 1.10+ kubectl binary to work. If you receive Please enter Username: when trying to use kubectl you need to update to the latest kubectl

Finally, once the server is set up you'll want to authenticate! You will still need a kubeconfig that has the public data about your cluster (cluster CA certificate, endpoint address). The users section of your configuration, however, should include an exec section (refer to the v1.10 docs)::

# [...]
- name: kubernetes-admin
      command: aws-iam-authenticator
        - "token"
        - "-i"
        - "CLUSTER_ID"
        - "-r"
        - "ROLE_ARN"
  # no client certificate/key needed here!

This means the kubeconfig is entirely public data and can be shared across all Authenticator users. It may make sense to upload it to a trusted public location such as AWS S3.

Make sure you have the aws-iam-authenticator binary installed. You can install it with go get -u -v

To authenticate, run kubectl --kubeconfig /path/to/kubeconfig" [...]. kubectl will exec the aws-iam-authenticator binary with the supplied params in your kubeconfig which will generate a token and pass it to the apiserver. The token is valid for 15 minutes (the shortest value AWS permits) and can be reused multiple times.

You can also omit -r ROLE_ARN to sign the token with your existing credentials without assuming a dedicated role. This is useful if you want to authenticate as an IAM user directly or if you want to authenticate using an EC2 instance role or a federated role.

Kops Usage

Clusters managed by Kops can be configured to use Authenticator. Both single and HA master cluster configurations are supported. Perform the following steps to setup Authenticator on a Kops cluster:

  1. Pre-generate the certificate, key, and kubeconfig and upload them to the kops state store.

    aws-iam-authenticator init -i $CLUSTER_NAME
    aws s3 cp cert.pem ${KOPS_STATE_STORE}/${CLUSTER_NAME}/addons/authenticator/cert.pem;
    aws s3 cp key.pem ${KOPS_STATE_STORE}/${CLUSTER_NAME}/addons/authenticator/key.pem;
    aws s3 cp aws-iam-authenticator.kubeconfig ${KOPS_STATE_STORE}/${CLUSTER_NAME}/addons/authenticator/kubeconfig.yaml;
  2. Add the following sections to the cluster spec, either using kops edit cluster ${CLUSTER_NAME} or editing the manifest yaml file. Be sure to replace KOPS_STATE_STORE and CLUSTER_NAME with their appropriate values since those environment variables are not available at runtime. This downloads the files from the state store on masters to a directory that is volume mounted by kube-apiserver. Kops does not support adding additional volumes to kube-apiserver so we must reuse the existing /srv/kubernetes hostPath volume.

    apiVersion: kops/v1alpha2
    kind: Cluster
        authenticationTokenWebhookConfigFile: /srv/kubernetes/aws-iam-authenticator/kubeconfig.yaml
      - name: kops-hook-authenticator-config.service
          - kubelet.service
        roles: [Master]
        manifest: |
          Description=Download AWS Authenticator configs from S3
          ExecStart=/bin/mkdir -p /srv/kubernetes/aws-iam-authenticator
          ExecStart=/usr/local/bin/aws s3 cp --recursive s3://KOPS_STATE_STORE/CLUSTER_NAME/addons/authenticator /srv/kubernetes/aws-iam-authenticator/

    If using a non-default AMI that does not have the AWS CLI, replace the second ExecStart statement with:

    ExecStart=/usr/bin/docker run --net=host --rm -v /srv/kubernetes/aws-iam-authenticator:/srv/kubernetes/aws-iam-authenticator aws s3 cp --recursive s3://KOPS_STATE_STORE/CLUSTER_NAME/addons/authenticator /srv/kubernetes/aws-iam-authenticator/
  3. Apply the changes with kops update cluster ${CLUSTER_NAME}. If the cluster already exists, roll the cluster with kops rolling-update cluster ${CLUSTER_NAME} in order to recreate the master nodes.

  4. Update the Authenticator DaemonSet's state and output volumes to both use /srv/kubernetes/aws-iam-authenticator/ for their hostPaths.

  5. Apply the DaemonSet and ConfigMap resource manifests to launch the Authenticator server on the cluster.

Note: Certain Kops commands will overwrite the exec configuration in kubeconfig so it may need to be restored manually. See kubernetes/kops#5051 for more information.

How does it work?

It works using the AWS sts:GetCallerIdentity API endpoint. This endpoint returns information about whatever AWS IAM credentials you use to connect to it.

Client side (aws-iam-authenticator token)

We use this API in a somewhat unusual way by having the Authenticator client generate and pre-sign a request to the endpoint. We serialize that request into a token that can pass through the Kubernetes authentication system.

Server side (aws-iam-authenticator server)

The token is passed through the Kubernetes API server and into the Authenticator server's /authenticate endpoint via a webhook configuration. The Authenticator server validates all the parameters of the pre-signed request to make sure nothing looks funny. It then submits the request to the real server, which validates the client's HMAC signature and returns information about the user. Now that the server knows the AWS identity of the client, it translates this identity into a Kubernetes user and groups via a simple static mapping.

This mechanism is borrowed with a few changes from Vault.

What is a cluster ID?

The Authenticator cluster ID is a unique-per-cluster identifier that prevents certain replay attacks. Specifically, it prevents one Authenticator server (e.g., in a dev environment) from using a client's token to authenticate to another Authenticator server in another cluster.

The cluster ID does need to be unique per-cluster, but it doesn't need to be a secret. Some good choices are:

  • A random ID such as from openssl rand 16 -hex
  • The domain name of your Kubernetes API server

The Vault documentation also explains this attack (see X-Vault-AWS-IAM-Server-ID).

Specifying Credentials & Using AWS Profiles

Credentials can be specified for use with aws-iam-authenticator via any of the methods available to the AWS SDK for Go. This includes specifying AWS credentials with enviroment variables or by utilizing a credentials file.

AWS named profiles are supported by aws-iam-authenticator via the AWS_PROFILE environment variable. For example, to authenticate with credentials specified in the dev profile the AWS_PROFILE can be exported or specified explictly (e.g., AWS_PROFILE=dev kubectl get all). If no AWS_PROFILE is set, the default profile is used.

The AWS_PROFILE can also be specified directly in the kubeconfig file as part of the exec flow. For example, to specify that credentials from the dev named profile should always be used by aws-iam-authenticator, your kubeconfig would include an env key thats sets the profile:

apiVersion: v1
- cluster:
    server: ${server}
    certificate-authority-data: ${cert}
  name: kubernetes
- context:
    cluster: kubernetes
    user: aws
  name: aws
current-context: aws
kind: Config
preferences: {}
- name: aws
      command: aws-iam-authenticator
      - name: "AWS_PROFILE"
        value: "dev"
        - "token"
        - "-i"
        - "mycluster"

This method allows the appropriate profile to be used implicitly. Note that any environment variables set as part of the exec flow will take precedence over what's already set in your environment.

Note for federated users:

Federated AWS users often will have a "meaningful" attribute mapped onto their assumed role, such as an email address, through the account's AWS configuration. These assumed sessions have a few parts, the role id and caller-specified-role-name. By default, when a federated user uses the --role option of aws-iam-authenticator to assume a new role the caller-specified-role-name will be converted to a random token and the role id carries through to the newly assumed role.

Using aws-iam-authenticator token ... --forward-session-name will map the original caller-specified-role-name attribute onto the new STS assumed session. This can be helpful for quickly attempting to associate "who performed action X on the K8 cluster".

Please note, this should not be considered definitive and needs to be cross referenced via the role id (which remains consistent) with CloudTrail logs as a user could potentially change this on the client side.


If your client fails with an error like could not get token: AccessDenied [...], you can try assuming the role with the AWS CLI directly:

# AWS CLI version of `aws-iam-authenticator token -r arn:aws:iam::ACCOUNT:role/ROLE`:
$ aws sts assume-role --role-arn arn:aws:iam::ACCOUNT:role/ROLE --role-session-name test

If that fails, there are a few possible problems to check for:

  • Make sure your base AWS credentials are available in your shell (aws sts get-caller-identity can help troubleshoot this).

  • Make sure the target role allows your source account access (in the role trust policy).

  • Make sure your source principal (user/role/group) has an IAM policy that allows sts:AssumeRole for the target role.

  • Make sure you don't have any explicit deny policies attached to your user, group, or in AWS Organizations that would prevent the sts:AssumeRole.

  • Try simulating the sts:AssumeRole call in the Policy Simulator.

Full Configuration Format

The client and server have the same configuration format. They can share the same exact configuration file, since there are no secrets stored in the configuration.

# a unique-per-cluster identifier to prevent replay attacks (see above)

# default IAM role to assume for `aws-iam-authenticator token`
defaultRole: arn:aws:iam::000000000000:role/KubernetesAdmin

# server listener configuration
  # localhost port where the server will serve the /authenticate endpoint
  port: 21362 # (default)

  # state directory for generated TLS certificate and private keys
  stateDir: /var/aws-iam-authenticator # (default)

  # output `path` where a generated webhook kubeconfig will be stored.
  generateKubeconfig: /etc/kubernetes/aws-iam-authenticator.kubeconfig # (default)

  # role to assume before querying EC2 API in order to discover metadata like EC2 private DNS Name
  ec2DescribeInstancesRoleARN: arn:aws:iam::000000000000:role/DescribeInstancesRole

  # each mapRoles entry maps an IAM role to a username and set of groups
  # Each username and group can optionally contain template parameters:
  #  1) "{{AccountID}}" is the 12 digit AWS ID.
  #  2) "{{SessionName}}" is the role session name.
  # statically map arn:aws:iam::000000000000:role/KubernetesAdmin to cluster admin
  - roleARN: arn:aws:iam::000000000000:role/KubernetesAdmin
    username: kubernetes-admin
    - system:masters

  # map EC2 instances in my "KubernetesNode" role to users like
  # "aws:000000000000:instance:i-0123456789abcdef0". Only use this if you
  # trust that the role can only be assumed by EC2 instances. If an IAM user
  # can assume this role directly (with sts:AssumeRole) they can control
  # SessionName.
  - roleARN: arn:aws:iam::000000000000:role/KubernetesNode
    username: aws:{{AccountID}}:instance:{{SessionName}}
    - system:bootstrappers
    - aws:instances

  # map nodes that should conform to the username "system:node:<private-DNS>".  This
  # requires the authenticator to query the EC2 API in order to discover the private
  # DNS of the EC2 instance originating the authentication request.  Optionally, you
  # may specify a role that should be assumed before querying the EC2 API with the
  # key "server.ec2DescribeInstancesRoleARN" (see above).
  - roleARN: arn:aws:iam::000000000000:role/KubernetesNode
    username: system:node:{{EC2PrivateDNSName}}
    - system:nodes
    - system:bootstrappers

  # map federated users in my "KubernetesAdmin" role to users like
  # "". The SessionName is an arbitrary role name
  # like an e-mail address passed by the identity provider. Note that if this
  # role is assumed directly by an IAM User (not via federation), the user
  # can control the SessionName.
  - roleARN: arn:aws:iam::000000000000:role/KubernetesAdmin
    username: admin:{{SessionName}}
    - system:masters

  # each mapUsers entry maps an IAM role to a static username and set of groups
  # map user IAM user Alice in 000000000000 to user "alice" in group "system:masters"
  - userARN: arn:aws:iam::000000000000:user/Alice
    username: alice
    - system:masters

  # automatically map IAM ARN from these accounts to username.
  # NOTE: Always use quotes to avoid the account numbers being recognized as numbers
  # instead of strings by the yaml parser.
  - "012345678901"
  - "456789012345"