# Data plane proxy authentication

To obtain a configuration from the control-plane, a data plane proxy must authenticate itself. There are several authentication methods availble.

# Service Account Token

On Kubernetes, a data plane proxy proves its identity with the Service Account Token (opens new window) that is mounted in every pod.

Keep in mind that if you don't explicitly specify serviceAccountTokenName in Deployment, the Pod is run with the default Service Account Token in the Namespace. This means that authentication scope is bound to a Namespace, so any Pod in the Namespace can authenticate as any other Pod in a Namespace.

To have a strict security bound to a Deployment, every Deployment should use unique Service Account Token. On top of that, users should not be able to modify serviceAccountTokenName in Deployment. This can be achieved for example with OPA Gatekeeper (opens new window).

Service Account Token is not bound to a mesh, see data plane proxy membership how to restrict which Pods can join a mesh.

# Data plane proxy token

On Universal, a data plane proxy must be explicitly configured with a unique security token (data plane proxy token) that will be used to prove its identity.

The data plane proxy token is a JWT token (opens new window) that contains:

  • Mesh to which data plane belongs (required)
  • The name of a data plane proxy (optional)
  • List of tags that are permitted to use (optional)
  • Expiration date of the token (required, 10 years if not specified)

The Data plane proxy token is signed by a signing key that is autogenerated when Mesh is created. Tokens are never stored in the control plane, the only thing that is stored are signing keys that are used to verify if a token is valid. The signing key is RSA256 encrypted.

You can check for the signing key:

kumactl get secrets
1

which returns something like:

MESH      NAME                                    AGE
default   dataplane-token-signing-key-default-1   49s
1
2

# Usage

You can generate token either by REST API

curl -XPOST \
  -H "Content-Type: application/json" \
  --data '{"name": "dp-echo-1", "mesh": "default", "tags": {"kuma.io/service": ["backend", "backend-admin"]}, "validFor": "720h"}' \
  http://localhost:5681/tokens/dataplane
1
2
3
4

or by using kumactl

kumactl generate dataplane-token \
  --name dp-echo-1 \
  --mesh default \
  --tag kuma.io/service=backend,backend-admin \
  --valid-for 720h > /tmp/kuma-dp-echo1-token
1
2
3
4
5

The token should be stored in a file and then used when starting kuma-dp

kuma-dp run \
  --cp-address=https://127.0.0.1:5678 \
  --dataplane-file=dp-backend.yaml \
  --dataplane-token-file=/tmp/kuma-dp-echo-1-token
1
2
3
4

You can also pass Dataplane Token inline as KUMA_DATAPLANE_RUNTIME_TOKEN Environment Variable.

# Data Plane Proxy Token boundary

As we can see in the example above, we can generate a token by passing a name, mesh, and a list of tags. The control plane will then verify the data plane proxy resources that are connecting to it against the token. This means we can generate a token by specifying:

  • Only mesh. By doing so we can reuse the token for all dataplanes in a given mesh.
  • mesh + tag (ex. kuma.io/service). This way we can use one token across all instances/replicas of the given service. Please keep in mind that we have to specify to include all the services that a data plane proxy is in charge of. For example, if we have a Dataplane with two inbounds, one valued with kuma.io/service: backend and one with kuma.io/service: backend-admin, we need to specify both values (--tag kuma.io/service=backend,backend-admin).
  • mesh + name + tag (ex. kuma.io/service). This way we can use one token for one instance/replica of the given service.

# Token Revocation

Kuma does not keep the list of issued tokens. Whenever the single token is compromised, we can add it to revocation list so it's no longer valid.

Every token has its own ID which is available in payload under jti key. You can extract ID from token using jwt.io or jwt-cli (opens new window) tool. Here is example of jti

0e120ec9-6b42-495d-9758-07b59fe86fb9
1

Specify list of revoked IDs separated by , and store it as Secret named dataplane-token-revocations-{mesh}

    # Signing key rotation

    If the signing key is compromised, we must rotate it and all the tokens that was signed by it.

    1. Generate new signing key The signing key is stored as a Secret with a name that looks like dataplane-token-signing-key-{mesh}-{serialNumber}.

      Make sure to generate the new signing key with a serial number greater than the serial number of the current signing key.

      • Regenerate tokens Create new data plane proxy tokens. These tokens are automatically created with the signing key that’s assigned the highest serial number, so they’re created with the new signing key. At this point, tokens signed by either new or old signing key are valid.

      • Remove the old signing key

          All new connections to the control plane now require tokens signed with the new signing key.

        # Multizone

        When running in multizone, mode we can generate data plane proxy token both on global and zone control plane. If the deployment pipeline is configured to generate data plane proxy token before running the proxy, it can rely on the Zone CP. This way Global CP is not a single point of failure. Signing key rotation or token revocation should be performed on the global control plane.

        # None

        We can turn off the authentication by setting KUMA_DP_SERVER_AUTH_TYPE to none.

        If we disable the authentication between the control plane and the data plane proxies, any data plane proxy will be able to impersonate any service, therefore this is not recommended in production.

        Last Updated: 4/7/2022, 3:45:22 PM