"Sealed Secrets" for Kubernetes

Downloads

Problem: "I can manage all my K8s config in git, except Secrets."

Solution: Encrypt your Secret into a SealedSecret, which is safe to store - even inside a public repository. The SealedSecret can be decrypted only by the controller running in the target cluster and nobody else (not even the original author) is able to obtain the original Secret from the SealedSecret.

Overview
Installation
- Controller
  - Kustomize
  - Helm Chart
- Kubeseal
Upgrade
Usage
Secret Rotation
Details (advanced)
- Crypto
Developing
FAQ
Community
- Related projects

Overview

Sealed Secrets is composed of two parts:

A cluster-side controller / operator
A client-side utility: kubeseal

The kubeseal utility uses asymmetric crypto to encrypt secrets that only the controller can decrypt.

These encrypted secrets are encoded in a SealedSecret resource, which you can see as a recipe for creating a secret. Here is how it looks:

apiVersion: bitnami.com/v1alpha1
kind: SealedSecret
metadata:
  name: mysecret
  namespace: mynamespace
spec:
  encryptedData:
    foo: AgBy3i4OJSWK+PiTySYZZA9rO43cGDEq.....

Once unsealed this will produce a secret equivalent to this:

apiVersion: v1
kind: Secret
metadata:
  name: mysecret
  namespace: mynamespace
data:
  foo: YmFy  # <- base64 encoded "bar"

This normal kubernetes secret will appear in the cluster after a few seconds you can use it as you would use any secret that you would have created directly (e.g. reference it from a Pod).

Jump to the Installation section to get up and running.

The Usage section explores in more detail how you craft SealedSecret resources.

SealedSecrets as templates for secrets

The previous example only focused on the encrypted secret items themselves, but the relationship between a SealedSecret custom resource and the Secret it unseals into is similar in many ways (but not in all of them) to the familiar Deployment vs Pod.

In particular, the annotations and labels of a SealedSecret resource are not the same as the annotations of the Secret that gets generated out of it.

To capture this distinction, the SealedSecret object has a template section which encodes all the fields you want the controller to put in the unsealed Secret.

The Sprig function library is available in addition to the default Go Text Template functions.

The metadata block is copied as is (the ownerReference field will be updated unless disabled).

Other secret fields are handled individually. The type and immutable fields are copied, and the data field can be used to template complex values on the Secret. All other fields are currently ignored.

apiVersion: bitnami.com/v1alpha1
kind: SealedSecret
metadata:
  name: mysecret
  namespace: mynamespace
  annotations:
    "kubectl.kubernetes.io/last-applied-configuration": ....
spec:
  encryptedData:
    .dockerconfigjson: AgBy3i4OJSWK+PiTySYZZA9rO43cGDEq.....
  template:
    type: kubernetes.io/dockerconfigjson
    immutable: true
    # this is an example of labels and annotations that will be added to the output secret
    metadata:
      labels:
        "jenkins.io/credentials-type": usernamePassword
      annotations:
        "jenkins.io/credentials-description": credentials from Kubernetes

The controller would unseal that into something like:

apiVersion: v1
kind: Secret
metadata:
  name: mysecret
  namespace: mynamespace
  labels:
    "jenkins.io/credentials-type": usernamePassword
  annotations:
    "jenkins.io/credentials-description": credentials from Kubernetes
  ownerReferences:
  - apiVersion: bitnami.com/v1alpha1
    controller: true
    kind: SealedSecret
    name: mysecret
    uid: 5caff6a0-c9ac-11e9-881e-42010aac003e
type: kubernetes.io/dockerconfigjson
immutable: true
data:
  .dockerconfigjson: ewogICJjcmVk...

As you can see, the generated Secret resource is a "dependent object" of the SealedSecret and as such it will be updated and deleted whenever the SealedSecret object gets updated or deleted.

Public key / Certificate

The key certificate (public key portion) is used for sealing secrets, and needs to be available wherever kubeseal is going to be used. The certificate is not secret information, although you need to ensure you are using the correct one.

kubeseal will fetch the certificate from the controller at runtime (requires secure access to the Kubernetes API server), which is convenient for interactive use, but it's known to be brittle when users have clusters with special configurations such as private GKE clusters that have firewalls between control plane and nodes.

An alternative workflow is to store the certificate somewhere (e.g. local disk) with kubeseal --fetch-cert >mycert.pem, and use it offline with kubeseal --cert mycert.pem. The certificate is also printed to the controller log on startup.

Since v0.9.x certificates get automatically renewed every 30 days. It's good practice that you and your team update your offline certificate periodically. To help you with that, since v0.9.2 kubeseal accepts URLs too. You can set up your internal automation to publish certificates somewhere you trust.

kubeseal --cert https://your.intranet.company.com/sealed-secrets/your-cluster.cert

It also recognizes the SEALED_SECRETS_CERT env var. (pro-tip: see also direnv).

NOTE: we are working on providing key management mechanisms that offload the encryption to HSM based modules or managed cloud crypto solutions such as KMS.

Scopes

SealedSecrets are from the POV of an end user a "write only" device.

The idea is that the SealedSecret can be decrypted only by the controller running in the target cluster and nobody else (not even the original author) is able to obtain the original Secret from the SealedSecret.

The user may or may not have direct access to the target cluster. More specifically, the user might or might not have access to the Secret unsealed by the controller.

There are many ways to configure RBAC on k8s, but it's quite common to forbid low-privilege users from reading Secrets. It's also common to give users one or more namespaces where they have higher privileges, which would allow them to create and read secrets (and/or create deployments that can reference those secrets).

Encrypted SealedSecret resources are designed to be safe to be looked at without gaining any knowledge about the secrets it conceals. This implies that we cannot allow users to read a SealedSecret meant for a namespace they wouldn't have access to and just push a copy of it in a namespace where they can read secrets from.

Sealed-secrets thus behaves as if each namespace had its own independent encryption key and thus once you seal a secret for a namespace, it cannot be moved in another namespace and decrypted there.

We don't technically use an independent private key for each namespace, but instead we include the namespace name during the encryption process, effectively achieving the same result.

Furthermore, namespaces are not the only level at which RBAC configurations can decide who can see which secret. In fact, it's possible that users can access a secret called foo in a given namespace but not any other secret in the same namespace. We cannot thus by default let users freely rename SealedSecret resources otherwise a malicious user would be able to decrypt any SealedSecret for that namespace by just renaming it to overwrite the one secret user does have access to. We use the same mechanism used to include the namespace in the encryption key to also include the secret name.

That said, there are many scenarios where you might not care about this level of protection. For example, the only people who have access to your clusters are either admins or they cannot read any Secret resource at all. You might have a use case for moving a sealed secret to other namespaces (e.g. you might not know the namespace name upfront), or you might not know the name of the secret (e.g. it could contain a unique suffix based on the hash of the contents etc).

These are the possible scopes:

strict (default): the secret must be sealed with exactly the same name and namespace. These attributes become part of the encrypted data and thus changing name and/or namespace would lead to "decryption error".
namespace-wide: you can freely rename the sealed secret within a given namespace.
cluster-wide: the secret can be unsealed in any namespace and can be given any name.

In contrast to the restrictions of name and namespace, secret items (i.e. JSON object keys like spec.encryptedData.my-key) can be renamed at will without losing the ability to decrypt the sealed secret.

The scope is selected with the --scope flag:

kubeseal --scope cluster-wide <secret.yaml >sealed-secret.json

It's also possible to request a scope via annotations in the input secret you pass to kubeseal:

sealedsecrets.bitnami.com/namespace-wide: "true" -> for namespace-wide
sealedsecrets.bitnami.com/cluster-wide: "true" -> for cluster-wide

The lack of any of such annotations means strict mode. If both are set, cluster-wide takes precedence.

NOTE: Next release will consolidate this into a single sealedsecrets.bitnami.com/scope annotation.

Installation

See https://github.com/bitnami-labs/sealed-secrets/releases for the latest release and detailed installation instructions.

Cloud platform specific notes and instructions:

Controller

Once you deploy the manifest it will create the SealedSecret resource and install the controller into kube-system namespace, create a service account and necessary RBAC roles.

After a few moments, the controller will start, generate a key pair, and be ready for operation. If it does not, check the controller logs.

Kustomize

The official controller manifest installation mechanism is just a YAML file.

In some cases you might need to apply your own customizations, like set a custom namespace or set some env variables.

kubectl has native support for that, see kustomize.

Helm Chart

The Sealed Secrets helm chart is now officially supported and hosted in this GitHub repo.

helm repo add sealed-secrets https://bitnami-labs.github.io/sealed-secrets

NOTE: The versioning scheme of the helm chart differs from the versioning scheme of the sealed secrets project itself.

Originally the helm chart was maintained by the community and the first version adopted a major version of 1 while the sealed secrets project itself is still at major 0. This is ok because the version of the helm chart itself is not meant to be necessarily the version of the app itself. However this is confusing, so our current versioning rule is:

The SealedSecret controller version scheme: 0.X.Y
The helm chart version scheme: 1.X.Y-rZ

There can be thus multiple revisions of the helm chart, with fixes that apply only to the helm chart without affecting the static YAML manifests or the controller image itself.

NOTE: The helm chart readme still contains a deprecation notice, but it no longer reflects reality and will be removed upon the next release.

NOTE: The helm chart by default installs the controller with the name sealed-secrets, while the kubeseal command line interface (CLI) tries to access the controller with the name sealed-secrets-controller. You can explicitly pass --controller-name to the CLI:

kubeseal --controller-name sealed-secrets <args>

Alternatively, you can set fullnameOverride when installing the chart to override the name. Note also that kubeseal assumes that the controller is installed within the kube-system namespace by default. So if you want to use the kubeseal CLI without having to pass the expected controller name and namespace you should install the Helm Chart like this:

helm install sealed-secrets -n kube-system --set-string fullnameOverride=sealed-secrets-controller sealed-secrets/sealed-secrets

Helm Chart on a restricted environment

In some companies you might be given access only to a single namespace, not a full cluster.

One of the most restrictive environments you can encounter is:

A namespace was allocated to you with some service account.
You do not have access to the rest of the cluster, not even cluster CRDs.
You may not even be able to create further service accounts or roles in your namespace.
You are required to include resource limits in all your deployments.

Even with these restrictions you can still install the sealed secrets Helm