Red Hat OpenShift Container Platform

Enabling Operators to support CCO-based workflows with AWS STS

As an Operator author designing your project to run on Operator Lifecycle Manager (OLM), you can enable your Operator to authenticate against AWS on STS-enabled Red Hat OpenShift Container Platform clusters by customizing your project to support the Cloud Credential Operator (CCO).

With this method, the Operator is responsible for and requires RBAC permissions for creating the CredentialsRequest object and reading the resulting Secret object.

By default, pods related to the Operator deployment mount a serviceAccountToken volume so that the service account token can be referenced in the resulting Secret object.

Prerequisites

Red Hat OpenShift Container Platform 4.14 or later
Cluster in STS mode
OLM-based Operator project

Procedure

Update your Operator project’s ClusterServiceVersion (CSV) object:

Ensure your Operator has RBAC permission to create CredentialsRequests objects:

Example clusterPermissions list

# ...
install:
  spec:
    clusterPermissions:
    - rules:
      - apiGroups:
        - "cloudcredential.openshift.io"
        resources:
        - credentialsrequests
        verbs:
        - create
        - delete
        - get
        - list
        - patch
        - update
        - watch

Add the following annotation to claim support for this method of CCO-based workflow with AWS STS:
```
# ...
metadata:
 annotations:
   features.operators.openshift.io/token-auth-aws: "true"
```

Update your Operator project code:

Get the role ARN from the environment variable set on the pod by the Subscription object. For example:

// Get ENV var
roleARN := os.Getenv("ROLEARN")
setupLog.Info("getting role ARN", "role ARN = ", roleARN)
webIdentityTokenPath := "/var/run/secrets/openshift/serviceaccount/token"

Ensure you have a CredentialsRequest object ready to be patched and applied. For example:

Example CredentialsRequest object creation

import (
   minterv1 "github.com/openshift/cloud-credential-operator/pkg/apis/cloudcredential/v1"
   corev1 "k8s.io/api/core/v1"
   metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
)

var in = minterv1.AWSProviderSpec{
   StatementEntries: []minterv1.StatementEntry{
      {
         Action: []string{
            "s3:*",
         },
         Effect:   "Allow",
         Resource: "arn:aws:s3:*:*:*",
      },
   },
	STSIAMRoleARN: "<role_arn>",
}

var codec = minterv1.Codec
var ProviderSpec, _ = codec.EncodeProviderSpec(in.DeepCopyObject())

const (
   name      = "<credential_request_name>"
   namespace = "<namespace_name>"
)

var CredentialsRequestTemplate = &minterv1.CredentialsRequest{
   ObjectMeta: metav1.ObjectMeta{
       Name:      name,
       Namespace: "openshift-cloud-credential-operator",
   },
   Spec: minterv1.CredentialsRequestSpec{
      ProviderSpec: ProviderSpec,
      SecretRef: corev1.ObjectReference{
         Name:      "<secret_name>",
         Namespace: namespace,
      },
      ServiceAccountNames: []string{
         "<service_account_name>",
      },
      CloudTokenPath:   "",
   },
}

Alternatively, if you are starting from a CredentialsRequest object in YAML form (for example, as part of your Operator project code), you can handle it differently:

Example CredentialsRequest object creation in YAML form

// CredentialsRequest is a struct that represents a request for credentials
type CredentialsRequest struct {
  APIVersion string `yaml:"apiVersion"`
  Kind       string `yaml:"kind"`
  Metadata   struct {
     Name      string `yaml:"name"`
     Namespace string `yaml:"namespace"`
  } `yaml:"metadata"`
  Spec struct {
     SecretRef struct {
        Name      string `yaml:"name"`
        Namespace string `yaml:"namespace"`
     } `yaml:"secretRef"`
     ProviderSpec struct {
        APIVersion     string `yaml:"apiVersion"`
        Kind           string `yaml:"kind"`
        StatementEntries []struct {
           Effect   string   `yaml:"effect"`
           Action   []string `yaml:"action"`
           Resource string   `yaml:"resource"`
        } `yaml:"statementEntries"`
        STSIAMRoleARN   string `yaml:"stsIAMRoleARN"`
     } `yaml:"providerSpec"`

     // added new field
      CloudTokenPath   string `yaml:"cloudTokenPath"`
  } `yaml:"spec"`
}

// ConsumeCredsRequestAddingTokenInfo is a function that takes a YAML filename and two strings as arguments
// It unmarshals the YAML file to a CredentialsRequest object and adds the token information.
func ConsumeCredsRequestAddingTokenInfo(fileName, tokenString, tokenPath string) (*CredentialsRequest, error) {
  // open a file containing YAML form of a CredentialsRequest
  file, err := os.Open(fileName)
  if err != nil {
     return nil, err
  }
  defer file.Close()

  // create a new CredentialsRequest object
  cr := &CredentialsRequest{}

  // decode the yaml file to the object
  decoder := yaml.NewDecoder(file)
  err = decoder.Decode(cr)
  if err != nil {
     return nil, err
  }

  // assign the string to the existing field in the object
  cr.Spec.CloudTokenPath = tokenPath

  // return the modified object
  return cr, nil
}

Adding a CredentialsRequest object to the Operator bundle is not currently supported.

Add the role ARN and web identity token path to the credentials request and apply it during Operator initialization:

Example applying CredentialsRequest object during Operator initialization

// apply CredentialsRequest on install
credReq := credreq.CredentialsRequestTemplate
credReq.Spec.CloudTokenPath = webIdentityTokenPath

c := mgr.GetClient()
if err := c.Create(context.TODO(), credReq); err != nil {
   if !errors.IsAlreadyExists(err) {
      setupLog.Error(err, "unable to create CredRequest")
      os.Exit(1)
   }
}

Ensure your Operator can wait for a Secret object to show up from the CCO, as shown in the following example, which is called along with the other items you are reconciling in your Operator:

Example wait for Secret object

// WaitForSecret is a function that takes a Kubernetes client, a namespace, and a v1 "k8s.io/api/core/v1" name as arguments
// It waits until the secret object with the given name exists in the given namespace
// It returns the secret object or an error if the timeout is exceeded
func WaitForSecret(client kubernetes.Interface, namespace, name string) (*v1.Secret, error) {
  // set a timeout of 10 minutes
  timeout := time.After(10 * time.Minute)

  // set a polling interval of 10 seconds
  ticker := time.NewTicker(10 * time.Second)

  // loop until the timeout or the secret is found
  for {
     select {
     case <-timeout:
        // timeout is exceeded, return an error
        return nil, fmt.Errorf("timed out waiting for secret %s in namespace %s", name, namespace)
           // add to this error with a pointer to instructions for following a manual path to a Secret that will work on STS
     case <-ticker.C:
        // polling interval is reached, try to get the secret
        secret, err := client.CoreV1().Secrets(namespace).Get(context.Background(), name, metav1.GetOptions{})
        if err != nil {
           if errors.IsNotFound(err) {
              // secret does not exist yet, continue waiting
              continue
           } else {
              // some other error occurred, return it
              return nil, err
           }
        } else {
           // secret is found, return it
           return secret, nil
        }
     }
  }
}

The timeout value is based on an estimate of how fast the CCO might detect an added CredentialsRequest object and generate a Secret object. You might consider lowering the time or creating custom feedback for cluster administrators that could be wondering why the Operator is not yet accessing the cloud resources.

Set up the AWS configuration by reading the secret created by the CCO from the credentials request and creating the AWS config file containing the data from that secret:

Example AWS configuration creation

func SharedCredentialsFileFromSecret(secret *corev1.Secret) (string, error) {
   var data []byte
   switch {
   case len(secret.Data["credentials"]) > 0:
       data = secret.Data["credentials"]
   default:
       return "", errors.New("invalid secret for aws credentials")
   }


   f, err := ioutil.TempFile("", "aws-shared-credentials")
   if err != nil {
       return "", errors.Wrap(err, "failed to create file for shared credentials")
   }
   defer f.Close()
   if _, err := f.Write(data); err != nil {
       return "", errors.Wrapf(err, "failed to write credentials to %s", f.Name())
   }
   return f.Name(), nil
}

The secret is assumed to exist, but your Operator code should wait and retry when using this secret to give time to the CCO to create the secret.

Additionally, the wait period should eventually time out and warn users that the Red Hat OpenShift Container Platform cluster version, and therefore the CCO, might be an earlier version that does not support the CredentialsRequest object workflow with STS detection. In such cases, instruct users that they must add a secret by using another method.

Configure the AWS SDK session, for example:

Example AWS SDK session configuration

sharedCredentialsFile, err := SharedCredentialsFileFromSecret(secret)
if err != nil {
   // handle error
}
options := session.Options{
   SharedConfigState: session.SharedConfigEnable,
   SharedConfigFiles: []string{sharedCredentialsFile},
}