Authorization using centralized ACLs

Note

Apache Kafka® ships with a pluggable, out-of-box Authorizer implementation that uses Apache ZooKeeper™ to store all the ACLs. If you are not using role-based access control (RBAC) on MDS, then refer to Authorization using Access Control Lists (ACLs) for details about authorization using ACLs (also known as ZooKeeper-based ACLs).

If you are using Confluent Platform RBAC-enabled Kafka clusters, then you can manage ACLs for multiple Kafka clusters using the shared, centralized MDS. This enables you to configure and store ACLs similar to the way you manage and store RBAC role bindings.

If you are planning to use centralized (MDS-based) ACLs, but will continue using ZooKeeper-based ACLs until you migrate, you can continue to use the AdminClient API to manage ACLs up until you migrate.

Prerequisites

Before creating and using ACLs in the centralized MDS, you must satisfy the following requirements:

  • You must configure RBAC on a new cluster (using confluent.authorizer.access.rule.providers=CONFLUENT) that does not have any ZooKeeper-based ACLs. You must configure MDS broker principals as super.users or broker.users.

    Note

    The broker.users option is a semicolon-separated list of principals of users who are allowed access to all resources on interbroker listeners. Unlike super.users, broker.users only allows requests originating from interbroker listeners. The primary purpose of this option is to bootstrap MDS clusters when configuring authorization using centralized ACLs. For details about this configuration, see Configuring the Confluent Server Authorizer.

  • If you have already configured RBAC on an existing cluster that is using ZooKeeper-based ACLs, then you must migrate the ZooKeeper-based ACLs to centralized storage in MDS.

  • Centralized ACLs are not supported for with KRaft-based MDS.

Migrate ZooKeeper-based ACLs to MDS

You must first enable RBAC on running clusters and configure the MDS broker principals as super.users or broker.users.

After enabling RBAC, brokers in your Kafka clusters can read ACLs from both ZooKeeper and MDS. To avoid inconsistent ACLs, do not add MDS-based ACLs until after this migration is complete.

To migrate the existing ZooKeeper-based ACLs to MDS:

  1. Confirm that confluent.authorizer.access.rule.providers=CONFLUENT, ZK_ACL is configured for all brokers in the server.properties file (as mentioned in Prerequisites).

  2. Specify confluent.authorizer.migrate.acls.from.zk=true on a single broker of choice (If the MDS cluster and Kafka cluster for which you are migrating ACLs are same, then select a broker that is not a MDS writer. You can identify the MDS writer broker in the writerMemberId, which is in the latest metadata-service.log log file). The non-MDS broker is the broker performing the ZooKeeper ACL migration. Make note of this broker’s broker.id, as it is required in the next step.

  3. Perform a rolling restart of the broker mentioned in Step 2. When restarted, this broker copies the ZooKeeper-based ACLs to MDS and also listens for any ACL updates during migration. After completing this step, you can add additional Kafka-based ACLs, which are applied to all brokers and stored in the MDS.

  4. After the migration completes (you can verify completion in the kafka.authorizer.log), verify that the centralized ACLs are in MDS:

    confluent iam acl list --kafka-cluster-id <kafka-cluster-id>

  5. Specify confluent.authorizer.access.rule.providers=CONFLUENT for all brokers.

  6. Specify confluent.authorizer.migrate.acls.from.zk=false for the broker from Step 2.

  7. Perform a rolling restart of all the brokers in the cluster, including the broker from Step 2, which should be the last broker to be restarted. This ensures that you do not miss any ACL updates that may have occurred during migration.

    From this point forward, you cannot update ZooKeeper using AdminClient, and any ACLs added directly to ZooKeeper are ignored.

Using centralized ACLs

Examples in this section use confluent iam rbac role-binding create to create, delete or list ACLs. For detailed information about the supported options, run confluent iam acl --help.

For details about ACL format and customizing user names refer to Authorization using Access Control Lists (ACLs).

Limitations

Centralized ACLs are not supported with KRaft-based MDS.

Centralized ACLs supports up to 1000 ACLs per cluster.

Creating centralized ACLs

In this example you are creating an ACL where Principal User:Bob is allowed to perform a Read operation from IP 198.51.100.0 on Topic test-topic from a specified Kafka cluster. You can do this by executing the following:

confluent iam acl create --allow --principal User:Bob --operation READ --host 198.51.100.0 --topic test-topic  --kafka-cluster-id  <kafka-cluster-id>

Note that confluent iam acl supports IPv6 addresses, but does not support IP ranges and subnets.

By default all principals that don’t have an explicit ACL allowing an operation to access a resource are denied. In rare cases where an ACL that allows access to all but some principal is desired, you can use the --deny option. For example, use the following commands to allow all users to Read from test-topic in a specified Kafka cluster, but deny only User:BadBob from the specified Kafka cluster:

confluent iam acl create --allow --principal User:'*' --operation READ --topic test-topic --kafka-cluster-id <kafka-cluster-id>
confluent iam acl create --deny --principal User:BadBob --operation READ --topic test-topic  --kafka-cluster-id <kafka-cluster-id>

Note that when multiple ACLs apply to a user, as shown above, the --deny option always overrides the --allow option.

The preceding examples create ACLs for a topic by specifying --topic [topic-name] as the resource pattern option. Similarly, you can create ACLs for a cluster by specifying --cluster-scope and to a group by specifying --group:[group-name]. In the event that you want to grant permission to all groups, you can do so by specifying --group='*' as shown in the following command:

confluent iam acl create --allow --principal User:'*' --operation READ --topic test --group='*' --kafka-cluster-id <kafka-cluster-id>

You can add ACLs on prefixed resource patterns too. For example, if you include --topic abc- and --prefix in the command, it will impact permissions on all resources whose name starts with abc-. The next example shows how to add an ACL for user Jane to access any topic whose name starts with test- in a specified Kafka cluster. You can do this using the CLI with following options:

confluent iam acl create --allow --principal User:Jane --prefix  --topic test- --kafka-cluster-id <kafka-cluster-id>

Deleting centralized ACLs

To delete the ACL added in the first example above, execute the following:

confluent iam acl delete --allow --principal User:Bob --operation READ --host 198.51.100.0 --topic test-topic  --kafka-cluster-id  <kafka-cluster-id>

If you want to remove the ACL that you created using a prefixed pattern, execute the following:

confluent iam acl create --allow --principal User:Jane --prefix  --topic test-topic --kafka-cluster-id <kafka-cluster-id>

Listing centralized ACLs

You can list the ACLs for a given Kafka cluster using the confluent iam acl list command and the cluster ID. For example, to list all ACLs for test-topic that use the prefix test- in a specified Kafka cluster, execute the following:

confluent iam acl list --topic test-topic --prefix --kafka-cluster-id <kafka-cluster-id>

Note that this will only return the ACLs that have been added to this exact prefix pattern.