Configuring KMS Access Control Lists (ACLs)

Cloudera’s framework for key management is based on enforcing a secure-by-default configuration based upon the KMS ACLs and roles described here.

ACLs are implemented in the upstream Hadoop KMS and apply to any variant of the service, whether it is through Key Trustee Server (KTS) or the Java Keystore (JKS). Understanding how these ACLs are defined and work enables you to more accurately create ACLs that meet both your needs and the needs of your customers.

The KMS ACL classes provide fine-grain control of key access at different decision points (for example, preventing a user from accessing the KMS altogether), and are applied either KMS-wide or key-specific:

There are five distinct classes of ACLs, each defined by the operations defined within.

Before diving into the details of how KMS ACLs are evaluated, it is critical that you understand the key rules that the KMS uses in performing this evaluation.

The better you understand these rules, the more likely it is that you will be successful creating and managing KMS ACL sets to achieve your desired goals.

The KMS evaluates hadoop.kms.acl.<OPERATION> and hadoop.kms.blacklist.<OPERATION> classes to determine whether or not access to a specific KMS feature or function is authorized.

In other words, a user must be allowed by hadoop.kms.acl.<OPERATION>, and not be disallowed by hadoop.kms.blacklist.<OPERATION>.

If a user is disallowed or otherwise not allowed access to a KMS-wide operation, then the flow halts and returns the result "Denied".

If a user is allowed access to a KMS-wide operation, then the evaluation flow proceeds.

The KMS evaluates whitelist.key.acl class.

The KMS ACL workflow evaluates the whitelist.key.acl.<OPERATION>, and if the user is allowed access, then it is granted ("Allowed") . If not, then the flow continues with the evaluation.

The KMS evaluates default.key.acl.<OPERATION>, and key.acl.<OPERATION> classes.

The KMS evaluates whether or not there is a key.acl.KEY.<OPERATION> that matches the action the user is attempting to perform. If there is, it then evaluates that value to determine whether or not the user can perform the requested operation.

Depending on the result of the KMS ACL evaluation, controls are applied to the key and results (Allowed or Denied).

Blacklist and Whitelist Syntax

Blocking User Access

If you wish to block access to an operation entirely, use the value of an empty space, or some non-existent values (for example,'NOUSERS NOGROUPS'). By doing this, you ensure that no user maps to a particular operation by default. Alternatively, you can restrict those features to Key Administrators only by setting the value to Keyadmin users and/or groups.

Group Membership in KMS ACLs

The group membership used by ACL entries depends upon the configured group mapping mechanism for HDFS. By default, group membership is determined on the local Linux system running the KMS service. If you have configured HDFS to use LDAP for group mapping, then group membership for the ACL entries is determined using the configured LDAP settings.

Minimum Required Role: Key Administrator (also provided by Full Administrator)

Cloudera recommends the following ACL definition for secure production settings. Replace keyadmin and keyadmingroup with the user and group responsible for maintaining encryption keys.

<configuration>

<!--
KMS ACLs control which users can perform various actions on the KMS,and which
users and groups have access to which keys.

This file includes the following sections:
  * ACLs for KMS operations
  ** Access to specific KMS operations
  ** Blacklists for specific operations
  * ACLs for keys
  ** Default ACLs for keys
  ** Whitelist ACLs for keys
  ** Key-specific ACLs
-->

<!--
KMS ACLs that govern access to specific key operations. If access is not granted
for an operation here, then the operation is forbidden, even if a key ACL allows it.

The ACL value should be either a username or a username and groupname separated
by whitespace.

A value of "*" (for the username or groupname) indicates that all users are
granted access to that operation. Any operation for which there is no ACL or an
empty (zero-length) ACL is treated as having an ACL with a value of "*".
To disallow all users, add an ACL with a value of " " (a single space).

Note: This convention applies only to the KMS-wide ACLs beginning with 'hadoop.kms.acl'.
-->
<property>
  <name>hadoop.kms.acl.CREATE</name>
  <value>keyadmin keyadmingroup</value>
  <description>
    ACL for create-key operations.
    If the user is not in the GET ACL, the key material is not returned
    as part of the response.
  </description>
</property>

<property>
  <name>hadoop.kms.acl.DELETE</name>
  <value>keyadmin keyadmingroup</value>
  <description>
   ACL for delete-key operations.
  </description>
</property>

<property>
  <name>hadoop.kms.acl.ROLLOVER</name>
  <value>keyadmin keyadmingroup</value>
  <description>
    ACL for rollover-key operations.
    If the user does is not in the GET ACL, the key material is not returned as part of the response.
  </description>
</property>

<property>
  <name>hadoop.kms.acl.GET</name>
  <value></value>
  <description>
    ACL for get-key-version and get-current-key operations.
  </description>
</property>

<property>
  <name>hadoop.kms.acl.GET_KEYS</name>
  <value>keyadmin keyadmingroup</value>
  <description>
    ACL for get-keys operations.
  </description>
</property>

<property>
  <name>hadoop.kms.acl.SET_KEY_MATERIAL</name>
  <value></value>
  <description>
    Complementary ACL for CREATE and ROLLOVER operations to allow the client to provide the
    key material when creating or rolling a key.
  </description>
</property>

<property>
  <name>hadoop.kms.acl.GENERATE_EEK</name>
  <value>hdfs supergroup</value>
  <description>
    ACL for generateEncryptedKey CryptoExtension operations.
  </description>
</property>

<!--
  KMS blacklists to prevent access to operations. These settings override the permissions granted by the KMS ACLs listed above.

  The blacklist value should be either a username or a username and groupname separated by whitespace.

  A blank value indicates that no user is blacklisted from the operation. A value of "*" (for either the username or
  groupname) indicates that all users are blacklisted from the operation. Any operation for which there is no blacklist
  will be treated as having a blacklist with an empty value.
-->

<!--
  In this template the hdfs user is blacklisted for everything except GET_METADATA, GET_KEYS, and GENERATE_EEK. The
  GET and SET_KEY_MATERIAL operations are blacklisted for all users because Hadoop users should not need to perform
  those operations, and access to the key material should be as restricted as possible.
-->

<property>
  <name>hadoop.kms.blacklist.CREATE</name>
  <value>hdfs supergroup</value>
</property>

<property>
  <name>hadoop.kms.blacklist.DELETE</name>
  <value>hdfs supergroup</value>
</property>

<property>
  <name>hadoop.kms.blacklist.ROLLOVER</name>
  <value>hdfs supergroup</value>
</property>

<property>
  <name>hadoop.kms.blacklist.GET</name>
  <value>*</value>
</property>

<property>
  <name>hadoop.kms.blacklist.GET_KEYS</name>
  <value></value>
</property>

<property>
  <name>hadoop.kms.blacklist.SET_KEY_MATERIAL</name>
  <value>*</value>
</property>

<property>
  <name>hadoop.kms.blacklist.DECRYPT_EEK</name>
  <value>hdfs supergroup</value>
</property>

<property>
  <name>keytrustee.kms.acl.UNDELETE</name>
  <value></value>
  <description>
    ACL that grants access to the UNDELETE operation on all keys.
    Only used by Key Trustee KMS.
  </description>
</property>

<property>
  <name>keytrustee.kms.acl.PURGE</name>
  <value></value>
  <description>
    ACL that grants access to the PURGE operation on all keys.
    Only used by Key Trustee KMS.
  </description>
</property>

<!--
  Default key ACLs that govern access to key operations for key-operation pairs that do not have a
  key-specific ACL already. Key-specific ACLs override the default key ACLs.

  The ACL value should be either a username or a username and group name separated by whitespace.

  An empty value for an ACL indicates that no user is granted access to that operation. A value
  of "*" (for the username or groupname) indicates that all users are granted access to that operation.
  Any operation for which there is no ACL will be treated as having an ACL with an empty value.
-->

<property>
  <name>default.key.acl.MANAGEMENT</name>
  <value></value>
  <description>
    Default ACL that grants access to the MANAGEMENT operation on all keys.
  </description>
</property>

<property>
  <name>default.key.acl.GENERATE_EEK</name>
  <value></value>
  <description>
    Default ACL that grants access to the GENERATE_EEK operation on all keys.
  </description>
</property>

<property>
  <name>default.key.acl.DECRYPT_EEK</name>
  <value></value>
  <description>
    Default ACL that grants access to the DECRYPT_EEK operation on all keys.
  </description>
</property>

<property>
  <name>default.key.acl.READ</name>
  <value></value>
  <description>
    Default ACL that grants access to the READ operation on all keys.
  </description>
</property>

<!--
  Whitelist key ACLs that grant access to key-specific operations. Any permissions granted here
  will be added to whatever permissions are granted by the specific key ACL or the default key ACL.
  Note that these whitelist ACLs grant access to operations on specific keys. If the operations
  are not allowed because of the KMS ACLs/blacklists, then the they will not be permitted,
  regardless of the whitelist settings.

  The ACL value should be either a username or a username and group name separated by whitespace.

  An empty value for an ACL indicates that no user is granted access to that operation. A value
  of "*" (for the username or groupname) indicates that all users are granted access to that
  operation. Any operation for which there is no ACL will
  be treated as having an ACL with an empty value.
-->

<property>
  <name>whitelist.key.acl.MANAGEMENT</name>
  <value>keyadmin keyadmingroup</value>
  <description>
    Whitelist ACL for MANAGEMENT operations for all keys.
  </description>
</property>

<property>
  <name>whitelist.key.acl.READ</name>
  <value>hdfs supergroup</value>
  <description>
    Whitelist ACL for READ operations for all keys.
  </description>
</property>

<property>
  <name>whitelist.key.acl.GENERATE_EEK</name>
  <value>hdfs supergroup</value>
  <description>
    Whitelist ACL for GENERATE_EEK operations for all keys.
  </description>
</property>

<property>
  <name>whitelist.key.acl.DECRYPT_EEK</name>
  <value>keyadmin keyadmingroup</value>
  <description>
    Whitelist ACL for DECRYPT_EEK operations for all keys.
  </description>
</property>

<!--
  Key ACLs that grant access to specific key operations. Any permissions granted here are added
  to whatever permissions are granted by the whitelists.
  The key ACL name should be key.acl.<keyname>.<OPERATION>.

  The ACL value should be either a username or a username and group name separated by whitespace.

  An empty value for an ACL indicates that no user is granted access to that operation. A value
  of "*" (for the username or group name) indicates that all users are granted access to that operation.
  Any key operation for which there is no ACL will default to the default ACL for the operation.

  Normally adding users or groups for a specific key and DECRYPT_EEK is sufficient to allow access
  to data protected with HDFS data at rest encryption.
-->

<!--
  The following ACLs are required for proper functioning of services. Cloudera Manager does not create keys or
  encryption zones; however, our best practices recommend encryption zones on certain directories.
  An assumption in these ACLs is that the user has followed the recommended naming scheme and named the keys
  according to documented best practices: "hive-key" for the Hive service,
  "hbase-key" for the Hbase service, etc. If the key names are different, then none of this will work
  out of the box, and you will need to edit these ACLs to match your key names.
-->

<property>
  <name>key.acl.hive-key.DECRYPT_EEK</name>
  <value>hive hive</value>
  <description>
    Gives the Hive user and the Hive group access to the key named "hive-key".
    This allows the Hive service to read and write files in /user/hive/.
    Also note that the Impala user ought to be a member of the Hive group to enjoy this same access.
  </description>
</property>

<property>
  <name>key.acl.hive-key.READ</name>
  <value>hive hive</value>
  <description>
    Required because Hive compares key strengths when joining tables.
  </description>
</property>

<property>
  <name>key.acl.hbase-key.DECRYPT_EEK</name>
  <value>hbase hbase</value>
  <description>
    Gives the hbase user and hbase group access to the key named "hbase-key".
    This allows the hbase service to read and write files in /hbase.
  </description>
</property>

<property>
  <name>key.acl.solr-key.DECRYPT_EEK</name>
  <value>solr solr</value>
  <description>
    Gives the solr user and solr group access to the key named "solr-key".
    This allows the solr service to read and write files in /solr.
  </description>
</property>

<property>
  <name>key.acl.mapred-key.DECRYPT_EEK</name>
  <value>mapred,yarn hadoop</value>
  <description>
    Gives the mapred user and mapred group access to the key named "mapred-key".
    This allows mapreduce to read and write files in /user/history.
    This is required by YARN.
  </description>
</property>

<property>
  <name>key.acl.hue-key.DECRYPT_EEK</name>
  <value>oozie,hue oozie,hue</value>
  <description>
    Gives the appropriate users and groups access to the key named "hue-key".
    This allows Hue and Oozie to read and write files in /user/hue.
    Oozie is required here because it will attempt to access workflows in /user/hue/oozie/workspaces.
  </description>
</property>



<!-- This example is required if there are encryption zones on user’s home directories. -->
<!--
<property>
  <name>key.acl.username-key.DECRYPT_EEK</name>
  <value>username username,hive,hbase,solr,oozie,hue,yarn</value>
  <description>
    Designed to be placed on a key that protects the EZ /user/username, and assumes that
    the key name is also "username-key"; this shows that a number of services can reach in
    to access data. Remove those that are not needed for your use case.
  </description>
</property>
-->

</configuration>