Configuring Kerberos for Authentication Using MapR Tickets
To use Kerberos to generate MapR tickets for users, enable Kerberos on the CLDB by creating a Kerberos identity on the Kerberos server used by the cluster and distributing that identity to the other CLDB nodes in the cluster.
MapR clusters do not provide Kerberos infrastructure. This section assumes you have a functioning Kerberos realm and your systems have the Kerberos client installed. The tips in this section assume a Linux-based Kerberos environment, and the specific commands for your environment may vary. Please consult with your Kerberos administrator for assistance.
Creating a Kerberos Identity for the CLDB
The CLDB requires a Kerberos server identity, but no other nodes do. By default, this identity takes the form mapr/<cluster name>. You can use
configure.sh or edit the
mapr-clusters.conf file to change this default. Use the following commands in a Linux-based Kerberos environment to set up the identity:
Copy the resulting
mapr.keytab file to the same location on every CLDB node. The
mapr.keytab file must be owned and readable only by the
mapr user. You can specify the location of the
mapr.keytab file in the
conf/mapr.login.conf file. The default location for
Updating the keytab File
You can use the
kadmin tool to update the server keys that are stored in the keytab file. Because the server tickets used to authenticate to the CLDB use the new keys immediately, you must copy the new keytab file to all the CLDB servers in the cluster immediately after updating the server keys.
To update the keytab file with a new key, run the following command:
The CLDB automatically detects changes to the keytab file on systems that use Java 7 or later. Systems that use Java 6 require a CLDB restart to detect changes to the keytab file.
After a Kerberos principal is created for the CLDB, that principal is added to the
mapr.keytab file, and the
mapr.keytab file is copied to all the CLDB servers, Kerberos user authentication is fully enabled for the MapR cluster.
configure.sh parameters are important for Kerberos:
-K|-kerberosEnable- lets the rest of the cluster know that Kerberos is enabled, so that clients can auto detect Kerberos tickets and use them to get MapR tickets.
-P "<cldbPrincipal>"- specifies the Kerberos instance which is used to form the CLDB Kerberos principal in the form of mapr/<instance-name>@<realm-name>. Enclose this value in quotes (").
configure.sh on each MapR cluster node, and each MapR client node that will communicate with one or more clusters. For more information on
configure.sh, see the configure.sh topic.
configure.sh on each node enters the Kerberos information into the local
clusters.conf file, so that the following command is all that is required for the client to access the cluster:
If you do not run
configure.sh on each node, the following two commands are required from the client:
Kerberos Command Summary
- kinit: Creates a Kerberos ticket. Prompts the user for userid and password. After validating, Kerberos creates a ticket file in /tmp that is owned by the user. Use the
-Roption to renew an existing ticket. Kerberos credentials expire in 8-10 hours. Expired credentials must be renewed or replaced. By default, tickets can be renewed for up to 24 hours.
- klist: Lists the contents of the user's ticket file.
- kdestroy: Destroys the contents of the user's ticket file. The user is no longer authenticated.
- kadmin: Used to administer Kerberos. The login for this command is implicitly
<userid>/admin, since administrator ids typically end in
- ktutil: Kerberos keytab maintenance utility. Used to combine, or alter Kerberos keytabs.
Java errors from Kerberos problems can be obscure and difficult to interpret. To see the Kerberos error messages, enable Kerberos debugging by adding these settings to your JVM:
You can also enable Kerberos debugging for the MapR-provided
maprcli and Hadoop clients by adding the following line to the
env.sh shell script:
Capture the Kerberos error to research the issue.
The following sections list common Kerberos error conditions:
Nodes often have multiple Java Virtual Machines (JVM) installed. The MapR
env.sh script automatically configures a JVM to use. To change the automatically configured JVM, set the value of the
JAVA_HOME environment variable in the
Incorrect Server Name
The following error message is caused by an incorrect CLDB server name in the
mapr.login.conf file. The error message mentions passwords, but the error condition is unrelated to password authentication.
Invalid or missing keytab file
keytab file must be consistent with the key versions of the Kerberos principal. The following example shows an inconsistent keytab file:
Note that the key versions in the Kerberos principal
/realm1 are 15, and the versions in the
keytab file are 14. This mismatch can result in errors about missing keys or mismatched encryption.
Incompatible encryption on Java runtime
Incompatible cryptography between the KDC and the JDK results in failed handshakes, leading to errors similar to the following:
With debugging active, the following message is displayed:
This debug message indicates that the problem is an unsupported key type.
Incompatible encryption errors can occur due to a
keytab file that is not present or contains outdated keys.
Be sure to update the Java jurisdiction policy file. Jurisdiction policy files are available from Oracle.
A persistent encryption incompatibility problem may require you to edit the
krb5.conf file to ensure compatible algorithms between Java and Kerberos.
Bugs in Java
The following error occurs in Java version 1.6.0_25. Upgrade to 1.6.0_45 to resolve the error.
Kerberos and PAM validation
Standard Kerberos implementations are predicated on access to elevated user privileges that are not present on secure MapR clusters. In a MapR cluster, the MCS console and other components call PAM as an ordinary user process. This discrepancy in expected and actual privileges can cause a variety of obscure file permisson errors. Since different Kerberos PAM modules are available, error reports can vary.
To diagnose this issue, attempt starting the MCS as the root user, or clear out the
/tmp folder. If there are no problems when starting MCS as root, or if clearing out the
/tmp folder enables a single login before errors appear again, the problem may lie in the Kerberos PAM configuration.
To resolve this condition, prevent Kerberos from creating a ticket file. MapR security does not use Kerberos tickets. The Kerberos KDC is used to validate passwords. Typically the configuration file for PAM is in the
/etc/pam.d directory. See the documentation for your specific Kerberos PAM module for more information.
Disabling Replay Detection for Kerberos Authentication
You can set an option in
mapr-clusters.conf to disable replay detection for Kerberos runtime authentication.
By default, this parameter is set to
false, meaning that MapR clients enable Kerberos replay detection. Typically, replay detection is enabled to prevent potential attacks such as the replay of Kerberos packets or multiple login attempts with the same user ID. Set this parameter to true only if you want MapR clients not to enforce this detection.
This parameter applies when users attempt an implicit or explicit maprlogin, such as by using the
maprlogin kerberos command or by submitting jobs and other operations with
kerberosEnable=true set in the
This parameter is used when applications connect to the cluster using Kerberos;
mapr-clusters.conf only needs to be updated when it is used by such applications. If all Kerberos access to the cluster is from clients outside the cluster, only the
mapr-clusters.conf file on those client machines has to be updated. If Kerberos is used from applications running on the cluster,
mapr-clusters.conf should be updated there as well.