Host configuration

All hosts running Kerberos software, whether they are clients, application servers, or KDCs, can be configured using krb5.conf. Here we describe some of the behavior changes you might want to make.

Default realm

In the [libdefaults] section, the default_realm realm relation sets the default Kerberos realm. For example:

    default_realm = ATHENA.MIT.EDU

The default realm affects Kerberos behavior in the following ways:

  • When a principal name is parsed from text, the default realm is used if no @REALM component is specified.
  • The default realm affects login authorization as described below.
  • For programs which operate on a Kerberos database, the default realm is used to determine which database to operate on, unless the -r parameter is given to specify a realm.
  • A server program may use the default realm when looking up its key in a keytab file, if its realm is not determined by [domain_realm] configuration or by the server program itself.
  • If kinit is passed the -n flag, it requests anonymous tickets from the default realm.

In some situations, these uses of the default realm might conflict. For example, it might be desirable for principal name parsing to use one realm by default, but for login authorization to use a second realm. In this situation, the first realm can be configured as the default realm, and auth_to_local relations can be used as described below to use the second realm for login authorization.

Login authorization

If a host runs a Kerberos-enabled login service such as OpenSSH with GSSAPIAuthentication enabled, login authorization rules determine whether a Kerberos principal is allowed to access a local account.

By default, a Kerberos principal is allowed access to an account if its realm matches the default realm and its name matches the account name. (For historical reasons, access is also granted by default if the name has two components and the second component matches the default realm; for instance, alice/ATHENA.MIT.EDU@ATHENA.MIT.EDU is granted access to the alice account if ATHENA.MIT.EDU is the default realm.)

The simplest way to control local access is using .k5login files. To use these, place a .k5login file in the home directory of each account listing the principal names which should have login access to that account. If it is not desirable to use .k5login files located in account home directories, the k5login_directory relation in the [libdefaults] section can specify a directory containing one file per account uname.

By default, if a .k5login file is present, it controls authorization both positively and negatively–any principal name contained in the file is granted access and any other principal name is denied access, even if it would have had access if the .k5login file didn’t exist. The k5login_authoritative relation in the [libdefaults] section can be set to false to make .k5login files provide positive authorization only.

The auth_to_local relation in the [realms] section for the default realm can specify pattern-matching rules to control login authorization. For example, the following configuration allows access to principals from a different realm than the default realm:

        # Allow access to principals from OTHER.REALM.
        # [1:$1@$0] matches single-component principal names and creates
        # a selection string containing the principal name and realm.
        # (.*@OTHER\.REALM) matches against the selection string, so that
        # only principals in OTHER.REALM are matched.
        # s/@OTHER\.REALM$// removes the realm name, leaving behind the
        # principal name as the account name.
        auth_to_local = RULE:[1:$1@$0](.*@OTHER\.REALM)s/@OTHER\.REALM$//

        # Also allow principals from the default realm.  Omit this line
        # to only allow access to principals in OTHER.REALM.
        auth_to_local = DEFAULT

The auth_to_local_names subsection of the [realms] section for the default realm can specify explicit mappings from principal names to local accounts. The key used in this subsection is the principal name without realm, so it is only safe to use in a Kerberos environment with a single realm or a tightly controlled set of realms. An example use of auth_to_local_names might be:

        auth_to_local_names = {
            # Careful, these match principals in any realm!
            host/ = hostaccount
            fred = localfred

Local authorization behavior can also be modified using plugin modules; see Host-to-realm interface (hostrealm) for details.

Plugin module configuration

Many aspects of Kerberos behavior, such as client preauthentication and KDC service location, can be modified through the use of plugin modules. For most of these behaviors, you can use the [plugins] section of krb5.conf to register third-party modules, and to switch off registered or built-in modules.

A plugin module takes the form of a Unix shared object ( or Windows DLL (modname.dll). If you have installed a third-party plugin module and want to register it, you do so using the module relation in the appropriate subsection of the [plugins] section. The value for module must give the module name and the path to the module, separated by a colon. The module name will often be the same as the shared object’s name, but in unusual cases (such as a shared object which implements multiple modules for the same interface) it might not be. For example, to register a client preauthentication module named mypreauth installed at /path/to/, you could write:

    clpreauth = {
        module = mypreauth:/path/to/

Many of the pluggable behaviors in MIT krb5 contain built-in modules which can be switched off. You can disable a built-in module (or one you have registered) using the disable directive in the appropriate subsection of the [plugins] section. For example, to disable the use of .k5identity files to select credential caches, you could write:

    ccselect = {
        disable = k5identity

If you want to disable multiple modules, specify the disable directive multiple times, giving one module to disable each time.

Alternatively, you can explicitly specify which modules you want to be enabled for that behavior using the enable_only directive. For example, to make kadmind check password quality using only a module you have registered, and no other mechanism, you could write:

    pwqual = {
        module = mymodule:/path/to/
        enable_only = mymodule

Again, if you want to specify multiple modules, specify the enable_only directive multiple times, giving one module to enable each time.

Some Kerberos interfaces use different mechanisms to register plugin modules.

KDC location modules

For historical reasons, modules to control how KDC servers are located are registered simply by placing the shared object or DLL into the “libkrb5” subdirectory of the krb5 plugin directory, which defaults to LIBDIR/krb5/plugins. For example, Samba’s winbind krb5 locator plugin would be registered by placing its shared object in LIBDIR/krb5/plugins/libkrb5/

GSSAPI mechanism modules

GSSAPI mechanism modules are registered using the file SYSCONFDIR/gss/mech or configuration files in the SYSCONFDIR/gss/mech.d directory with a .conf suffix. Each line in these files has the form:

name  oid  pathname  [options]  <type>

Only the name, oid, and pathname are required. name is the mechanism name, which may be used for debugging or logging purposes. oid is the object identifier of the GSSAPI mechanism to be registered. pathname is a path to the module shared object or DLL. options (if present) are options provided to the plugin module, surrounded in square brackets. type (if present) can be used to indicate a special type of module. Currently the only special module type is “interposer”, for a module designed to intercept calls to other mechanisms.

If the environment variable GSS_MECH_CONFIG is set, its value is used as the sole mechanism configuration filename.

Configuration profile modules

A configuration profile module replaces the information source for krb5.conf itself. To use a profile module, begin krb5.conf with the line:


where PATHNAME is a path to the module shared object or DLL, and STRING is a string to provide to the module. The module will then take over, and the rest of krb5.conf will be ignored.