Realm configuration decisions

Before installing Kerberos V5, it is necessary to consider the following issues:

  • The name of your Kerberos realm (or the name of each realm, if you need more than one).
  • How you will assign your hostnames to Kerberos realms.
  • Which ports your KDC and and kadmind services will use, if they will not be using the default ports.
  • How many slave KDCs you need and where they should be located.
  • The hostnames of your master and slave KDCs.
  • How frequently you will propagate the database from the master KDC to the slave KDCs.

Realm name

Although your Kerberos realm can be any ASCII string, convention is to make it the same as your domain name, in upper-case letters.

For example, hosts in the domain example.com would be in the Kerberos realm:

EXAMPLE.COM

If you need multiple Kerberos realms, MIT recommends that you use descriptive names which end with your domain name, such as:

BOSTON.EXAMPLE.COM
HOUSTON.EXAMPLE.COM

Mapping hostnames onto Kerberos realms

Mapping hostnames onto Kerberos realms is done in one of three ways.

The first mechanism works through a set of rules in the [domain_realm] section of krb5.conf. You can specify mappings for an entire domain or on a per-hostname basis. Typically you would do this by specifying the mappings for a given domain or subdomain and listing the exceptions.

The second mechanism is to use KDC host-based service referrals. With this method, the KDC’s krb5.conf has a full [domain_realm] mapping for hosts, but the clients do not, or have mappings for only a subset of the hosts they might contact. When a client needs to contact a server host for which it has no mapping, it will ask the client realm’s KDC for the service ticket, and will receive a referral to the appropriate service realm.

To use referrals, clients must be running MIT krb5 1.6 or later, and the KDC must be running MIT krb5 1.7 or later. The host_based_services and no_host_referral variables in the [realms] section of kdc.conf can be used to fine-tune referral behavior on the KDC.

It is also possible for clients to use DNS TXT records, if dns_lookup_realm is enabled in krb5.conf. Such lookups are disabled by default because DNS is an insecure protocol and security holes could result if DNS records are spoofed. If enabled, the client will try to look up a TXT record formed by prepending the prefix _kerberos to the hostname in question. If that record is not found, the client will attempt a lookup by prepending _kerberos to the host’s domain name, then its parent domain, up to the top-level domain. For the hostname boston.engineering.example.com, the names looked up would be:

_kerberos.boston.engineering.example.com
_kerberos.engineering.example.com
_kerberos.example.com
_kerberos.com

The value of the first TXT record found is taken as the realm name.

Even if you do not choose to use this mechanism within your site, you may wish to set it up anyway, for use when interacting with other sites.

Ports for the KDC and admin services

The default ports used by Kerberos are port 88 for the KDC and port 749 for the admin server. You can, however, choose to run on other ports, as long as they are specified in each host’s krb5.conf files or in DNS SRV records, and the kdc.conf file on each KDC. For a more thorough treatment of port numbers used by the Kerberos V5 programs, refer to the Configuring your firewall to work with Kerberos V5.

Slave KDCs

Slave KDCs provide an additional source of Kerberos ticket-granting services in the event of inaccessibility of the master KDC. The number of slave KDCs you need and the decision of where to place them, both physically and logically, depends on the specifics of your network.

Kerberos authentication requires that each client be able to contact a KDC. Therefore, you need to anticipate any likely reason a KDC might be unavailable and have a slave KDC to take up the slack.

Some considerations include:

  • Have at least one slave KDC as a backup, for when the master KDC is down, is being upgraded, or is otherwise unavailable.
  • If your network is split such that a network outage is likely to cause a network partition (some segment or segments of the network to become cut off or isolated from other segments), have a slave KDC accessible to each segment.
  • If possible, have at least one slave KDC in a different building from the master, in case of power outages, fires, or other localized disasters.

Hostnames for KDCs

MIT recommends that your KDCs have a predefined set of CNAME records (DNS hostname aliases), such as kerberos for the master KDC and kerberos-1, kerberos-2, … for the slave KDCs. This way, if you need to swap a machine, you only need to change a DNS entry, rather than having to change hostnames.

As of MIT krb5 1.4, clients can locate a realm’s KDCs through DNS using SRV records (RFC 2782), assuming the Kerberos realm name is also a DNS domain name. These records indicate the hostname and port number to contact for that service, optionally with weighting and prioritization. The domain name used in the SRV record name is the realm name. Several different Kerberos-related service names are used:

_kerberos._udp
This is for contacting any KDC by UDP. This entry will be used the most often. Normally you should list port 88 on each of your KDCs.
_kerberos._tcp
This is for contacting any KDC by TCP. The MIT KDC by default will not listen on any TCP ports, so unless you’ve changed the configuration or you’re running another KDC implementation, you should leave this unspecified. If you do enable TCP support, normally you should use port 88.
_kerberos-master._udp

This entry should refer to those KDCs, if any, that will immediately see password changes to the Kerberos database. If a user is logging in and the password appears to be incorrect, the client will retry with the master KDC before failing with an “incorrect password” error given.

If you have only one KDC, or for whatever reason there is no accessible KDC that would get database changes faster than the others, you do not need to define this entry.

_kerberos-adm._tcp
This should list port 749 on your master KDC. Support for it is not complete at this time, but it will eventually be used by the kadmin program and related utilities. For now, you will also need the admin_server variable in krb5.conf.
_kpasswd._udp
This should list port 464 on your master KDC. It is used when a user changes her password. If this entry is not defined but a _kerberos-adm._tcp entry is defined, the client will use the _kerberos-adm._tcp entry with the port number changed to 749.

The DNS SRV specification requires that the hostnames listed be the canonical names, not aliases. So, for example, you might include the following records in your (BIND-style) zone file:

$ORIGIN foobar.com.
_kerberos               TXT       "FOOBAR.COM"
kerberos                CNAME     daisy
kerberos-1              CNAME     use-the-force-luke
kerberos-2              CNAME     bunny-rabbit
_kerberos._udp          SRV       0 0 88 daisy
                        SRV       0 0 88 use-the-force-luke
                        SRV       0 0 88 bunny-rabbit
_kerberos-master._udp   SRV       0 0 88 daisy
_kerberos-adm._tcp      SRV       0 0 749 daisy
_kpasswd._udp           SRV       0 0 464 daisy

Clients can also be configured with the explicit location of services using the kdc, master_kdc, admin_server, and kpasswd_server variables in the [realms] section of krb5.conf. Even if some clients will be configured with explicit server locations, providing SRV records will still benefit unconfigured clients, and be useful for other sites.

KDC Discovery

As of MIT krb5 1.15, clients can also locate KDCs in DNS through URI records (RFC 7553). Limitations with the SRV record format may result in extra DNS queries in situations where a client must failover to other transport types, or find a master server. The URI record can convey more information about a realm’s KDCs with a single query.

The client performs a query for the following URI records:

  • _kerberos.REALM for finding KDCs.
  • _kerberos-adm.REALM for finding kadmin services.
  • _kpasswd.REALM for finding password services.

The URI record includes a priority, weight, and a URI string that consists of case-insensitive colon separated fields, in the form scheme:[flags]:transport:residual.

  • scheme defines the registered URI type. It should always be krb5srv.
  • flags contains zero or more flag characters. Currently the only valid flag is m, which indicates that the record is for a master server.
  • transport defines the transport type of the residual URL or address. Accepted values are tcp, udp, or kkdcp for the MS-KKDCP type.
  • residual contains the hostname, IP address, or URL to be contacted using the specified transport, with an optional port extension. The MS-KKDCP transport type uses a HTTPS URL, and can include a port and/or path extension.

An example of URI records in a zone file:

_kerberos.EXAMPLE.COM  URI  10 1 krb5srv:m:tcp:kdc1.example.com
                       URI  20 1 krb5srv:m:udp:kdc2.example.com:89
                       URI  40 1 krb5srv::udp:10.10.0.23
                       URI  30 1 krb5srv::kkdcp:https://proxy:89/auth

URI lookups are enabled by default, and can be disabled by setting dns_uri_lookup in the [libdefaults] section of krb5.conf to False. When enabled, URI lookups take precedence over SRV lookups, falling back to SRV lookups if no URI records are found.

Database propagation

The Kerberos database resides on the master KDC, and must be propagated regularly (usually by a cron job) to the slave KDCs. In deciding how frequently the propagation should happen, you will need to balance the amount of time the propagation takes against the maximum reasonable amount of time a user should have to wait for a password change to take effect.

If the propagation time is longer than this maximum reasonable time (e.g., you have a particularly large database, you have a lot of slaves, or you experience frequent network delays), you may wish to cut down on your propagation delay by performing the propagation in parallel. To do this, have the master KDC propagate the database to one set of slaves, and then have each of these slaves propagate the database to additional slaves.

See also Incremental database propagation