Next: , Previous: (dir), Up: (dir)


This manual is last updated $Date: 2007-12-06 18:26:30 +0100 (Thu, 06 Dec 2007) $ for version 1.1 of Heimdal.

--- The Detailed Node Listing ---

Setting up a realm


Authentication modules

Kerberos 4 issues

Windows 2000 compatability

Programming with Kerberos

Next: , Previous: Top, Up: Top

1 Introduction

What is Heimdal?

Heimdal is a free implementation of Kerberos 5. The goals are to:


Heimdal has the following features (this does not mean any of this works):

Bug reports

If you find bugs in this software, make sure it is a genuine bug and not just a part of the code that isn't implemented.

Bug reports should be sent to heimdal-bugs@h5l.org. Please include information on what machine and operating system (including version) you are running, what you are trying to do, what happens, what you think should have happened, an example for us to repeat, the output you get when trying the example, and a patch for the problem if you have one. Please make any patches with diff -u or diff -c.

Suggestions, comments and other non bug reports are also welcome.

Mailing list

There are two mailing lists with talk about Heimdal. heimdal-announce@sics.se is a low-volume announcement list, while heimdal-discuss@sics.se is for general discussion. Send a message to majordomo@sics.se to subscribe.

Heimdal source code, binaries and the manual

The source code for heimdal, links to binaries and the manual (this document) can be found on our web-page at http://www.pdc.kth.se/heimdal/.

Next: , Previous: Introduction, Up: Top

2 What is Kerberos?

Now this Cerberus had three heads of dogs, the tail of a dragon, and on his back the heads of all sorts of snakes. — Pseudo-Apollodorus Library 2.5.12

Kerberos is a system for authenticating users and services on a network. It is built upon the assumption that the network is “unsafe”. For example, data sent over the network can be eavesdropped and altered, and addresses can also be faked. Therefore they cannot be used for authentication purposes. Kerberos is a trusted third-party service. That means that there is a third party (the kerberos server) that is trusted by all the entities on the network (users and services, usually called principals). All principals share a secret password (or key) with the kerberos server and this enables principals to verify that the messages from the kerberos server are authentic. Thus trusting the kerberos server, users and services can authenticate each other.

2.1 Basic mechanism

Note This discussion is about Kerberos version 4, but version 5 works similarly.

In Kerberos, principals use tickets to prove that they are who they claim to be. In the following example, A is the initiator of the authentication exchange, usually a user, and B is the service that A wishes to use.

To obtain a ticket for a specific service, A sends a ticket request to the kerberos server. The request contains A's and B's names (along with some other fields). The kerberos server checks that both A and B are valid principals.

Having verified the validity of the principals, it creates a packet containing A's and B's names, A's network address (Aaddr), the current time (tissue), the lifetime of the ticket (life), and a secret session key (KAB). This packet is encrypted with B's secret key (KB). The actual ticket (TAB) looks like this: ({A, B, Aaddr, tissue, life, KAB}KB).

The reply to A consists of the ticket (TAB), B's name, the current time, the lifetime of the ticket, and the session key, all encrypted in A's secret key ({B, tissue, life, KAB, TAB}KA). A decrypts the reply and retains it for later use.

Before sending a message to B, A creates an authenticator consisting of A's name, A's address, the current time, and a “checksum” chosen by A, all encrypted with the secret session key ({A, Aaddr, tcurrent, checksum}KAB). This is sent together with the ticket received from the kerberos server to B. Upon reception, B decrypts the ticket using B's secret key. Since the ticket contains the session key that the authenticator was encrypted with, B can now also decrypt the authenticator. To verify that A really is A, B now has to compare the contents of the ticket with that of the authenticator. If everything matches, B now considers A as properly authenticated.

2.2 Different attacks

Impersonating A

An impostor, C could steal the authenticator and the ticket as it is transmitted across the network, and use them to impersonate A. The address in the ticket and the authenticator was added to make it more difficult to perform this attack. To succeed C will have to either use the same machine as A or fake the source addresses of the packets. By including the time stamp in the authenticator, C does not have much time in which to mount the attack.

Impersonating B

C can hijack B's network address, and when A sends her credentials, C just pretend to verify them. C can't be sure that she is talking to A.

2.3 Defence strategies

It would be possible to add a replay cache to the server side. The idea is to save the authenticators sent during the last few minutes, so that B can detect when someone is trying to retransmit an already used message. This is somewhat impractical (mostly regarding efficiency), and is not part of Kerberos 4; MIT Kerberos 5 contains it.

To authenticate B, A might request that B sends something back that proves that B has access to the session key. An example of this is the checksum that A sent as part of the authenticator. One typical procedure is to add one to the checksum, encrypt it with the session key and send it back to A. This is called mutual authentication.

The session key can also be used to add cryptographic checksums to the messages sent between A and B (known as message integrity). Encryption can also be added (message confidentiality). This is probably the best approach in all cases.

2.4 Further reading

The original paper on Kerberos from 1988 is Kerberos: An Authentication Service for Open Network Systems, by Jennifer Steiner, Clifford Neuman and Jeffrey I. Schiller.

A less technical description can be found in Designing an Authentication System: a Dialogue in Four Scenes by Bill Bryant, also from 1988.

These documents can be found on our web-page at http://www.pdc.kth.se/kth-krb/.

Next: , Previous: What is Kerberos?, Up: Top

3 Building and Installing

Heimdal uses GNU Autoconf to configure for specific hosts, and GNU Automake to manage makefiles. If this is new to you, the short instruction is to run the configure script in the top level directory, and when that finishes make.

If you want to build the distribution in a different directory from the source directory, you will need a make that implements VPATH correctly, such as GNU make.

You will need to build the distribution:

When everything is built, you can install by doing make install. The default location for installation is /usr/heimdal, but this can be changed by running configure with `--prefix=/some/other/place'.

If you need to change the default behaviour, configure understands the following options:

DB is preferred before NDBM, but if you for some reason want to use NDBM instead, you can use this option.
Gives the location of Kerberos 4 libraries and headers. This enables Kerberos 4 support in the applications (telnet, rsh, popper, etc) and the KDC. It is automatically found if present under /usr/athena. If you keep libraries and headers in different places, you can instead give the path to each with the --with-krb4-lib=dir, and --with-krb4-include=dir options.

You will need a fairly recent version of our Kerberos 4 distribution for rshd and popper to support version 4 clients.

Enables support for getting DCE credentials and tokens. See the README files in appl/dceutils for more information.
By default some of the application programs will build with support for one-time passwords (OTP). Use this option to disable that support.
Enable some C2 support for OSF/Digital Unix/Tru64. Use this option if you are running your OSF operating system in C2 mode.
Gives the path for the GNU Readline library, which will be used in some programs. If no readline library is found, the (simpler) editline library will be used instead.
Enables hesiod support in push.
Add support for using netinfo to lookup configuration information. Probably only useful (and working) on NextStep/Mac OS X.
Disable the IPv6 support.
Compile Heimdal with support for storing the database in LDAP. Requires OpenLDAP http://www.openldap.org. See http://www.padl.com/Research/Heimdal.html for more information.
Normally, the build process will figure out by itself if the machine is big or little endian. It might fail in some cases when cross-compiling. If it does fail to figure it out, use the relevant of these two options.
On Irix there are three different ABIs that can be used (`32', `n32', or `64'). This option allows you to override the automatic selection.
Do not use the mmap system call. Normally, configure detects if there is a working mmap and it is only used if there is one. Only try this option if it fails to work anyhow.

Next: , Previous: Building and Installing, Up: Top

4 Setting up a realm

A realm is an administrative domain. The name of a Kerberos realm is usually the Internet domain name in uppercase. Call your realm the same as your Internet domain name if you do not have strong reasons for not doing so. It will make life easier for you and everyone else.

Next: , Previous: Setting up a realm, Up: Setting up a realm

4.1 Configuration file

To setup a realm you will first have to create a configuration file: /etc/krb5.conf. The krb5.conf file can contain many configuration options, some of which are described here.

There is a sample krb5.conf supplied with the distribution.

The configuration file is a hierarchical structure consisting of sections, each containing a list of bindings (either variable assignments or subsections). A section starts with `[`section-name']'. A binding consists of a left hand side, an equal sign (`=') and a right hand side (the left hand side tag must be separated from the equal sign with some whitespace). Subsections have a `{' as the first non-whitespace character after the equal sign. All other bindings are treated as variable assignments. The value of a variable extends to the end of the line.

             a-subsection = {
                     var = value1
                     other-var = value with {}
                     sub-sub-section = {
                             var = 123
             var = some other value
             var = yet another value

In this manual, names of sections and bindings will be given as strings separated by slashes (`/'). The `other-var' variable will thus be `section1/a-subsection/other-var'.

For in-depth information about the contents of the configuration file, refer to the krb5.conf manual page. Some of the more important sections are briefly described here.

The `libdefaults' section contains a list of library configuration parameters, such as the default realm and the timeout for KDC responses. The `realms' section contains information about specific realms, such as where they hide their KDC. This section serves the same purpose as the Kerberos 4 krb.conf file, but can contain more information. Finally the `domain_realm' section contains a list of mappings from domains to realms, equivalent to the Kerberos 4 krb.realms file.

To continue with the realm setup, you will have to create a configuration file, with contents similar to the following.

             default_realm = MY.REALM
             MY.REALM = {
                     kdc = my.kdc my.slave.kdc
                     kdc = my.third.kdc
             .my.domain = MY.REALM

If you use a realm name equal to your domain name, you can omit the `libdefaults', and `domain_realm', sections. If you have a DNS SRV-record for your realm, or your Kerberos server has DNS CNAME `kerberos.my.realm', you can omit the `realms' section too.

Next: , Previous: Configuration file, Up: Setting up a realm

4.2 Creating the database

The database library will look for the database in the directory /var/heimdal, so you should probably create that directory. Make sure the directory has restrictive permissions.

     # mkdir /var/heimdal

The keys of all the principals are stored in the database. If you choose to, these can be encrypted with a master key. You do not have to remember this key (or password), but just to enter it once and it will be stored in a file (/var/heimdal/m-key). If you want to have a master key, run `kstash' to create this master key:

     # kstash
     Master key:
     Verifying password - Master key:

If you want to generate a random master key you can use the --random-key flag to kstash. This will make sure you have a good key on which attackers can't do a dictionary attack.

If you have a master key, make sure you make a backup of your master key file; without it backups of the database are of no use.

To initialise the database use the kadmin program, with the -l option (to enable local database mode). First issue a init MY.REALM command. This will create the database and insert default principals for that realm. You can have more than one realm in one database, so `init' does not destroy any old database.

Before creating the database, `init' will ask you some questions about maximum ticket lifetimes.

After creating the database you should probably add yourself to it. You do this with the `add' command. It takes as argument the name of a principal. The principal should contain a realm, so if you haven't set up a default realm, you will need to explicitly include the realm.

     # kadmin -l
     kadmin> init MY.REALM
     Realm max ticket life [unlimited]:
     Realm max renewable ticket life [unlimited]:
     kadmin> add me
     Max ticket life [unlimited]:
     Max renewable life [unlimited]:
     Attributes []:
     Verifying password - Password:

Now start the KDC and try getting a ticket.

     # kdc &
     # kinit me
     me@MY.REALMS's Password:
     # klist
     Credentials cache: /tmp/krb5cc_0
             Principal: me@MY.REALM
       Issued           Expires          Principal
     Aug 25 07:25:55  Aug 25 17:25:55  krbtgt/MY.REALM@MY.REALM

If you are curious you can use the `dump' command to list all the entries in the database. It should look something similar to the following example (note that the entries here are truncated for typographical reasons):

     kadmin> dump
     me@MY.REALM 1:0:1:0b01d3cb7c293b57:-:0:7:8aec316b9d1629e3baf8 ...
     kadmin/admin@MY.REALM 1:0:1:e5c8a2675b37a443:-:0:7:cb913ebf85 ...
     krbtgt/MY.REALM@MY.REALM 1:0:1:52b53b61c875ce16:-:0:7:c8943be ...
     kadmin/changepw@MY.REALM 1:0:1:f48c8af2b340e9fb:-:0:7:e3e6088 ...

Next: , Previous: Creating the database, Up: Setting up a realm

4.3 Modifying the database

All modifications of principals are done with with kadmin.

A principal has several attributes and lifetimes associated with it.

Principals are added, renamed, modified, and deleted with the kadmin commands `add', `rename', `modify', `delete'. Both interactive editing and command line flags can be used (use –help to list the available options).

There are different kinds of types for the fields in the database; attributes, absolute time times and relative times.

4.3.1 Attributes

When doing interactive editing, attributes are listed with `?'.

The attributes are given in a comma (`,') separated list. Attributes are removed from the list by prefixing them with `-'.

     kadmin> modify me
     Max ticket life [1 day]:
     Max renewable life [1 week]:
     Principal expiration time [never]:
     Password expiration time [never]:
     Attributes [disallow-renewable]: requires-pre-auth,-disallow-renewable
     kadmin> get me
                 Principal: me@MY.REALM
                Attributes: requires-pre-auth

4.3.2 Absolute times

The format for absolute times are any of the following:

     YYYY-mm-dd HH:MM:SS

4.3.3 Relative times

The format for relative times are any of the following combined:

     N year
     M month
     O day
     P hour
     Q minute
     R second

Next: , Previous: Modifying the database, Up: Setting up a realm

4.4 Checking the setup

There are two tools that can check the consistency of the Kerberos configuration file and the Kerberos database.

The Kerberos configuration file is checked using verify_krb5_conf. The tool checks for common errors, but commonly there are several uncommon configuration entries that are never added to the tool and thus generates “unknown entry” warnings. This is usually nothing to worry about.

The database check is built into the kadmin tool. It will check for common configuration error that will cause problems later. Common check are for existence and flags on important principals. The database check by run by the following command :

     kadmin check REALM.EXAMPLE.ORG

Next: , Previous: Checking the setup, Up: Setting up a realm

4.5 keytabs

To extract a service ticket from the database and put it in a keytab, you need to first create the principal in the database with `ank' (using the --random-key flag to get a random key) and then extract it with `ext_keytab'.

     kadmin> add --random-key host/my.host.name
     Max ticket life [unlimited]:
     Max renewable life [unlimited]:
     Attributes []:
     kadmin> ext host/my.host.name
     kadmin> exit
     # ktutil list
     Version  Type             Principal
          1   des-cbc-md5      host/my.host.name@MY.REALM
          1   des-cbc-md4      host/my.host.name@MY.REALM
          1   des-cbc-crc      host/my.host.name@MY.REALM
          1   des3-cbc-sha1    host/my.host.name@MY.REALM

Next: , Previous: keytabs, Up: Setting up a realm

4.6 Serving Kerberos 4/524/kaserver

Heimdal can be configured to support 524, Kerberos 4 or kaserver. All these services are turned off by default. Kerberos 4 is always supported by the KDC, but the Kerberos 4 client support also depends on Kerberos 4 support having been included at compile-time, using --with-krb4=dir.

4.6.1 524

524 is a service that allows the KDC to convert Kerberos 5 tickets to Kerberos 4 tickets for backward compatibility. See also Using 2b tokens with AFS in See Things in search for a better place.

524 can be turned on by adding this to the configuration file

     	enable-524 = yes

4.6.2 Kerberos 4

Kerberos 4 is the predecessor to to Kerberos 5. It only supports single DES. You should only enable Kerberos 4 support if you have needs for compatibility with an installed base of Kerberos 4 clients/servers.

Kerberos 4 can be turned on by adding this to the configuration file

     	enable-kerberos4 = yes

4.6.3 kaserver

Kaserver is a Kerberos 4 that is used in AFS. The protocol has some extra features over plain Kerberos 4, but like Kerberos 4, only uses single DES.

You should only enable Kaserver support if you have needs for compatibility with an installed base of AFS machines.

Kaserver can be turned on by adding this to the configuration file

     	enable-kaserver = yes

Next: , Previous: Serving Kerberos 4/524/kaserver, Up: Setting up a realm

4.7 Remote administration

The administration server, kadmind, can be started by inetd (which isn't recommended) or run as a normal daemon. If you want to start it from inetd you should add a line similar to the one below to your /etc/inetd.conf.

     kerberos-adm stream     tcp     nowait  root /usr/heimdal/libexec/kadmind kadmind

You might need to add `kerberos-adm' to your /etc/services as `749/tcp'.

Access to the administration server is controlled by an ACL file, (default /var/heimdal/kadmind.acl.) The file has the following syntax:

     principal       [priv1,priv2,...]       [glob-pattern]

The matching is from top to bottom for matching principals (and if given, glob-pattern). When there is a match, the access rights of that line are applied.

The privileges you can assign to a principal are: `add', `change-password' (or `cpw' for short), `delete', `get', `list', and `modify', or the special privilege `all'. All of these roughly correspond to the different commands in kadmin.

If a glob-pattern is given on a line, it restricts the access rights for the principal to only apply for subjects that match the pattern. The patterns are of the same type as those used in shell globbing, see fnmatch(3).

In the example below `lha/admin' can change every principal in the database. `jimmy/admin' can only modify principals that belong to the realm `E.KTH.SE'. `mille/admin' is working at the help desk, so he should only be able to change the passwords for single component principals (ordinary users). He will not be able to change any `/admin' principal.

     lha/admin@E.KTH.SE	all
     jimmy/admin@E.KTH.SE	all		*@E.KTH.SE
     jimmy/admin@E.KTH.SE	all		*/*@E.KTH.SE
     mille/admin@E.KTH.SE	change-password	*@E.KTH.SE

Next: , Previous: Remote administration, Up: Setting up a realm

4.8 Password changing

To allow users to change their passwords, you should run kpasswdd. It is not run from inetd.

You might need to add `kpasswd' to your /etc/services as `464/udp'.

4.8.1 Password quality assurance

It is important that users have good passwords, both to make it harder to guess them and to avoid off-line attacks (although pre-authentication provides some defence against off-line attacks). To ensure that the users choose good passwords, you can enable password quality controls in kpasswdd and kadmind. The controls themselves are done in a shared library or an external program that is used by kpasswdd. To configure in these controls, add lines similar to the following to your /etc/krb5.conf:

     	policies = external-check builtin:minimum-length module:policyname
     	external_program = /bin/false
     	policy_libraries = library1.so library2.so

In `[password_quality]policies' the module name is optional if the policy name is unique in all modules (members of `policy_libraries').

The built-in polices are

If you want to write your own shared object to check password policies, see the manual page kadm5_pwcheck(3).

Code for a password quality checking function that uses the cracklib library can be found in lib/kadm5/sample_password_check.c in the source code distribution. It requires that the cracklib library be built with the patch available at ftp://ftp.pdc.kth.se/pub/krb/src/cracklib.patch.

A sample policy external program is included in lib/kadm5/check-cracklib.pl.

If no password quality checking function is configured, the only check performed is that the password is at least six characters long.

To check the password policy settings, use the command password-quality in kadmin program. The password verification is only performed locally, on the client. It may be convenient to set the environment variable `KRB5_CONFIG' to point to a test version of krb5.conf while you're testing the `[password_quality]' stanza that way.

Next: , Previous: Password changing, Up: Setting up a realm

4.9 Testing clients and servers

Now you should be able to run all the clients and servers. Refer to the appropriate man pages for information on how to use them.

Next: , Previous: Testing clients and servers, Up: Setting up a realm

4.10 Slave servers, Incremental propagation, Testing clients and servers, Setting up a realm

It is desirable to have at least one backup (slave) server in case the master server fails. It is possible to have any number of such slave servers but more than three usually doesn't buy much more redundancy.

All Kerberos servers for a realm must have the same database so that they present the same service to the users. The hprop program, running on the master, will propagate the database to the slaves, running hpropd processes.

Every slave needs a database directory, the master key (if it was used for the database) and a keytab with the principal `hprop/hostname'. Add the principal with the ktutil command and start hpropd, as follows:

     slave# ktutil get -p foo/admin hprop/`hostname`
     slave# mkdir /var/heimdal
     slave# hpropd

The master will use the principal `kadmin/hprop' to authenticate to the slaves. This principal should be added when running kadmin -l init but if you do not have it in your database for whatever reason, please add it with kadmin -l add.

Then run hprop on the master:

     master# hprop slave

This was just an hands-on example to make sure that everything was working properly. Doing it manually is of course the wrong way, and to automate this you will want to start hpropd from inetd on the slave(s) and regularly run hprop on the master to regularly propagate the database. Starting the propagation once an hour from cron is probably a good idea.

Next: , Previous: Slave Servers, Up: Setting up a realm

4.11 Incremental propagation

There is also a newer, and still somewhat experimental, mechanism for doing incremental propagation in Heimdal. Instead of sending the whole database regularly, it sends the changes as they happen on the master to the slaves. The master keeps track of all the changes by assigning a version number to every change to the database. The slaves know which was the latest version they saw and in this way it can be determined if they are in sync or not. A log of all the changes is kept on the master, and when a slave is at an older version than the oldest one in the log, the whole database has to be sent.

Protocol-wise, all the slaves connect to the master and as a greeting tell it the latest version that they have (`IHAVE' message). The master then responds by sending all the changes between that version and the current version at the master (a series of `FORYOU' messages) or the whole database in a `TELLYOUEVERYTHING' message. There is also a keep-alive protocol that makes sure all slaves are up and running.

4.11.1 Configuring incremental propagation

The program that runs on the master is ipropd-master and all clients run ipropd-slave.

Create the file /var/heimdal/slaves on the master containing all the slaves that the database should be propagated to. Each line contains the full name of the principal (for example `iprop/hemligare.foo.se@FOO.SE').

You should already have `iprop/tcp' defined as 2121, in your /etc/services. Otherwise, or if you need to use a different port for some peculiar reason, you can use the --port option. This is useful when you have multiple realms to distribute from one server.

Then you need to create those principals that you added in the configuration file. Create one `iprop/hostname' for the master and for every slave.

     master# /usr/heimdal/sbin/ktutil get iprop/`hostname`

The next step is to start the ipropd-master process on the master server. The ipropd-master listens on the UNIX domain socket /var/heimdal/signal to know when changes have been made to the database so they can be propagated to the slaves. There is also a safety feature of testing the version number regularly (every 30 seconds) to see if it has been modified by some means that do not raise this signal. Then, start ipropd-slave on all the slaves:

     master# /usr/heimdal/libexec/ipropd-master &
     slave#  /usr/heimdal/libexec/ipropd-slave master &

To manage the iprop log file you should use the iprop-log command. With it you can dump, truncate and replay the logfile.

Next: , Previous: Incremental propagation, Up: Setting up a realm

4.12 Encryption types and salting

The encryption types that the KDC is going to assign by default is possible to change. Since the keys used for user authentication is salted the encryption types are described together with the salt strings.

Salting is used to make it harder to pre-calculate all possible keys. Using a salt increases the search space to make it almost impossible to pre-calculate all keys. Salting is the process of mixing a public string (the salt) with the password, then sending it through an encryption type specific string-to-key function that will output the fixed size encryption key.

In Kerberos 5 the salt is determined by the encryption type, except in some special cases.

In des there is the Kerberos 4 salt (none at all) or the afs-salt (using the cell (realm in AFS lingo)).

In arcfour (the encryption type that Microsoft Windows 2000 uses) there is no salt. This is to be compatible with NTLM keys in Windows NT 4.

[kadmin]default_keys in krb5.conf controls what salting to use.

The syntax of [kadmin]default_keys is `[etype:]salt-type[:salt-string]'. `etype' is the encryption type (des-cbc-crc, arcfour-hmac-md5, aes256-cts-hmac-sha1-96), salt-type is the type of salt (pw-salt or afs3-salt), and the salt-string is the string that will be used as salt (remember that if the salt is appended/prepended, the empty salt "" is the same thing as no salt at all).

Common types of salting include

Next: , Previous: Encryption types and salting, Up: Setting up a realm

4.13 Cross realm

Suppose you reside in the realm `MY.REALM', how do you authenticate to a server in `OTHER.REALM'? Having valid tickets in `MY.REALM' allows you to communicate with Kerberised services in that realm. However, the computer in the other realm does not have a secret key shared with the Kerberos server in your realm.

It is possible to share keys between two realms that trust each other. When a client program, such as telnet or ssh, finds that the other computer is in a different realm, it will try to get a ticket granting ticket for that other realm, but from the local Kerberos server. With that ticket granting ticket, it will then obtain service tickets from the Kerberos server in the other realm.

For a two way trust between `MY.REALM' and `OTHER.REALM' add the following principals to each realm. The principals should be `krbtgt/OTHER.REALM@MY.REALM' and `krbtgt/MY.REALM@OTHER.REALM' in `MY.REALM', and `krbtgt/MY.REALM@OTHER.REALM' and `krbtgt/OTHER.REALM@MY.REALM'in `OTHER.REALM'.

In Kerberos 5 the trust can be configured to be one way. So that users from `MY.REALM' can authenticate to services in `OTHER.REALM', but not the opposite. In the example above, the `krbtgt/MY.REALM@OTHER.REALM' then should be removed.

The two principals must have the same key, key version number, and the same set of encryption types. Remember to transfer the two keys in a safe manner.

     vr$ klist
     Credentials cache: FILE:/tmp/krb5cc_913.console
             Principal: lha@E.KTH.SE
       Issued           Expires          Principal
     May  3 13:55:52  May  3 23:55:54  krbtgt/E.KTH.SE@E.KTH.SE
     vr$ telnet -l lha hummel.it.su.se
     Trying 2001:6b0:5:1095:250:fcff:fe24:dbf...
     Connected to hummel.it.su.se.
     Escape character is '^]'.
     Waiting for encryption to be negotiated...
     [ Trying mutual KERBEROS5 (host/hummel.it.su.se@SU.SE)... ]
     [ Kerberos V5 accepts you as ``lha@E.KTH.SE'' ]
     Encryption negotiated.
     Last login: Sat May  3 14:11:47 from vr.l.nxs.se
     hummel$ exit
     vr$ klist
     Credentials cache: FILE:/tmp/krb5cc_913.console
             Principal: lha@E.KTH.SE
       Issued           Expires          Principal
     May  3 13:55:52  May  3 23:55:54  krbtgt/E.KTH.SE@E.KTH.SE
     May  3 13:55:56  May  3 23:55:54  krbtgt/SU.SE@E.KTH.SE
     May  3 14:10:54  May  3 23:55:54  host/hummel.it.su.se@SU.SE

Next: , Previous: Cross realm, Up: Setting up a realm

4.14 Transit policy

If you want to use cross realm authentication through an intermediate realm, it must be explicitly allowed by either the KDCs or the server receiving the request. This is done in krb5.conf in the [capaths] section.

When the ticket transits through a realm to another realm, the destination realm adds its peer to the "transited-realms" field in the ticket. The field is unordered, since there is no way to know if know if one of the transited-realms changed the order of the list.

The syntax for [capaths] section:

             CLIENT-REALM = {
                     SERVER-REALM = PERMITTED-CROSS-REALMS ...

The realm STACKEN.KTH.SE allows clients from SU.SE and DSV.SU.SE to cross it. Since STACKEN.KTH.SE only has direct cross realm setup with KTH.SE, and DSV.SU.SE only has direct cross realm setup with SU.SE they need to use both SU.SE and KTH.SE as transit realms.

     	SU.SE = {
                         STACKEN.KTH.SE = KTH.SE
     	DSV.SU.SE = {
                         STACKEN.KTH.SE = SU.SE KTH.SE

The order of the PERMITTED-CROSS-REALMS is not important when doing transit cross realm verification.

However, the order is important when the [capaths] section is used to figure out the intermediate realm to go to when doing multi-realm transit. When figuring out the next realm, the first realm of the list of PERMITTED-CROSS-REALMS is chosen. This is done in both the client kerberos library and the KDC.

Next: , Previous: Transit policy, Up: Setting up a realm

4.15 Setting up DNS

4.15.1 Using DNS to find KDC

If there is information about where to find the KDC or kadmind for a realm in the krb5.conf for a realm, that information will be preferred, and DNS will not be queried.

Heimdal will try to use DNS to find the KDCs for a realm. First it will try to find a SRV resource record (RR) for the realm. If no SRV RRs are found, it will fall back to looking for an A RR for a machine named kerberos.REALM, and then kerberos-1.REALM, etc

Adding this information to DNS minimises the client configuration (in the common case, resulting in no configuration needed) and allows the system administrator to change the number of KDCs and on what machines they are running without caring about clients.

The downside of using DNS is that the client might be fooled to use the wrong server if someone fakes DNS replies/data, but storing the IP addresses of the KDC on all the clients makes it very hard to change the infrastructure.

An example of the configuration for the realm EXAMPLE.COM:

     $ORIGIN example.com.
     _kerberos._tcp          SRV     10 1 88 kerberos.example.com.
     _kerberos._udp          SRV     10 1 88 kerberos.example.com.
     _kerberos._tcp          SRV     10 1 88 kerberos-1.example.com.
     _kerberos._udp          SRV     10 1 88 kerberos-1.example.com.
     _kpasswd._udp           SRV     10 1 464 kerberos.example.com.
     _kerberos-adm._tcp	SRV	10 1 749 kerberos.example.com.

More information about DNS SRV resource records can be found in RFC-2782 (A DNS RR for specifying the location of services (DNS SRV)).

4.15.2 Using DNS to map hostname to Kerberos realm

Heimdal also supports a way to lookup a realm from a hostname. This to minimise configuration needed on clients. Using this has the drawback that clients can be redirected by an attacker to realms within the same cross realm trust and made to believe they are talking to the right server (since Kerberos authentication will succeed).

An example configuration that informs clients that for the realms it.example.com and srv.example.com, they should use the realm EXAMPLE.COM:

     $ORIGIN example.com.
     _kerberos.it		TXT     "EXAMPLE.COM"
     _kerberos.srv		TXT     "EXAMPLE.COM"

Next: , Previous: Setting up DNS, Up: Setting up a realm

4.16 Using LDAP to store the database

This document describes how to install the LDAP backend for Heimdal. Note that before attempting to configure such an installation, you should be aware of the implications of storing private information (such as users' keys) in a directory service primarily designed for public information. Nonetheless, with a suitable authorisation policy, it is possible to set this up in a secure fashion. A knowledge of LDAP, Kerberos, and C is necessary to install this backend. The HDB schema was devised by Leif Johansson.


4.16.1 Troubleshooting guide


4.16.2 Using Samba LDAP password database

The Samba domain and the Kerberos realm can have different names since arcfour's string to key functions principal/realm independent. So now will be your first and only chance name your Kerberos realm without needing to deal with old configuration files.

First, you should set up Samba and get that working with LDAP backend.

Now you can proceed as in See Using LDAP to store the database. Heimdal will pick up the Samba LDAP entries if they are in the same search space as the Kerberos entries.

Next: , Previous: Using LDAP to store the database, Up: Setting up a realm

4.17 Providing Kerberos credentials to servers and programs

Some services require Kerberos credentials when they start to make connections to other services or need to use them when they have started.

The easiest way to get tickets for a service is to store the key in a keytab. Both ktutil get and kadmin ext can be used to get a keytab. ktutil get is better in that way it changes the key/password for the user. This is also the problem with ktutil. If ktutil is used for the same service principal on several hosts, they keytab will only be useful on the last host. In that case, run the extract command on one host and then securely copy the keytab around to all other hosts that need it.

     host# ktutil -k /etc/krb5-service.keytab \
           get -p lha/admin@EXAMPLE.ORG service-principal@EXAMPLE.ORG
     lha/admin@EXAMPLE.ORG's Password:

To get a Kerberos credential file for the service, use kinit in the --keytab mode. This will not ask for a password but instead fetch the key from the keytab.

     service@host$ kinit --cache=/var/run/service_krb5_cache \
                    --keytab=/etc/krb5-service.keytab \

Long running services might need credentials longer then the expiration time of the tickets. kinit can run in a mode that refreshes the tickets before they expire. This is useful for services that write into AFS and other distributed file systems using Kerberos. To run the long running script, just append the program and arguments (if any) after the principal. kinit will stop refreshing credentials and remove the credentials when the script-to-start-service exits.

     service@host$ kinit --cache=/var/run/service_krb5_cache \
            --keytab=/etc/krb5-service.keytab \
            service-principal@EXAMPLE.ORG \
            script-to-start-service argument1 argument2

Previous: Providing Kerberos credentials to servers and programs, Up: Setting up a realm

4.18 Setting up PK-INIT

PK-INIT is levering the existing PKI infrastructure to use certificates to get the initial ticket, that is usually the krbtgt.

To use PK-INIT you must first have a PKI, so if you don't have one, it is time to create it. Note that you should read the whole chapter of the document to see the requirements on the CA software.

There needs to exist a mapping between the certificate and what principals that certificate is allowed to use. There are several ways to do this. The administrator can use a configuration file, storing the principal in the SubjectAltName extension of the certificate, or store the mapping in the principals entry in the kerberos database.

4.19 Certificates

This section documents the requirements on the KDC and client certificates and the format used in the id-pkinit-san OtherName extention.

4.19.1 KDC certificate

The certificate for the KDC have serveral requirements.

First the certificate should have an Extended Key Usage (EKU) id-pkkdcekuoid ( set. Second there must be a subjectAltName otherName using oid id-pkinit-san ( in the type field and a DER encoded KRB5PrincipalName that matches the name of the TGS of the target realm.

Both of these two requirements are not required by the standard to be checked by the client if it have external information what the certificate the KDC is supposed to be used. So it's in the interest of minimum amount of configuration on the clients they should be included.

Remember that if the client would accept any certificate as the KDC's certificate, the client could be fooled into trusting something that isn't a KDC and thus expose the user to giving away information (like password or other private information) that it is supposed to secret.

Also, if the certificate has a nameConstraints extention with a Generalname with dNSName or iPAdress it must match the hostname or adress of the KDC.

4.19.2 Client certificate

The client certificate may need to have a EKU id-pkekuoid ( set depending on the certifiate on the KDC.

It possible to store the principal (if allowed by the KDC) in the certificate and thus delegate responsibility to do the mapping between certificates and principals to the CA. Using KRB5PrincipalName in id-pkinit-san

OtherName extention in the GeneralName is used to do the mapping between certifiate and principal in the certifiate or storing the krbtgt principal in the KDC certificate.

The principal is stored in a SubjectAltName in the certificate using OtherName. The oid in the type is id-pkinit-san.

     id-pkinit-san OBJECT IDENTIFIER ::= { iso (1) org (3) dod (6)
     internet (1) security (5) kerberosv5 (2) 2 }

The data part of the OtherName is filled with the following DER encoded ASN.1 structure:

     KRB5PrincipalName ::= SEQUENCE {
     	realm [0] Realm,
     	principalName [1] PrincipalName

where Realm and PrincipalName is defined by the Kerberos ASN.1 specification.

4.20 Naming certificate using hx509

hx509 is the X.509 software used in Heimdal to handle certificates. hx509 uses different syntaxes to specify the different formats the certificates are stored in and what formats they exist in.

There are several formats that can be used, PEM, embedded into PKCS12 files, embedded into PKCS11 devices and raw DER encoded certificates. Below is a list of types to use.

DIR is reading all certificates in a directory that is DER or PEM formatted.

The main feature of DIR is that the directory is read on demand when iterating over certificates, that way applictions can for some cases avoid to store all certificates in memory. It's very useful for tests that iterate over larger amount of certificates.

Syntax is:


FILE: is used to have the lib pick up a certificate chain and a private key. The file can be either a PEM (openssl) file or a raw DER encoded certificate. If it's a PEM file it can contain several keys and certificates and the code will try to match the private key and certificate together.

Its useful to have one PEM file that contains all the trust anchors.

Syntax is:


PKCS11: is used to handle smartcards via PKCS11 drivers, for example soft-token, opensc, or muscle. The default is to use all slots on the device/token.

Syntax is:


PKCS12: is used to handle PKCS12 files. PKCS12 files commonly have the extension pfx or p12.

Syntax is:


4.21 Configure the Kerberos software

First configure the client's trust anchors and what parameters to verify, see subsection below how to do that. Now you can use kinit to get yourself tickets. One example how that can look like is:

     $ kinit -C FILE:$HOME/.certs/lha.crt,$HOME/.certs/lha.key lha@EXAMPLE.ORG
     Enter your private key passphrase:
     : lha@nutcracker ; klist
     Credentials cache: FILE:/tmp/krb5cc_19100a
             Principal: lha@EXAMPLE.ORG
       Issued           Expires          Principal
     Apr 20 02:08:08  Apr 20 12:08:08  krbtgt/EXAMPLE.ORG@EXAMPLE.ORG

Using PKCS11 it can look like this instead:

     $ kinit -C PKCS11:/tmp/pkcs11/lib/soft-pkcs11.so lha@EXAMPLE.ORG
     PIN code for SoftToken (slot):
     $ klist
     Credentials cache: API:4
             Principal: lha@EXAMPLE.ORG
       Issued           Expires          Principal
     Mar 26 23:40:10  Mar 27 09:40:10  krbtgt/EXAMPLE.ORG@EXAMPLE.ORG

Write about the kdc.

4.22 Configure the client

     	pkinit_anchors = FILE:/path/to/trust-anchors.pem
             EXAMPLE.COM = {
     		pkinit_require_eku = true
     		pkinit_require_krbtgt_otherName = true
     		pkinit_win2k = no
     		pkinit_win2k_require_binding = yes

4.23 Configure the KDC

     	enable-pkinit = yes
     	pkinit_identity = FILE:/secure/kdc.crt,/secure/kdc.key
     	pkinit_anchors = FILE:/path/to/trust-anchors.pem
     	pkinit_pool = PKCS12:/path/to/useful-intermediate-certs.pfx
     	pkinit_pool = FILE:/path/to/other-useful-intermediate-certs.pem
     	pkinit_allow_proxy_certificate = false
     	pkinit_win2k_require_binding = yes

4.23.1 Using pki-mapping file

Note that the file name is space sensitive.

     # cat /var/heimdal/pki-mapping
     # comments starts with #
     lha@EXAMPLE.ORG:C=SE,O=Stockholm universitet,CN=Love,UID=lha

4.23.2 Using the Kerberos database

4.24 Use hxtool to create certificates

4.24.1 Generate certificates

First you need to generate a CA certificate, change the –subject to something appropriate, the CA certificate will be valid for 10 years.

You need to change –subject in the command below.

     hxtool issue-certificate \
         --self-signed \
         --issue-ca \
         --generate-key=rsa \
         --subject="CN=CA,DC=test,DC=h5l,DC=se" \
         --lifetime=10years \

The KDC needs to have a certificate, so generate a certificate of the type “pkinit-kdc” and set the PK-INIT specifial SubjectAltName to the name of the krbtgt of the realm.

You need to change –subject and –pk-init-principal in the command below.

     hxtool issue-certificate \
         --ca-certificate=FILE:ca.pem \
         --generate-key=rsa \
         --type="pkinit-kdc" \
         --pk-init-principal="krbtgt/TEST.H5L.SE@TEST.H5L.SE" \
         --subject="uid=kdc,DC=test,DC=h5l,DC=se" \

The users also needs to have a certificate, so generate a certificate of the type “pkinit-client”. The client doesn't need to have the PK-INIT SubjectAltName set, you can have the Subject DN in the ACL file (pki-mapping) instead.

You need to change –subject and –pk-init-principal in the command below.

     hxtool issue-certificate \
         --ca-certificate=FILE:ca.pem \
         --generate-key=rsa \
         --type="pkinit-client" \
         --pk-init-principal="lha@TEST.H5L.SE" \
         --subject="uid=lha,DC=test,DC=h5l,DC=se" \

4.24.2 Validate the certificate

hxtool also contains a tool that will validate certificates according to rules from the PKIX document. These checks are not complete, but a good test to check if you got all of the basic bits right in your certificates.

     hxtool validate FILE:user.pem

4.25 Use OpenSSL to create certificates

This section tries to give the CA owners hints how to create certificates using OpenSSL (or CA software based on OpenSSL).

4.25.1 Using OpenSSL to create certificates with krb5PrincipalName

To make OpenSSL create certificates with krb5PrincipalName use openssl.cnf as described below. To see a complete example of creating client and KDC certificates, see the test-data generation script lib/hx509/data/gen-req.sh in the source-tree. The certicates it creates are used to test the PK-INIT functionality in tests/kdc/check-kdc.in.

To use this example you have to use OpenSSL 0.9.8a or later.

     realm = EXP:0, GeneralString:MY.REALM
     principal_name = EXP:1, SEQUENCE:principal_seq
     name_type = EXP:0, INTEGER:1
     name_string = EXP:1, SEQUENCE:principals
     princ1 = GeneralString:userid

Command usage

     openssl x509 -extensions user_certificate
     openssl ca -extensions user_certificate

4.26 Using PK-INIT with Windows

4.26.1 Client configration

Clients using a Windows KDC with PK-INIT need configuration since windows uses pre-standard format and this can't be autodetected.

The pkinit_win2k_require_binding option requires the reply for the KDC to be of the new, secure, type that binds the request to reply. Before clients should fake the reply from the KDC. To use this option you have to apply a fix from Microsoft.

             MY.MS.REALM = {
                     pkinit_win2k = yes
                     pkinit_win2k_require_binding = no

4.26.2 Certificates

The client certificates need to have the extended keyusage “Microsoft Smartcardlogin” (openssl have the oid shortname msSmartcardLogin).

See Microsoft Knowledge Base Article - 281245 “Guidelines for Enabling Smart Card Logon with Third-Party Certification Authorities” for a more extensive description of how set setup an external CA to it includes all information that will make a Windows KDC happy.

4.26.3 Configure Windows 2000 CA

To enable Microsoft Smartcardlogin> for certificates in your Windows 2000 CA, you want to look at Microsoft Knowledge Base Article - 313274 “HOW TO: Configure a Certification Authority to Issue Smart Card Certificates in Windows”.

Next: , Previous: Setting up a realm, Up: Top

5 Applications

Next: , Previous: Applications, Up: Applications

5.1 Authentication modules

The problem of having different authentication mechanisms has been recognised by several vendors, and several solutions have appeared. In most cases these solutions involve some kind of shared modules that are loaded at run-time. Modules for some of these systems can be found in lib/auth. Presently there are modules for Digital's SIA, and IRIX' login and xdm (in lib/auth/afskauthlib).

Next: , Previous: Authentication modules, Up: Authentication modules

5.1.1 Digital SIA

How to install the SIA module depends on which OS version you're running. Tru64 5.0 has a new command, siacfg, which makes this process quite simple. If you have this program, you should just be able to run:

     siacfg -a KRB5 /usr/athena/lib/libsia_krb5.so

On older versions, or if you want to do it by hand, you have to do the following (not tested by us on Tru64 5.0):

Users with local passwords (like `root') should be able to login safely.

When using Digital's xdm the `KRB5CCNAME' environment variable isn't passed along as it should (since xdm zaps the environment). Instead you have to set `KRB5CCNAME' to the correct value in /usr/lib/X11/xdm/Xsession. Add a line similar to

     KRB5CCNAME=FILE:/tmp/krb5cc`id -u`_`ps -o ppid= -p $$`; export KRB5CCNAME

If you use CDE, dtlogin allows you to specify which additional environment variables it should export. To add `KRB5CCNAME' to this list, edit /usr/dt/config/Xconfig, and look for the definition of `exportList'. You want to add something like:

     Dtlogin.exportList:     KRB5CCNAME
Notes to users with Enhanced security

Digital's `ENHANCED' (C2) security, and Kerberos solve two different problems. C2 deals with local security, adds better control of who can do what, auditing, and similar things. Kerberos deals with network security.

To make C2 security work with Kerberos you will have to do the following.

At present `su' does not accept the vouching flag, so it will not work as expected.

Also, kerberised ftp will not work with C2 passwords. You can solve this by using both Digital's ftpd and our on different ports.

Remember, if you do these changes you will get a system that most certainly does not fulfil the requirements of a C2 system. If C2 is what you want, for instance if someone else is forcing you to use it, you're out of luck. If you use enhanced security because you want a system that is more secure than it would otherwise be, you probably got an even more secure system. Passwords will not be sent in the clear, for instance.

Previous: Digital SIA, Up: Authentication modules

5.1.2 IRIX

The IRIX support is a module that is compatible with Transarc's afskauthlib.so. It should work with all programs that use this library. This should include login and xdm.

The interface is not very documented but it seems that you have to copy libkafs.so, libkrb.so, and libdes.so to /usr/lib, or build your afskauthlib.so statically.

The afskauthlib.so itself is able to reside in /usr/vice/etc, /usr/afsws/lib, or the current directory (wherever that is).

IRIX 6.4 and newer seem to have all programs (including xdm and login) in the N32 object format, whereas in older versions they were O32. For it to work, the afskauthlib.so library has to be in the same object format as the program that tries to load it. This might require that you have to configure and build for O32 in addition to the default N32.

Apart from this it should “just work”; there are no configuration files.

Note that recent Irix 6.5 versions (at least 6.5.22) have PAM, including a pam_krb5.so module. Not all relevant programs use PAM, though, e.g. ssh. In particular, for console graphical login you need to turn off `visuallogin' and turn on `xdm' with chkconfig.

Previous: Authentication modules, Up: Applications

5.2 AFS

AFS is a distributed filesystem that uses Kerberos for authentication.

For more information about AFS see OpenAFS http://www.openafs.org/ and Arla http://www.stacken.kth.se/projekt/arla/.

5.2.1 How to get a KeyFile

ktutil -k AFSKEYFILE:KeyFile get afs@MY.REALM

or you can extract it with kadmin

     kadmin> ext -k AFSKEYFILE:/usr/afs/etc/KeyFile afs@My.CELL.NAME

You have to make sure you have a des-cbc-md5 encryption type since that is the enctype that will be converted.

5.2.2 How to convert a srvtab to a KeyFile

You need a /usr/vice/etc/ThisCell containing the cellname of your AFS-cell.

ktutil copy krb4:/root/afs-srvtab AFSKEYFILE:/usr/afs/etc/KeyFile.

If keyfile already exists, this will add the new key in afs-srvtab to KeyFile.

5.3 Using 2b tokens with AFS

5.3.1 What is 2b ?

2b is the name of the proposal that was implemented to give basic Kerberos 5 support to AFS in rxkad. It's not real Kerberos 5 support since it still uses fcrypt for data encryption and not Kerberos encryption types.

Its only possible (in all cases) to do this for DES encryption types because only then the token (the AFS equivalent of a ticket) will be smaller than the maximum size that can fit in the token cache in the OpenAFS/Transarc client. It is a so tight fit that some extra wrapping on the ASN1/DER encoding is removed from the Kerberos ticket.

2b uses a Kerberos 5 EncTicketPart instead of a Kerberos 4 ditto for the part of the ticket that is encrypted with the service's key. The client doesn't know what's inside the encrypted data so to the client it doesn't matter.

To differentiate between Kerberos 4 tickets and Kerberos 5 tickets, 2b uses a special kvno, 213 for 2b tokens and 255 for Kerberos 5 tokens.

Its a requirement that all AFS servers that support 2b also support native Kerberos 5 in rxkad.

5.3.2 Configuring a Heimdal kdc to use 2b tokens

Support for 2b tokens in the kdc are turned on for specific principals by adding them to the string list option [kdc]use_2b in the kdc's krb5.conf file.

     	use_2b = {
     		afs@SU.SE = yes
     		afs/it.su.se@SU.SE = yes

5.3.3 Configuring AFS clients for 2b support

There is no need to configure AFS clients for 2b support. The only software that needs to be installed/upgrade is a Kerberos 5 enabled afslog.

Next: , Previous: Applications, Up: Top

6 Things in search for a better place

6.1 Making things work on Ciscos

Modern versions of Cisco IOS has some support for authenticating via Kerberos 5. This can be used both by having the router get a ticket when you login (boring), and by using Kerberos authenticated telnet to access your router (less boring). The following has been tested on IOS 11.2(12), things might be different with other versions. Old versions are known to have bugs.

To make this work, you will first have to configure your router to use Kerberos (this is explained in the documentation). A sample configuration looks like the following:

     aaa new-model
     aaa authentication login default krb5-telnet krb5 enable
     aaa authorization exec krb5-instance
     kerberos local-realm FOO.SE
     kerberos srvtab entry host/router.foo.se 0 891725446 4 1 8 012345678901234567
     kerberos server FOO.SE
     kerberos instance map admin 15

This tells you (among other things) that when logging in, the router should try to authenticate with kerberised telnet, and if that fails try to verify a plain text password via a Kerberos ticket exchange (as opposed to a local database, RADIUS or something similar), and if that fails try the local enable password. If you're not careful when you specify the `login default' authentication mechanism, you might not be able to login at all. The `instance map' and `authorization exec' lines says that people with `admin' instances should be given `enabled' shells when logging in.

The numbers after the principal on the `srvtab' line are principal type, time stamp (in seconds since 1970), key version number (4), keytype (1 == des), key length (always 8 with des), and then the key.

To make the Heimdal KDC produce tickets that the Cisco can decode you might have to turn on the `encode_as_rep_as_tgs_rep' flag in the KDC. You will also have to specify that the router can't handle anything but `des-cbc-crc'. This can be done with the `del_enctype' command of `kadmin'.

This all fine and so, but unless you have an IOS version with encryption (available only in the U.S) it doesn't really solve any problems. Sure you don't have to send your password over the wire, but since the telnet connection isn't protected it's still possible for someone to steal your session. This won't be fixed until someone adds integrity to the telnet protocol.

A working solution would be to hook up a machine with a real operating system to the console of the Cisco and then use it as a backwards terminal server.

Next: , Previous: Things in search for a better place, Up: Top

7 Kerberos 4 issues

The KDC has built-in version 4 support. It is not enabled by default, see setup how to set it up.

The KDC will also have kaserver emulation and be able to handle AFS-clients that use klog.

Next: , Previous: Kerberos 4 issues, Up: Kerberos 4 issues

7.1 Principal conversion issues

First, Kerberos 4 and Kerberos 5 principals are different. A version 4 principal consists of a name, an instance, and a realm. A version 5 principal has one or more components, and a realm (the terms “name” and “instance” are still used, for the first and second component, respectively). Also, in some cases the name of a version 4 principal differs from the first component of the corresponding version 5 principal. One notable example is the “host” type principals, where the version 4 name is `rcmd' (for “remote command”), and the version 5 name is `host'. For the class of principals that has a hostname as instance, there is an other major difference, Kerberos 4 uses only the first component of the hostname, whereas Kerberos 5 uses the fully qualified hostname.

Because of this it can be hard or impossible to correctly convert a version 4 principal to a version 5 principal 1. The biggest problem is to know if the conversion resulted in a valid principal. To give an example, suppose you want to convert the principal `rcmd.foo'.

The `rcmd' name suggests that the instance is a hostname (even if there are exceptions to this rule). To correctly convert the instance `foo' to a hostname, you have to know which host it is referring to. You can to this by either guessing (from the realm) which domain name to append, or you have to have a list of possible hostnames. In the simplest cases you can cover most principals with the first rule. If you have several domains sharing a single realm this will not usually work. If the exceptions are few you can probably come by with a lookup table for the exceptions.

In a complex scenario you will need some kind of host lookup mechanism. Using DNS for this is tempting, but DNS is error prone, slow and unsafe 2.

Fortunately, the KDC has a trump on hand: it can easily tell if a principal exists in the database. The KDC will use krb5_425_conv_principal_ext to convert principals when handling to version 4 requests.

Next: , Previous: Principal conversion issues, Up: Kerberos 4 issues

7.2 Converting a version 4 database

If you want to convert an existing version 4 database, the principal conversion issue arises too.

If you decide to convert your database once and for all, you will only have to do this conversion once. It is also possible to run a version 5 KDC as a slave to a version 4 KDC. In this case this conversion will happen every time the database is propagated. When doing this conversion, there are a few things to look out for. If you have stale entries in the database, these entries will not be converted. This might be because these principals are not used anymore, or it might be just because the principal couldn't be converted.

You might also see problems with a many-to-one mapping of principals. For instance, if you are using DNS lookups and you have two principals `rcmd.foo' and `rcmd.bar', where `foo' is a CNAME for `bar', the resulting principals will be the same. Since the conversion function can't tell which is correct, these conflicts will have to be resolved manually.

7.2.1 Conversion example

Given the following set of hosts and services:

     foo.se          rcmd
     mail.foo.se     rcmd, pop
     ftp.bar.se      rcmd, ftp

you have a database that consists of the following principals:

`rcmd.foo', `rcmd.mail', `pop.mail', `rcmd.ftp', and `ftp.ftp'.

lets say you also got these extra principals: `rcmd.gone', `rcmd.old-mail', where `gone.foo.se' was a machine that has now passed away, and `old-mail.foo.se' was an old mail machine that is now a CNAME for `mail.foo.se'.

When you convert this database you want the following conversions to be done:

     rcmd.foo         host/foo.se
     rcmd.mail        host/mail.foo.se
     pop.mail         pop/mail.foo.se
     rcmd.ftp         host/ftp.bar.se
     ftp.ftp          ftp/ftp.bar.se
     rcmd.gone        removed
     rcmd.old-mail    removed

A krb5.conf that does this looks like:

             FOO.SE = {
                     v4_name_convert = {
                             host = {
                                     ftp = ftp
                                     pop = pop
                                     rcmd = host
                     v4_instance_convert = {
                             foo = foo.se
                             ftp = ftp.bar.se
                     default_domain = foo.se

The `v4_name_convert' section says which names should be considered having an instance consisting of a hostname, and it also says how the names should be converted (for instance `rcmd' should be converted to `host'). The `v4_instance_convert' section says how a hostname should be qualified (this is just a hosts-file in disguise). Host-instances that aren't covered by `v4_instance_convert' are qualified by appending the contents of the `default_domain'.

Actually, this example doesn't work. Or rather, it works to well. Since it has no way of knowing which hostnames are valid and which are not, it will happily convert `rcmd.gone' to `host/gone.foo.se'. This isn't a big problem, but if you have run your kerberos realm for a few years, chances are big that you have quite a few `junk' principals.

If you don't want this you can remove the `default_domain' statement, but then you will have to add entries for all your hosts in the `v4_instance_convert' section.

Instead of doing this you can use DNS to convert instances. This is not a solution without problems, but it is probably easier than adding lots of static host entries.

To enable DNS lookup you should turn on `v4_instance_resolve' in the `[libdefaults]' section.

7.2.2 Converting a database

The database conversion is done with `hprop'. You can run this command to propagate the database to the machine called `slave-server' (which should be running a `hpropd').

     hprop --source=krb4-db --master-key=/.m slave-server

This command can also be to use for converting the v4 database on the server:

     hprop -n --source=krb4-db -d /var/kerberos/principal --master-key=/.m | hpropd -n

7.3 Version 4 Kadmin

`kadmind' can act as a version 4 kadmind, and you can do most operations, but with some restrictions (since the version 4 kadmin protocol is, lets say, very ad hoc.) One example is that it only passes des keys when creating principals and changing passwords (modern kpasswd clients do send the password, so it's possible to to password quality checks). Because of this you can only create principals with des keys, and you can't set any flags or do any other fancy stuff.

To get this to work, you have to add another entry to inetd (since version 4 uses port 751, not 749).

And then there are a many more things you can do; more on this in a later version of this manual. Until then, UTSL.

Previous: Converting a version 4 database, Up: Kerberos 4 issues

7.4 kaserver

7.4.1 kaserver emulation

The Heimdal kdc can emulate a kaserver. The kaserver is a Kerberos 4 server with pre-authentication using Rx as the on-wire protocol. The kdc contains a minimalistic Rx implementation.

There are three parts of the kaserver; KAA (Authentication), KAT (Ticket Granting), and KAM (Maintenance). The KAA interface and KAT interface both passes over DES encrypted data-blobs (just like the Kerberos-protocol) and thus do not need any other protection. The KAM interface uses rxkad (Kerberos authentication layer for Rx) for security and data protection, and is used for example for changing passwords. This part is not implemented in the kdc.

Another difference between the ka-protocol and the Kerberos 4 protocol is that the pass-phrase is salted with the cellname in the string to key function in the ka-protocol, while in the Kerberos 4 protocol there is no salting of the password at all. To make sure AFS-compatible keys are added to each principals when they are created or their password are changed, `afs3-salt' should be added to `[kadmin]default_keys'.

7.4.2 Transarc AFS Windows client

The Transarc Windows client uses Kerberos 4 to obtain tokens, and thus does not need a kaserver. The Windows client assumes that the Kerberos server is on the same machine as the AFS-database server. If you do not like to do that you can add a small program that runs on the database servers that forward all kerberos requests to the real kerberos server. A program that does this is krb-forward (ftp://ftp.stacken.kth.se/pub/projekts/krb-forward).

Next: , Previous: Kerberos 4 issues, Up: Top

8 Windows 2000 compatability

Windows 2000 (formerly known as Windows NT 5) from Microsoft implements Kerberos 5. Their implementation, however, has some quirks, peculiarities, and bugs. This chapter is a short summary of the things that we have found out while trying to test Heimdal against Windows 2000. Another big problem with the Kerberos implementation in Windows 2000 is that the available documentation is more focused on getting things to work rather than how they work, and not that useful in figuring out how things really work.

This information should apply to Heimdal 1.1 and Windows 2000 Professional. It's of course subject to change all the time and mostly consists of our not so inspired guesses. Hopefully it's still somewhat useful.

Next: , Previous: Windows 2000 compatability, Up: Windows 2000 compatability

8.1 Configuring Windows 2000 to use a Heimdal KDC

You need the command line program called ksetup.exe which is available in the file SUPPORT/TOOLS/SUPPORT.CAB on the Windows 2000 Professional CD-ROM. This program is used to configure the Kerberos settings on a Workstation.

Ksetup store the domain information under the registry key: HKEY_LOCAL_MACHINE\System\CurrentControlSet\Control\LSA\Kerberos\Domains.

Use the kadmin program in Heimdal to create a host principal in the Kerberos realm.

     unix% kadmin
     kadmin> ank --password=password host/datan.example.com

The name `datan.example.com' should be replaced with DNS name of the workstation.

You must configure the workstation as a member of a workgroup, as opposed to a member in an NT domain, and specify the KDC server of the realm as follows:

     C:> ksetup /setdomain EXAMPLE.COM
     C:> ksetup /addkdc EXAMPLE.COM kdc.example.com

Set the machine password, i.e. create the local keytab:

     C:> ksetup /SetComputerPassword password

The password used in ksetup /setmachpassword must be the same as the password used in the kadmin ank command.

The workstation must now be rebooted.

A mapping between local NT users and Kerberos principals must be specified. You have two choices. First:

     C:> ksetup /mapuser user@MY.REALM nt_user

This will map a user to a specific principal; this allows you to have other usernames in the realm than in your NT user database. (Don't ask me why on earth you would want that....)

You can also say:

     C:> ksetup /mapuser * *

The Windows machine will now map any user to the corresponding principal, for example `nisse' to the principal `nisse@MY.REALM'. (This is most likely what you want.)

Next: , Previous: Configuring Windows 2000 to use a Heimdal KDC, Up: Windows 2000 compatability

8.2 Inter-Realm keys (trust) between Windows 2000 and a Heimdal KDC

See also the Step-by-Step guide from Microsoft, referenced below.

Install Windows 2000, and create a new controller (Active Directory Server) for the domain.

By default the trust will be non-transitive. This means that only users directly from the trusted domain may authenticate. This can be changed to transitive by using the netdom.exe tool. netdom.exe can also be used to add the trust between two realms.

You need to tell Windows 2000 on what hosts to find the KDCs for the non-Windows realm with ksetup, see See Configuring Windows 2000 to use a Heimdal KDC.

This needs to be done on all computers that want enable cross-realm login with Mapped Names.

Then you need to add the inter-realm keys on the Windows KDC. Start the Domain Tree Management tool (found in Programs, Administrative tools, Active Directory Domains and Trusts).

Right click on Properties of your domain, select the Trust tab. Press Add on the appropriate trust windows and enter domain name and password. When prompted if this is a non-Windows Kerberos realm, press OK.

Do not forget to add trusts in both directions (if that's what you want).

If you want to use netdom.exe instead of the Domain Tree Management tool, you do it like this:

     netdom trust NT.REALM.EXAMPLE.COM /Domain:EXAMPLE.COM /add /realm /passwordt:TrustPassword

You also need to add the inter-realm keys to the Heimdal KDC. Make sure you have matching encryption types (DES, Arcfour and AES in case of Longhorn)

Another issue is salting. Since Windows 2000 does not seem to understand Kerberos 4 salted hashes you might need to turn off anything similar to the following if you have it, at least while adding the principals that are going to share keys with Windows 2000.

             default_keys = v5 v4

So remove v4 from default keys.

What you probably want to use is this:

             default_keys = des-cbc-crc:pw-salt arcfour-hmac-md5:pw-salt

Once that is also done, you can add the required inter-realm keys:

     kadmin add krbtgt/NT.REALM.EXAMPLE.COM@EXAMPLE.COM
     kadmin add krbtgt/REALM.EXAMPLE.COM@NT.EXAMPLE.COM

Use the same passwords for both keys.

Do not forget to reboot before trying the new realm-trust (after running ksetup). It looks like it might work, but packets are never sent to the non-Windows KDC.

Next: , Previous: Inter-Realm keys (trust) between Windows 2000 and a Heimdal KDC, Up: Windows 2000 compatability

8.3 Create account mappings

Start the Active Directory Users and Computers tool. Select the View menu, that is in the left corner just below the real menu (or press Alt-V), and select Advanced Features. Right click on the user that you are going to do a name mapping for and choose Name mapping.

Click on the Kerberos Names tab and add a new principal from the non-Windows domain.

This adds `authorizationNames' entry to the users LDAP entry to the Active Directory LDAP catalog. When you create users by script you can add this entry instead.

Next: , Previous: Create account mappings, Up: Windows 2000 compatability

8.4 Encryption types

Windows 2000 supports both the standard DES encryptions (`des-cbc-crc' and `des-cbc-md5') and its own proprietary encryption that is based on MD4 and RC4 that is documented in and is supposed to be described in draft-brezak-win2k-krb-rc4-hmac-03.txt. New users will get both MD4 and DES keys. Users that are converted from a NT4 database, will only have MD4 passwords and will need a password change to get a DES key.

Next: , Previous: Encryption types, Up: Windows 2000 compatability

8.5 Authorisation data

The Windows 2000 KDC also adds extra authorisation data in tickets. It is at this point unclear what triggers it to do this. The format of this data is only available under a “secret” license from Microsoft, which prohibits you implementing it.

A simple way of getting hold of the data to be able to understand it better is described here.

  1. Find the client example on using the SSPI in the SDK documentation.
  2. Change “AuthSamp” in the source code to lowercase.
  3. Build the program.
  4. Add the “authsamp” principal with a known password to the database. Make sure it has a DES key.
  5. Run ktutil add to add the key for that principal to a keytab.
  6. Run appl/test/nt_gss_server -p 2000 -s authsamp --dump-auth=file where file is an appropriate file.
  7. It should authenticate and dump for you the authorisation data in the file.
  8. The tool lib/asn1/asn1_print is somewhat useful for analysing the data.

Next: , Previous: Authorisation data, Up: Windows 2000 compatability

8.6 Quirks of Windows 2000 KDC

There are some issues with salts and Windows 2000. Using an empty salt—which is the only one that Kerberos 4 supported, and is therefore known as a Kerberos 4 compatible salt—does not work, as far as we can tell from out experiments and users' reports. Therefore, you have to make sure you keep around keys with all the different types of salts that are required. Microsoft have fixed this issue post Windows 2003.

Microsoft seems also to have forgotten to implement the checksum algorithms `rsa-md4-des' and `rsa-md5-des'. This can make Name mapping (see Create account mappings) fail if a `des-cbc-md5' key is used. To make the KDC return only `des-cbc-crc' you must delete the `des-cbc-md5' key from the kdc using the kadmin del_enctype command.

     kadmin del_enctype lha des-cbc-md5

You should also add the following entries to the krb5.conf file:

     	default_etypes = des-cbc-crc
     	default_etypes_des = des-cbc-crc

These configuration options will make sure that no checksums of the unsupported types are generated.

Previous: Quirks of Windows 2000 KDC, Up: Windows 2000 compatability

8.7 Useful links when reading about the Windows 2000

See also our paper presented at the 2001 Usenix Annual Technical Conference, available in the proceedings or at http://www.usenix.org/publications/library/proceedings/usenix01/freenix01/westerlund.html.

There are lots of texts about Kerberos on Microsoft's web site, here is a short list of the interesting documents that we have managed to find.

Other useful programs include these:

Next: , Previous: Windows 2000 compatability, Up: Top

9 Programming with Kerberos

First you need to know how the Kerberos model works, go read the introduction text (see What is Kerberos?).

Next: , Previous: Programming with Kerberos, Up: Programming with Kerberos

9.1 Kerberos 5 API Overview

All functions are documented in manual pages. This section tries to give an overview of the major components used in Kerberos library, and point to where to look for a specific function.

9.1.1 Kerberos context

A kerberos context (krb5_context) holds all per thread state. All global variables that are context specific are stored in this structure, including default encryption types, credential cache (for example, a ticket file), and default realms.

See the manual pages for krb5_context(3) and krb5_init_context(3).

9.1.2 Kerberos authentication context

Kerberos authentication context (krb5_auth_context) holds all context related to an authenticated connection, in a similar way to the kerberos context that holds the context for the thread or process.

The krb5_auth_context is used by various functions that are directly related to authentication between the server/client. Example of data that this structure contains are various flags, addresses of client and server, port numbers, keyblocks (and subkeys), sequence numbers, replay cache, and checksum types.

See the manual page for krb5_auth_context(3).

9.1.3 Kerberos principal

The Kerberos principal is the structure that identifies a user or service in Kerberos. The structure that holds the principal is the krb5_principal. There are function to extract the realm and elements of the principal, but most applications have no reason to inspect the content of the structure.

The are several ways to create a principal (with different degree of portability), and one way to free it.

See manual page for krb5_principal(3) for more information about the functions.

9.1.4 Credential cache

A credential cache holds the tickets for a user. A given user can have several credential caches, one for each realm where the user have the initial tickets (the first krbtgt).

The credential cache data can be stored internally in different way, each of them for different proposes. File credential (FILE) caches and processes based (KCM) caches are for permanent storage. While memory caches (MEMORY) are local caches to the local process.

Caches are opened with krb5_cc_resolve(3) or created with krb5_cc_gen_unique(3).

If the cache needs to be opened again (using krb5_cc_resolve(3)) krb5_cc_close(3) will close the handle, but not the remove the cache. krb5_cc_destroy(3) will zero out the cache, remove the cache so it can no longer be referenced.

See also manual page for krb5_ccache(3)

9.1.5 Kerberos errors

Kerberos errors are based on the com_err library. All error codes are 32-bit signed numbers, the first 24 bits define what subsystem the error originates from, and last 8 bits are 255 error codes within the library. Each error code have fixed string associated with it. For example, the error-code -1765328383 have the symbolic name KRB5KDC_ERR_NAME_EXP, and associated error string “Client's entry in database has expired”.

This is a great improvement compared to just getting one of the unix error-codes back. However, Heimdal have an extention to pass back customised errors messages. Instead of getting “Key table entry not found”, the user might back “failed to find host/host.example.com@EXAMLE.COM(kvno 3) in keytab /etc/krb5.keytab (des-cbc-crc)”. This improves the chance that the user find the cause of the error so you should use the customised error message whenever it's available.

See also manual page for krb5_get_error_string(3) and krb5_get_err_text(3).

9.1.6 Keytab management

A keytab is a storage for locally stored keys. Heimdal includes keytab support for Kerberos 5 keytabs, Kerberos 4 srvtab, AFS-KeyFile's, and for storing keys in memory.

Keytabs are used for servers and long-running services.

See also manual page for krb5_keytab(3)

9.1.7 Kerberos crypto

Heimdal includes a implementation of the Kerberos crypto framework, all crypto operations.

See also manual page for krb5_crypto_init(3), krb5_keyblock(3), krb5_create_checksum(3), and krb5_encrypt(3).

Next: , Previous: Kerberos 5 API Overview, Up: Programming with Kerberos

9.2 Walkthrough of a sample Kerberos 5 client

This example contains parts of a sample TCP Kerberos 5 clients, if you want a real working client, please look in appl/test directory in the Heimdal distribution.

All Kerberos error-codes that are returned from kerberos functions in this program are passed to krb5_err, that will print a descriptive text of the error code and exit. Graphical programs can convert error-code to a human readable error-string with the krb5_get_err_text(3) function.

Note that you should not use any Kerberos function before krb5_init_context() have completed successfully. That is the reason err() is used when krb5_init_context() fails.

First the client needs to call krb5_init_context to initialise the Kerberos 5 library. This is only needed once per thread in the program. If the function returns a non-zero value it indicates that either the Kerberos implementation is failing or it's disabled on this host.

     #include <krb5.h>
     main(int argc, char **argv)
             krb5_context context;
             if (krb5_context(&context))
                     errx (1, "krb5_context");

Now the client wants to connect to the host at the other end. The preferred way of doing this is using getaddrinfo(3) (for operating system that have this function implemented), since getaddrinfo is neutral to the address type and can use any protocol that is available.

             struct addrinfo *ai, *a;
             struct addrinfo hints;
             int error;
             memset (&hints, 0, sizeof(hints));
             hints.ai_socktype = SOCK_STREAM;
             hints.ai_protocol = IPPROTO_TCP;
             error = getaddrinfo (hostname, "pop3", &hints, &ai);
             if (error)
                     errx (1, "%s: %s", hostname, gai_strerror(error));
             for (a = ai; a != NULL; a = a->ai_next) {
                     int s;
                     s = socket (a->ai_family, a->ai_socktype, a->ai_protocol);
                     if (s < 0)
                     if (connect (s, a->ai_addr, a->ai_addrlen) < 0) {
                             warn ("connect(%s)", hostname);
                                 close (s);
                     freeaddrinfo (ai);
                     ai = NULL;
             if (ai) {
                         freeaddrinfo (ai);
                         errx ("failed to contact %s", hostname);

Before authenticating, an authentication context needs to be created. This context keeps all information for one (to be) authenticated connection (see krb5_auth_context(3)).

             status = krb5_auth_con_init (context, &auth_context);
             if (status)
                     krb5_err (context, 1, status, "krb5_auth_con_init");

For setting the address in the authentication there is a help function krb5_auth_con_setaddrs_from_fd that does everything that is needed when given a connected file descriptor to the socket.

             status = krb5_auth_con_setaddrs_from_fd (context,
             if (status)
                     krb5_err (context, 1, status,

The next step is to build a server principal for the service we want to connect to. (See also krb5_sname_to_principal(3).)

             status = krb5_sname_to_principal (context,
             if (status)
                     krb5_err (context, 1, status, "krb5_sname_to_principal");

The client principal is not passed to krb5_sendauth(3) function, this causes the krb5_sendauth function to try to figure it out itself.

The server program is using the function krb5_recvauth(3) to receive the Kerberos 5 authenticator.

In this case, mutual authentication will be tried. That means that the server will authenticate to the client. Using mutual authentication is good since it enables the user to verify that they are talking to the right server (a server that knows the key).

If you are using a non-blocking socket you will need to do all work of krb5_sendauth yourself. Basically you need to send over the authenticator from krb5_mk_req(3) and, in case of mutual authentication, verifying the result from the server with krb5_rd_rep(3).

             status = krb5_sendauth (context,
             if (status)
                     krb5_err (context, 1, status, "krb5_sendauth");

Once authentication has been performed, it is time to send some data. First we create a krb5_data structure, then we sign it with krb5_mk_safe(3) using the auth_context that contains the session-key that was exchanged in the krb5_sendauth(3)/krb5_recvauth(3) authentication sequence.

             data.data   = "hej";
             data.length = 3;
             krb5_data_zero (&packet);
             status = krb5_mk_safe (context,
             if (status)
                     krb5_err (context, 1, status, "krb5_mk_safe");

And send it over the network.

             len = packet.length;
             net_len = htonl(len);
             if (krb5_net_write (context, &sock, &net_len, 4) != 4)
                     err (1, "krb5_net_write");
             if (krb5_net_write (context, &sock, packet.data, len) != len)
                     err (1, "krb5_net_write");

To send encrypted (and signed) data krb5_mk_priv(3) should be used instead. krb5_mk_priv(3) works the same way as krb5_mk_safe(3), with the exception that it encrypts the data in addition to signing it.

             data.data   = "hemligt";
             data.length = 7;
             krb5_data_free (&packet);
             status = krb5_mk_priv (context,
             if (status)
                     krb5_err (context, 1, status, "krb5_mk_priv");

And send it over the network.

             len = packet.length;
             net_len = htonl(len);
             if (krb5_net_write (context, &sock, &net_len, 4) != 4)
                     err (1, "krb5_net_write");
             if (krb5_net_write (context, &sock, packet.data, len) != len)
                     err (1, "krb5_net_write");

The server is using krb5_rd_safe(3) and krb5_rd_priv(3) to verify the signature and decrypt the packet.

Next: , Previous: Walkthrough of a sample Kerberos 5 client, Up: Programming with Kerberos

9.3 Validating a password in an application

See the manual page for krb5_verify_user(3).

Next: , Previous: Validating a password in a server application, Up: Programming with Kerberos

9.4 API differences to MIT Kerberos

This section is somewhat disorganised, but so far there is no overall structure to the differences, though some of the have their root in that Heimdal uses an ASN.1 compiler and MIT doesn't.

9.4.1 Principal and realms

Heimdal stores the realm as a krb5_realm, that is a char *. MIT Kerberos uses a krb5_data to store a realm.

In Heimdal krb5_principal doesn't contain the component name_type; it's instead stored in component name.name_type. To get and set the nametype in Heimdal, use krb5_principal_get_type(3) and krb5_principal_set_type(3).

For more information about principal and realms, see krb5_principal(3).

9.4.2 Error messages

To get the error string, Heimdal uses krb5_get_error_string(3) or, if NULL is returned, krb5_get_err_text(3). This is to return custom error messages (like “Can't find host/datan.example.com@EXAMPLE.COM in /etc/krb5.conf.” instead of a “Key table entry not found” that error_message(3) returns.

Heimdal uses a threadsafe(r) version of the com_err interface; the global com_err table isn't initialised. Then error_message(3) returns quite a boring error string (just the error code itself).

Previous: API differences to MIT Kerberos, Up: Programming with Kerberos

9.5 File formats

This section documents the diffrent file formats that are used in Heimdal and other Kerberos implementations.

9.5.1 keytab

The keytab binary format is not a standard format. The format has evolved and may continue to. It is however understood by several Kerberos implementations including Heimdal, MIT, Sun's Java ktab and are created by the ktpass.exe utility from Windows. So it has established itself as the defacto format for storing Kerberos keys.

The following C-like structure definitions illustrate the MIT keytab file format. All values are in network byte order. All text is ASCII.

       keytab {
           uint16_t file_format_version;                    /* 0x502 */
           keytab_entry entries[*];
       keytab_entry {
           int32_t size;
           uint16_t num_components;   /* subtract 1 if version 0x501 */
           counted_octet_string realm;
           counted_octet_string components[num_components];
           uint32_t name_type;       /* not present if version 0x501 */
           uint32_t timestamp;
           uint8_t vno8;
           keyblock key;
           uint32_t vno; /* only present if >= 4 bytes left in entry */
       counted_octet_string {
           uint16_t length;
           uint8_t data[length];
       keyblock {
           uint16_t type;

All numbers are stored in network byteorder (big endian) format.

The keytab file format begins with the 16 bit file_format_version which at the time this document was authored is 0x502. The format of older keytabs is described at the end of this document.

The file_format_version is immediately followed by an array of keytab_entry structures which are prefixed with a 32 bit size indicating the number of bytes that follow in the entry. Note that the size should be evaluated as signed. This is because a negative value indicates that the entry is in fact empty (e.g. it has been deleted) and that the negative value of that negative value (which is of course a positive value) is the offset to the next keytab_entry. Based on these size values alone the entire keytab file can be traversed.

The size is followed by a 16 bit num_components field indicating the number of counted_octet_string components in the components array.

The num_components field is followed by a counted_octet_string representing the realm of the principal.

A counted_octet_string is simply an array of bytes prefixed with a 16 bit length. For the realm and name components, the counted_octet_string bytes are ASCII encoded text with no zero terminator.

Following the realm is the components array that represents the name of the principal. The text of these components may be joined with slashs to construct the typical SPN representation. For example, the service principal HTTP/www.foo.net@FOO.NET would consist of name components "HTTP" followed by "www.foo.net".

Following the components array is the 32 bit name_type (e.g. 1 is KRB5_NT_PRINCIPAL, 2 is KRB5_NT_SRV_INST, 5 is KRB5_NT_UID, etc). In practice the name_type is almost certainly 1 meaning KRB5_NT_PRINCIPAL.

The 32 bit timestamp indicates the time the key was established for that principal. The value represents the number of seconds since Jan 1, 1970.

The 8 bit vno8 field is the version number of the key. This value is overridden by the 32 bit vno field if it is present. The vno8 field is filled with the lower 8 bits of the 32 bit protocol kvno field.

The keyblock structure consists of a 16 bit value indicating the encryption type and is a counted_octet_string containing the key. The encryption type is the same as the Kerberos standard (e.g. 3 is des-cbc-md5, 23 is arcfour-hmac-md5, etc).

The last field of the keytab_entry structure is optional. If the size of the keytab_entry indicates that there are at least 4 bytes remaining, a 32 bit value representing the key version number is present. This value supersedes the 8 bit vno8 value preceeding the keyblock.

Older keytabs with a file_format_version of 0x501 are different in three ways:

All integers are in host byte order [1].
The num_components field is 1 too large (i.e. after decoding, decrement by 1).
The 32 bit name_type field is not present.

[1] The file_format_version field should really be treated as two separate 8 bit quantities representing the major and minor version number respectively.

9.5.2 Heimdal database dump file

Format of the Heimdal text dump file as of Heimdal 0.6.3:

Each line in the dump file is one entry in the database.

Each field of a line is separated by one or more spaces, with the exception of fields consisting of principals containing spaces, where space can be quoted with \ and \ is quoted by \.

Fields and their types are:

     	Quoted princial (quote character is \) [string]
     	Keys [keys]
     	Created by [event]
     	Modified by [event optional]
     	Valid start time [time optional]
     	Valid end time [time optional]
     	Password end valid time [time optional]
     	Max lifetime of ticket [time optional]
     	Max renew time of ticket [integer optional]
     	Flags [hdb flags]
     	Generation number [generation optional]
     	Extensions [extentions optional]

Fields following these silently are ignored.

All optional fields will be skipped if they fail to parse (or comprise the optional field marker of "-", w/o quotes).


     fred@EXAMPLE.COM 27:1:16:e8b4c8fc7e60b9e641dcf4cff3f08a701d982a2f89ba373733d26ca59ba6c789666f6b8bfcf169412bb1e5dceb9b33cda29f3412:-:1:3:4498a933881178c744f4232172dcd774c64e81fa6d05ecdf643a7e390624a0ebf3c7407a:-:1:2:b01934b13eb795d76f3a80717d469639b4da0cfb644161340ef44fdeb375e54d684dbb85:-:1:1:ea8e16d8078bf60c781da90f508d4deccba70595258b9d31888d33987cd31af0c9cced2e:- 20020415130120:admin@EXAMPLE.COM 20041221112428:fred@EXAMPLE.COM - - - 86400 604800 126 20020415130120:793707:28 -

Encoding of types are as follows:

          kvno:[masterkvno:keytype:keydata:salt]{zero or more separated by :}

kvno is the key version number.

keydata is hex-encoded

masterkvno is the kvno of the database master key. If this field is empty, the kadmin load and merge operations will encrypt the key data with the master key if there is one. Otherwise the key data will be imported asis.

salt is encoded as "-" (no/default salt) or

          salt-type /
          salt-type / "string"
          salt-type / hex-encoded-data

keytype is the protocol enctype number; see enum ENCTYPE in include/krb5_asn1.h for values.


          kvno=27,{key: masterkvno=1,keytype=des3-cbc-sha1,keydata=..., default salt}...

Format of the time is: YYYYmmddHHMMSS, corresponding to strftime format "%Y%m%d%k%M%S".

Time is expressed in UTC.

Time can be optional (using -), when the time 0 is used.




time is as given in format time

principal is a string. Not quoting it may not work in earlier versions of Heimdal.



hdb flags
Integer encoding of HDB flags, see HDBFlags in lib/hdb/hdb.asn1. Each bit in the integer is the same as the bit in the specification.

usec is a the microsecond, integer. gen is generation number, integer.

The generation can be defaulted (using '-') or the empty string


HDB-extension is encoded the DER encoded HDB-Extension from lib/hdb/hdb.asn1. Consumers HDB extensions should be aware that unknown entires needs to be preserved even thought the ASN.1 data content might be unknown. There is a critical flag in the data to show to the KDC that the entry MUST be understod if the entry is to be used.

Next: , Previous: Programming with Kerberos, Up: Top

10 Migration

10.1 General issues

When migrating from a Kerberos 4 KDC.

10.2 Order in what to do things:

Previous: Migration, Up: Top

Appendix A Acknowledgments

Eric Young wrote “libdes”. Heimdal used to use libdes, without it kth-krb would never have existed. Since there are no longer any Eric Young code left in the library, we renamed it to libhcrypto.

All functions in libhcrypto have been re-implemented or used available public domain code. The core AES function where written by Vincent Rijmen, Antoon Bosselaers and Paulo Barreto. The core DES SBOX transformation was written by Richard Outerbridge. imath that is used for public key crypto support is written by Michael J. Fromberger.

The University of California at Berkeley initially wrote telnet, and telnetd. The authentication and encryption code of telnet and telnetd was added by David Borman (then of Cray Research, Inc). The encryption code was removed when this was exported and then added back by Juha Eskelinen.

The popper was also a Berkeley program initially.

Some of the functions in libroken also come from Berkeley by way of NetBSD/FreeBSD.

editline was written by Simmule Turner and Rich Salz. Heimdal contains a modifed copy.

The getifaddrs implementation for Linux was written by Hideaki YOSHIFUJI for the Usagi project.

The pkcs11.h headerfile was written by the Scute project.

Bugfixes, documentation, encouragement, and code has been contributed by:

Alexander Boström
Andreaw Bartlett
Björn Sandell
Brandon S. Allbery KF8NH
Brian A May
Chaskiel M Grundman
Cizzi Storm
Daniel Kouril
David Love
Derrick J Brashear
Douglas E Engert
Frank van der Linden
Jason McIntyre
Johan Ihrén
Jun-ichiro itojun Hagino
Ken Hornstein
Magnus Ahltorp
Marc Horowitz
Mario Strasser
Mark Eichin
Mattias Amnefelt
Michael B Allen
Michael Fromberger
Michal Vocu
Miroslav Ruda
Petr Holub
Phil Fisher
Rafal Malinowski
Richard Nyberg
Åke Sandgren
and we hope that those not mentioned here will forgive us.

All bugs were introduced by ourselves.

Table of Contents


[1] the other way is not always trivial either, but usually easier

[2] at least until secure DNS is commonly available