This page explains why you want a root instance
joeuser/root@ATHENA.MIT.EDU and an extra instance
joeuser/extra@ATHENA.MIT.EDU, how to get them, and how to use them
There are three parts of a Kerberos name: a principal, an optional
instance, and a realm. The principal is typically your username
(for Kerberos identities belonging to a user), and the realm, at MIT,
is usually ATHENA.MIT.EDU. (Other realms you may see are
CSAIL.MIT.EDU, ZONE.MIT.EDU, CS.CMU.EDU, etc.) For the Kerberos
identity you typically regard as your own, the one that you use to log
in to Athena with your regular password, the instance is null
(empty). However, you can ask for additional instances, usually a
"root" or "extra" instance. You can use them in places where you
wouldn't want to use your regular Athena password. You usually write a
non-null instance as, e.g.,
The entire triplet is also often referred to as a principal or instance, depending on context: "There are two Kerberos principals that can log in to this server, namely, my extra instance and my root instance. My null instance can't log in."
Root instances are often used when logging in to servers that have some security import. Most students regularly log in to Athena cluster workstations and quickstations and often type their password on other people's laptops to SSH or get to webmail. This puts the password at huge risk for theft. Athena passwords have been stolen from clusters in the past, so it's not the wisest idea to let servers that you're asking hundreds of people to log into and use be controlled by a password you're typing everywhere.
So you get a root instance, and only type the root instance password on computers you trust and only when you need to. You should only type your password on a computer that you trust, such a private laptop or workstation. If you have access to physical hardware that you trust, but not a trusted installation, you can use a printed LiveCD, or a one-time LiveUSB as created by hardware that you trust.
You can also make things in Moira or AFS owned by your root instance,
if you don't want your null instance to be able to mess with mailing
lists or lockers. For Moira, make them owned by
KERBEROS:yourname/root@ATHENA.MIT.EDU. For AFS, ask accounts or
afsreq to get you a 'pts id', basically an account with the AFS
servers, and then you can give bits to yourname.root and start
blanching your root instance onto AFS groups.
To use another instance, just specify it to the kinit command, e.g.,
Because you would want to use your null instance tickets most of the time but your root instance tickets occasionally, a couple of people have developed shell scripts to make it easy to switch between them.
nelhage has the krbroot
command, with which you
use syntax like "krbroot ssh linerva" when you want to use your
root instance for a command. You can also "krbroot shell". adehnert extended it to add a
krbroot screen subcommand, use
ATHENA_USER, and support arbitrary principals.
quentin and broder wrote kdo, which is similar in spirit to krbroot, but designed for Mac OS X. It takes advantage of the fact that OS X's Kerberos implementation is better at handling multiple tickets.
geofft has kpagsh, a way of configuring your .bashrc to prompt you for tickets (null instance by default) if you start a shell and don't have tickets. If you want to switch tickets, you start a new shell, and also a new PAG, which lets you use multiple AFS credentials at once, too. It also modifies your prompt.
These aliases are also careful to get shorter lifetime tickets that are marked nonforwardable. Some versions of SSH try to forward tickets by default. Since you might let your root instance tickets access many servers, but not trust all of these servers equally, you don't want your tickets to be forwardable. (Thankfully, recent Debian, Ubuntu, and OS X have turned off this default, but it's a good precaution.)
Another thing you might want is an extra instance. Some people use
these just like another root instance, with slightly lower
security. But a common use is something less secure than your null
instance. For example, if you're writing a zephyrbot to run on a
shared server like
scripts.mit.edu, the zephyrbot will need Kerberos
tickets to subscribe to zephyrs. But you don't want to leave your
Kerberos password in a file in your locker, so you can leave your
extra instance's password instead.
You need to show up in person to IS&T User Accounts in E17 during business hours with a photo ID to obtain new Kerberos identities. For the reasons described above, being in control of your null instance and sending a zephyr or authenticated e-mail with it does not mean that you can go ahead and make changes to your root or extra instance too. While you're there, be sure to ask for a pts id, if you want to use your tickets with AFS.
You should change your root instance’s password with a command like this, to upgrade your key from critically weak DES encryption algorithm to strong AES encryption:
kadmin -p andersk/root -q 'cpw -e aes256-cts:normal,aes128-cts:normal andersk/root'
(Note: This previously made your password incompatible with a handful of services that you should not have been using with your root instance in the first place, but these services have now been fixed.) You can confirm the change with
kadmin -p andersk/root -q 'getprinc andersk/root'
which should list a line like
Key: vno 4, aes256-cts-hmac-sha1-96, no salt
If you change your password again, you will need to specify your desired enctypes with the -e option; otherwise, they will be reset to the defaults.