Kerberos is a frequently misunderstood monster that lives in darkness, and eating unaware explorers. This guide is your 5 minute introduction to krb5, and how to manage it effectively
What Is Kerberos
Kerberos is a single-sign on service. When a user is autheticated by kerberos, they can access all kerberosized services without having to re-autheticate. This allows for things like node-to-node scps to work without having to deal with setting up per user public/private keys. The main unit in kerberos is the principle, which is simply an account which tickets can be issued for. Users and hosts have kerberos accounts to autheticate with themselves and other nodes across the network. Tickets are granted to users which are handed to kerberosed services to facility authetication.
As of writing, the only kerberosized service we're running is SSH.
Principles are generated through the kadmin tool; you have to kinit as an krb admin before you can run this command, they're a string, and a password; with the string determining what it is, here are some examples
- mcasadevall@LI694-22 - a user
- krb/admin@LI694-22 - krb admin (admin accounts end in /admin, normal admins should use krb/admin for admin work)
- host/hydrogen.li694-22@LI694-22 - a host. Hosts must be FQDN against the internal domain to allow them to resolve correctly. hosts always have a random password
Straight forward, right?
Keytabs are used for host-to-host authetication and preauthorization. For the most part, all you need to know is a host needs to have its own key in its keytab. If you following the new host instructions, this was done during KRB5 setup.
To access kadmin, you need to autheticate as an administrator and get a ticket. You should use the krb/admin account, the password is in the master_password file on helium.
mcasadevall@lithium:~# kinit krb/admin # (password is in master_passwd on ldap-master) Password for krb/admin@LI694-22: mcasadevall@lithium:~$ kadmin Authenticating as principal krb/admin@LI694-22 with password. Password for krb/admin@LI694-22: kadmin:
Kerberos admins are controlled by an ACL on the master server (helium) in /etc/krb5kdc/kadmin.acl
NOTE: if you're on helium, you can also use kadmin.local to edit the kerberos realm. This edits the realm directly instead of of going through the DB. This is useful if we're locked out due to kerberos loosing its mind. It functions identically to kadmin expect you don't need to kinit first.
Seeing What's In Kerberos
The list_principles command is used for this purpose
kadmin: list_principals K/M@LI694-22 audioguy@LI694-22 host/boron.li694-22@LI694-22 host/carbon.li694-22@LI694-22 ...
Creating users is the most common task, and doing so is very simple. Just add_principle, set a password, and you're done
kadmin: add_principal martyb WARNING: no policy specified for martyb@LI694-22; defaulting to no policy Enter password for principal "martyb@LI694-22": Re-enter password for principal "martyb@LI694-22": Principal "martyb@LI694-22" created. kadmin: q root@helium:/root#
NOTE: You shouldn't need to do this under normal cirmstances except during host setup. Changing a hosts principle will break tickets issued and require users to kdestroy them.
See the node setup guide for a more in-depth instructions. Creating a host principle is similar to a user except we generate a random password (which isn't used for anything since hosts use keytabs), then generate a keytab. You must be root to write the keytab file
add_principal -randkey host/carbon.li694-22@LI694-22 ktadd host/carbon.li694-22@LI694-22
kadmin: add_principal -randkey host/carbon.li694-22@LI694-22 WARNING: no policy specified for host/carbon.li694-22@LI694-22; defaulting to no policy add_principal: Principal or policy already exists while creating "host/carbon.li694-22@LI694-22". kadmin: ktadd host/carbon.li694-22@LI694-22 (lots of information) kadmin: quit
Deletion is done with the delete_principle command. This will cause tickets to fail to validate, and for hosts, require a new keytab be generated. This shouldn't be a frequent operation
TBD, I have it semi-working, *but* replication disabled. Document it when its fully setup