Cross domain authentication

Cross domain authentication has been implemented in two different forms recently (Jul 2005), however there is still discussion about the "best" solution.

IMPLEMENTATION 2

Central authentication server http://groups-beta.google.com/group/comp.os.plan9/msg/4625ee53672ef106?dmode=source&hl=en

IMPLEMENTATION 1

Distributed authentication http://groups-beta.google.com/group/comp.os.plan9/msg/3c379056e750c514?dmode=source&hl=en

A EARLIER, SIMILAR PROPOSAL

(Taken from a post by David Presotto to the 9grid list: http://nwn.definitive.org/pipermail/9grid/2004-September/000269.html)

In the current scheme, when a user authenticates, he/she gets a list of domains that the server will accept and picks one to use (i.e. one he has a key for). That should provide the second part of the name, i.e., if I'm presotto authenticating via the domain bignose.com then I should be presotto@bignose.com. Tacking on the @bignose.com is currently missing.

That of course leaves the server to map the domain bignose.com to an auth server that it trusts (i.e. has a key for and trusts others to authenticate via). Currently that's done via the lib/ndb entry

	authdom=bignose.com auth=1.2.3.4

So we have a way to trust other authentication domains but not yet a way to tag the resulting name. However, that's like 1 line of code in factotum.

However, this requires that there be a shared key twixt each server and each domain that it trusts. I'll leave it up to you if that's too much.

If the server is a cpu server, then the caller can export its factotum to its process on that cpu server so that it can authenticate further. Is this a good idea? Would we be letting too much access to the factotum? Might we want the user to start a second less capable factotum for 9net access?

The client coming into the cpu server will need at least readonly access to binaries and to enough directory structure to set up a name space on that machine. I already had to make that work for the deep lens survey stuff. I made ken's group 'noworld' access work for that. If we put each 9net user in the 'noworld' group, then we can point him to a file server and he can only get at files that are accessible by his uid or by the 'noworld' group. You could even give him some write access if you wanted to. It keeps such users from accessing anything that was accidentally made world readable or writable. It does mean that you have to change the file server so that anyone coming in as a@b is just treated as a 'noworld' group user.

My 2 cents.