[OpenAFS-devel] Re: patch: make jafs work again (at least the libadmin portion)

[Marcus Watts]

<< mumble mumble must remember never to send mail to "megacz@cs.berkeley.edu
it only bounces mail mumble mumble >>...

> Marcus Watts <mdw@spam.ifs.umich.edu> writes:
> > I'm seriously considering coming up with a new lighter-weight
> > "from-scratch" JNI implementation to solve some of the things folks
> > want to do here at UM.
> 
> How feasible is it to implement something like this using JNI
> interfaces to the kernel CM rather than UKERNEL?
> 
>   - a

Quite feasible - depending on what the goal is.

In my immediate case, I won't actually care about UKERNEL,
and I'd like not to care about the kernel CM.  I want something that
will allow me to perform administrative actions on pts, particularly
"createu".  For that, all I need is something that can get an
AFS service ticket, followed by something that can do the logical
equivalent of pr_Initialize & pr_CreateUser, etc.  Since
pr_Initialize uses static globals (pruclient) architecturally
I'd ideally rather use something else, perhaps something like
ugen_ClientInit and ubik_PR_NewEntry.  Since ugen_ClientInit
likes to use the kernel CM for tokens, that's not ideal.

So maybe I ought to have my list of goals for such an
interface.  This is not complete, but for starters:
	> simple 1-1 or at least predictable and controllable
	relationship between any calls made to it, and any
	network calls it makes in turn.

	> least complicated data possible passed between C & JNI.

	> reasonable flexibility in terms of credentials
	management.  I'd like to be able to use any of a keytab,
	k5 credentials cache, or afs cm token, and I'd like
	various additional choices and options.

	> simple java configuration.  jaas is just terrible here.
	I should not need -Dthis and -Dthat to make things work.
	
				-Marcus Watts