Re: [nfsv4] [FedFS] proposed ADMIN protocol procedure to enumerate server's NSDB store

[Nicolas Williams <[e-mail]>]

On Thu, Jul 08, 2010 at 01:13:32PM -0400, Chuck Lever wrote:
> On 07/ 8/10 12:35 PM, Everhart, Craig wrote:
> >Have I simply forgotten precisely why this is being proposed?
> >(Wouldn't be a surprise, here in vacation season.)
> >
> >My naïve mental model for an NFSv4 server would not necessarily
> >include a data structure that would match the thing being queried in
> >this op.  I'm not sure of the value-add of trying to include it.
> >
> >I'm sure we can discuss later today, but a written record is *so*
> >handy.
> 
> See below:
> 
> "Such a procedure can be used to check for stale NSDB name entries or
> misspellings, or to generate a menu of NSDB names in a graphical ADMIN
> client (ie for browsing the NSDB list on a server)."
> 
> The ADMIN server, not the NFSv4 server, maintains the NSDB list.
> Given that FEDFS_CREATE_JUNCTION must fail if the ADMIN server
> doesn't have connection information for the presented NSDB, the
> ADMIN server therefore must keep a permanent list of NSDB names.

I agree that the ADMIN server "therefore must keep" such a list, even
without an RPC for listing them.  For some implementors that might not
be a complete list, but I suppose that's not a problem we should be
concerned with here ("don't do that").

But it's worth pointing this implication out explicitly.

There's no additional implication regarding any ability to list
junctions, nor reference counting of NSDBs referred to by junctions.
But it'd be nice to have an option for listing junctions (for
implementors that can do it) and/or reference counting NSDBs (for
implementors that can do it).

Nico
--