Re: [NFS] Re: [PATCH][RFC] NFS: Improving the access cache

[Trond Myklebust <trond.myklebust@fys.uio.no>]

On Mon, 2006-05-08 at 10:07 -0400, Peter Staubach wrote:
> Steve Dickson wrote:
> 
> > Peter Staubach wrote:
> >
> >>>>
> >>>> 2.  Use a radix tree per inode.  
> >>>
> >>>
> >>>
> >>> Using radix trees actual makes things much simpler... Good idea, imo.
> >>>
> >>
> >> It does seem like a good idea, although just using the uid is not 
> >> sufficient
> >> to identify the correct access cache entry.  The group and groups 
> >> list must
> >> also be used.  Can the radix tree implementation support this?
> >
> > We could use (nfsi + uid) as the index... but its not clear what that
> > would buys us... And the group and group list were never part of the
> > cache in the first place.... is this something you feel needs to be
> > added or am I just not understanding.... 
> 
> 
> Yes, I believe that the entire user identification, uid, gid, groups list,
> needs to be used to identify the correct cache entry.  An easy example of
> differing access rights, but with the same uid, is an application which is
> setgid.
> 
> I believe that the "key" to the cache entry should be the entire user
> identification that the NFS server sees and uses when calculating access
> rights.  Using the uid as part of a hash key is okay and probably a good
> idea, but I don't think that the uid is sufficient to completely identify
> a specific cache entry.
> 
> Given this, then I suspect that the current access cache is broken...

No it is not: it uses the full RPC cred as the index.

Cheers,
  Trond