Re: [PATCH 00/59] Introduce credentials

["Paul E. McKenney" <paulmck@linux.vnet.ibm.com>]

On Thu, Aug 28, 2008 at 10:45:03AM +0100, Alan Cox wrote:
> On Thu, 28 Aug 2008 10:19:44 +0100
> David Howells <dhowells@redhat.com> wrote:
> 
> > Alan Cox <alan@lxorguk.ukuu.org.uk> wrote:
> > 
> > > > We get a number of things:
> > > > 
> > > >  (1) Multiple credential changes all happen simultaneously (setresuid() for
> > > >      example).  The new set of credentials is committed with a single RCU
> > > >      assignment.
> > > 
> > > Makes sense - except for the question of bounding memory utilisation.
> > 
> > Would it make sense to call synchronise_rcu() from commit_creds() or from
> > setuid()/setresuid()/setgroups()/etc. to make sure that some user process
> > doing:
> > 
> > 	while (1)
> > 		alter_credentials();
> > 
> > doesn't run the system out of memory by having loads of frees waiting in RCU's
> > queues because put_cred() uses call_rcu() to defer the destruction.
> 
> I suspect you to - or every "nth" event - building up a small queue as
> would occur in normal usage probably isn't a problem.

Makes sense to me, given that we should not expect real-time determinism
out of alter_credentials().  ;-)

							Thanx, Paul