Return-Path: linux-nfs-owner@vger.kernel.org Received: from mx1.redhat.com ([209.132.183.28]:1548 "EHLO mx1.redhat.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1754805Ab2GYP5K (ORCPT ); Wed, 25 Jul 2012 11:57:10 -0400 From: David Howells In-Reply-To: <20120725155347.24392.44505.stgit@warthog.procyon.org.uk> References: <20120725155347.24392.44505.stgit@warthog.procyon.org.uk> <20120725155336.24392.25186.stgit@warthog.procyon.org.uk> To: Trond.Myklebust@netapp.com Cc: dhowells@redhat.com, bjschuma@netapp.com, linux-nfs@vger.kernel.org, steved@redhat.com, jlayton@redhat.com Subject: Re: [PATCH 2/2] NFS: Combine the idmapper key types Date: Wed, 25 Jul 2012 16:57:05 +0100 Message-ID: <24518.1343231825@warthog.procyon.org.uk> Sender: linux-nfs-owner@vger.kernel.org List-ID: Hi Trond, David Howells wrote: > The NFS idmapper has two key types (normal and legacy) but should only use one > if it can - otherwise it risks having twice as many keys as it would otherwise > need. > > Get rid of the legacy key type and have the normal key type have a > .request_key() op. The choice of which instantiation method is then made by > the upcaller, in order: > > (1) If there's no auxdata, the normal method is called, invoking > /sbin/request-key. > > (2) If there's something attached to the idmapper pipe (rpc.idmapd) then use > that. > > (3) Fall back to (1). > > Note that this does change the prioritisation of normal vs legacy if both are > available. I'm not sure whether this will be a problem, so it needs careful consideration. David