Return-Path: Received: from mx1.redhat.com ([209.132.183.28]:50426 "EHLO mx1.redhat.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1750779AbdLGPim (ORCPT ); Thu, 7 Dec 2017 10:38:42 -0500 Subject: Re: [RFC PATCH] rpc.svcgssd: add ability to override hostname To: Leigh Brown Cc: linux-nfs@vger.kernel.org References: <93e03b4df2fecbc505547d80935d5be7@doppler.thel33t.co.uk> From: Steve Dickson Message-ID: <7cf41533-124c-b1be-e527-79fd768383d0@RedHat.com> Date: Thu, 7 Dec 2017 10:38:40 -0500 MIME-Version: 1.0 In-Reply-To: Content-Type: text/plain; charset=utf-8 Sender: linux-nfs-owner@vger.kernel.org List-ID: On 12/06/2017 04:25 PM, Leigh Brown wrote: > Hi Steve, > > On 2017-11-21 17:03, Steve Dickson wrote: >> Hello, >> >> On 11/20/2017 09:27 AM, Leigh Brown wrote: >>> Add the -h option to rpc.svcgssd to allow the hostname to be overridden. >>> This is useful in clustered configurations using NVSv4 and Kerberos to >>> ensure the hostname is set to the service name of the cluster. >> A couple things... >> >> 1) The patch did not apply for krb5_util.c or svcgssd.c. Not >>     clear why.. but they didn't >> 2) The patch cause a "implicit declaration of function" >>    warning because the new routines were not added to gss_util.h >> 3) Since the return value of gssd_sethostname() is never checked >>    why not make it void and log an warning when something >>    goes wrong. >> >> Finally, adding a command line argument is always a touchy thing, >> supporting unnecessary flags is the last thing we want to do. So.. >> Please give me an example how this will be used, I know you say in >> a cluster configuration, but what does that mean... A little >> context please. Also is there any around not adding this flag >> and achieving the same results. >> >> I'm not totally against adding this flag I just want to >> investigate all avenues.. > > TL;DR Sorry for wasting your time, please ignore the patch. > > Apologies for the delay in getting back to you.  I have been using this patch > for the last three years on my server at home.  I have two N40L Microservers > running Xen.  I set up an NFS cluster using NVSv4 with Kerberos authentication, > DRBD and Pacemaker.  When I tested it back in 2015 or so, it would not fail over > cleanly when I mounted the NFS mount on the service NFS name.  After messing > around with setting the hostname in /etc/init.d/nfs-kernel-server  I eventually > came up with the patch to rpc.svcgssd and it fixed the issue. > > Fast forward to 2017, and I thought I might be a good idea to send this patch > for other people to use.  Anyway, when I got your email I thought I had better > create a couple of test VMs, set them up like my working setup and show how > things don't work at first without the patch and then show how they work with > the patch......except it worked perfectly. > > This is quite embarrassing, actually.  I spent a few days trying to find out > why it now worked without success, eventually I installed the stock package on > my normal server and it still worked (to be fair I've upgraded Debian on the VMs > once or twice in that time) .  I'm too lazy to have done all the work for no reason > so I'm hoping that back then there was a genuine reason why it wouldn't work > and that in the interim something has changed somewhere that fixes the issue. > > Anyway, thanks very much for the feedback and sorry for wasting your time. > Not a problem... Thank you for the explanation... steved.