Re: [pnfs] [PATCH v2 06/47] nfsd41: Add Kconfig symbols for NFSv4.1

["J. Bruce Fields" <bfields@fieldses.org>]

On Thu, Apr 02, 2009 at 09:41:33PM +0300, Benny Halevy wrote:
> On Apr. 02, 2009, 20:58 +0300, "J. Bruce Fields" <bfields@fieldses.org> wrote:
> > On Thu, Apr 02, 2009 at 01:55:17PM -0400, bfields wrote:
> >> On Thu, Apr 02, 2009 at 08:52:40PM +0300, Benny Halevy wrote:
> >>> On Apr. 02, 2009, 20:45 +0300, Benny Halevy <bhalevy@panasas.com> wrote:
> >>>> On Apr. 02, 2009, 19:54 +0300, "J. Bruce Fields" <bfields@fieldses.org> wrote:
> >>>>> On Thu, Apr 02, 2009 at 07:34:29PM +0300, Benny Halevy wrote:
> >>>>>> On Apr. 02, 2009, 17:31 +0300, "J. Bruce Fields" <bfields@fieldses.org> wrote:
> >>>>>>> On Thu, Apr 02, 2009 at 10:25:55AM -0400, Trond Myklebust wrote:
> >>>>>>>> On Thu, 2009-04-02 at 10:22 -0400, J. Bruce Fields wrote:
> >>>>>>>>> On Thu, Apr 02, 2009 at 10:16:40AM -0400, Trond Myklebust wrote:
> >>>>>>>>>> On Thu, 2009-04-02 at 16:46 +0300, Benny Halevy wrote:
> >>>>>>>>>>> Trond, please speak up if you want to remove CONFIG_NFS_V4_1 as well.
> >>>>>>>>>>> On the client side minorversion 1 will be used only if the user
> >>>>>>>>>>> explicitly asked for it with mount -o minorversion=1.
> >>>>>>>>>> I'd feel more comfortable with being able to compile it out until the
> >>>>>>>>>> stability of the code has been established. I'd certainly want to be
> >>>>>>>>>> able to do that on the server side, since it has no other means to
> >>>>>>>>>> restrict the protocol version should it turn out that NFSv4.1 has some
> >>>>>>>>>> fatal condition.
> >>>>>>>>> I think it's acceptable given an interface that allows choosing the
> >>>>>>>>> supported minorversion at runtime (and that defaults 4.1 to off).
> >>>>>>>> Is there such an interface on the server?
> >>>>>>> That's the patch Benny just posted.  It seems like a pretty simple
> >>>>>>> extension of the existing version-choosing interface
> >>>>>>> (/proc/fs/nfsd/versions), though I think the version he posted defaults
> >>>>>>> 4.1 to on?  I need to take another look.
> >>>>>> That's right.  This can be changed trivially,
> >>>>>> but I'll have to find and work the reason why the settings reset on
> >>>>>> service nfs restart, otherwise it will reset itself.
> >>>>> Maybe you're starting and stopping the server using distro init scripts
> >>>>> that load and unload the nfsd module?
> >>>> I'm using redhat's /etc/init.d/nfs
> >>>> My impression was that it doesn't unload the nfsd module when the service
> >>>> is stopped.  However it does run modprobe on "start".  I wonder if that's
> >>>> causing the module to reload.
> >>> Hmm, after "stop" nfsd is up and must stay up to keep serving
> >>> /proc/fs/nfsd/versions.
> >>>
> >>> Giving modprobe the --first-time option reports:
> >>> FATAL: Module nfsd already in kernel.
> >> Yeah, actually, /proc/fs/nfsd/versions is probably getting written
> >> somewhere in there.  That may be the effect of the --no-nfs-version
> >> mountd argument.
> > 
> > Whoops, sorry, rpc.nfsd: it should be the --no-nfs-version argument to
> > rpc.nfsd.  See nfssvc_versbits() in nfs-utils/support/nfs/nfssvc.c.
> 
> OK. I see that now.
> It looks like rpc.nfsd's command line syntax is all negative.
> I.e. all versions are on by default and they can be turned off
> using the -N option.  I can either add an option to enable minorversion
> or have it enabled by default and add an option to disable it.

I'm not a big fan of that commandline syntax--seems to me it should give
a range rather than a list of versions not to support.

For now the important thing is that the kernel default minorversion 1 to
off.  We can worry about what rpc.nfsd should do later--I guess a
commandline option to turn on specific minor versions would work....
Maybe allow --nfs-version as well as --no-nfs-version?

--b.