Re: nfs4 mounts failing with 3.6.0-rc1

[Jeff Layton <jlayton@redhat.com>]

On Tue, 7 Aug 2012 22:36:12 +0000
"Myklebust, Trond" <Trond.Myklebust@netapp.com> wrote:

> On Tue, 2012-08-07 at 18:26 -0400, Jeff Layton wrote:
> > On Tue, 7 Aug 2012 21:23:34 +0000
> > "Myklebust, Trond" <Trond.Myklebust@netapp.com> wrote:
> > 
> > > On Tue, 2012-08-07 at 16:25 -0400, J. Bruce Fields wrote:
> > > > On Tue, Aug 07, 2012 at 04:10:49PM -0400, Bryan Schumaker wrote:
> > > > > On 08/07/2012 03:57 PM, J. Bruce Fields wrote:
> > > > > > On Tue, Aug 07, 2012 at 03:44:53PM -0400, Bryan Schumaker wrote:
> > > > > >> On 08/07/2012 03:42 PM, J. Bruce Fields wrote:
> > > > > >>> On Tue, Aug 07, 2012 at 01:09:32PM -0400, Jeff Layton wrote:
> > > > > >>>> On Sat, 4 Aug 2012 15:01:04 -0400
> > > > > >>>> "J. Bruce Fields" <bfields@fieldses.org> wrote:
> > > > > >>>>
> > > > > >>>>> On Fri, Aug 03, 2012 at 10:00:39PM -0400, Jeff Layton wrote:
> > > > > >>>>>> On Fri, 3 Aug 2012 20:08:19 -0400
> > > > > >>>>>> "J. Bruce Fields" <bfields@fieldses.org> wrote:
> > > > > >>>>>>
> > > > > >>>>>>> I'm getting
> > > > > >>>>>>>
> > > > > >>>>>>> 	# mount -tnfs -onfsvers=4 pip1:/exports /mnt/
> > > > > >>>>>>>
> > > > > >>>>>>> (OK, admittedly that's with 3.6.0-rc1 + a few experimental patches, but
> > > > > >>>>>>> I doubt they're related.)
> > > > > >>>>>>>
> > > > > >>>>>>> Also:
> > > > > >>>>>>>
> > > > > >>>>>>> 	[root@pip2 ~]# modprobe nfs4
> > > > > >>>>>>> 	[root@pip2 ~]# lsmod|grep nfs4
> > > > > >>>>>>> 	[root@pip2 ~]# 
> > > > > >>>>>>>
> > > > > >>>>>>> --b.
> > > > > >>>>>>
> > > > > >>>>>> I hit the same problem...
> > > > > >>>>>>
> > > > > >>>>>> Try removing /usr/lib/modprobe.d/nfs.conf (assuming you're running
> > > > > >>>>>> Fedora).
> > > > > >>>>>
> > > > > >>>>> Oog, right.
> > > > > >>>>>
> > > > > >>>>> But, without testing--won't that make v4 mounts fail on older kernels?
> > > > > >>>>
> > > > > >>>> Actually, now that I look, this does not seem to break on older kernels
> > > > > >>>> as long as you use a syntax like:
> > > > > >>>>
> > > > > >>>>     # mount -t nfs server:/export /mnt/point -o vers=4
> > > > > >>>>
> > > > > >>>> ...if, however you use a syntax like:
> > > > > >>>>
> > > > > >>>>     # mount -t nfs4 server:/export /mnt/point
> > > > > >>>>
> > > > > >>>> ...then it fails without the above file in place. I guess the question
> > > > > >>>> we have to answer is: Do we want to continue to support the "-t nfs4"
> > > > > >>>> mount syntax?
> > > > > >>>
> > > > > >>> I think you're right that we want to deprecate it.
> > > > > >>>
> > > > > >>> Though this is a bit of a harsh way to do it--would have been nice to
> > > > > >>> have some transition period with a warning or something.
> > > > > >>
> > > > > >> I didn't expect this to be broken, both ways of mounting still work on my VMs so I expected them to work for everybody else too.
> > > > > > 
> > > > > > Huh.  Just checked on an old kernel without an "alias nfs4 nfs" in
> > > > > > modprobe configuration, and sure enough I get "No such device".
> > > > > > 
> > > > > > Maybe you have some initscripts or something else that's loading the
> > > > > > nfs module for you before the mount?
> > > > > 
> > > > > My nfs-common daemon script loads sunrpc, nfs
> > > > 
> > > > Yep, that's why you're not seeing it.
> > > > 
> > > > > and nfsd but not nfs2, nfs3 or nfs4.
> > > > > 
> > > > > Could we rename the module to avoid the alias name collision?  Something like this (untested) maybe?
> > > > 
> > > > I don't think that will help.
> > > 
> > > It should if we also add back in
> > > 
> > > struct file_system_type nfs4_fs_type = {
> > >         .owner          = THIS_MODULE,
> > >         .name           = "nfs4",
> > >         .mount          = nfs_fs_mount,
> > >         .kill_sb        = nfs_kill_super,
> > >         .fs_flags       = FS_RENAME_DOES_D_MOVE|FS_REVAL_DOT|FS_BINARY_MOUNTDATA,
> > > };
> > > 
> > > and then add that to register_nfs_fs()/unregister_nfs_fs(). As far as I
> > > can see, that will trigger all the right incantations in
> > > nfs_validate_mount_data() to mount an NFSv4 filesystem.
> > > 
> > 
> > So, move the nfs4 fstype definition back into nfs.ko? That would
> > ensure that you could still do a "-t nfs4" mount with that modprobe
> > alias in place.
> 
> No. I mean to add a separate nfs4_fstype definition in nfs.ko, and
> register it so that the VFS recognises the 'nfs4' filesystem name.
> 
> > I think Bryan is correct though. We'll also need to rename nfs4.ko or
> > you won't ever be able to call request_module for it in order to plug
> > it in if you have that module alias in place.
> 
> You did note my use of the word "also" above?
> 
> > Does this mean that we plan to support the "-t nfs4" mount syntax in
> > perpetuity?
> 
> No, but I also agree with Bruce's point that we shouldn't pull the plug
> without some prior notice. We should at least keep an entry in
> Documentation/feature-removal-schedule.txt for a few kernel revisions.
> 

Ok, I follow you now...

Fair warning is reasonable, perhaps we should shoot for removing
support in 3.9 or 3.10?

I think we'll also need to consider retiring the nfs4_fs_type
altogether, or at least make them all show up as "nfs"
in /proc/self/mounts. Otherwise remounts could be problematic...

-- 
Jeff Layton <jlayton@redhat.com>