Re: nfs-utils-1.2.9 does not play well with linux 3.10.x

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

On Mon, Dec 16, 2013 at 11:24:03PM +0100, Joakim Tjernlund wrote:
> "J. Bruce Fields" <bfields@fieldses.org> wrote on 2013/12/16 21:23:45:
> 
> > From: "J. Bruce Fields" <bfields@fieldses.org>
> > To: Joakim Tjernlund <joakim.tjernlund@transmode.se>, 
> > Cc: linux-nfs@vger.kernel.org, steved@redhat.com
> > Date: 2013/12/16 21:23
> > Subject: Re: nfs-utils-1.2.9 does not play well with linux 3.10.x
> > 
> > On Mon, Dec 16, 2013 at 09:21:15PM +0100, Joakim Tjernlund wrote:
> > > "J. Bruce Fields" <bfields@fieldses.org> wrote on 2013/12/16 21:03:01:
> > > 
> > > > From: "J. Bruce Fields" <bfields@fieldses.org>
> > > > To: Joakim Tjernlund <joakim.tjernlund@transmode.se>, 
> > > > Cc: linux-nfs@vger.kernel.org, steved@redhat.com
> > > > Date: 2013/12/16 21:03
> > > > Subject: Re: nfs-utils-1.2.9 does not play well with linux 3.10.x
> > > > 
> > > > On Mon, Dec 16, 2013 at 01:54:19PM -0500, bfields wrote:
> > > > > On Mon, Dec 16, 2013 at 07:16:00PM +0100, Joakim Tjernlund wrote:
> > > > > > rpc.nfsd insists on adding "-4.2" when writing 
> > > /proc/fs/nfsd/versions :
> > > > > >   rpc.nfsd: Writing version string to kernel: +4.1 -4.2 -2 +3 +4
> > > > > > Which causes Linux to return an EIVAL error as 3.10 has no 
> support 
> > > for 4.2 
> > > > > > and
> > > > > > does not accept any reference to 4.2
> > > > > > 
> > > > > > It seems reasonable to me that Linux should accpect -4.2 as a 
> noop 
> > > and 
> > > > > > continue
> > > > > > processing the rest of the options but I am just guessing.
> > > > > > Anyhow, just to test I applied this commit to my 3.10.24 kernel:
> > > > > > 
> > > > > > 
> > > 
> http://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git/commit/?id=4bdc33ed5bd9fbaa243bda6fdccb22674aed6305
> 
> > > 
> > > > > > and now it accepts the "-4.2" but I have no idea if this messes 
> up 
> > > > > > something else.
> > > > > 
> > > > > That should be perfectly safe.
> > > 
> > > Thanks, I will keep this then until nfs-utils is working.
> > > 
> > > > > 
> > > > > I agree that we should teach the kernel to treat "-4.x" at least 
> as a
> > > > > no-op for unknown .x.  But nfs-utils also has to keep working with 
> 
> > > older
> > > > > kernels which don't do that.
> > > > > 
> > > > > The problem was introduced by 
> 12a590f8d556c00a9502eeebaa763d906222d521
> > > > > "rpc.nfsd: Allow v4.2 server support with the -V option".  That 
> should
> > > > > be using an array of ints not a bit array, so it can make the
> > > > > distinction between "off", "on", and "don't care".
> > > > 
> > > > So, something like this (untested).--b.
> > > 
> > > I tested this on my system(which has the above kernel patch) and I 
> noticed
> > > a difference:
> > 
> > Thanks!
> > 
> > >   rpc.nfsd: Writing version string to kernel: -2 +3 +4 
> > > which is different than previous
> > >  rpc.nfsd: Writing version string to kernel: +4.1 -4.2 -2 +3 +4
> > > 
> > > The "+4.1" is missing.
> > 
> > Yes, that's intentional.  Is it causing you any problem?
> 
> I don't know yet, but I don't think it would be a problem for me.
> However, are you not changing the defaults here? In that
> case someone else relying on 4.1 might have a problem I guess.

That's just restoring the behavior we had before
12a590f8d556c00a9502eeebaa763d906222d521, and will still result in 4.1
being turned on if the kernel is recent enough.

--b.