Date: Fri, 25 Jun 2010 18:35:13 -0400
From: Jeff Layton <jlayton@redhat.com>
To: Chuck Lever <chuck.lever@oracle.com>
Cc: linux-nfs@vger.kernel.org, trond.myklebust@fys.uio.no,
        Staubach_Peter@emc.com, bfields@fieldses.org
Subject: Re: [PATCH 1/2] sunrpc: split the vs_hidden flag into TCP and UDP
 variants (try #2)
Message-ID: <20100625183513.365f168a@tlielax.poochiereds.net>
In-Reply-To: <4C2510C6.902@oracle.com>
References: <1277480946-23844-1-git-send-email-jlayton@redhat.com>
	<4C24D057.4090403@oracle.com>
	<20100625120710.2ead8998@tlielax.poochiereds.net>
	<4C24D95F.6090301@oracle.com>
	<20100625134754.51e313fa@tlielax.poochiereds.net>
	<4C2510C6.902@oracle.com>
Content-Type: text/plain; charset=US-ASCII
Sender: linux-nfs-owner@vger.kernel.org
MIME-Version: 1.0

On Fri, 25 Jun 2010 16:25:42 -0400
Chuck Lever <chuck.lever@oracle.com> wrote:

> On 06/25/10 01:47 PM, Jeff Layton wrote:
> > nfsd shares the same sockets between all NFS versions. We can't change
> > that since you don't know what NFS version you're getting a call for
> > until you read in the data. There's really no way around this.
> 
> That is worth mentioning in a documenting comment before nfsd4_dispatch().
> 
> I feel that we're adding yet another kludge to the NFSD infrastructure 
> to preserve an aging user space API and transport design that was built 
> for an era when all NFS versions would always use every transport and 
> address family.
> 
> I agree with Peter that we should make an effort to use the RPC 
> infrastructure as it was intended, in order to advertise the correct 
> services and transport capabilities; and update our user space API to 
> allow user space to have more fine-grained control over what the kernel 
> will advertise and run.  But perhaps that's for another day.

Now that I've started poking at this, there are some other
irregularities. For instance doing this:

$ rpcinfo -n 2049 -u server 100003 1
rpcinfo: RPC: Program/version mismatch; low version = 2, high version = 4
program 100003 version 1 is not available

That looks reasonable, but those low/high versions are set at service
create time and don't seem to be governed by the write_versions
interfaces. We could certainly fix that with another set of hacks, but
maybe you and Peter have a point here and it would be better to fix
this at a more fundamental level.

I think I need to step back and ponder this a bit. Thanks to all for
the comments so far...
-- 
Jeff Layton <jlayton@redhat.com>