Received-SPF: pass (google.com: domain of linux-nfs-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Content-Type: text/plain; charset="utf-8"
Content-Transfer-Encoding: quoted-printable
MIME-Version: 1.0
From:   "NeilBrown" <neilb@suse.de>
To:     "John Stoffel" <john@stoffel.org>
Cc:     "Al Viro" <viro@zeniv.linux.org.uk>,
        "Linus Torvalds" <torvalds@linux-foundation.org>,
        "Daire Byrne" <daire@dneg.com>,
        "Trond Myklebust" <trond.myklebust@hammerspace.com>,
        "Chuck Lever" <chuck.lever@oracle.com>,
        "Linux NFS Mailing List" <linux-nfs@vger.kernel.org>,
        linux-fsdevel@vger.kernel.org,
        "LKML" <linux-kernel@vger.kernel.org>
Subject: Re: [PATCH 10/10] NFS: support parallel updates in the one directory.
In-reply-to: <25352.59221.528023.534884@quad.stoffel.home>
References: <166147828344.25420.13834885828450967910.stgit@noble.brown>,
 <166147984378.25420.4023980607067991846.stgit@noble.brown>,
 <25352.59221.528023.534884@quad.stoffel.home>
Date:   Sat, 27 Aug 2022 09:13:31 +1000
Message-id: <166155561138.27490.7464518594320056247@noble.neil.brown.name>
Precedence: bulk

On Sat, 27 Aug 2022, John Stoffel wrote:
> >>>>> "NeilBrown" =3D=3D NeilBrown  <neilb@suse.de> writes:
>=20
> NeilBrown> NFS can easily support parallel updates as the locking is done o=
n the
> NeilBrown> server, so this patch enables parallel updates for NFS.
>=20
> Just curious, how is this handled if I have a server with an EXT#
> filesystem which is exported via NFS to multiple clients.  If those
> NFS clients are doing lots of changes in a single directory, I can see
> how the NFS server handles that, but what about if at the same time
> someone does multiple changes on the server in that EXT# behind the
> NFSd's back, how are the conflicts handled?

Currently, an exclusive lock is taken by the VFS or NFSD before any
create request gets the the EXT# filesystem.  That doesn't change (yet).

When multiple NFS clients try the create files in a directory, the nfsd
threads handling those requests will be serialized by the exclusive
lock.

Before the patch, if there were multiple threads on a single NFS client,
then they are all serialized on the client, and then again on the
server.
With my patch, the threads on the one client are NOT serialized, but
once the requests get to the server they will be.  So an exclusive lock
is only held of a smaller part of the total time.

>=20
> It would seem that this all really needs to purely happen at the VFS
> layer, but I'm probably missing something.

Serialization has to happen somewhere.  It doesn't have to be at the VFS
layer.  Doing locking at the VFS layer is easy and slow.  Doing it deep
in the filesystem is difficult and fast.  This exercise is really just
pushing locking deeper into the filesystem.

>=20
> I ask this because my home server exports /home to a couple of other
> systems via NFS, but I still work in /home on the server. =20
>=20

Thanks,
NeilBrown