Received-SPF: pass (google.com: domain of linux-nfs-owner@vger.kernel.org designates 23.128.96.38 as permitted sender) client-ip=23.128.96.38;
Content-Type: text/plain; charset="utf-8"
Content-Transfer-Encoding: 7bit
MIME-Version: 1.0
From:   "NeilBrown" <neilb@suse.de>
To:     "Chuck Lever III" <chuck.lever@oracle.com>
Cc:     "Jeff Layton" <jlayton@kernel.org>,
        "Linux NFS Mailing List" <linux-nfs@vger.kernel.org>,
        "Olga Kornievskaia" <kolga@netapp.com>,
        "Dai Ngo" <dai.ngo@oracle.com>, "Tom Talpey" <tom@talpey.com>
Subject: Re: [PATCH 0/5] sunrpc: not refcounting svc_serv
In-reply-to: <C869423C-B578-4D5D-B171-444EDF2D0D02@oracle.com>
References: <20231030011247.9794-1-neilb@suse.de>,
 <9BA28F70-A3FC-4EC4-A638-A27C1867F221@oracle.com>,
 <169878484259.24305.17786556797570881562@noble.neil.brown.name>,
 <C869423C-B578-4D5D-B171-444EDF2D0D02@oracle.com>
Date:   Wed, 01 Nov 2023 07:46:46 +1100
Message-id: <169878520677.24305.4940476987470825179@noble.neil.brown.name>
Precedence: bulk

On Wed, 01 Nov 2023, Chuck Lever III wrote:
> 
> > On Oct 31, 2023, at 1:40 PM, NeilBrown <neilb@suse.de> wrote:
> > 
> > On Wed, 01 Nov 2023, Chuck Lever III wrote:
> >> 
> >>> On Oct 29, 2023, at 6:08 PM, NeilBrown <neilb@suse.de> wrote:
> >>> 
> >>> This patch set continues earlier work of improving how threads and
> >>> services are managed.  Specifically it drop the refcount.
> >>> 
> >>> The refcount is always changed under the mutex, and almost always is
> >>> exactly equal to the number of threads.  Those few cases where it is
> >>> more than the number of threads can usefully be handled other ways as
> >>> see in the patches.
> >>> 
> >>> The first patches fixes a potential use-after-free when adding a socket
> >>> fails.  This might be the UAF that Jeff mentioned recently.
> >>> 
> >>> The second patch which removes the use of a refcount in pool_stats
> >>> handling is more complex than I would have liked, but I think it is
> >>> worth if for the result seen in 4/5 of substantial simplification.
> >> 
> >> So I need a v2 of this series, then...
> >> 
> >> - Move 4/5 to the beginning of the series (patch 1 or 2, doesn't matter)
> > 
> > I don't understand....  2 and 3 are necessary prerequisites for 4.  The
> > remove places where the refcount is used.
> 
> Hrm. that's what I was afraid of.
> 
> Isn't there a fix in 4/5 that we want applied to stable kernels,
> or did I misunderstand the email exchange between you and Jeff?
> 
That is 
 Commit bf32075256e9 ("NFSD: simplify error paths in nfsd_svc()")
which is already in nfsd-next.

NeilBrown


> 
> --
> Chuck Lever
> 
> 
>