Received-SPF: pass (google.com: domain of linux-nfs-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
MIME-Version: 1.0
References: <20210202184244.288898-1-Anna.Schumaker@Netapp.com>
 <20210202184244.288898-5-Anna.Schumaker@Netapp.com> <4eaf0c288d97a2d03c5cd2a7ed728a73085b2719.camel@hammerspace.com>
 <67497ea4a7d22726112e0083893e85a17f1ca681.camel@hammerspace.com>
In-Reply-To: <67497ea4a7d22726112e0083893e85a17f1ca681.camel@hammerspace.com>
From:   Anna Schumaker <schumaker.anna@gmail.com>
Date:   Fri, 5 Feb 2021 13:33:52 -0500
Message-ID: <CAFX2JfkGZbvgvxKJyPuxO9NG_RJGj-3BNH6jJDYGJV0pVhwipA@mail.gmail.com>
Subject: Re: [PATCH v2 4/5] sunrpc: Prepare xs_connect() for taking NULL tasks
To:     Trond Myklebust <trondmy@hammerspace.com>
Cc:     "linux-nfs@vger.kernel.org" <linux-nfs@vger.kernel.org>
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

On Tue, Feb 2, 2021 at 4:59 PM Trond Myklebust <trondmy@hammerspace.com> wrote:
>
> On Tue, 2021-02-02 at 16:49 -0500, Trond Myklebust wrote:
> > On Tue, 2021-02-02 at 13:42 -0500, schumaker.anna@gmail.com wrote:
> > > From: Anna Schumaker <Anna.Schumaker@Netapp.com>
> > >
> > > We won't have a task structure when we go to change IP addresses,
> > > so
> > > check for one before calling the WARN_ON() to avoid crashing.
> > >
> > > Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
> > > ---
> > >  net/sunrpc/xprtsock.c | 3 ++-
> > >  1 file changed, 2 insertions(+), 1 deletion(-)
> > >
> > > diff --git a/net/sunrpc/xprtsock.c b/net/sunrpc/xprtsock.c
> > > index c56a66cdf4ac..250abf1aa018 100644
> > > --- a/net/sunrpc/xprtsock.c
> > > +++ b/net/sunrpc/xprtsock.c
> > > @@ -2311,7 +2311,8 @@ static void xs_connect(struct rpc_xprt *xprt,
> > > struct rpc_task *task)
> > >         struct sock_xprt *transport = container_of(xprt, struct
> > > sock_xprt, xprt);
> > >         unsigned long delay = 0;
> > >
> > > -       WARN_ON_ONCE(!xprt_lock_connect(xprt, task, transport));
> > > +       if (task)
> > > +               WARN_ON_ONCE(!xprt_lock_connect(xprt, task,
> > > transport));
> >
> > Nit: That's the same as
> >    WARN_ON_ONCE(task && !xprt_lock_connect(xprt, task, transport));
> >
> > >
> > >         if (transport->sock != NULL) {
> > >                 dprintk("RPC:       xs_connect delayed xprt %p for
> > > %lu "
> >
>
> So, the problem with this patch is that you're deliberately
> circumventing the process of locking the socket. That's not going to
> work.
> What you could do is follow the example set by xprt_destroy() and
> xs_enable_swap(), to call wait_on_bit_lock() when there is no task.
> That should work, but you'd better make sure that your process holds a
> reference to the xprt->kref before doing that, or else you could easily
> end up deadlocking with xprt_destroy().

I've tried this, and the kref isn't a problem but no matter where I
put the wait_on_bit_lock() call it ends up deadlocking. I think it's
happening in the xs_tcp_setup_socket() function, but I'm still trying
to figure out exactly where.

Anna
>
> --
> Trond Myklebust
> Linux NFS client maintainer, Hammerspace
> trond.myklebust@hammerspace.com
>
>