Received-SPF: pass (google.com: best guess record for domain of linux-nfs-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
MIME-Version: 1.0
References: <20200301232145.1465430-1-trond.myklebust@hammerspace.com>
 <20200301232145.1465430-2-trond.myklebust@hammerspace.com>
 <20200301232145.1465430-3-trond.myklebust@hammerspace.com>
 <20200301232145.1465430-4-trond.myklebust@hammerspace.com>
 <20200301232145.1465430-5-trond.myklebust@hammerspace.com>
 <20200301232145.1465430-6-trond.myklebust@hammerspace.com>
 <20200301232145.1465430-7-trond.myklebust@hammerspace.com>
 <20200301232145.1465430-8-trond.myklebust@hammerspace.com> <20200301232145.1465430-9-trond.myklebust@hammerspace.com>
In-Reply-To: <20200301232145.1465430-9-trond.myklebust@hammerspace.com>
From:   David Wysochanski <dwysocha@redhat.com>
Date:   Mon, 2 Mar 2020 11:27:09 -0500
Message-ID: <CALF+zOkJPkYaXjCn-tj0dPCQUAjA05zyzxm5fzwoB-XP0SGYvw@mail.gmail.com>
Subject: Re: [PATCH 8/8] sunrpc: Drop the connection when the server drops a request
To:     Trond Myklebust <trondmy@gmail.com>
Cc:     "J. Bruce Fields" <bfields@redhat.com>,
        linux-nfs <linux-nfs@vger.kernel.org>
Content-Type: text/plain; charset="UTF-8"
Sender: linux-nfs-owner@vger.kernel.org
Precedence: bulk

On Sun, Mar 1, 2020 at 6:25 PM Trond Myklebust <trondmy@gmail.com> wrote:
>
> If a server wants to drop a request, then it should also drop the
> connection, in order to let the client know.
>
> Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>
> ---
>  net/sunrpc/svc_xprt.c | 10 ++++++++++
>  1 file changed, 10 insertions(+)
>
> diff --git a/net/sunrpc/svc_xprt.c b/net/sunrpc/svc_xprt.c
> index de3c077733a7..83a527e56c87 100644
> --- a/net/sunrpc/svc_xprt.c
> +++ b/net/sunrpc/svc_xprt.c
> @@ -873,6 +873,13 @@ int svc_recv(struct svc_rqst *rqstp, long timeout)
>  }
>  EXPORT_SYMBOL_GPL(svc_recv);
>
> +static void svc_drop_connection(struct svc_xprt *xprt)
> +{
> +       if (test_bit(XPT_TEMP, &xprt->xpt_flags) &&
> +           !test_and_set_bit(XPT_CLOSE, &xprt->xpt_flags))
> +               svc_xprt_enqueue(xprt);
> +}
> +
>  /*
>   * Drop request
>   */
> @@ -880,6 +887,8 @@ void svc_drop(struct svc_rqst *rqstp)
>  {
>         trace_svc_drop(rqstp);
>         dprintk("svc: xprt %p dropped request\n", rqstp->rq_xprt);
> +       /* Close the connection when dropping a request */
> +       svc_drop_connection(rqstp->rq_xprt);
>         svc_xprt_release(rqstp);
>  }
>  EXPORT_SYMBOL_GPL(svc_drop);
> @@ -1148,6 +1157,7 @@ static void svc_revisit(struct cache_deferred_req *dreq, int too_many)
>         if (too_many || test_bit(XPT_DEAD, &xprt->xpt_flags)) {
>                 spin_unlock(&xprt->xpt_lock);
>                 dprintk("revisit canceled\n");
> +               svc_drop_connection(xprt);
>                 svc_xprt_put(xprt);
>                 trace_svc_drop_deferred(dr);
>                 kfree(dr);
> --
> 2.24.1
>

Trond, back in 2014 you had this NFSv4 only patch that took a more
surgical approach:
https://marc.info/?l=linux-nfs&m=141414531832768&w=2

It looks like discussion died out on it after it was ineffective to
solve a different problem.
Is there a reason why you don't want to do that approach now?