Date: Thu, 27 Jul 2017 20:13:38 -0700
From: Jakub Kicinski <kubakici@wp.pl>
To: Jason Wang <jasowang@redhat.com>
Cc: davem@davemloft.net, mst@redhat.com, netdev@vger.kernel.org,
        linux-kernel@vger.kernel.org
Subject: Re: [PATCH net-next 3/3] tap: XDP support
Message-ID: <20170727201338.65d56b2a@cakuba.netronome.com>
In-Reply-To: <1501147533-12368-4-git-send-email-jasowang@redhat.com>
References: <1501147533-12368-1-git-send-email-jasowang@redhat.com>
        <1501147533-12368-4-git-send-email-jasowang@redhat.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=US-ASCII
Content-Transfer-Encoding: 7bit
Sender: linux-kernel-owner@vger.kernel.org
Content-Length: 2024
Lines: 60

On Thu, 27 Jul 2017 17:25:33 +0800, Jason Wang wrote:
> This patch tries to implement XDP for tun. The implementation was
> split into two parts:
> 
> - fast path: small and no gso packet. We try to do XDP at page level
>   before build_skb(). For XDP_TX, since creating/destroying queues
>   were completely under control of userspace, it was implemented
>   through generic XDP helper after skb has been built. This could be
>   optimized in the future.
> - slow path: big or gso packet. We try to do it after skb was created
>   through generic XDP helpers.
> 
> XDP_REDIRECT was not implemented, it could be done on top.
> 
> xdp1 test shows 47.6% improvement:
> 
> Before: ~2.1Mpps
> After:  ~3.1Mpps
> 
> Suggested-by: Michael S. Tsirkin <mst@redhat.com>
> Signed-off-by: Jason Wang <jasowang@redhat.com>

> @@ -1008,6 +1016,56 @@ tun_net_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
>  	stats->tx_dropped = tx_dropped;
>  }
>  
> +static int tun_xdp_set(struct net_device *dev, struct bpf_prog *prog,
> +		       struct netlink_ext_ack *extack)
> +{
> +	struct tun_struct *tun = netdev_priv(dev);
> +	struct bpf_prog *old_prog;
> +
> +	/* We will shift the packet that can't be handled to generic
> +	 * XDP layer.
> +	 */
> +
> +	old_prog = rtnl_dereference(tun->xdp_prog);
> +	if (old_prog)
> +		bpf_prog_put(old_prog);
> +	rcu_assign_pointer(tun->xdp_prog, prog);

Is this OK?  Could this lead to the program getting freed and then
datapath accessing a stale pointer?  I mean in the scenario where the
process gets pre-empted between the bpf_prog_put() and
rcu_assign_pointer()?

> +	if (prog) {
> +		prog = bpf_prog_add(prog, 1);
> +		if (IS_ERR(prog))
> +			return PTR_ERR(prog);
> +	}

I don't think you need this extra reference here.  dev_change_xdp_fd()
will call bpf_prog_get_type() which means driver gets the program with
a reference already taken, drivers does have to free that reference when
program is removed (or device is freed, as you correctly do).

> +	return 0;
> +}
> +