Re: [PATCH RFC v4 bpf-next 09/11] tun: Support xdp in the Tx path for xdp_frames

[Jesper Dangaard Brouer <brouer@redhat.com>]

On Wed, 26 Feb 2020 20:20:11 -0700
David Ahern <dsahern@kernel.org> wrote:

> From: David Ahern <dahern@digitalocean.com>
> 
> Add support to run Tx path program on packets arriving at a tun
> device via XDP redirect.
> 
> XDP_TX return code means move the packet to the Tx path of the device.
> For a program run in the Tx / egress path, XDP_TX is essentially the
> same as "continue on" which is XDP_PASS.
> 
> Conceptually, XDP_REDIRECT for this path can work the same as it
> does for the Rx path, but that return code is left for a follow
> on series.
> 
> Signed-off-by: Prashant Bhole <prashantbhole.linux@gmail.com>
> Signed-off-by: David Ahern <dahern@digitalocean.com>
> ---
>  drivers/net/tun.c | 49 +++++++++++++++++++++++++++++++++++++++++++++--
>  1 file changed, 47 insertions(+), 2 deletions(-)
> 
> diff --git a/drivers/net/tun.c b/drivers/net/tun.c
> index dcae6521a39d..d3fc7e921c85 100644
> --- a/drivers/net/tun.c
> +++ b/drivers/net/tun.c
> @@ -1359,10 +1359,50 @@ static void __tun_xdp_flush_tfile(struct tun_file *tfile)
>  	tfile->socket.sk->sk_data_ready(tfile->socket.sk);
>  }
>  
> +static u32 tun_do_xdp_tx(struct tun_struct *tun, struct tun_file *tfile,
> +			 struct xdp_frame *frame, struct xdp_txq_info *txq)
> +{
> +	struct bpf_prog *xdp_prog;
> +	u32 act = XDP_PASS;
> +
> +	xdp_prog = rcu_dereference(tun->xdp_egress_prog);
> +	if (xdp_prog) {
> +		struct xdp_buff xdp;
> +
> +		xdp.data_hard_start = frame->data - frame->headroom;

This is correct, only because frame->headroom have been reduced with
sizeof(*xdp_frame), as we want to avoid that the BPF-prog have access
to xdp_frame memory.  Remember that memory storing xdp_frame in located
in the top of the payload/page.


> +		xdp.data = frame->data;
> +		xdp.data_end = xdp.data + frame->len;
> +		xdp_set_data_meta_invalid(&xdp);
> +		xdp.txq = txq;
> +
> +		act = bpf_prog_run_xdp(xdp_prog, &xdp);

The BPF-prog can change/adjust headroom and tailroom (tail only shrink,
but I'm working on extending this).  Thus, you need to adjust the
xdp_frame accordingly afterwards.

(The main use-case is pushing on a header, right?)

> +		switch (act) {
> +		case XDP_TX:    /* for Tx path, XDP_TX == XDP_PASS */
> +			act = XDP_PASS;
> +			break;
> +		case XDP_PASS:
> +			break;
> +		case XDP_REDIRECT:
> +			/* fall through */
> +		default:
> +			bpf_warn_invalid_xdp_action(act);
> +			/* fall through */
> +		case XDP_ABORTED:
> +			trace_xdp_exception(tun->dev, xdp_prog, act);
> +			/* fall through */
> +		case XDP_DROP:
> +			break;
> +		}
> +	}
> +
> +	return act;
> +}
> +
>  static int tun_xdp_xmit(struct net_device *dev, int n,
>  			struct xdp_frame **frames, u32 flags)
>  {
>  	struct tun_struct *tun = netdev_priv(dev);
> +	struct xdp_txq_info txq = { .dev = dev };
>  	struct tun_file *tfile;
>  	u32 numqueues;
>  	int drops = 0;
> @@ -1389,12 +1429,17 @@ static int tun_xdp_xmit(struct net_device *dev, int n,
>  	spin_lock(&tfile->tx_ring.producer_lock);
>  	for (i = 0; i < n; i++) {
>  		struct xdp_frame *xdp = frames[i];
> +		void *frame;
> +
> +		if (tun_do_xdp_tx(tun, tfile, xdp, &txq) != XDP_PASS)
> +			goto drop;
> +
>  		/* Encode the XDP flag into lowest bit for consumer to differ
>  		 * XDP buffer from sk_buff.
>  		 */
> -		void *frame = tun_xdp_to_ptr(xdp);
> -
> +		frame = tun_xdp_to_ptr(xdp);
>  		if (__ptr_ring_produce(&tfile->tx_ring, frame)) {
> +drop:
>  			this_cpu_inc(tun->pcpu_stats->tx_dropped);
>  			xdp_return_frame_rx_napi(xdp);
>  			drops++;



-- 
Best regards,
  Jesper Dangaard Brouer
  MSc.CS, Principal Kernel Engineer at Red Hat
  LinkedIn: http://www.linkedin.com/in/brouer