Re: [RFC PATCH v2 4/5] mlx4: add support for fast rx drop bpf program

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

On Thu,  7 Apr 2016 21:48:49 -0700 Brenden Blanco <bblanco@plumgrid.com> wrote:

> +int mlx4_call_bpf(struct bpf_prog *prog, void *data, unsigned int length)
> +{
> +	struct sk_buff *skb = this_cpu_ptr(&percpu_bpf_phys_dev_md);
> +	int ret;
> +
> +	build_bpf_phys_dev_md(skb, data, length);
> +
> +	rcu_read_lock();
> +	ret = BPF_PROG_RUN(prog, (void *)skb);
> +	rcu_read_unlock();
> +
> +	return ret;
> +}
[...]

> diff --git a/drivers/net/ethernet/mellanox/mlx4/en_rx.c b/drivers/net/ethernet/mellanox/mlx4/en_rx.c
> index 86bcfe5..287da02 100644
> --- a/drivers/net/ethernet/mellanox/mlx4/en_rx.c
> +++ b/drivers/net/ethernet/mellanox/mlx4/en_rx.c
> @@ -840,6 +843,23 @@ int mlx4_en_process_rx_cq(struct net_device *dev, struct mlx4_en_cq *cq, int bud
>  		l2_tunnel = (dev->hw_enc_features & NETIF_F_RXCSUM) &&
>  			(cqe->vlan_my_qpn & cpu_to_be32(MLX4_CQE_L2_TUNNEL));
>  
> +		/* A bpf program gets first chance to drop the packet. It may
> +		 * read bytes but not past the end of the frag.
> +		 */
> +		if (prog) {
> +			struct ethhdr *ethh;
> +			dma_addr_t dma;
> +
> +			dma = be64_to_cpu(rx_desc->data[0].addr);
> +			dma_sync_single_for_cpu(priv->ddev, dma, sizeof(*ethh),
> +						DMA_FROM_DEVICE);
> +			ethh = page_address(frags[0].page) +
> +							frags[0].page_offset;
> +			if (mlx4_call_bpf(prog, ethh, frags[0].page_size) ==
> +							BPF_PHYS_DEV_DROP)
> +				goto next;
> +		}
> +

Do the program need to know if the packet-page we handover is
considered 'read-only' at the call site?  Or do we rely on my idea
requiring the registration to "know" this...

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