[Intel-wired-lan] [net-next PATCH v4 1/2] ixgbe: add XDP support for pass and drop actions

[John Fastabend <john.fastabend@gmail.com>]

On 17-03-11 07:49 AM, William Tu wrote:
> On Fri, Mar 10, 2017 at 11:11 AM, John Fastabend
> <john.fastabend@gmail.com> wrote:
>> Basic XDP drop support for ixgbe. Uses READ_ONCE/xchg semantics on XDP
>> programs instead of rcu primitives as suggested by Daniel Borkmann and
>> Alex Duyck.
>>
>> Signed-off-by: John Fastabend <john.r.fastabend@intel.com>
>> ---

[...]

>>  /**
>>   * ixgbe_clean_rx_irq - Clean completed descriptors from Rx ring - bounce buf
>>   * @q_vector: structure containing interrupt and ring information
>> @@ -2184,6 +2230,7 @@ static int ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector,
>>                 union ixgbe_adv_rx_desc *rx_desc;
>>                 struct ixgbe_rx_buffer *rx_buffer;
>>                 struct sk_buff *skb;
>> +               struct xdp_buff xdp;
>>                 unsigned int size;
>>
>>                 /* return some buffers to hardware, one at a time is too slow */
>> @@ -2205,15 +2252,29 @@ static int ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector,
>>
>>                 rx_buffer = ixgbe_get_rx_buffer(rx_ring, rx_desc, &skb, size);
>>
>> -               /* retrieve a buffer from the ring */
>> -               if (skb)
>> +               if (!skb) {
>> +                       xdp.data = page_address(rx_buffer->page) +
>> +                                               rx_buffer->page_offset;
>> +                       xdp.data_hard_start = xdp.data -
>> +                                               ixgbe_rx_offset(rx_ring);

We have ixgbe_rx_offset(rx_ring) headroom to support adding headers.

>> +                       xdp.data_end = xdp.data + size;
>> +
>> +                       skb = ixgbe_run_xdp(rx_ring, &xdp);
>> +               }
>> +
>> +               if (IS_ERR(skb)) {
>> +                       total_rx_packets++;
>> +                       total_rx_bytes += size;
>> +                       rx_buffer->pagecnt_bias++;
>> +               } else if (skb) {
>>                         ixgbe_add_rx_frag(rx_ring, rx_buffer, skb, size);
>> -               else if (ring_uses_build_skb(rx_ring))
>> +               } else if (ring_uses_build_skb(rx_ring)) {
>>                         skb = ixgbe_build_skb(rx_ring, rx_buffer,
>> -                                             rx_desc, size);
>> -               else
>> +                                             &xdp, rx_desc);
>> +               } else {
>>                         skb = ixgbe_construct_skb(rx_ring, rx_buffer,
>> -                                                 rx_desc, size);
>> +                                                 &xdp, rx_desc);
>> +               }
>>
>>                 /* exit if we failed to retrieve a buffer */
>>                 if (!skb) {

[...]

>> +static int ixgbe_xdp_setup(struct net_device *dev, struct bpf_prog *prog)
>> +{
>> +       int i, frame_size = dev->mtu + ETH_HLEN + ETH_FCS_LEN + VLAN_HLEN;
>> +       struct ixgbe_adapter *adapter = netdev_priv(dev);
>> +       struct bpf_prog *old_prog;
>> +
>> +       if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
>> +               return -EINVAL;
>> +
>> +       if (adapter->flags & IXGBE_FLAG_DCB_ENABLED)
>> +               return -EINVAL;
>> +
>> +       /* verify ixgbe ring attributes are sufficient for XDP */
>> +       for (i = 0; i < adapter->num_rx_queues; i++) {
>> +               struct ixgbe_ring *ring = adapter->rx_ring[i];
>> +
>> +               if (ring_is_rsc_enabled(ring))
>> +                       return -EINVAL;
>> +
>> +               if (frame_size > ixgbe_rx_bufsz(ring))
>> +                       return -EINVAL;
>> +       }
>> +
>> +       old_prog = xchg(&adapter->xdp_prog, prog);
>> +       for (i = 0; i < adapter->num_rx_queues; i++)
>> +               xchg(&adapter->rx_ring[i]->xdp_prog, adapter->xdp_prog);
>> +
>> +       if (old_prog)
>> +               bpf_prog_put(old_prog);
>> +
>> +       return 0;
>> +}
> 
> Since the patch does not support xdp_adjust_head() yet, should we
> detect and return -EOPNOTSUPP?
> 

It actually should support xdp_adjust_head() :)

At least I tested the xdp tunnel program in ./bpf/samples/ and it worked.
Also I do a standard test where I push the packet up the stack after adding
headers and that appears to work although I wonder a bit about some of the
skb metadata. By working in this case I just use tshark and verify the pkt
is received with the new header by the stack.

> --William