Re: AF_PACKET: tx_ring mirrored in rx_ring?

[Willem de Bruijn <willemb@google.com>]

>>> What'd you mean by local traffic? The packets which are replicated are
>>> destined to remote machine(s).
>>
>>
>> Sure, but you are sending them out via your packet socket.
>
>
> Well yes. It's just that I interpreted local as if they were not going out
> of the machine. But in fact they do.

That is a semantic issue. The technical point is that packet
sockets read not only incoming packets, but also outgoing
ones. The tap in the egress path (dev_queue_xmit_nit) is taken
for almost all transmitted packets, included those transmitted
by a packet socket.

There is logic to avoid looping outgoing packets back into the
originating socket (and fanout group) by detecting the source
socket (skb_loop_sk). Other packet sockets will receive the
outgoing packets, however. This is correct behavior, as it is
how tcpdump can log all traffic, among others.

You can use PACKET_QDISC_BYPASS on your transmit
packet socket, as Daniel mentions, or attach a BPF filter to
your receive socket that filters on !PACKET_OUTGOING, e.g.,:

  struct sock_filter bpf_filter[] = {
    {BPF_LD | BPF_B | BPF_ABS, 0, 0, (uint32_t) (SKF_AD_OFF + SKF_AD_PKTTYPE)},
    {BPF_JMP | BPF_JEQ, 1, 0, PACKET_OUTGOING},
    {BPF_RET, 0, 0, 0x00000000},
    {BPF_RET, 0, 0, 0x0000ffff},
  };
  struct sock_fprog bpf_prog;

  bpf_prog.filter = bpf_filter;
  bpf_prog.len = sizeof(bpf_filter) / sizeof(struct sock_filter);
  if (setsockopt(fd, SOL_SOCKET, SO_ATTACH_FILTER, &bpf_prog,
                 sizeof(bpf_prog))) {
    error(1, errno, "setsockopt filter");
  }