コード例 #1
0
ファイル: bier_disp_dispatch_node.c プロジェクト: chrisy/vpp
always_inline uword
bier_disp_dispatch_inline (vlib_main_t * vm,
                         vlib_node_runtime_t * node,
                         vlib_frame_t * from_frame)
{
    u32 n_left_from, next_index, * from, * to_next;

    from = vlib_frame_vector_args (from_frame);
    n_left_from = from_frame->n_vectors;

    next_index = node->cached_next_index;

    while (n_left_from > 0)
    {
        u32 n_left_to_next;

        vlib_get_next_frame(vm, node, next_index, to_next, n_left_to_next);

        while (n_left_from > 0 && n_left_to_next > 0)
        {
            bier_hdr_proto_id_t pproto0;
            bier_disp_entry_t *bde0;
            u32 next0, bi0, bdei0;
            const dpo_id_t *dpo0;
            vlib_buffer_t * b0;
            bier_hdr_t *hdr0;
            u32 entropy0;

            bi0 = from[0];
            to_next[0] = bi0;
            from += 1;
            to_next += 1;
            n_left_from -= 1;
            n_left_to_next -= 1;

            b0 = vlib_get_buffer (vm, bi0);
            bdei0 = vnet_buffer(b0)->ip.adj_index[VLIB_TX];
            hdr0 = vlib_buffer_get_current(b0);
            bde0 = bier_disp_entry_get(bdei0);
            vnet_buffer(b0)->ip.adj_index[VLIB_RX] = BIER_RX_ITF;

            /*
             * header is in network order - flip it, we are about to
             * consume it anyway
             */
            bier_hdr_ntoh(hdr0);
            pproto0 = bier_hdr_get_proto_id(hdr0);
            entropy0 = bier_hdr_get_entropy(hdr0);

            /*
             * strip the header and copy the entropy value into
             * the packets flow-hash field
             * DSCP mumble mumble...
             */
            vlib_buffer_advance(b0, (vnet_buffer(b0)->mpls.bier.n_bytes +
                                     sizeof(*hdr0)));
            vnet_buffer(b0)->ip.flow_hash = entropy0;

            /*
             * use the payload proto to dispatch to the
             * correct stacked DPO.
             */
            dpo0 = &bde0->bde_fwd[pproto0].bde_dpo;
            next0 = dpo0->dpoi_next_node;
            vnet_buffer(b0)->ip.adj_index[VLIB_TX] = dpo0->dpoi_index;
            vnet_buffer(b0)->ip.rpf_id = bde0->bde_fwd[pproto0].bde_rpf_id;

            if (PREDICT_FALSE(b0->flags & VLIB_BUFFER_IS_TRACED))
            {
                bier_disp_dispatch_trace_t *tr =
                    vlib_add_trace (vm, node, b0, sizeof (*tr));
                tr->pproto = pproto0;
                tr->rpf_id = vnet_buffer(b0)->ip.rpf_id;
            }

            vlib_validate_buffer_enqueue_x1(vm, node, next_index, to_next,
                                            n_left_to_next, bi0, next0);
        }
        vlib_put_next_frame (vm, node, next_index, n_left_to_next);
    }
    return from_frame->n_vectors;
}
コード例 #2
0
ファイル: ip6_sixrd.c プロジェクト: Venkattk/vpp
/*
 * ip6_sixrd
 */
static uword
ip6_sixrd (vlib_main_t *vm,
	   vlib_node_runtime_t *node,
	   vlib_frame_t *frame)
{
  u32 n_left_from, *from, next_index, *to_next, n_left_to_next;
  vlib_node_runtime_t *error_node = vlib_node_get_runtime(vm, ip6_sixrd_node.index);
  u32 encap = 0;
  from = vlib_frame_vector_args(frame);
  n_left_from = frame->n_vectors;
  next_index = node->cached_next_index;

  while (n_left_from > 0) {
    vlib_get_next_frame(vm, node, next_index, to_next, n_left_to_next);

    while (n_left_from > 0 && n_left_to_next > 0) {
      u32 pi0;
      vlib_buffer_t *p0;
      sixrd_domain_t *d0;
      u8 error0 = SIXRD_ERROR_NONE;
      ip6_header_t *ip60;
      ip4_header_t *ip4h0;
      u32 next0 = IP6_SIXRD_NEXT_IP4_LOOKUP;
      u32 sixrd_domain_index0 = ~0;

      pi0 = to_next[0] = from[0];
      from += 1;
      n_left_from -= 1;
      to_next +=1;
      n_left_to_next -= 1;

      p0 = vlib_get_buffer(vm, pi0);
      ip60 = vlib_buffer_get_current(p0);
      //      p0->current_length = clib_net_to_host_u16(ip40->length);
      d0 = ip6_sixrd_get_domain(vnet_buffer(p0)->ip.adj_index[VLIB_TX], &sixrd_domain_index0);
      ASSERT(d0);

      /* SIXRD calc */
      u64 dal60 = clib_net_to_host_u64(ip60->dst_address.as_u64[0]);
      u32 da40 = sixrd_get_addr(d0, dal60);
      u16 len = clib_net_to_host_u16(ip60->payload_length) + 60;
      if (da40 == 0) error0 = SIXRD_ERROR_UNKNOWN;

      /* construct ipv4 header */
      vlib_buffer_advance(p0, - (sizeof(ip4_header_t)));
      ip4h0 = vlib_buffer_get_current(p0);
      vnet_buffer(p0)->sw_if_index[VLIB_TX] = (u32)~0;
      ip4h0->ip_version_and_header_length = 0x45;
      ip4h0->tos = 0;
      ip4h0->length = clib_host_to_net_u16(len);
      ip4h0->fragment_id = 0;
      ip4h0->flags_and_fragment_offset = 0;
      ip4h0->ttl = 0x40;
      ip4h0->protocol = IP_PROTOCOL_IPV6;
      ip4h0->src_address = d0->ip4_src;
      ip4h0->dst_address.as_u32 = clib_host_to_net_u32(da40);
      ip4h0->checksum = ip4_header_checksum(ip4h0);

      next0 = error0 == SIXRD_ERROR_NONE ? IP6_SIXRD_NEXT_IP4_LOOKUP : IP6_SIXRD_NEXT_DROP;

      if (PREDICT_FALSE(p0->flags & VLIB_BUFFER_IS_TRACED)) {
	sixrd_trace_t *tr = vlib_add_trace(vm, node, p0, sizeof(*tr));
	tr->sixrd_domain_index = sixrd_domain_index0;
      }

      p0->error = error_node->errors[error0];
      if (PREDICT_TRUE(error0 == SIXRD_ERROR_NONE)) encap++;

      vlib_validate_buffer_enqueue_x1(vm, node, next_index, to_next, n_left_to_next, pi0, next0);
    }
    vlib_put_next_frame(vm, node, next_index, n_left_to_next);  
  }
  vlib_node_increment_counter(vm, ip6_sixrd_node.index, SIXRD_ERROR_ENCAPSULATED, encap);

  return frame->n_vectors;
}
コード例 #3
0
ファイル: bier_imp_node.c プロジェクト: vpp-dev/vpp
always_inline uword
bier_imp_dpo_inline (vlib_main_t * vm,
                     vlib_node_runtime_t * node,
                     vlib_frame_t * from_frame,
                     fib_protocol_t fproto,
                     bier_hdr_proto_id_t bproto)
{
    u32 n_left_from, next_index, * from, * to_next;

    from = vlib_frame_vector_args (from_frame);
    n_left_from = from_frame->n_vectors;

    next_index = node->cached_next_index;

    while (n_left_from > 0)
    {
        u32 n_left_to_next;

        vlib_get_next_frame(vm, node, next_index, to_next, n_left_to_next);

        while (n_left_from > 0 && n_left_to_next > 0)
        {
            vlib_buffer_t * b0;
            bier_imp_t *bimp0;
            bier_hdr_t *hdr0;
            u32 bi0, bii0;
            u32 next0;

            bi0 = from[0];
            to_next[0] = bi0;
            from += 1;
            to_next += 1;
            n_left_from -= 1;
            n_left_to_next -= 1;

            b0 = vlib_get_buffer (vm, bi0);

            bii0 = vnet_buffer(b0)->ip.adj_index[VLIB_TX];
            bimp0 = bier_imp_get(bii0);

            if (FIB_PROTOCOL_IP4 == fproto)
            {
                /*
                 * decrement the TTL on ingress to the BIER domain
                 */
                ip4_header_t * ip0 = vlib_buffer_get_current(b0);
                u32 checksum0;

                checksum0 = ip0->checksum + clib_host_to_net_u16 (0x0100);
                checksum0 += checksum0 >= 0xffff;

                ip0->checksum = checksum0;
                ip0->ttl -= 1;

                /*
                 * calculate an entropy
                 */
                if (0 == vnet_buffer(b0)->ip.flow_hash)
                {
                    vnet_buffer(b0)->ip.flow_hash =
                        ip4_compute_flow_hash (ip0, IP_FLOW_HASH_DEFAULT);
                }
            }
            if (FIB_PROTOCOL_IP6 == fproto)
            {
                /*
                 * decrement the TTL on ingress to the BIER domain
                 */
                ip6_header_t * ip0 = vlib_buffer_get_current(b0);

                ip0->hop_limit -= 1;

                /*
                 * calculate an entropy
                 */
                if (0 == vnet_buffer(b0)->ip.flow_hash)
                {
                    vnet_buffer(b0)->ip.flow_hash =
                        ip6_compute_flow_hash (ip0, IP_FLOW_HASH_DEFAULT);
                }
            }

            /* Paint the BIER header */
            vlib_buffer_advance(b0, -(sizeof(bier_hdr_t) +
                                      bier_hdr_len_id_to_num_bytes(bimp0->bi_tbl.bti_hdr_len)));
            hdr0 = vlib_buffer_get_current(b0);

            /* RPF check */
            if (PREDICT_FALSE(BIER_RX_ITF == vnet_buffer(b0)->ip.adj_index[VLIB_RX]))
            {
                next0 = 0;
            }
            else
            {
                clib_memcpy_fast(hdr0, &bimp0->bi_hdr,
                            (sizeof(bier_hdr_t) +
                             bier_hdr_len_id_to_num_bytes(bimp0->bi_tbl.bti_hdr_len)));
                /*
                 * Fixup the entropy and protocol, both of which have a
                 * zero value post the paint job
                 */
                hdr0->bh_oam_dscp_proto |=
                    clib_host_to_net_u16(bproto << BIER_HDR_PROTO_FIELD_SHIFT);
                hdr0->bh_first_word |=
                    clib_host_to_net_u32((vnet_buffer(b0)->ip.flow_hash &
                                          BIER_HDR_ENTROPY_FIELD_MASK) <<
                                         BIER_HDR_ENTROPY_FIELD_SHIFT);

                /*
                 * use TTL 64 for the post enacp MPLS label/BIFT-ID
                 * this we be decremeted in bier_output node.
                 */
                vnet_buffer(b0)->mpls.ttl = 65;

                /* next node */
                next0 = bimp0->bi_dpo[fproto].dpoi_next_node;
                vnet_buffer(b0)->ip.adj_index[VLIB_TX] =
                    bimp0->bi_dpo[fproto].dpoi_index;
            }

            if (PREDICT_FALSE(b0->flags & VLIB_BUFFER_IS_TRACED))
            {
                bier_imp_trace_t *tr =
                    vlib_add_trace (vm, node, b0, sizeof (*tr));
                tr->imp = bii0;
                tr->hdr = *hdr0;
            }

            vlib_validate_buffer_enqueue_x1(vm, node, next_index, to_next,
                                            n_left_to_next, bi0, next0);
        }
        vlib_put_next_frame (vm, node, next_index, n_left_to_next);
    }
    return from_frame->n_vectors;
}
コード例 #4
0
ファイル: adj_midchain.c プロジェクト: vpp-dev/vpp
always_inline uword
adj_midchain_tx_inline (vlib_main_t * vm,
			vlib_node_runtime_t * node,
			vlib_frame_t * frame,
			int interface_count)
{
    u32 * from, * to_next, n_left_from, n_left_to_next;
    u32 next_index;
    vnet_main_t *vnm = vnet_get_main ();
    vnet_interface_main_t *im = &vnm->interface_main;
    u32 thread_index = vm->thread_index;

    /* Vector of buffer / pkt indices we're supposed to process */
    from = vlib_frame_vector_args (frame);

    /* Number of buffers / pkts */
    n_left_from = frame->n_vectors;

    /* Speculatively send the first buffer to the last disposition we used */
    next_index = node->cached_next_index;

    while (n_left_from > 0)
    {
	/* set up to enqueue to our disposition with index = next_index */
	vlib_get_next_frame (vm, node, next_index, to_next, n_left_to_next);

	while (n_left_from >= 8 && n_left_to_next > 4)
	{
	    u32 bi0, adj_index0, next0;
	    const ip_adjacency_t * adj0;
	    const dpo_id_t *dpo0;
	    vlib_buffer_t * b0;
	    u32 bi1, adj_index1, next1;
	    const ip_adjacency_t * adj1;
	    const dpo_id_t *dpo1;
	    vlib_buffer_t * b1;
	    u32 bi2, adj_index2, next2;
	    const ip_adjacency_t * adj2;
	    const dpo_id_t *dpo2;
	    vlib_buffer_t * b2;
	    u32 bi3, adj_index3, next3;
	    const ip_adjacency_t * adj3;
	    const dpo_id_t *dpo3;
	    vlib_buffer_t * b3;

	    /* Prefetch next iteration. */
	    {
		vlib_buffer_t * p4, * p5;
		vlib_buffer_t * p6, * p7;

		p4 = vlib_get_buffer (vm, from[4]);
		p5 = vlib_get_buffer (vm, from[5]);
		p6 = vlib_get_buffer (vm, from[6]);
		p7 = vlib_get_buffer (vm, from[7]);

		vlib_prefetch_buffer_header (p4, LOAD);
		vlib_prefetch_buffer_header (p5, LOAD);
		vlib_prefetch_buffer_header (p6, LOAD);
		vlib_prefetch_buffer_header (p7, LOAD);
	    }

	    bi0 = from[0];
	    to_next[0] = bi0;
	    bi1 = from[1];
	    to_next[1] = bi1;
	    bi2 = from[2];
	    to_next[2] = bi2;
	    bi3 = from[3];
	    to_next[3] = bi3;

	    from += 4;
	    to_next += 4;
	    n_left_from -= 4;
	    n_left_to_next -= 4;

	    b0 = vlib_get_buffer(vm, bi0);
	    b1 = vlib_get_buffer(vm, bi1);
	    b2 = vlib_get_buffer(vm, bi2);
	    b3 = vlib_get_buffer(vm, bi3);

	    /* Follow the DPO on which the midchain is stacked */
	    adj_index0 = vnet_buffer(b0)->ip.adj_index[VLIB_TX];
	    adj_index1 = vnet_buffer(b1)->ip.adj_index[VLIB_TX];
	    adj_index2 = vnet_buffer(b2)->ip.adj_index[VLIB_TX];
	    adj_index3 = vnet_buffer(b3)->ip.adj_index[VLIB_TX];

	    adj0 = adj_get(adj_index0);
	    adj1 = adj_get(adj_index1);
	    adj2 = adj_get(adj_index2);
	    adj3 = adj_get(adj_index3);

	    dpo0 = &adj0->sub_type.midchain.next_dpo;
	    dpo1 = &adj1->sub_type.midchain.next_dpo;
	    dpo2 = &adj2->sub_type.midchain.next_dpo;
	    dpo3 = &adj3->sub_type.midchain.next_dpo;

	    next0 = dpo0->dpoi_next_node;
	    next1 = dpo1->dpoi_next_node;
	    next2 = dpo2->dpoi_next_node;
	    next3 = dpo3->dpoi_next_node;

            vnet_buffer(b1)->ip.adj_index[VLIB_TX] = dpo1->dpoi_index;
            vnet_buffer(b0)->ip.adj_index[VLIB_TX] = dpo0->dpoi_index;
            vnet_buffer(b2)->ip.adj_index[VLIB_TX] = dpo2->dpoi_index;
            vnet_buffer(b3)->ip.adj_index[VLIB_TX] = dpo3->dpoi_index;

	    if (interface_count)
	    {
		vlib_increment_combined_counter (im->combined_sw_if_counters
						 + VNET_INTERFACE_COUNTER_TX,
						 thread_index,
						 adj0->rewrite_header.sw_if_index,
						 1,
						 vlib_buffer_length_in_chain (vm, b0));
		vlib_increment_combined_counter (im->combined_sw_if_counters
						 + VNET_INTERFACE_COUNTER_TX,
						 thread_index,
						 adj1->rewrite_header.sw_if_index,
						 1,
						 vlib_buffer_length_in_chain (vm, b1));
		vlib_increment_combined_counter (im->combined_sw_if_counters
						 + VNET_INTERFACE_COUNTER_TX,
						 thread_index,
						 adj2->rewrite_header.sw_if_index,
						 1,
						 vlib_buffer_length_in_chain (vm, b2));
		vlib_increment_combined_counter (im->combined_sw_if_counters
						 + VNET_INTERFACE_COUNTER_TX,
						 thread_index,
						 adj3->rewrite_header.sw_if_index,
						 1,
						 vlib_buffer_length_in_chain (vm, b3));
	    }

	    if (PREDICT_FALSE(b0->flags & VLIB_BUFFER_IS_TRACED))
	    {
		adj_midchain_tx_trace_t *tr = vlib_add_trace (vm, node,
							      b0, sizeof (*tr));
		tr->ai = adj_index0;
	    }
	    if (PREDICT_FALSE(b1->flags & VLIB_BUFFER_IS_TRACED))
	    {
		adj_midchain_tx_trace_t *tr = vlib_add_trace (vm, node,
							      b1, sizeof (*tr));
		tr->ai = adj_index1;
	    }
	    if (PREDICT_FALSE(b2->flags & VLIB_BUFFER_IS_TRACED))
	    {
		adj_midchain_tx_trace_t *tr = vlib_add_trace (vm, node,
							      b2, sizeof (*tr));
		tr->ai = adj_index2;
	    }
	    if (PREDICT_FALSE(b3->flags & VLIB_BUFFER_IS_TRACED))
	    {
		adj_midchain_tx_trace_t *tr = vlib_add_trace (vm, node,
							      b3, sizeof (*tr));
		tr->ai = adj_index3;
	    }

	    vlib_validate_buffer_enqueue_x4 (vm, node, next_index,
					     to_next, n_left_to_next,
					     bi0, bi1, bi2, bi3,
					     next0, next1, next2, next3);
	}
	while (n_left_from > 0 && n_left_to_next > 0)
	{
	    u32 bi0, adj_index0, next0;
	    const ip_adjacency_t * adj0;
	    const dpo_id_t *dpo0;
	    vlib_buffer_t * b0;

	    bi0 = from[0];
	    to_next[0] = bi0;
	    from += 1;
	    to_next += 1;
	    n_left_from -= 1;
	    n_left_to_next -= 1;

	    b0 = vlib_get_buffer(vm, bi0);

	    /* Follow the DPO on which the midchain is stacked */
	    adj_index0 = vnet_buffer(b0)->ip.adj_index[VLIB_TX];
	    adj0 = adj_get(adj_index0);
	    dpo0 = &adj0->sub_type.midchain.next_dpo;
	    next0 = dpo0->dpoi_next_node;
            vnet_buffer(b0)->ip.adj_index[VLIB_TX] = dpo0->dpoi_index;

	    if (interface_count)
	    {
		vlib_increment_combined_counter (im->combined_sw_if_counters
						 + VNET_INTERFACE_COUNTER_TX,
						 thread_index,
						 adj0->rewrite_header.sw_if_index,
						 1,
						 vlib_buffer_length_in_chain (vm, b0));
	    }

	    if (PREDICT_FALSE(b0->flags & VLIB_BUFFER_IS_TRACED))
	    {
		adj_midchain_tx_trace_t *tr = vlib_add_trace (vm, node,
							      b0, sizeof (*tr));
		tr->ai = adj_index0;
	    }

	    vlib_validate_buffer_enqueue_x1 (vm, node, next_index,
					     to_next, n_left_to_next,
					     bi0, next0);
	}

	vlib_put_next_frame (vm, node, next_index, n_left_to_next);
    }

    return frame->n_vectors;
}
コード例 #5
0
ファイル: gtpu_encap.c プロジェクト: chrisy/vpp
always_inline uword
gtpu_encap_inline (vlib_main_t * vm,
		    vlib_node_runtime_t * node,
		    vlib_frame_t * from_frame,
		    u32 is_ip4)
{
  u32 n_left_from, next_index, * from, * to_next;
  gtpu_main_t * gtm = &gtpu_main;
  vnet_main_t * vnm = gtm->vnet_main;
  vnet_interface_main_t * im = &vnm->interface_main;
  u32 pkts_encapsulated = 0;
  u16 old_l0 = 0, old_l1 = 0, old_l2 = 0, old_l3 = 0;
  u32 thread_index = vlib_get_thread_index();
  u32 stats_sw_if_index, stats_n_packets, stats_n_bytes;
  u32 sw_if_index0 = 0, sw_if_index1 = 0, sw_if_index2 = 0, sw_if_index3 = 0;
  u32 next0 = 0, next1 = 0, next2 = 0, next3 = 0;
  vnet_hw_interface_t * hi0, * hi1, * hi2, * hi3;
  gtpu_tunnel_t * t0 = NULL, * t1 = NULL, * t2 = NULL, * t3 = NULL;

  from = vlib_frame_vector_args (from_frame);
  n_left_from = from_frame->n_vectors;

  next_index = node->cached_next_index;
  stats_sw_if_index = node->runtime_data[0];
  stats_n_packets = stats_n_bytes = 0;

  while (n_left_from > 0)
    {
      u32 n_left_to_next;

      vlib_get_next_frame (vm, node, next_index,
			   to_next, n_left_to_next);

      while (n_left_from >= 8 && n_left_to_next >= 4)
	{
          u32 bi0, bi1, bi2, bi3;
	  vlib_buffer_t * b0, * b1, * b2, * b3;
          u32 flow_hash0, flow_hash1, flow_hash2, flow_hash3;
	  u32 len0, len1, len2, len3;
          ip4_header_t * ip4_0, * ip4_1, * ip4_2, * ip4_3;
          ip6_header_t * ip6_0, * ip6_1, * ip6_2, * ip6_3;
          udp_header_t * udp0, * udp1, * udp2, * udp3;
          gtpu_header_t * gtpu0, * gtpu1, * gtpu2, * gtpu3;
          u64 * copy_src0, * copy_dst0;
          u64 * copy_src1, * copy_dst1;
          u64 * copy_src2, * copy_dst2;
          u64 * copy_src3, * copy_dst3;
          u32 * copy_src_last0, * copy_dst_last0;
          u32 * copy_src_last1, * copy_dst_last1;
          u32 * copy_src_last2, * copy_dst_last2;
          u32 * copy_src_last3, * copy_dst_last3;
          u16 new_l0, new_l1, new_l2, new_l3;
          ip_csum_t sum0, sum1, sum2, sum3;

	  /* Prefetch next iteration. */
	  {
	    vlib_buffer_t * p4, * p5, * p6, * p7;

	    p4 = vlib_get_buffer (vm, from[4]);
	    p5 = vlib_get_buffer (vm, from[5]);
	    p6 = vlib_get_buffer (vm, from[6]);
	    p7 = vlib_get_buffer (vm, from[7]);

	    vlib_prefetch_buffer_header (p4, LOAD);
	    vlib_prefetch_buffer_header (p5, LOAD);
	    vlib_prefetch_buffer_header (p6, LOAD);
	    vlib_prefetch_buffer_header (p7, LOAD);

	    CLIB_PREFETCH (p4->data, 2*CLIB_CACHE_LINE_BYTES, LOAD);
	    CLIB_PREFETCH (p5->data, 2*CLIB_CACHE_LINE_BYTES, LOAD);
	    CLIB_PREFETCH (p6->data, 2*CLIB_CACHE_LINE_BYTES, LOAD);
	    CLIB_PREFETCH (p7->data, 2*CLIB_CACHE_LINE_BYTES, LOAD);
	  }

	  bi0 = from[0];
	  bi1 = from[1];
	  bi2 = from[2];
	  bi3 = from[3];
	  to_next[0] = bi0;
	  to_next[1] = bi1;
	  to_next[2] = bi2;
	  to_next[3] = bi3;
	  from += 4;
	  to_next += 4;
	  n_left_to_next -= 4;
	  n_left_from -= 4;

	  b0 = vlib_get_buffer (vm, bi0);
	  b1 = vlib_get_buffer (vm, bi1);
	  b2 = vlib_get_buffer (vm, bi2);
	  b3 = vlib_get_buffer (vm, bi3);

          flow_hash0 = vnet_l2_compute_flow_hash (b0);
          flow_hash1 = vnet_l2_compute_flow_hash (b1);
          flow_hash2 = vnet_l2_compute_flow_hash (b2);
          flow_hash3 = vnet_l2_compute_flow_hash (b3);

	  /* Get next node index and adj index from tunnel next_dpo */
	  sw_if_index0 = vnet_buffer(b0)->sw_if_index[VLIB_TX];
	  sw_if_index1 = vnet_buffer(b1)->sw_if_index[VLIB_TX];
	  sw_if_index2 = vnet_buffer(b2)->sw_if_index[VLIB_TX];
	  sw_if_index3 = vnet_buffer(b3)->sw_if_index[VLIB_TX];
	  hi0 = vnet_get_sup_hw_interface (vnm, sw_if_index0);
	  hi1 = vnet_get_sup_hw_interface (vnm, sw_if_index1);
	  hi2 = vnet_get_sup_hw_interface (vnm, sw_if_index2);
	  hi3 = vnet_get_sup_hw_interface (vnm, sw_if_index3);
	  t0 = &gtm->tunnels[hi0->dev_instance];
	  t1 = &gtm->tunnels[hi1->dev_instance];
	  t2 = &gtm->tunnels[hi2->dev_instance];
	  t3 = &gtm->tunnels[hi3->dev_instance];

	  /* Note: change to always set next0 if it may be set to drop */
	  next0 = t0->next_dpo.dpoi_next_node;
          vnet_buffer(b0)->ip.adj_index[VLIB_TX] = t0->next_dpo.dpoi_index;
	  next1 = t1->next_dpo.dpoi_next_node;
          vnet_buffer(b1)->ip.adj_index[VLIB_TX] = t1->next_dpo.dpoi_index;
	  next2 = t2->next_dpo.dpoi_next_node;
          vnet_buffer(b2)->ip.adj_index[VLIB_TX] = t2->next_dpo.dpoi_index;
	  next3 = t3->next_dpo.dpoi_next_node;
          vnet_buffer(b3)->ip.adj_index[VLIB_TX] = t3->next_dpo.dpoi_index;

          /* Apply the rewrite string. $$$$ vnet_rewrite? */
          vlib_buffer_advance (b0, -(word)_vec_len(t0->rewrite));
          vlib_buffer_advance (b1, -(word)_vec_len(t1->rewrite));
          vlib_buffer_advance (b2, -(word)_vec_len(t2->rewrite));
          vlib_buffer_advance (b3, -(word)_vec_len(t3->rewrite));

	  if (is_ip4)
	    {
	      ip4_0 = vlib_buffer_get_current(b0);
	      ip4_1 = vlib_buffer_get_current(b1);
	      ip4_2 = vlib_buffer_get_current(b2);
	      ip4_3 = vlib_buffer_get_current(b3);

	      /* Copy the fixed header */
	      copy_dst0 = (u64 *) ip4_0;
	      copy_src0 = (u64 *) t0->rewrite;
	      copy_dst1 = (u64 *) ip4_1;
	      copy_src1 = (u64 *) t1->rewrite;
	      copy_dst2 = (u64 *) ip4_2;
	      copy_src2 = (u64 *) t2->rewrite;
	      copy_dst3 = (u64 *) ip4_3;
	      copy_src3 = (u64 *) t3->rewrite;

	      /* Copy first 32 octets 8-bytes at a time */
#define _(offs) copy_dst0[offs] = copy_src0[offs];
	      foreach_fixed_header4_offset;
#undef _
#define _(offs) copy_dst1[offs] = copy_src1[offs];
	      foreach_fixed_header4_offset;
#undef _
#define _(offs) copy_dst2[offs] = copy_src2[offs];
	      foreach_fixed_header4_offset;
#undef _
#define _(offs) copy_dst3[offs] = copy_src3[offs];
	      foreach_fixed_header4_offset;
#undef _
	      /* Last 4 octets. Hopefully gcc will be our friend */
              copy_dst_last0 = (u32 *)(&copy_dst0[4]);
              copy_src_last0 = (u32 *)(&copy_src0[4]);
              copy_dst_last0[0] = copy_src_last0[0];
              copy_dst_last1 = (u32 *)(&copy_dst1[4]);
              copy_src_last1 = (u32 *)(&copy_src1[4]);
              copy_dst_last1[0] = copy_src_last1[0];
              copy_dst_last2 = (u32 *)(&copy_dst2[4]);
              copy_src_last2 = (u32 *)(&copy_src2[4]);
              copy_dst_last2[0] = copy_src_last2[0];
              copy_dst_last3 = (u32 *)(&copy_dst3[4]);
              copy_src_last3 = (u32 *)(&copy_src3[4]);
              copy_dst_last3[0] = copy_src_last3[0];

	      /* Fix the IP4 checksum and length */
	      sum0 = ip4_0->checksum;
	      new_l0 = /* old_l0 always 0, see the rewrite setup */
                clib_host_to_net_u16 (vlib_buffer_length_in_chain (vm, b0));
              sum0 = ip_csum_update (sum0, old_l0, new_l0, ip4_header_t,
				     length /* changed member */);
	      ip4_0->checksum = ip_csum_fold (sum0);
	      ip4_0->length = new_l0;
	      sum1 = ip4_1->checksum;
	      new_l1 = /* old_l1 always 0, see the rewrite setup */
                clib_host_to_net_u16 (vlib_buffer_length_in_chain (vm, b1));
              sum1 = ip_csum_update (sum1, old_l1, new_l1, ip4_header_t,
				     length /* changed member */);
	      ip4_1->checksum = ip_csum_fold (sum1);
	      ip4_1->length = new_l1;
	      sum2 = ip4_2->checksum;
	      new_l2 = /* old_l0 always 0, see the rewrite setup */
                clib_host_to_net_u16 (vlib_buffer_length_in_chain (vm, b2));
              sum2 = ip_csum_update (sum2, old_l2, new_l2, ip4_header_t,
				     length /* changed member */);
	      ip4_2->checksum = ip_csum_fold (sum2);
	      ip4_2->length = new_l2;
	      sum3 = ip4_3->checksum;
	      new_l3 = /* old_l1 always 0, see the rewrite setup */
                clib_host_to_net_u16 (vlib_buffer_length_in_chain (vm, b3));
              sum3 = ip_csum_update (sum3, old_l3, new_l3, ip4_header_t,
				     length /* changed member */);
	      ip4_3->checksum = ip_csum_fold (sum3);
	      ip4_3->length = new_l3;

	      /* Fix UDP length and set source port */
	      udp0 = (udp_header_t *)(ip4_0+1);
	      new_l0 = clib_host_to_net_u16 (vlib_buffer_length_in_chain(vm, b0)
					     - sizeof (*ip4_0));
	      udp0->length = new_l0;
	      udp0->src_port = flow_hash0;
	      udp1 = (udp_header_t *)(ip4_1+1);
	      new_l1 = clib_host_to_net_u16 (vlib_buffer_length_in_chain(vm, b1)
					     - sizeof (*ip4_1));
	      udp1->length = new_l1;
	      udp1->src_port = flow_hash1;
	      udp2 = (udp_header_t *)(ip4_2+1);
	      new_l2 = clib_host_to_net_u16 (vlib_buffer_length_in_chain(vm, b2)
					     - sizeof (*ip4_2));
	      udp2->length = new_l2;
	      udp2->src_port = flow_hash2;
	      udp3 = (udp_header_t *)(ip4_3+1);
	      new_l3 = clib_host_to_net_u16 (vlib_buffer_length_in_chain(vm, b3)
					     - sizeof (*ip4_3));
	      udp3->length = new_l3;
	      udp3->src_port = flow_hash3;

	      /* Fix GTPU length */
	      gtpu0 = (gtpu_header_t *)(udp0+1);
	      new_l0 = clib_host_to_net_u16 (vlib_buffer_length_in_chain(vm, b0)
					     - sizeof (*ip4_0) - sizeof(*udp0)
					     - GTPU_V1_HDR_LEN);
	      gtpu0->length = new_l0;
	      gtpu1 = (gtpu_header_t *)(udp1+1);
	      new_l1 = clib_host_to_net_u16 (vlib_buffer_length_in_chain(vm, b1)
					     - sizeof (*ip4_1) - sizeof(*udp1)
					     - GTPU_V1_HDR_LEN);
	      gtpu1->length = new_l1;
	      gtpu2 = (gtpu_header_t *)(udp2+1);
	      new_l2 = clib_host_to_net_u16 (vlib_buffer_length_in_chain(vm, b2)
					     - sizeof (*ip4_2) - sizeof(*udp2)
					     - GTPU_V1_HDR_LEN);
	      gtpu2->length = new_l2;
	      gtpu3 = (gtpu_header_t *)(udp3+1);
	      new_l3 = clib_host_to_net_u16 (vlib_buffer_length_in_chain(vm, b3)
					     - sizeof (*ip4_3) - sizeof(*udp3)
					     - GTPU_V1_HDR_LEN);
	      gtpu3->length = new_l3;
	    }
	  else /* ipv6 */
	    {
              int bogus = 0;

	      ip6_0 = vlib_buffer_get_current(b0);
	      ip6_1 = vlib_buffer_get_current(b1);
	      ip6_2 = vlib_buffer_get_current(b2);
	      ip6_3 = vlib_buffer_get_current(b3);

	      /* Copy the fixed header */
	      copy_dst0 = (u64 *) ip6_0;
	      copy_src0 = (u64 *) t0->rewrite;
	      copy_dst1 = (u64 *) ip6_1;
	      copy_src1 = (u64 *) t1->rewrite;
	      copy_dst2 = (u64 *) ip6_2;
	      copy_src2 = (u64 *) t2->rewrite;
	      copy_dst3 = (u64 *) ip6_3;
	      copy_src3 = (u64 *) t3->rewrite;
	      /* Copy first 56 (ip6) octets 8-bytes at a time */
#define _(offs) copy_dst0[offs] = copy_src0[offs];
	      foreach_fixed_header6_offset;
#undef _
#define _(offs) copy_dst1[offs] = copy_src1[offs];
	      foreach_fixed_header6_offset;
#undef _
#define _(offs) copy_dst2[offs] = copy_src2[offs];
	      foreach_fixed_header6_offset;
#undef _
#define _(offs) copy_dst3[offs] = copy_src3[offs];
	      foreach_fixed_header6_offset;
#undef _
	      /* Fix IP6 payload length */
	      new_l0 =
                clib_host_to_net_u16 (vlib_buffer_length_in_chain (vm, b0)
				      - sizeof(*ip6_0));
	      ip6_0->payload_length = new_l0;
	      new_l1 =
                clib_host_to_net_u16 (vlib_buffer_length_in_chain (vm, b1)
				      - sizeof(*ip6_1));
	      ip6_1->payload_length = new_l1;
	      new_l2 =
                clib_host_to_net_u16 (vlib_buffer_length_in_chain (vm, b2)
				      - sizeof(*ip6_2));
	      ip6_2->payload_length = new_l2;
	      new_l3 =
                clib_host_to_net_u16 (vlib_buffer_length_in_chain (vm, b3)
				      - sizeof(*ip6_3));
	      ip6_3->payload_length = new_l3;

	      /* Fix UDP length  and set source port */
	      udp0 = (udp_header_t *)(ip6_0+1);
	      udp0->length = new_l0;
	      udp0->src_port = flow_hash0;
	      udp1 = (udp_header_t *)(ip6_1+1);
	      udp1->length = new_l1;
	      udp1->src_port = flow_hash1;
	      udp2 = (udp_header_t *)(ip6_2+1);
	      udp2->length = new_l2;
	      udp2->src_port = flow_hash2;
	      udp3 = (udp_header_t *)(ip6_3+1);
	      udp3->length = new_l3;
	      udp3->src_port = flow_hash3;

	      /* Fix GTPU length */
	      gtpu0 = (gtpu_header_t *)(udp0+1);
	      new_l0 = clib_host_to_net_u16 (vlib_buffer_length_in_chain(vm, b0)
					     - sizeof (*ip6_0) - sizeof(*udp0)
					     - GTPU_V1_HDR_LEN);
	      gtpu0->length = new_l0;
	      gtpu1 = (gtpu_header_t *)(udp1+1);
	      new_l1 = clib_host_to_net_u16 (vlib_buffer_length_in_chain(vm, b1)
					     - sizeof (*ip6_1) - sizeof(*udp1)
					     - GTPU_V1_HDR_LEN);
	      gtpu1->length = new_l1;
	      gtpu2 = (gtpu_header_t *)(udp2+1);
	      new_l2 = clib_host_to_net_u16 (vlib_buffer_length_in_chain(vm, b2)
					     - sizeof (*ip6_2) - sizeof(*udp2)
					     - GTPU_V1_HDR_LEN);
	      gtpu2->length = new_l2;
	      gtpu3 = (gtpu_header_t *)(udp3+1);
	      new_l3 = clib_host_to_net_u16 (vlib_buffer_length_in_chain(vm, b3)
					     - sizeof (*ip6_3) - sizeof(*udp3)
					     - GTPU_V1_HDR_LEN);
	      gtpu3->length = new_l3;

	      /* IPv6 UDP checksum is mandatory */
	      udp0->checksum = ip6_tcp_udp_icmp_compute_checksum(vm, b0,
								 ip6_0, &bogus);
	      if (udp0->checksum == 0)
		udp0->checksum = 0xffff;
	      udp1->checksum = ip6_tcp_udp_icmp_compute_checksum(vm, b1,
								 ip6_1, &bogus);
	      if (udp1->checksum == 0)
		udp1->checksum = 0xffff;
	      udp2->checksum = ip6_tcp_udp_icmp_compute_checksum(vm, b2,
								 ip6_2, &bogus);
	      if (udp2->checksum == 0)
		udp2->checksum = 0xffff;
	      udp3->checksum = ip6_tcp_udp_icmp_compute_checksum(vm, b3,
								 ip6_3, &bogus);
	      if (udp3->checksum == 0)
		udp3->checksum = 0xffff;

	    }

          pkts_encapsulated += 4;
 	  len0 = vlib_buffer_length_in_chain (vm, b0);
 	  len1 = vlib_buffer_length_in_chain (vm, b1);
 	  len2 = vlib_buffer_length_in_chain (vm, b2);
 	  len3 = vlib_buffer_length_in_chain (vm, b3);
	  stats_n_packets += 4;
	  stats_n_bytes += len0 + len1 + len2 + len3;

	  /* Batch stats increment on the same gtpu tunnel so counter is not
	     incremented per packet. Note stats are still incremented for deleted
	     and admin-down tunnel where packets are dropped. It is not worthwhile
	     to check for this rare case and affect normal path performance. */
	  if (PREDICT_FALSE ((sw_if_index0 != stats_sw_if_index) ||
			     (sw_if_index1 != stats_sw_if_index) ||
			     (sw_if_index2 != stats_sw_if_index) ||
			     (sw_if_index3 != stats_sw_if_index) ))
	    {
	      stats_n_packets -= 4;
	      stats_n_bytes -= len0 + len1 + len2 + len3;
	      if ( (sw_if_index0 == sw_if_index1 ) &&
		   (sw_if_index1 == sw_if_index2 ) &&
		   (sw_if_index2 == sw_if_index3 ) )
	        {
		  if (stats_n_packets)
		    vlib_increment_combined_counter
		      (im->combined_sw_if_counters + VNET_INTERFACE_COUNTER_TX,
		       thread_index, stats_sw_if_index,
		       stats_n_packets, stats_n_bytes);
		  stats_sw_if_index = sw_if_index0;
		  stats_n_packets = 4;
		  stats_n_bytes = len0 + len1 + len2 + len3;
	        }
	      else
	        {
		  vlib_increment_combined_counter
		      (im->combined_sw_if_counters + VNET_INTERFACE_COUNTER_TX,
		       thread_index, sw_if_index0, 1, len0);
		  vlib_increment_combined_counter
		      (im->combined_sw_if_counters + VNET_INTERFACE_COUNTER_TX,
		       thread_index, sw_if_index1, 1, len1);
		  vlib_increment_combined_counter
		      (im->combined_sw_if_counters + VNET_INTERFACE_COUNTER_TX,
		       thread_index, sw_if_index2, 1, len2);
		  vlib_increment_combined_counter
		      (im->combined_sw_if_counters + VNET_INTERFACE_COUNTER_TX,
		       thread_index, sw_if_index3, 1, len3);
		}
	    }

	  if (PREDICT_FALSE(b0->flags & VLIB_BUFFER_IS_TRACED))
            {
              gtpu_encap_trace_t *tr =
                vlib_add_trace (vm, node, b0, sizeof (*tr));
              tr->tunnel_index = t0 - gtm->tunnels;
              tr->teid = t0->teid;
           }

          if (PREDICT_FALSE(b1->flags & VLIB_BUFFER_IS_TRACED))
            {
              gtpu_encap_trace_t *tr =
                vlib_add_trace (vm, node, b1, sizeof (*tr));
              tr->tunnel_index = t1 - gtm->tunnels;
              tr->teid = t1->teid;
            }

	  vlib_validate_buffer_enqueue_x4 (vm, node, next_index,
					   to_next, n_left_to_next,
					   bi0, bi1, bi2, bi3,
					   next0, next1, next2, next3);
	}

      while (n_left_from > 0 && n_left_to_next > 0)
	{
	  u32 bi0;
	  vlib_buffer_t * b0;
          u32 flow_hash0;
	  u32 len0;
          ip4_header_t * ip4_0;
          ip6_header_t * ip6_0;
          udp_header_t * udp0;
          gtpu_header_t * gtpu0;
          u64 * copy_src0, * copy_dst0;
          u32 * copy_src_last0, * copy_dst_last0;
          u16 new_l0;
          ip_csum_t sum0;

	  bi0 = from[0];
	  to_next[0] = bi0;
	  from += 1;
	  to_next += 1;
	  n_left_from -= 1;
	  n_left_to_next -= 1;

	  b0 = vlib_get_buffer (vm, bi0);

          flow_hash0 = vnet_l2_compute_flow_hash(b0);

	  /* Get next node index and adj index from tunnel next_dpo */
	  sw_if_index0 = vnet_buffer(b0)->sw_if_index[VLIB_TX];
	  hi0 = vnet_get_sup_hw_interface (vnm, sw_if_index0);
	  t0 = &gtm->tunnels[hi0->dev_instance];
	  /* Note: change to always set next0 if it may be set to drop */
	  next0 = t0->next_dpo.dpoi_next_node;
	  vnet_buffer(b0)->ip.adj_index[VLIB_TX] = t0->next_dpo.dpoi_index;

          /* Apply the rewrite string. $$$$ vnet_rewrite? */
          vlib_buffer_advance (b0, -(word)_vec_len(t0->rewrite));

	  if (is_ip4)
	    {
	      ip4_0 = vlib_buffer_get_current(b0);

	      /* Copy the fixed header */
	      copy_dst0 = (u64 *) ip4_0;
	      copy_src0 = (u64 *) t0->rewrite;
	      /* Copy first 32 octets 8-bytes at a time */
#define _(offs) copy_dst0[offs] = copy_src0[offs];
	      foreach_fixed_header4_offset;
#undef _
	      /* Last 4 octets. Hopefully gcc will be our friend */
              copy_dst_last0 = (u32 *)(&copy_dst0[4]);
              copy_src_last0 = (u32 *)(&copy_src0[4]);
              copy_dst_last0[0] = copy_src_last0[0];

	      /* Fix the IP4 checksum and length */
	      sum0 = ip4_0->checksum;
	      new_l0 = /* old_l0 always 0, see the rewrite setup */
                clib_host_to_net_u16 (vlib_buffer_length_in_chain (vm, b0));
              sum0 = ip_csum_update (sum0, old_l0, new_l0, ip4_header_t,
				     length /* changed member */);
	      ip4_0->checksum = ip_csum_fold (sum0);
	      ip4_0->length = new_l0;

	      /* Fix UDP length and set source port */
	      udp0 = (udp_header_t *)(ip4_0+1);
	      new_l0 = clib_host_to_net_u16 (vlib_buffer_length_in_chain(vm, b0)
					     - sizeof (*ip4_0));
	      udp0->length = new_l0;
	      udp0->src_port = flow_hash0;

	      /* Fix GTPU length */
	      gtpu0 = (gtpu_header_t *)(udp0+1);
	      new_l0 = clib_host_to_net_u16 (vlib_buffer_length_in_chain(vm, b0)
					     - sizeof (*ip4_0) - sizeof(*udp0)
					     - GTPU_V1_HDR_LEN);
	      gtpu0->length = new_l0;
	    }

	  else /* ip6 path */
	    {
              int bogus = 0;

	      ip6_0 = vlib_buffer_get_current(b0);
	      /* Copy the fixed header */
	      copy_dst0 = (u64 *) ip6_0;
	      copy_src0 = (u64 *) t0->rewrite;
	      /* Copy first 56 (ip6) octets 8-bytes at a time */
#define _(offs) copy_dst0[offs] = copy_src0[offs];
	      foreach_fixed_header6_offset;
#undef _
	      /* Fix IP6 payload length */
	      new_l0 =
                clib_host_to_net_u16 (vlib_buffer_length_in_chain (vm, b0)
				      - sizeof(*ip6_0));
	      ip6_0->payload_length = new_l0;

	      /* Fix UDP length  and set source port */
	      udp0 = (udp_header_t *)(ip6_0+1);
	      udp0->length = new_l0;
	      udp0->src_port = flow_hash0;

	      /* Fix GTPU length */
	      gtpu0 = (gtpu_header_t *)(udp0+1);
	      new_l0 = clib_host_to_net_u16 (vlib_buffer_length_in_chain(vm, b0)
					     - sizeof (*ip4_0) - sizeof(*udp0)
					     - GTPU_V1_HDR_LEN);
	      gtpu0->length = new_l0;

	      /* IPv6 UDP checksum is mandatory */
	      udp0->checksum = ip6_tcp_udp_icmp_compute_checksum(vm, b0,
								 ip6_0, &bogus);
	      if (udp0->checksum == 0)
		udp0->checksum = 0xffff;
	    }

          pkts_encapsulated ++;
	  len0 = vlib_buffer_length_in_chain (vm, b0);
	  stats_n_packets += 1;
	  stats_n_bytes += len0;

	  /* Batch stats increment on the same gtpu tunnel so counter is not
	     incremented per packet. Note stats are still incremented for deleted
	     and admin-down tunnel where packets are dropped. It is not worthwhile
	     to check for this rare case and affect normal path performance. */
	  if (PREDICT_FALSE (sw_if_index0 != stats_sw_if_index))
	    {
	      stats_n_packets -= 1;
	      stats_n_bytes -= len0;
	      if (stats_n_packets)
		vlib_increment_combined_counter
		  (im->combined_sw_if_counters + VNET_INTERFACE_COUNTER_TX,
		   thread_index, stats_sw_if_index,
		   stats_n_packets, stats_n_bytes);
	      stats_n_packets = 1;
	      stats_n_bytes = len0;
	      stats_sw_if_index = sw_if_index0;
	    }

          if (PREDICT_FALSE(b0->flags & VLIB_BUFFER_IS_TRACED))
            {
              gtpu_encap_trace_t *tr =
                vlib_add_trace (vm, node, b0, sizeof (*tr));
              tr->tunnel_index = t0 - gtm->tunnels;
              tr->teid = t0->teid;
            }
	  vlib_validate_buffer_enqueue_x1 (vm, node, next_index,
					   to_next, n_left_to_next,
					   bi0, next0);
	}

      vlib_put_next_frame (vm, node, next_index, n_left_to_next);
    }

  /* Do we still need this now that tunnel tx stats is kept? */
  vlib_node_increment_counter (vm, node->node_index,
                               GTPU_ENCAP_ERROR_ENCAPSULATED,
                               pkts_encapsulated);

  /* Increment any remaining batch stats */
  if (stats_n_packets)
    {
      vlib_increment_combined_counter
	(im->combined_sw_if_counters + VNET_INTERFACE_COUNTER_TX,
	 thread_index, stats_sw_if_index, stats_n_packets, stats_n_bytes);
      node->runtime_data[0] = stats_sw_if_index;
    }

  return from_frame->n_vectors;
}
コード例 #6
0
ファイル: encap.c プロジェクト: chrisy/vpp
always_inline uword
geneve_encap_inline (vlib_main_t * vm,
		     vlib_node_runtime_t * node,
		     vlib_frame_t * from_frame, u32 is_ip4)
{
  u32 n_left_from, next_index, *from, *to_next;
  geneve_main_t *vxm = &geneve_main;
  vnet_main_t *vnm = vxm->vnet_main;
  vnet_interface_main_t *im = &vnm->interface_main;
  u32 pkts_encapsulated = 0;
  u16 old_l0 = 0, old_l1 = 0;
  u32 thread_index = vm->thread_index;
  u32 stats_sw_if_index, stats_n_packets, stats_n_bytes;
  u32 sw_if_index0 = ~0, sw_if_index1 = ~0;
  u32 next0 = 0, next1 = 0;
  vnet_hw_interface_t *hi0, *hi1;
  geneve_tunnel_t *t0 = NULL, *t1 = NULL;

  from = vlib_frame_vector_args (from_frame);
  n_left_from = from_frame->n_vectors;

  next_index = node->cached_next_index;
  stats_sw_if_index = node->runtime_data[0];
  stats_n_packets = stats_n_bytes = 0;

  while (n_left_from > 0)
    {
      u32 n_left_to_next;

      vlib_get_next_frame (vm, node, next_index, to_next, n_left_to_next);

      while (n_left_from >= 4 && n_left_to_next >= 2)
	{
	  u32 bi0, bi1;
	  vlib_buffer_t *b0, *b1;
	  u32 flow_hash0, flow_hash1;
	  u32 len0, len1;
	  ip4_header_t *ip4_0, *ip4_1;
	  ip6_header_t *ip6_0, *ip6_1;
	  udp_header_t *udp0, *udp1;
	  u64 *copy_src0, *copy_dst0;
	  u64 *copy_src1, *copy_dst1;
	  u32 *copy_src_last0, *copy_dst_last0;
	  u32 *copy_src_last1, *copy_dst_last1;
	  u16 new_l0, new_l1;
	  ip_csum_t sum0, sum1;

	  /* Prefetch next iteration. */
	  {
	    vlib_buffer_t *p2, *p3;

	    p2 = vlib_get_buffer (vm, from[2]);
	    p3 = vlib_get_buffer (vm, from[3]);

	    vlib_prefetch_buffer_header (p2, LOAD);
	    vlib_prefetch_buffer_header (p3, LOAD);

	    CLIB_PREFETCH (p2->data, 2 * CLIB_CACHE_LINE_BYTES, LOAD);
	    CLIB_PREFETCH (p3->data, 2 * CLIB_CACHE_LINE_BYTES, LOAD);
	  }

	  bi0 = from[0];
	  bi1 = from[1];
	  to_next[0] = bi0;
	  to_next[1] = bi1;
	  from += 2;
	  to_next += 2;
	  n_left_to_next -= 2;
	  n_left_from -= 2;

	  b0 = vlib_get_buffer (vm, bi0);
	  b1 = vlib_get_buffer (vm, bi1);

	  flow_hash0 = vnet_l2_compute_flow_hash (b0);
	  flow_hash1 = vnet_l2_compute_flow_hash (b1);

	  /* Get next node index and adj index from tunnel next_dpo */
	  if (sw_if_index0 != vnet_buffer (b0)->sw_if_index[VLIB_TX])
	    {
	      sw_if_index0 = vnet_buffer (b0)->sw_if_index[VLIB_TX];
	      hi0 = vnet_get_sup_hw_interface (vnm, sw_if_index0);
	      t0 = &vxm->tunnels[hi0->dev_instance];
	      /* Note: change to always set next0 if it may be set to drop */
	      next0 = t0->next_dpo.dpoi_next_node;
	    }

	  ASSERT (t0 != NULL);

	  vnet_buffer (b0)->ip.adj_index[VLIB_TX] = t0->next_dpo.dpoi_index;

	  /* Get next node index and adj index from tunnel next_dpo */
	  if (sw_if_index1 != vnet_buffer (b1)->sw_if_index[VLIB_TX])
	    {
	      sw_if_index1 = vnet_buffer (b1)->sw_if_index[VLIB_TX];
	      hi1 = vnet_get_sup_hw_interface (vnm, sw_if_index1);
	      t1 = &vxm->tunnels[hi1->dev_instance];
	      /* Note: change to always set next1 if it may be set to drop */
	      next1 = t1->next_dpo.dpoi_next_node;
	    }

	  ASSERT (t1 != NULL);

	  vnet_buffer (b1)->ip.adj_index[VLIB_TX] = t1->next_dpo.dpoi_index;

	  /* Apply the rewrite string. $$$$ vnet_rewrite? */
	  vlib_buffer_advance (b0, -(word) _vec_len (t0->rewrite));
	  vlib_buffer_advance (b1, -(word) _vec_len (t1->rewrite));

	  if (is_ip4)
	    {
	      u8 ip4_geneve_base_header_len =
		sizeof (ip4_header_t) + sizeof (udp_header_t) +
		GENEVE_BASE_HEADER_LENGTH;
	      u8 ip4_geneve_header_total_len0 = ip4_geneve_base_header_len;
	      u8 ip4_geneve_header_total_len1 = ip4_geneve_base_header_len;
#if SUPPORT_OPTIONS_HEADER==1
	      ip4_geneve_header_total_len0 += t0->options_len;
	      ip4_geneve_header_total_len1 += t1->options_len;
#endif
	      ASSERT (vec_len (t0->rewrite) == ip4_geneve_header_total_len0);
	      ASSERT (vec_len (t1->rewrite) == ip4_geneve_header_total_len1);

	      ip4_0 = vlib_buffer_get_current (b0);
	      ip4_1 = vlib_buffer_get_current (b1);

	      /* Copy the fixed header */
	      copy_dst0 = (u64 *) ip4_0;
	      copy_src0 = (u64 *) t0->rewrite;
	      copy_dst1 = (u64 *) ip4_1;
	      copy_src1 = (u64 *) t1->rewrite;
	      /* Copy first 32 octets 8-bytes at a time */
#define _(offs) copy_dst0[offs] = copy_src0[offs];
	      foreach_fixed_header4_offset;
#undef _
#define _(offs) copy_dst1[offs] = copy_src1[offs];
	      foreach_fixed_header4_offset;
#undef _
	      /* Last 4 octets. Hopefully gcc will be our friend */
	      copy_dst_last0 = (u32 *) (&copy_dst0[4]);
	      copy_src_last0 = (u32 *) (&copy_src0[4]);
	      copy_dst_last0[0] = copy_src_last0[0];
	      copy_dst_last1 = (u32 *) (&copy_dst1[4]);
	      copy_src_last1 = (u32 *) (&copy_src1[4]);
	      copy_dst_last1[0] = copy_src_last1[0];

	      /* Fix the IP4 checksum and length */
	      sum0 = ip4_0->checksum;
	      new_l0 =		/* old_l0 always 0, see the rewrite setup */
		clib_host_to_net_u16 (vlib_buffer_length_in_chain (vm, b0));
	      sum0 = ip_csum_update (sum0, old_l0, new_l0, ip4_header_t,
				     length /* changed member */ );
	      ip4_0->checksum = ip_csum_fold (sum0);
	      ip4_0->length = new_l0;
	      sum1 = ip4_1->checksum;
	      new_l1 =		/* old_l1 always 0, see the rewrite setup */
		clib_host_to_net_u16 (vlib_buffer_length_in_chain (vm, b1));
	      sum1 = ip_csum_update (sum1, old_l1, new_l1, ip4_header_t,
				     length /* changed member */ );
	      ip4_1->checksum = ip_csum_fold (sum1);
	      ip4_1->length = new_l1;

	      /* Fix UDP length and set source port */
	      udp0 = (udp_header_t *) (ip4_0 + 1);
	      new_l0 =
		clib_host_to_net_u16 (vlib_buffer_length_in_chain (vm, b0) -
				      sizeof (*ip4_0));
	      udp0->length = new_l0;
	      udp0->src_port = flow_hash0;
	      udp1 = (udp_header_t *) (ip4_1 + 1);
	      new_l1 =
		clib_host_to_net_u16 (vlib_buffer_length_in_chain (vm, b1) -
				      sizeof (*ip4_1));
	      udp1->length = new_l1;
	      udp1->src_port = flow_hash1;
	    }
	  else			/* ipv6 */
	    {
	      int bogus = 0;

	      u8 ip6_geneve_base_header_len =
		sizeof (ip6_header_t) + sizeof (udp_header_t) +
		GENEVE_BASE_HEADER_LENGTH;
	      u8 ip6_geneve_header_total_len0 = ip6_geneve_base_header_len;
	      u8 ip6_geneve_header_total_len1 = ip6_geneve_base_header_len;
#if SUPPORT_OPTIONS_HEADER==1
	      ip6_geneve_header_total_len0 += t0->options_len;
	      ip6_geneve_header_total_len1 += t1->options_len;
#endif
	      ASSERT (vec_len (t0->rewrite) == ip6_geneve_header_total_len0);
	      ASSERT (vec_len (t1->rewrite) == ip6_geneve_header_total_len1);

	      ip6_0 = vlib_buffer_get_current (b0);
	      ip6_1 = vlib_buffer_get_current (b1);

	      /* Copy the fixed header */
	      copy_dst0 = (u64 *) ip6_0;
	      copy_src0 = (u64 *) t0->rewrite;
	      copy_dst1 = (u64 *) ip6_1;
	      copy_src1 = (u64 *) t1->rewrite;
	      /* Copy first 56 (ip6) octets 8-bytes at a time */
#define _(offs) copy_dst0[offs] = copy_src0[offs];
	      foreach_fixed_header6_offset;
#undef _
#define _(offs) copy_dst1[offs] = copy_src1[offs];
	      foreach_fixed_header6_offset;
#undef _
	      /* Fix IP6 payload length */
	      new_l0 =
		clib_host_to_net_u16 (vlib_buffer_length_in_chain (vm, b0)
				      - sizeof (*ip6_0));
	      ip6_0->payload_length = new_l0;
	      new_l1 =
		clib_host_to_net_u16 (vlib_buffer_length_in_chain (vm, b1)
				      - sizeof (*ip6_1));
	      ip6_1->payload_length = new_l1;

	      /* Fix UDP length  and set source port */
	      udp0 = (udp_header_t *) (ip6_0 + 1);
	      udp0->length = new_l0;
	      udp0->src_port = flow_hash0;
	      udp1 = (udp_header_t *) (ip6_1 + 1);
	      udp1->length = new_l1;
	      udp1->src_port = flow_hash1;

	      /* IPv6 UDP checksum is mandatory */
	      udp0->checksum = ip6_tcp_udp_icmp_compute_checksum (vm, b0,
								  ip6_0,
								  &bogus);
	      ASSERT (bogus == 0);
	      if (udp0->checksum == 0)
		udp0->checksum = 0xffff;
	      udp1->checksum = ip6_tcp_udp_icmp_compute_checksum (vm, b1,
								  ip6_1,
								  &bogus);
	      ASSERT (bogus == 0);
	      if (udp1->checksum == 0)
		udp1->checksum = 0xffff;
	    }

	  pkts_encapsulated += 2;
	  len0 = vlib_buffer_length_in_chain (vm, b0);
	  len1 = vlib_buffer_length_in_chain (vm, b1);
	  stats_n_packets += 2;
	  stats_n_bytes += len0 + len1;

	  /* Batch stats increment on the same geneve tunnel so counter is not
	     incremented per packet. Note stats are still incremented for deleted
	     and admin-down tunnel where packets are dropped. It is not worthwhile
	     to check for this rare case and affect normal path performance. */
	  if (PREDICT_FALSE ((sw_if_index0 != stats_sw_if_index) ||
			     (sw_if_index1 != stats_sw_if_index)))
	    {
	      stats_n_packets -= 2;
	      stats_n_bytes -= len0 + len1;
	      if (sw_if_index0 == sw_if_index1)
		{
		  if (stats_n_packets)
		    vlib_increment_combined_counter
		      (im->combined_sw_if_counters +
		       VNET_INTERFACE_COUNTER_TX, thread_index,
		       stats_sw_if_index, stats_n_packets, stats_n_bytes);
		  stats_sw_if_index = sw_if_index0;
		  stats_n_packets = 2;
		  stats_n_bytes = len0 + len1;
		}
	      else
		{
		  vlib_increment_combined_counter
		    (im->combined_sw_if_counters + VNET_INTERFACE_COUNTER_TX,
		     thread_index, sw_if_index0, 1, len0);
		  vlib_increment_combined_counter
		    (im->combined_sw_if_counters + VNET_INTERFACE_COUNTER_TX,
		     thread_index, sw_if_index1, 1, len1);
		}
	    }

	  if (PREDICT_FALSE (b0->flags & VLIB_BUFFER_IS_TRACED))
	    {
	      geneve_encap_trace_t *tr =
		vlib_add_trace (vm, node, b0, sizeof (*tr));
	      tr->tunnel_index = t0 - vxm->tunnels;
	      tr->vni = t0->vni;
	    }

	  if (PREDICT_FALSE (b1->flags & VLIB_BUFFER_IS_TRACED))
	    {
	      geneve_encap_trace_t *tr =
		vlib_add_trace (vm, node, b1, sizeof (*tr));
	      tr->tunnel_index = t1 - vxm->tunnels;
	      tr->vni = t1->vni;
	    }

	  vlib_validate_buffer_enqueue_x2 (vm, node, next_index,
					   to_next, n_left_to_next,
					   bi0, bi1, next0, next1);
	}

      while (n_left_from > 0 && n_left_to_next > 0)
	{
	  u32 bi0;
	  vlib_buffer_t *b0;
	  u32 flow_hash0;
	  u32 len0;
	  ip4_header_t *ip4_0;
	  ip6_header_t *ip6_0;
	  udp_header_t *udp0;
	  u64 *copy_src0, *copy_dst0;
	  u32 *copy_src_last0, *copy_dst_last0;
	  u16 new_l0;
	  ip_csum_t sum0;

	  bi0 = from[0];
	  to_next[0] = bi0;
	  from += 1;
	  to_next += 1;
	  n_left_from -= 1;
	  n_left_to_next -= 1;

	  b0 = vlib_get_buffer (vm, bi0);

	  flow_hash0 = vnet_l2_compute_flow_hash (b0);

	  /* Get next node index and adj index from tunnel next_dpo */
	  if (sw_if_index0 != vnet_buffer (b0)->sw_if_index[VLIB_TX])
	    {
	      sw_if_index0 = vnet_buffer (b0)->sw_if_index[VLIB_TX];
	      hi0 = vnet_get_sup_hw_interface (vnm, sw_if_index0);
	      t0 = &vxm->tunnels[hi0->dev_instance];
	      /* Note: change to always set next0 if it may be set to drop */
	      next0 = t0->next_dpo.dpoi_next_node;
	    }
	  vnet_buffer (b0)->ip.adj_index[VLIB_TX] = t0->next_dpo.dpoi_index;

	  /* Apply the rewrite string. $$$$ vnet_rewrite? */
	  vlib_buffer_advance (b0, -(word) _vec_len (t0->rewrite));

	  if (is_ip4)
	    {
	      u8 ip4_geneve_base_header_len =
		sizeof (ip4_header_t) + sizeof (udp_header_t) +
		GENEVE_BASE_HEADER_LENGTH;
	      u8 ip4_geneve_header_total_len0 = ip4_geneve_base_header_len;
#if SUPPORT_OPTIONS_HEADER==1
	      ip4_geneve_header_total_len0 += t0->options_len;
#endif
	      ASSERT (vec_len (t0->rewrite) == ip4_geneve_header_total_len0);

	      ip4_0 = vlib_buffer_get_current (b0);

	      /* Copy the fixed header */
	      copy_dst0 = (u64 *) ip4_0;
	      copy_src0 = (u64 *) t0->rewrite;
	      /* Copy first 32 octets 8-bytes at a time */
#define _(offs) copy_dst0[offs] = copy_src0[offs];
	      foreach_fixed_header4_offset;
#undef _
	      /* Last 4 octets. Hopefully gcc will be our friend */
	      copy_dst_last0 = (u32 *) (&copy_dst0[4]);
	      copy_src_last0 = (u32 *) (&copy_src0[4]);
	      copy_dst_last0[0] = copy_src_last0[0];

	      /* Fix the IP4 checksum and length */
	      sum0 = ip4_0->checksum;
	      new_l0 =		/* old_l0 always 0, see the rewrite setup */
		clib_host_to_net_u16 (vlib_buffer_length_in_chain (vm, b0));
	      sum0 = ip_csum_update (sum0, old_l0, new_l0, ip4_header_t,
				     length /* changed member */ );
	      ip4_0->checksum = ip_csum_fold (sum0);
	      ip4_0->length = new_l0;

	      /* Fix UDP length and set source port */
	      udp0 = (udp_header_t *) (ip4_0 + 1);
	      new_l0 =
		clib_host_to_net_u16 (vlib_buffer_length_in_chain (vm, b0) -
				      sizeof (*ip4_0));
	      udp0->length = new_l0;
	      udp0->src_port = flow_hash0;
	    }

	  else			/* ip6 path */
	    {
	      int bogus = 0;

	      u8 ip6_geneve_base_header_len =
		sizeof (ip6_header_t) + sizeof (udp_header_t) +
		GENEVE_BASE_HEADER_LENGTH;
	      u8 ip6_geneve_header_total_len0 = ip6_geneve_base_header_len;
#if SUPPORT_OPTIONS_HEADER==1
	      ip6_geneve_header_total_len0 += t0->options_len;
#endif
	      ASSERT (vec_len (t0->rewrite) == ip6_geneve_header_total_len0);

	      ip6_0 = vlib_buffer_get_current (b0);
	      /* Copy the fixed header */
	      copy_dst0 = (u64 *) ip6_0;
	      copy_src0 = (u64 *) t0->rewrite;
	      /* Copy first 56 (ip6) octets 8-bytes at a time */
#define _(offs) copy_dst0[offs] = copy_src0[offs];
	      foreach_fixed_header6_offset;
#undef _
	      /* Fix IP6 payload length */
	      new_l0 =
		clib_host_to_net_u16 (vlib_buffer_length_in_chain (vm, b0)
				      - sizeof (*ip6_0));
	      ip6_0->payload_length = new_l0;

	      /* Fix UDP length  and set source port */
	      udp0 = (udp_header_t *) (ip6_0 + 1);
	      udp0->length = new_l0;
	      udp0->src_port = flow_hash0;

	      /* IPv6 UDP checksum is mandatory */
	      udp0->checksum = ip6_tcp_udp_icmp_compute_checksum (vm, b0,
								  ip6_0,
								  &bogus);
	      ASSERT (bogus == 0);
	      if (udp0->checksum == 0)
		udp0->checksum = 0xffff;
	    }

	  pkts_encapsulated++;
	  len0 = vlib_buffer_length_in_chain (vm, b0);
	  stats_n_packets += 1;
	  stats_n_bytes += len0;

	  /* Batch stats increment on the same geneve tunnel so counter is not
	     incremented per packet. Note stats are still incremented for deleted
	     and admin-down tunnel where packets are dropped. It is not worthwhile
	     to check for this rare case and affect normal path performance. */
	  if (PREDICT_FALSE (sw_if_index0 != stats_sw_if_index))
	    {
	      stats_n_packets -= 1;
	      stats_n_bytes -= len0;
	      if (stats_n_packets)
		vlib_increment_combined_counter
		  (im->combined_sw_if_counters + VNET_INTERFACE_COUNTER_TX,
		   thread_index, stats_sw_if_index,
		   stats_n_packets, stats_n_bytes);
	      stats_n_packets = 1;
	      stats_n_bytes = len0;
	      stats_sw_if_index = sw_if_index0;
	    }

	  if (PREDICT_FALSE (b0->flags & VLIB_BUFFER_IS_TRACED))
	    {
	      geneve_encap_trace_t *tr =
		vlib_add_trace (vm, node, b0, sizeof (*tr));
	      tr->tunnel_index = t0 - vxm->tunnels;
	      tr->vni = t0->vni;
	    }
	  vlib_validate_buffer_enqueue_x1 (vm, node, next_index,
					   to_next, n_left_to_next,
					   bi0, next0);
	}

      vlib_put_next_frame (vm, node, next_index, n_left_to_next);
    }

  /* Do we still need this now that tunnel tx stats is kept? */
  vlib_node_increment_counter (vm, node->node_index,
			       GENEVE_ENCAP_ERROR_ENCAPSULATED,
			       pkts_encapsulated);

  /* Increment any remaining batch stats */
  if (stats_n_packets)
    {
      vlib_increment_combined_counter
	(im->combined_sw_if_counters + VNET_INTERFACE_COUNTER_TX,
	 thread_index, stats_sw_if_index, stats_n_packets, stats_n_bytes);
      node->runtime_data[0] = stats_sw_if_index;
    }

  return from_frame->n_vectors;
}