From ef486b1545d892f9f0e0d35e7e57cb0ca04d7ff7 Mon Sep 17 00:00:00 2001 From: Hongjun Ni Date: Wed, 12 Apr 2017 19:21:16 +0800 Subject: Add GTP-U plugin. VPP-694 Basic GTP-U feature Change-Id: I31226f890a92c5303ac06e112ed7820cae52d9bd Signed-off-by: Hongjun Ni --- src/plugins/gtpu/gtpu_encap.c | 569 ++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 569 insertions(+) create mode 100644 src/plugins/gtpu/gtpu_encap.c (limited to 'src/plugins/gtpu/gtpu_encap.c') diff --git a/src/plugins/gtpu/gtpu_encap.c b/src/plugins/gtpu/gtpu_encap.c new file mode 100644 index 00000000..a2cfc646 --- /dev/null +++ b/src/plugins/gtpu/gtpu_encap.c @@ -0,0 +1,569 @@ +/* + * Copyright (c) 2015 Cisco and/or its affiliates. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ +#include +#include +#include +#include +#include +#include + +/* Statistics (not all errors) */ +#define foreach_gtpu_encap_error \ +_(ENCAPSULATED, "good packets encapsulated") + +static char * gtpu_encap_error_strings[] = { +#define _(sym,string) string, + foreach_gtpu_encap_error +#undef _ +}; + +typedef enum { +#define _(sym,str) GTPU_ENCAP_ERROR_##sym, + foreach_gtpu_encap_error +#undef _ + GTPU_ENCAP_N_ERROR, +} gtpu_encap_error_t; + +typedef enum { + GTPU_ENCAP_NEXT_DROP, + GTPU_ENCAP_N_NEXT, +} gtpu_encap_next_t; + +typedef struct { + u32 tunnel_index; + u32 teid; +} gtpu_encap_trace_t; + +u8 * format_gtpu_encap_trace (u8 * s, va_list * args) +{ + CLIB_UNUSED (vlib_main_t * vm) = va_arg (*args, vlib_main_t *); + CLIB_UNUSED (vlib_node_t * node) = va_arg (*args, vlib_node_t *); + gtpu_encap_trace_t * t + = va_arg (*args, gtpu_encap_trace_t *); + + s = format (s, "GTPU encap to gtpu_tunnel%d teid %d", + t->tunnel_index, t->teid); + return s; +} + + +#define foreach_fixed_header4_offset \ + _(0) _(1) _(2) _(3) + +#define foreach_fixed_header6_offset \ + _(0) _(1) _(2) _(3) _(4) _(5) _(6) + +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 = >pu_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; + 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; + u32 next0 = 0, next1 = 0; + vnet_hw_interface_t * hi0, * hi1; + gtpu_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; + gtpu_header_t * gtpu0, * gtpu1; + 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 = >m->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; + + /* 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 = >m->tunnels[hi1->dev_instance]; + /* Note: change to always set next1 if it may be set to drop */ + next1 = t1->next_dpo.dpoi_next_node; + } + 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) + { + 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 *)(©_dst0[4]); + copy_src_last0 = (u32 *)(©_src0[4]); + copy_dst_last0[0] = copy_src_last0[0]; + copy_dst_last1 = (u32 *)(©_dst1[4]); + copy_src_last1 = (u32 *)(©_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; + + /* 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)); + 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)); + gtpu1->length = new_l1; + } + else /* ipv6 */ + { + int bogus = 0; + + 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); + 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; + + /* 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)); + 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)); + gtpu1->length = new_l1; + } + + 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 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))) + { + 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)) + { + 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_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; + 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 */ + 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 = >m->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 *)(©_dst0[4]); + copy_src_last0 = (u32 *)(©_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)); + 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; + + /* 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; + + /* 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)); + gtpu0->length = new_l0; + } + + 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; +} + +static uword +gtpu4_encap (vlib_main_t * vm, + vlib_node_runtime_t * node, + vlib_frame_t * from_frame) +{ + return gtpu_encap_inline (vm, node, from_frame, /* is_ip4 */ 1); +} + +static uword +gtpu6_encap (vlib_main_t * vm, + vlib_node_runtime_t * node, + vlib_frame_t * from_frame) +{ + return gtpu_encap_inline (vm, node, from_frame, /* is_ip4 */ 0); +} + +VLIB_REGISTER_NODE (gtpu4_encap_node) = { + .function = gtpu4_encap, + .name = "gtpu4-encap", + .vector_size = sizeof (u32), + .format_trace = format_gtpu_encap_trace, + .type = VLIB_NODE_TYPE_INTERNAL, + .n_errors = ARRAY_LEN(gtpu_encap_error_strings), + .error_strings = gtpu_encap_error_strings, + .n_next_nodes = GTPU_ENCAP_N_NEXT, + .next_nodes = { + [GTPU_ENCAP_NEXT_DROP] = "error-drop", + }, +}; + +VLIB_NODE_FUNCTION_MULTIARCH (gtpu4_encap_node, gtpu4_encap) + +VLIB_REGISTER_NODE (gtpu6_encap_node) = { + .function = gtpu6_encap, + .name = "gtpu6-encap", + .vector_size = sizeof (u32), + .format_trace = format_gtpu_encap_trace, + .type = VLIB_NODE_TYPE_INTERNAL, + .n_errors = ARRAY_LEN(gtpu_encap_error_strings), + .error_strings = gtpu_encap_error_strings, + .n_next_nodes = GTPU_ENCAP_N_NEXT, + .next_nodes = { + [GTPU_ENCAP_NEXT_DROP] = "error-drop", + }, +}; + +VLIB_NODE_FUNCTION_MULTIARCH (gtpu6_encap_node, gtpu6_encap) + -- cgit 1.2.3-korg From e60959a2ad5b36e913c2ca92bcd52325ed9ecc89 Mon Sep 17 00:00:00 2001 From: Hongjun Ni Date: Fri, 19 May 2017 23:52:34 +0800 Subject: Fix GTP-U Coverity issues Change-Id: I982e0dfb96c371672aa636aa7e40f06a84e2bb69 Signed-off-by: Hongjun Ni --- src/plugins/gtpu/gtpu_encap.c | 39 +++++++++++++++------------------------ 1 file changed, 15 insertions(+), 24 deletions(-) (limited to 'src/plugins/gtpu/gtpu_encap.c') diff --git a/src/plugins/gtpu/gtpu_encap.c b/src/plugins/gtpu/gtpu_encap.c index a2cfc646..ef109c66 100644 --- a/src/plugins/gtpu/gtpu_encap.c +++ b/src/plugins/gtpu/gtpu_encap.c @@ -145,25 +145,19 @@ gtpu_encap_inline (vlib_main_t * vm, 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 = >m->tunnels[hi0->dev_instance]; - /* Note: change to always set next0 if it may be set to drop */ - next0 = t0->next_dpo.dpoi_next_node; - } + sw_if_index0 = vnet_buffer(b0)->sw_if_index[VLIB_TX]; + hi0 = vnet_get_sup_hw_interface (vnm, sw_if_index0); + t0 = >m->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; /* 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 = >m->tunnels[hi1->dev_instance]; - /* Note: change to always set next1 if it may be set to drop */ - next1 = t1->next_dpo.dpoi_next_node; - } + sw_if_index1 = vnet_buffer(b1)->sw_if_index[VLIB_TX]; + hi1 = vnet_get_sup_hw_interface (vnm, sw_if_index1); + t1 = >m->tunnels[hi1->dev_instance]; + /* Note: change to always set next1 if it may be set to drop */ + next1 = t1->next_dpo.dpoi_next_node; vnet_buffer(b1)->ip.adj_index[VLIB_TX] = t1->next_dpo.dpoi_index; /* Apply the rewrite string. $$$$ vnet_rewrite? */ @@ -376,14 +370,11 @@ gtpu_encap_inline (vlib_main_t * vm, flow_hash0 = vnet_l2_compute_flow_hash(b0); /* Get next node index and adj index from tunnel next_dpo */ - if (sw_if_index0 != vnet_buffer(b0)->sw_if_index[VLIB_TX]) - { - sw_if_index0 = vnet_buffer(b0)->sw_if_index[VLIB_TX]; - hi0 = vnet_get_sup_hw_interface (vnm, sw_if_index0); - t0 = >m->tunnels[hi0->dev_instance]; - /* Note: change to always set next0 if it may be set to drop */ - next0 = t0->next_dpo.dpoi_next_node; - } + sw_if_index0 = vnet_buffer(b0)->sw_if_index[VLIB_TX]; + hi0 = vnet_get_sup_hw_interface (vnm, sw_if_index0); + t0 = >m->tunnels[hi0->dev_instance]; + /* Note: change to always set next0 if it may be set to drop */ + next0 = t0->next_dpo.dpoi_next_node; vnet_buffer(b0)->ip.adj_index[VLIB_TX] = t0->next_dpo.dpoi_index; /* Apply the rewrite string. $$$$ vnet_rewrite? */ -- cgit 1.2.3-korg From ccd4a6588c6f486f7616286b054a5ac123f5c502 Mon Sep 17 00:00:00 2001 From: Hongjun Ni Date: Thu, 1 Jun 2017 00:24:12 +0800 Subject: Add l3 routing for GTP-U decap Change-Id: Ic3c8cd58cb2b02622299cf74c9efcabd2ef2b513 Signed-off-by: Hongjun Ni --- src/plugins/gtpu/gtpu.api | 6 +++--- src/plugins/gtpu/gtpu.c | 13 +++++++++++-- src/plugins/gtpu/gtpu.h | 6 ++++-- src/plugins/gtpu/gtpu_all_api_h.h | 2 +- src/plugins/gtpu/gtpu_api.c | 2 +- src/plugins/gtpu/gtpu_decap.c | 2 +- src/plugins/gtpu/gtpu_encap.c | 2 +- src/plugins/gtpu/gtpu_error.def | 2 +- src/plugins/gtpu/gtpu_msg_enum.h | 2 +- src/plugins/gtpu/gtpu_test.c | 2 +- 10 files changed, 25 insertions(+), 14 deletions(-) (limited to 'src/plugins/gtpu/gtpu_encap.c') diff --git a/src/plugins/gtpu/gtpu.api b/src/plugins/gtpu/gtpu.api index b11670b3..55ba0390 100644 --- a/src/plugins/gtpu/gtpu.api +++ b/src/plugins/gtpu/gtpu.api @@ -1,5 +1,5 @@ /* - * Copyright (c) 2017 Cisco and/or its affiliates. + * Copyright (c) 2017 Intel and/or its affiliates. * Licensed under the Apache License, Version 2.0 (the "License"); * you may not use this file except in compliance with the License. * You may obtain a copy of the License at: @@ -13,7 +13,7 @@ * limitations under the License. */ -/** \brief /** \brief Set or delete an GTPU tunnel +/** \brief Set or delete an GTPU tunnel @param client_index - opaque cookie to identify the sender @param context - sender context, to match reply w/ request @param is_add - add address if non-zero, else delete @@ -63,7 +63,7 @@ define gtpu_tunnel_dump u32 sw_if_index; }; -/** \brief /** \brief dump details of an GTPU tunnel +/** \brief dump details of an GTPU tunnel @param context - sender context, to match reply w/ request @param sw_if_index - software index of the interface @param is_ipv6 - src_address and dst_address is ipv6 or not diff --git a/src/plugins/gtpu/gtpu.c b/src/plugins/gtpu/gtpu.c index 6a5c5a34..c91612ef 100644 --- a/src/plugins/gtpu/gtpu.c +++ b/src/plugins/gtpu/gtpu.c @@ -1,6 +1,6 @@ /* *------------------------------------------------------------------ - * Copyright (c) 2016 Cisco and/or its affiliates. + * Copyright (c) 2017 Intel and/or its affiliates. * Licensed under the Apache License, Version 2.0 (the "License"); * you may not use this file except in compliance with the License. * You may obtain a copy of the License at: @@ -45,6 +45,10 @@ format_decap_next (u8 * s, va_list * args) return format (s, "drop"); case GTPU_INPUT_NEXT_L2_INPUT: return format (s, "l2"); + case GTPU_INPUT_NEXT_IP4_INPUT: + return format (s, "ip4"); + case GTPU_INPUT_NEXT_IP6_INPUT: + return format (s, "ip6"); default: return format (s, "index %d", next_index); } @@ -649,12 +653,17 @@ unformat_decap_next (unformat_input_t * input, va_list * args) if (unformat (input, "l2")) *result = GTPU_INPUT_NEXT_L2_INPUT; + else if (unformat (input, "ip4")) + *result = GTPU_INPUT_NEXT_IP4_INPUT; + else if (unformat (input, "ip6")) + *result = GTPU_INPUT_NEXT_IP6_INPUT; else if (unformat (input, "node %U", unformat_vlib_node, vm, &node_index)) *result = get_decap_next_for_node (node_index, ipv4_set); else if (unformat (input, "%d", &tmp)) *result = tmp; else return 0; + return 1; } @@ -871,7 +880,7 @@ VLIB_CLI_COMMAND (create_gtpu_tunnel_command, static) = { .short_help = "create gtpu tunnel src " " {dst |group } teid " - " [encap-vrf-id ] [decap-next [l2|node ]] [del]", + " [encap-vrf-id ] [decap-next [l2|ip4|ip6|node ]] [del]", .function = gtpu_add_del_tunnel_command_fn, }; /* *INDENT-ON* */ diff --git a/src/plugins/gtpu/gtpu.h b/src/plugins/gtpu/gtpu.h index 86861974..744d21d4 100644 --- a/src/plugins/gtpu/gtpu.h +++ b/src/plugins/gtpu/gtpu.h @@ -1,6 +1,6 @@ /* *------------------------------------------------------------------ - * Copyright (c) 2016 Intel and/or its affiliates. + * Copyright (c) 2017 Intel and/or its affiliates. * Licensed under the Apache License, Version 2.0 (the "License"); * you may not use this file except in compliance with the License. * You may obtain a copy of the License at: @@ -172,7 +172,9 @@ typedef struct #define foreach_gtpu_input_next \ _(DROP, "error-drop") \ -_(L2_INPUT, "l2-input") +_(L2_INPUT, "l2-input") \ +_(IP4_INPUT, "ip4-input") \ +_(IP6_INPUT, "ip6-input" ) typedef enum { diff --git a/src/plugins/gtpu/gtpu_all_api_h.h b/src/plugins/gtpu/gtpu_all_api_h.h index a4181ae3..dbfe0397 100644 --- a/src/plugins/gtpu/gtpu_all_api_h.h +++ b/src/plugins/gtpu/gtpu_all_api_h.h @@ -1,7 +1,7 @@ /* * gtpu_all_api_h.h - plug-in api #include file * - * Copyright (c) Intel and/or its affiliates. + * Copyright (c) 2017 Intel and/or its affiliates. * Licensed under the Apache License, Version 2.0 (the "License"); * you may not use this file except in compliance with the License. * You may obtain a copy of the License at: diff --git a/src/plugins/gtpu/gtpu_api.c b/src/plugins/gtpu/gtpu_api.c index 58f48c71..49a5053d 100644 --- a/src/plugins/gtpu/gtpu_api.c +++ b/src/plugins/gtpu/gtpu_api.c @@ -2,7 +2,7 @@ *------------------------------------------------------------------ * gtpu_api.c - gtpu api * - * Copyright (c) 2016 Cisco and/or its affiliates. + * Copyright (c) 2017 Intel and/or its affiliates. * Licensed under the Apache License, Version 2.0 (the "License"); * you may not use this file except in compliance with the License. * You may obtain a copy of the License at: diff --git a/src/plugins/gtpu/gtpu_decap.c b/src/plugins/gtpu/gtpu_decap.c index b0be32ec..fc74e7cb 100644 --- a/src/plugins/gtpu/gtpu_decap.c +++ b/src/plugins/gtpu/gtpu_decap.c @@ -1,7 +1,7 @@ /* * decap.c: gtpu tunnel decap packet processing * - * Copyright (c) 2013 Cisco and/or its affiliates. + * Copyright (c) 2017 Intel and/or its affiliates. * Licensed under the Apache License, Version 2.0 (the "License"); * you may not use this file except in compliance with the License. * You may obtain a copy of the License at: diff --git a/src/plugins/gtpu/gtpu_encap.c b/src/plugins/gtpu/gtpu_encap.c index ef109c66..adacf4c9 100644 --- a/src/plugins/gtpu/gtpu_encap.c +++ b/src/plugins/gtpu/gtpu_encap.c @@ -1,5 +1,5 @@ /* - * Copyright (c) 2015 Cisco and/or its affiliates. + * Copyright (c) 2017 Intel and/or its affiliates. * Licensed under the Apache License, Version 2.0 (the "License"); * you may not use this file except in compliance with the License. * You may obtain a copy of the License at: diff --git a/src/plugins/gtpu/gtpu_error.def b/src/plugins/gtpu/gtpu_error.def index a55b2026..093a886f 100644 --- a/src/plugins/gtpu/gtpu_error.def +++ b/src/plugins/gtpu/gtpu_error.def @@ -1,5 +1,5 @@ /* - * Copyright (c) 2015 Cisco and/or its affiliates. + * Copyright (c) 2017 Intel and/or its affiliates. * Licensed under the Apache License, Version 2.0 (the "License"); * you may not use this file except in compliance with the License. * You may obtain a copy of the License at: diff --git a/src/plugins/gtpu/gtpu_msg_enum.h b/src/plugins/gtpu/gtpu_msg_enum.h index 6a2ea448..358a220a 100644 --- a/src/plugins/gtpu/gtpu_msg_enum.h +++ b/src/plugins/gtpu/gtpu_msg_enum.h @@ -1,7 +1,7 @@ /* * gtpu_msg_enum.h - vpp engine plug-in message enumeration * - * Copyright (c) + * Copyright (c) 2017 Intel and/or its affiliates. * Licensed under the Apache License, Version 2.0 (the "License"); * you may not use this file except in compliance with the License. * You may obtain a copy of the License at: diff --git a/src/plugins/gtpu/gtpu_test.c b/src/plugins/gtpu/gtpu_test.c index 91c5c616..e7fd0d54 100644 --- a/src/plugins/gtpu/gtpu_test.c +++ b/src/plugins/gtpu/gtpu_test.c @@ -1,5 +1,5 @@ /* - * Copyright (c) 2016 Intel and/or its affiliates. + * Copyright (c) 2017 Intel and/or its affiliates. * Licensed under the Apache License, Version 2.0 (the "License"); * you may not use this file except in compliance with the License. * You may obtain a copy of the License at: -- cgit 1.2.3-korg From c847368bc4ec4969241d5cc65bf085bcf0514059 Mon Sep 17 00:00:00 2001 From: Hongjun Ni Date: Thu, 24 Aug 2017 22:53:42 +0800 Subject: Use Quad-loop to improve gtpu-encap performance Change-Id: I440fae09e77502392af5cde45dd9c532afa5d976 Signed-off-by: Hongjun Ni --- src/plugins/gtpu/gtpu_encap.c | 243 +++++++++++++++++++++++++++++++++--------- 1 file changed, 194 insertions(+), 49 deletions(-) (limited to 'src/plugins/gtpu/gtpu_encap.c') diff --git a/src/plugins/gtpu/gtpu_encap.c b/src/plugins/gtpu/gtpu_encap.c index adacf4c9..8ad53c53 100644 --- a/src/plugins/gtpu/gtpu_encap.c +++ b/src/plugins/gtpu/gtpu_encap.c @@ -36,8 +36,15 @@ typedef enum { GTPU_ENCAP_N_ERROR, } gtpu_encap_error_t; +#define foreach_gtpu_encap_next \ +_(DROP, "error-drop") \ +_(IP4_LOOKUP, "ip4-lookup") \ +_(IP6_LOOKUP, "ip6-lookup") + typedef enum { GTPU_ENCAP_NEXT_DROP, + GTPU_ENCAP_NEXT_IP4_LOOKUP, + GTPU_ENCAP_NEXT_IP6_LOOKUP, GTPU_ENCAP_N_NEXT, } gtpu_encap_next_t; @@ -76,13 +83,13 @@ gtpu_encap_inline (vlib_main_t * vm, 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; + 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; - u32 next0 = 0, next1 = 0; - vnet_hw_interface_t * hi0, * hi1; - gtpu_tunnel_t * t0 = NULL, * t1 = NULL; + 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; @@ -98,88 +105,129 @@ gtpu_encap_inline (vlib_main_t * vm, vlib_get_next_frame (vm, node, next_index, to_next, n_left_to_next); - while (n_left_from >= 4 && n_left_to_next >= 2) + while (n_left_from >= 8 && n_left_to_next >= 4) { - 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; - gtpu_header_t * gtpu0, * gtpu1; + 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; - u16 new_l0, new_l1; - ip_csum_t sum0, sum1; + 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 * 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); + 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; - from += 2; - to_next += 2; - n_left_to_next -= 2; - n_left_from -= 2; + 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 = >m->tunnels[hi0->dev_instance]; + t1 = >m->tunnels[hi1->dev_instance]; + t2 = >m->tunnels[hi2->dev_instance]; + t3 = >m->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; - - /* Get next node index and adj index from tunnel next_dpo */ - sw_if_index1 = vnet_buffer(b1)->sw_if_index[VLIB_TX]; - hi1 = vnet_get_sup_hw_interface (vnm, sw_if_index1); - t1 = >m->tunnels[hi1->dev_instance]; - /* Note: change to always set next1 if it may be set to drop */ 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 *)(©_dst0[4]); @@ -188,6 +236,12 @@ gtpu_encap_inline (vlib_main_t * vm, copy_dst_last1 = (u32 *)(©_dst1[4]); copy_src_last1 = (u32 *)(©_src1[4]); copy_dst_last1[0] = copy_src_last1[0]; + copy_dst_last2 = (u32 *)(©_dst2[4]); + copy_src_last2 = (u32 *)(©_src2[4]); + copy_dst_last2[0] = copy_src_last2[0]; + copy_dst_last3 = (u32 *)(©_dst3[4]); + copy_src_last3 = (u32 *)(©_src3[4]); + copy_dst_last3[0] = copy_src_last3[0]; /* Fix the IP4 checksum and length */ sum0 = ip4_0->checksum; @@ -204,6 +258,20 @@ gtpu_encap_inline (vlib_main_t * vm, 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); @@ -216,6 +284,16 @@ gtpu_encap_inline (vlib_main_t * vm, - 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); @@ -226,6 +304,14 @@ gtpu_encap_inline (vlib_main_t * vm, new_l1 = clib_host_to_net_u16 (vlib_buffer_length_in_chain(vm, b1) - sizeof (*ip4_1) - sizeof(*udp1)); 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)); + gtpu2->length = new_l2; + gtpu3 = (gtpu_header_t *)(udp1+3); + new_l3 = clib_host_to_net_u16 (vlib_buffer_length_in_chain(vm, b3) + - sizeof (*ip4_3) - sizeof(*udp3)); + gtpu3->length = new_l3; } else /* ipv6 */ { @@ -233,18 +319,30 @@ gtpu_encap_inline (vlib_main_t * vm, 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 = @@ -255,6 +353,14 @@ gtpu_encap_inline (vlib_main_t * vm, 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); @@ -263,6 +369,12 @@ gtpu_encap_inline (vlib_main_t * vm, 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; /* IPv6 UDP checksum is mandatory */ udp0->checksum = ip6_tcp_udp_icmp_compute_checksum(vm, b0, @@ -273,6 +385,14 @@ gtpu_encap_inline (vlib_main_t * vm, 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; /* Fix GTPU length */ gtpu0 = (gtpu_header_t *)(udp0+1); @@ -283,24 +403,38 @@ gtpu_encap_inline (vlib_main_t * vm, new_l1 = clib_host_to_net_u16 (vlib_buffer_length_in_chain(vm, b1) - sizeof (*ip4_1) - sizeof(*udp1)); 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)); + 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)); + gtpu3->length = new_l3; } - pkts_encapsulated += 2; + pkts_encapsulated += 4; 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; + 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_index1 != stats_sw_if_index) || + (sw_if_index2 != stats_sw_if_index) || + (sw_if_index3 != stats_sw_if_index) )) { - stats_n_packets -= 2; - stats_n_bytes -= len0 + len1; - if (sw_if_index0 == sw_if_index1) + 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 @@ -308,8 +442,8 @@ gtpu_encap_inline (vlib_main_t * vm, 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; + stats_n_packets = 4; + stats_n_bytes = len0 + len1 + len2 + len3; } else { @@ -319,6 +453,12 @@ gtpu_encap_inline (vlib_main_t * vm, 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); } } @@ -338,9 +478,10 @@ gtpu_encap_inline (vlib_main_t * vm, tr->teid = t1->teid; } - vlib_validate_buffer_enqueue_x2 (vm, node, next_index, + vlib_validate_buffer_enqueue_x4 (vm, node, next_index, to_next, n_left_to_next, - bi0, bi1, next0, next1); + bi0, bi1, bi2, bi3, + next0, next1, next2, next3); } while (n_left_from > 0 && n_left_to_next > 0) @@ -536,7 +677,9 @@ VLIB_REGISTER_NODE (gtpu4_encap_node) = { .error_strings = gtpu_encap_error_strings, .n_next_nodes = GTPU_ENCAP_N_NEXT, .next_nodes = { - [GTPU_ENCAP_NEXT_DROP] = "error-drop", +#define _(s,n) [GTPU_ENCAP_NEXT_##s] = n, + foreach_gtpu_encap_next +#undef _ }, }; @@ -552,7 +695,9 @@ VLIB_REGISTER_NODE (gtpu6_encap_node) = { .error_strings = gtpu_encap_error_strings, .n_next_nodes = GTPU_ENCAP_N_NEXT, .next_nodes = { - [GTPU_ENCAP_NEXT_DROP] = "error-drop", +#define _(s,n) [GTPU_ENCAP_NEXT_##s] = n, + foreach_gtpu_encap_next +#undef _ }, }; -- cgit 1.2.3-korg