gtpu.c revision 75c72369
1/*
2 *------------------------------------------------------------------
3 * Copyright (c) 2017 Intel and/or its affiliates.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
7 *
8 *     http://www.apache.org/licenses/LICENSE-2.0
9 *
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
15 *------------------------------------------------------------------
16 */
17#include <stdint.h>
18#include <net/if.h>
19#include <sys/ioctl.h>
20#include <inttypes.h>
21
22#include <vlib/vlib.h>
23#include <vlib/unix/unix.h>
24#include <vnet/ethernet/ethernet.h>
25#include <vnet/fib/fib_entry.h>
26#include <vnet/fib/fib_table.h>
27#include <vnet/fib/fib_entry_track.h>
28#include <vnet/mfib/mfib_table.h>
29#include <vnet/adj/adj_mcast.h>
30#include <vnet/dpo/dpo.h>
31#include <vnet/plugin/plugin.h>
32#include <vpp/app/version.h>
33#include <gtpu/gtpu.h>
34
35
36gtpu_main_t gtpu_main;
37
38/* *INDENT-OFF* */
39VNET_FEATURE_INIT (ip4_gtpu_bypass, static) = {
40  .arc_name = "ip4-unicast",
41  .node_name = "ip4-gtpu-bypass",
42  .runs_before = VNET_FEATURES ("ip4-lookup"),
43};
44
45VNET_FEATURE_INIT (ip6_gtpu_bypass, static) = {
46  .arc_name = "ip6-unicast",
47  .node_name = "ip6-gtpu-bypass",
48  .runs_before = VNET_FEATURES ("ip6-lookup"),
49};
50/* *INDENT-on* */
51
52u8 * format_gtpu_encap_trace (u8 * s, va_list * args)
53{
54  CLIB_UNUSED (vlib_main_t * vm) = va_arg (*args, vlib_main_t *);
55  CLIB_UNUSED (vlib_node_t * node) = va_arg (*args, vlib_node_t *);
56  gtpu_encap_trace_t * t
57      = va_arg (*args, gtpu_encap_trace_t *);
58
59  s = format (s, "GTPU encap to gtpu_tunnel%d teid %d",
60	      t->tunnel_index, t->teid);
61  return s;
62}
63
64static u8 *
65format_decap_next (u8 * s, va_list * args)
66{
67  u32 next_index = va_arg (*args, u32);
68
69  switch (next_index)
70    {
71    case GTPU_INPUT_NEXT_DROP:
72      return format (s, "drop");
73    case GTPU_INPUT_NEXT_L2_INPUT:
74      return format (s, "l2");
75    case GTPU_INPUT_NEXT_IP4_INPUT:
76      return format (s, "ip4");
77    case GTPU_INPUT_NEXT_IP6_INPUT:
78      return format (s, "ip6");
79    default:
80      return format (s, "index %d", next_index);
81    }
82  return s;
83}
84
85u8 *
86format_gtpu_tunnel (u8 * s, va_list * args)
87{
88  gtpu_tunnel_t *t = va_arg (*args, gtpu_tunnel_t *);
89  gtpu_main_t *ngm = &gtpu_main;
90
91  s = format (s, "[%d] src %U dst %U teid %d fib-idx %d sw-if-idx %d ",
92	      t - ngm->tunnels,
93	      format_ip46_address, &t->src, IP46_TYPE_ANY,
94	      format_ip46_address, &t->dst, IP46_TYPE_ANY,
95	      t->teid,  t->encap_fib_index, t->sw_if_index);
96
97  s = format (s, "encap-dpo-idx %d ", t->next_dpo.dpoi_index);
98  s = format (s, "decap-next-%U ", format_decap_next, t->decap_next_index);
99
100  if (PREDICT_FALSE (ip46_address_is_multicast (&t->dst)))
101    s = format (s, "mcast-sw-if-idx %d ", t->mcast_sw_if_index);
102
103  return s;
104}
105
106static u8 *
107format_gtpu_name (u8 * s, va_list * args)
108{
109  u32 dev_instance = va_arg (*args, u32);
110  return format (s, "gtpu_tunnel%d", dev_instance);
111}
112
113static clib_error_t *
114gtpu_interface_admin_up_down (vnet_main_t * vnm, u32 hw_if_index, u32 flags)
115{
116  u32 hw_flags = (flags & VNET_SW_INTERFACE_FLAG_ADMIN_UP) ?
117    VNET_HW_INTERFACE_FLAG_LINK_UP : 0;
118  vnet_hw_interface_set_flags (vnm, hw_if_index, hw_flags);
119
120  return /* no error */ 0;
121}
122
123/* *INDENT-OFF* */
124VNET_DEVICE_CLASS (gtpu_device_class,static) = {
125  .name = "GTPU",
126  .format_device_name = format_gtpu_name,
127  .format_tx_trace = format_gtpu_encap_trace,
128  .admin_up_down_function = gtpu_interface_admin_up_down,
129};
130/* *INDENT-ON* */
131
132static u8 *
133format_gtpu_header_with_length (u8 * s, va_list * args)
134{
135  u32 dev_instance = va_arg (*args, u32);
136  s = format (s, "unimplemented dev %u", dev_instance);
137  return s;
138}
139
140/* *INDENT-OFF* */
141VNET_HW_INTERFACE_CLASS (gtpu_hw_class) =
142{
143  .name = "GTPU",
144  .format_header = format_gtpu_header_with_length,
145  .build_rewrite = default_build_rewrite,
146  .flags = VNET_HW_INTERFACE_CLASS_FLAG_P2P,
147};
148/* *INDENT-ON* */
149
150static void
151gtpu_tunnel_restack_dpo (gtpu_tunnel_t * t)
152{
153  dpo_id_t dpo = DPO_INVALID;
154  u32 encap_index = ip46_address_is_ip4 (&t->dst) ?
155    gtpu4_encap_node.index : gtpu6_encap_node.index;
156  fib_forward_chain_type_t forw_type = ip46_address_is_ip4 (&t->dst) ?
157    FIB_FORW_CHAIN_TYPE_UNICAST_IP4 : FIB_FORW_CHAIN_TYPE_UNICAST_IP6;
158
159  fib_entry_contribute_forwarding (t->fib_entry_index, forw_type, &dpo);
160  dpo_stack_from_node (encap_index, &t->next_dpo, &dpo);
161  dpo_reset (&dpo);
162}
163
164static gtpu_tunnel_t *
165gtpu_tunnel_from_fib_node (fib_node_t * node)
166{
167  return ((gtpu_tunnel_t *) (((char *) node) -
168			     STRUCT_OFFSET_OF (gtpu_tunnel_t, node)));
169}
170
171/**
172 * Function definition to backwalk a FIB node -
173 * Here we will restack the new dpo of GTPU DIP to encap node.
174 */
175static fib_node_back_walk_rc_t
176gtpu_tunnel_back_walk (fib_node_t * node, fib_node_back_walk_ctx_t * ctx)
177{
178  gtpu_tunnel_restack_dpo (gtpu_tunnel_from_fib_node (node));
179  return (FIB_NODE_BACK_WALK_CONTINUE);
180}
181
182/**
183 * Function definition to get a FIB node from its index
184 */
185static fib_node_t *
186gtpu_tunnel_fib_node_get (fib_node_index_t index)
187{
188  gtpu_tunnel_t *t;
189  gtpu_main_t *gtm = &gtpu_main;
190
191  t = pool_elt_at_index (gtm->tunnels, index);
192
193  return (&t->node);
194}
195
196/**
197 * Function definition to inform the FIB node that its last lock has gone.
198 */
199static void
200gtpu_tunnel_last_lock_gone (fib_node_t * node)
201{
202  /*
203   * The GTPU tunnel is a root of the graph. As such
204   * it never has children and thus is never locked.
205   */
206  ASSERT (0);
207}
208
209/*
210 * Virtual function table registered by GTPU tunnels
211 * for participation in the FIB object graph.
212 */
213const static fib_node_vft_t gtpu_vft = {
214  .fnv_get = gtpu_tunnel_fib_node_get,
215  .fnv_last_lock = gtpu_tunnel_last_lock_gone,
216  .fnv_back_walk = gtpu_tunnel_back_walk,
217};
218
219
220#define foreach_copy_field                      \
221_(teid)                                          \
222_(mcast_sw_if_index)                            \
223_(encap_fib_index)                              \
224_(decap_next_index)                             \
225_(src)                                          \
226_(dst)
227
228static void
229ip_udp_gtpu_rewrite (gtpu_tunnel_t * t, bool is_ip6)
230{
231  union
232  {
233    ip4_gtpu_header_t *h4;
234    ip6_gtpu_header_t *h6;
235    u8 *rw;
236  } r =
237  {
238  .rw = 0};
239  int len = is_ip6 ? sizeof *r.h6 : sizeof *r.h4;
240
241  vec_validate_aligned (r.rw, len - 1, CLIB_CACHE_LINE_BYTES);
242
243  udp_header_t *udp;
244  gtpu_header_t *gtpu;
245  /* Fixed portion of the (outer) ip header */
246  if (!is_ip6)
247    {
248      ip4_header_t *ip = &r.h4->ip4;
249      udp = &r.h4->udp;
250      gtpu = &r.h4->gtpu;
251      ip->ip_version_and_header_length = 0x45;
252      ip->ttl = 254;
253      ip->protocol = IP_PROTOCOL_UDP;
254
255      ip->src_address = t->src.ip4;
256      ip->dst_address = t->dst.ip4;
257
258      /* we fix up the ip4 header length and checksum after-the-fact */
259      ip->checksum = ip4_header_checksum (ip);
260    }
261  else
262    {
263      ip6_header_t *ip = &r.h6->ip6;
264      udp = &r.h6->udp;
265      gtpu = &r.h6->gtpu;
266      ip->ip_version_traffic_class_and_flow_label =
267	clib_host_to_net_u32 (6 << 28);
268      ip->hop_limit = 255;
269      ip->protocol = IP_PROTOCOL_UDP;
270
271      ip->src_address = t->src.ip6;
272      ip->dst_address = t->dst.ip6;
273    }
274
275  /* UDP header, randomize src port on something, maybe? */
276  udp->src_port = clib_host_to_net_u16 (2152);
277  udp->dst_port = clib_host_to_net_u16 (UDP_DST_PORT_GTPU);
278
279  /* GTPU header */
280  gtpu->ver_flags = GTPU_V1_VER | GTPU_PT_GTP;
281  gtpu->type = GTPU_TYPE_GTPU;
282  gtpu->teid = clib_host_to_net_u32 (t->teid);
283
284  t->rewrite = r.rw;
285  /* Now only support 8-byte gtpu header. TBD */
286  _vec_len (t->rewrite) = sizeof (ip4_gtpu_header_t) - 4;
287
288  return;
289}
290
291static bool
292gtpu_decap_next_is_valid (gtpu_main_t * gtm, u32 is_ip6, u32 decap_next_index)
293{
294  vlib_main_t *vm = gtm->vlib_main;
295  u32 input_idx = (!is_ip6) ? gtpu4_input_node.index : gtpu6_input_node.index;
296  vlib_node_runtime_t *r = vlib_node_get_runtime (vm, input_idx);
297
298  return decap_next_index < r->n_next_nodes;
299}
300
301static uword
302vtep_addr_ref (ip46_address_t * ip)
303{
304  uword *vtep = ip46_address_is_ip4 (ip) ?
305    hash_get (gtpu_main.vtep4, ip->ip4.as_u32) :
306    hash_get_mem (gtpu_main.vtep6, &ip->ip6);
307  if (vtep)
308    return ++(*vtep);
309  ip46_address_is_ip4 (ip) ?
310    hash_set (gtpu_main.vtep4, ip->ip4.as_u32, 1) :
311    hash_set_mem_alloc (&gtpu_main.vtep6, &ip->ip6, 1);
312  return 1;
313}
314
315static uword
316vtep_addr_unref (ip46_address_t * ip)
317{
318  uword *vtep = ip46_address_is_ip4 (ip) ?
319    hash_get (gtpu_main.vtep4, ip->ip4.as_u32) :
320    hash_get_mem (gtpu_main.vtep6, &ip->ip6);
321  ASSERT (vtep);
322  if (--(*vtep) != 0)
323    return *vtep;
324  ip46_address_is_ip4 (ip) ?
325    hash_unset (gtpu_main.vtep4, ip->ip4.as_u32) :
326    hash_unset_mem_free (&gtpu_main.vtep6, &ip->ip6);
327  return 0;
328}
329
330typedef CLIB_PACKED (union
331		     {
332		     struct
333		     {
334		     fib_node_index_t mfib_entry_index;
335		     adj_index_t mcast_adj_index;
336		     }; u64 as_u64;
337		     }) mcast_shared_t;
338
339static inline mcast_shared_t
340mcast_shared_get (ip46_address_t * ip)
341{
342  ASSERT (ip46_address_is_multicast (ip));
343  uword *p = hash_get_mem (gtpu_main.mcast_shared, ip);
344  ASSERT (p);
345  return (mcast_shared_t)
346  {
347  .as_u64 = *p};
348}
349
350static inline void
351mcast_shared_add (ip46_address_t * dst, fib_node_index_t mfei, adj_index_t ai)
352{
353  mcast_shared_t new_ep = {
354    .mcast_adj_index = ai,
355    .mfib_entry_index = mfei,
356  };
357
358  hash_set_mem_alloc (&gtpu_main.mcast_shared, dst, new_ep.as_u64);
359}
360
361static inline void
362mcast_shared_remove (ip46_address_t * dst)
363{
364  mcast_shared_t ep = mcast_shared_get (dst);
365
366  adj_unlock (ep.mcast_adj_index);
367  mfib_table_entry_delete_index (ep.mfib_entry_index, MFIB_SOURCE_GTPU);
368
369  hash_unset_mem_free (&gtpu_main.mcast_shared, dst);
370}
371
372int vnet_gtpu_add_del_tunnel
373  (vnet_gtpu_add_del_tunnel_args_t * a, u32 * sw_if_indexp)
374{
375  gtpu_main_t *gtm = &gtpu_main;
376  gtpu_tunnel_t *t = 0;
377  vnet_main_t *vnm = gtm->vnet_main;
378  uword *p;
379  u32 hw_if_index = ~0;
380  u32 sw_if_index = ~0;
381  gtpu4_tunnel_key_t key4;
382  gtpu6_tunnel_key_t key6;
383  bool is_ip6 = !ip46_address_is_ip4 (&a->dst);
384
385  if (!is_ip6)
386    {
387      key4.src = a->dst.ip4.as_u32;	/* decap src in key is encap dst in config */
388      key4.teid = clib_host_to_net_u32 (a->teid);
389      p = hash_get (gtm->gtpu4_tunnel_by_key, key4.as_u64);
390    }
391  else
392    {
393      key6.src = a->dst.ip6;
394      key6.teid = clib_host_to_net_u32 (a->teid);
395      p = hash_get_mem (gtm->gtpu6_tunnel_by_key, &key6);
396    }
397
398  if (a->is_add)
399    {
400      l2input_main_t *l2im = &l2input_main;
401
402      /* adding a tunnel: tunnel must not already exist */
403      if (p)
404	return VNET_API_ERROR_TUNNEL_EXIST;
405
406      /*if not set explicitly, default to l2 */
407      if (a->decap_next_index == ~0)
408	a->decap_next_index = GTPU_INPUT_NEXT_L2_INPUT;
409      if (!gtpu_decap_next_is_valid (gtm, is_ip6, a->decap_next_index))
410	return VNET_API_ERROR_INVALID_DECAP_NEXT;
411
412      pool_get_aligned (gtm->tunnels, t, CLIB_CACHE_LINE_BYTES);
413      clib_memset (t, 0, sizeof (*t));
414
415      /* copy from arg structure */
416#define _(x) t->x = a->x;
417      foreach_copy_field;
418#undef _
419
420      ip_udp_gtpu_rewrite (t, is_ip6);
421
422      /* copy the key */
423      if (is_ip6)
424	hash_set_mem_alloc (&gtm->gtpu6_tunnel_by_key, &key6,
425			    t - gtm->tunnels);
426      else
427	hash_set (gtm->gtpu4_tunnel_by_key, key4.as_u64, t - gtm->tunnels);
428
429      vnet_hw_interface_t *hi;
430      if (vec_len (gtm->free_gtpu_tunnel_hw_if_indices) > 0)
431	{
432	  vnet_interface_main_t *im = &vnm->interface_main;
433	  hw_if_index = gtm->free_gtpu_tunnel_hw_if_indices
434	    [vec_len (gtm->free_gtpu_tunnel_hw_if_indices) - 1];
435	  _vec_len (gtm->free_gtpu_tunnel_hw_if_indices) -= 1;
436
437	  hi = vnet_get_hw_interface (vnm, hw_if_index);
438	  hi->dev_instance = t - gtm->tunnels;
439	  hi->hw_instance = hi->dev_instance;
440
441	  /* clear old stats of freed tunnel before reuse */
442	  sw_if_index = hi->sw_if_index;
443	  vnet_interface_counter_lock (im);
444	  vlib_zero_combined_counter
445	    (&im->combined_sw_if_counters[VNET_INTERFACE_COUNTER_TX],
446	     sw_if_index);
447	  vlib_zero_combined_counter (&im->combined_sw_if_counters
448				      [VNET_INTERFACE_COUNTER_RX],
449				      sw_if_index);
450	  vlib_zero_simple_counter (&im->sw_if_counters
451				    [VNET_INTERFACE_COUNTER_DROP],
452				    sw_if_index);
453	  vnet_interface_counter_unlock (im);
454	}
455      else
456	{
457	  hw_if_index = vnet_register_interface
458	    (vnm, gtpu_device_class.index, t - gtm->tunnels,
459	     gtpu_hw_class.index, t - gtm->tunnels);
460	  hi = vnet_get_hw_interface (vnm, hw_if_index);
461	}
462
463      /* Set gtpu tunnel output node */
464      u32 encap_index = !is_ip6 ?
465	gtpu4_encap_node.index : gtpu6_encap_node.index;
466      vnet_set_interface_output_node (vnm, hw_if_index, encap_index);
467
468      t->hw_if_index = hw_if_index;
469      t->sw_if_index = sw_if_index = hi->sw_if_index;
470
471      vec_validate_init_empty (gtm->tunnel_index_by_sw_if_index, sw_if_index,
472			       ~0);
473      gtm->tunnel_index_by_sw_if_index[sw_if_index] = t - gtm->tunnels;
474
475      /* setup l2 input config with l2 feature and bd 0 to drop packet */
476      vec_validate (l2im->configs, sw_if_index);
477      l2im->configs[sw_if_index].feature_bitmap = L2INPUT_FEAT_DROP;
478      l2im->configs[sw_if_index].bd_index = 0;
479
480      vnet_sw_interface_t *si = vnet_get_sw_interface (vnm, sw_if_index);
481      si->flags &= ~VNET_SW_INTERFACE_FLAG_HIDDEN;
482      vnet_sw_interface_set_flags (vnm, sw_if_index,
483				   VNET_SW_INTERFACE_FLAG_ADMIN_UP);
484
485      fib_node_init (&t->node, gtm->fib_node_type);
486      fib_prefix_t tun_dst_pfx;
487      vnet_flood_class_t flood_class = VNET_FLOOD_CLASS_TUNNEL_NORMAL;
488
489      fib_prefix_from_ip46_addr (&t->dst, &tun_dst_pfx);
490      if (!ip46_address_is_multicast (&t->dst))
491	{
492	  /* Unicast tunnel -
493	   * Track the FIB entry for the tunnel's destination.
494	   * The tunnel will then get poked
495	   * when the forwarding for the entry updates, and the tunnel can
496	   * re-stack accordingly
497	   */
498	  vtep_addr_ref (&t->src);
499	  t->fib_entry_index = fib_entry_track (t->encap_fib_index,
500						&tun_dst_pfx,
501						gtm->fib_node_type,
502						t - gtm->tunnels,
503						&t->sibling_index);
504	  gtpu_tunnel_restack_dpo (t);
505	}
506      else
507	{
508	  /* Multicast tunnel -
509	   * as the same mcast group can be used for multiple mcast tunnels
510	   * with different VNIs, create the output adjacency only if
511	   * it does not already exist
512	   */
513	  fib_protocol_t fp = fib_ip_proto (is_ip6);
514
515	  if (vtep_addr_ref (&t->dst) == 1)
516	    {
517	      fib_node_index_t mfei;
518	      adj_index_t ai;
519	      fib_route_path_t path = {
520		.frp_proto = fib_proto_to_dpo (fp),
521		.frp_addr = zero_addr,
522		.frp_sw_if_index = 0xffffffff,
523		.frp_fib_index = ~0,
524		.frp_weight = 1,
525		.frp_flags = FIB_ROUTE_PATH_LOCAL,
526		.frp_mitf_flags = MFIB_ITF_FLAG_FORWARD,
527	      };
528	      const mfib_prefix_t mpfx = {
529		.fp_proto = fp,
530		.fp_len = (is_ip6 ? 128 : 32),
531		.fp_grp_addr = tun_dst_pfx.fp_addr,
532	      };
533
534	      /*
535	       * Setup the (*,G) to receive traffic on the mcast group
536	       *  - the forwarding interface is for-us
537	       *  - the accepting interface is that from the API
538	       */
539	      mfib_table_entry_path_update (t->encap_fib_index,
540					    &mpfx, MFIB_SOURCE_GTPU, &path);
541
542	      path.frp_sw_if_index = a->mcast_sw_if_index;
543	      path.frp_flags = FIB_ROUTE_PATH_FLAG_NONE;
544	      path.frp_mitf_flags = MFIB_ITF_FLAG_ACCEPT;
545	      mfei = mfib_table_entry_path_update (t->encap_fib_index,
546						   &mpfx,
547						   MFIB_SOURCE_GTPU, &path);
548
549	      /*
550	       * Create the mcast adjacency to send traffic to the group
551	       */
552	      ai = adj_mcast_add_or_lock (fp,
553					  fib_proto_to_link (fp),
554					  a->mcast_sw_if_index);
555
556	      /*
557	       * create a new end-point
558	       */
559	      mcast_shared_add (&t->dst, mfei, ai);
560	    }
561
562	  dpo_id_t dpo = DPO_INVALID;
563	  mcast_shared_t ep = mcast_shared_get (&t->dst);
564
565	  /* Stack shared mcast dst mac addr rewrite on encap */
566	  dpo_set (&dpo, DPO_ADJACENCY_MCAST,
567		   fib_proto_to_dpo (fp), ep.mcast_adj_index);
568
569	  dpo_stack_from_node (encap_index, &t->next_dpo, &dpo);
570
571	  dpo_reset (&dpo);
572	  flood_class = VNET_FLOOD_CLASS_TUNNEL_MASTER;
573	}
574
575      vnet_get_sw_interface (vnet_get_main (), sw_if_index)->flood_class =
576	flood_class;
577    }
578  else
579    {
580      /* deleting a tunnel: tunnel must exist */
581      if (!p)
582	return VNET_API_ERROR_NO_SUCH_ENTRY;
583
584      t = pool_elt_at_index (gtm->tunnels, p[0]);
585      sw_if_index = t->sw_if_index;
586
587      vnet_sw_interface_set_flags (vnm, t->sw_if_index, 0 /* down */ );
588      vnet_sw_interface_t *si = vnet_get_sw_interface (vnm, t->sw_if_index);
589      si->flags |= VNET_SW_INTERFACE_FLAG_HIDDEN;
590
591      /* make sure tunnel is removed from l2 bd or xconnect */
592      set_int_l2_mode (gtm->vlib_main, vnm, MODE_L3, t->sw_if_index, 0,
593		       L2_BD_PORT_TYPE_NORMAL, 0, 0);
594      vec_add1 (gtm->free_gtpu_tunnel_hw_if_indices, t->hw_if_index);
595
596      gtm->tunnel_index_by_sw_if_index[t->sw_if_index] = ~0;
597
598      if (!is_ip6)
599	hash_unset (gtm->gtpu4_tunnel_by_key, key4.as_u64);
600      else
601	hash_unset_mem_free (&gtm->gtpu6_tunnel_by_key, &key6);
602
603      if (!ip46_address_is_multicast (&t->dst))
604	{
605	  vtep_addr_unref (&t->src);
606	  fib_entry_untrack (t->fib_entry_index, t->sibling_index);
607	}
608      else if (vtep_addr_unref (&t->dst) == 0)
609	{
610	  mcast_shared_remove (&t->dst);
611	}
612
613      fib_node_deinit (&t->node);
614      vec_free (t->rewrite);
615      pool_put (gtm->tunnels, t);
616    }
617
618  if (sw_if_indexp)
619    *sw_if_indexp = sw_if_index;
620
621  if (a->is_add)
622    {
623      /* register udp ports */
624      if (!is_ip6 && !udp_is_valid_dst_port (UDP_DST_PORT_GTPU, 1))
625	udp_register_dst_port (gtm->vlib_main, UDP_DST_PORT_GTPU,
626			       gtpu4_input_node.index, /* is_ip4 */ 1);
627      if (is_ip6 && !udp_is_valid_dst_port (UDP_DST_PORT_GTPU6, 0))
628	udp_register_dst_port (gtm->vlib_main, UDP_DST_PORT_GTPU6,
629			       gtpu6_input_node.index, /* is_ip4 */ 0);
630    }
631
632  return 0;
633}
634
635static uword
636get_decap_next_for_node (u32 node_index, u32 ipv4_set)
637{
638  gtpu_main_t *gtm = &gtpu_main;
639  vlib_main_t *vm = gtm->vlib_main;
640  uword input_node = (ipv4_set) ? gtpu4_input_node.index :
641    gtpu6_input_node.index;
642
643  return vlib_node_add_next (vm, input_node, node_index);
644}
645
646static uword
647unformat_decap_next (unformat_input_t * input, va_list * args)
648{
649  u32 *result = va_arg (*args, u32 *);
650  u32 ipv4_set = va_arg (*args, int);
651  gtpu_main_t *gtm = &gtpu_main;
652  vlib_main_t *vm = gtm->vlib_main;
653  u32 node_index;
654  u32 tmp;
655
656  if (unformat (input, "l2"))
657    *result = GTPU_INPUT_NEXT_L2_INPUT;
658  else if (unformat (input, "ip4"))
659    *result = GTPU_INPUT_NEXT_IP4_INPUT;
660  else if (unformat (input, "ip6"))
661    *result = GTPU_INPUT_NEXT_IP6_INPUT;
662  else if (unformat (input, "node %U", unformat_vlib_node, vm, &node_index))
663    *result = get_decap_next_for_node (node_index, ipv4_set);
664  else if (unformat (input, "%d", &tmp))
665    *result = tmp;
666  else
667    return 0;
668
669  return 1;
670}
671
672static clib_error_t *
673gtpu_add_del_tunnel_command_fn (vlib_main_t * vm,
674				unformat_input_t * input,
675				vlib_cli_command_t * cmd)
676{
677  unformat_input_t _line_input, *line_input = &_line_input;
678  ip46_address_t src, dst;
679  u8 is_add = 1;
680  u8 src_set = 0;
681  u8 dst_set = 0;
682  u8 grp_set = 0;
683  u8 ipv4_set = 0;
684  u8 ipv6_set = 0;
685  u32 encap_fib_index = 0;
686  u32 mcast_sw_if_index = ~0;
687  u32 decap_next_index = GTPU_INPUT_NEXT_L2_INPUT;
688  u32 teid = 0;
689  u32 tmp;
690  int rv;
691  vnet_gtpu_add_del_tunnel_args_t _a, *a = &_a;
692  u32 tunnel_sw_if_index;
693  clib_error_t *error = NULL;
694
695  /* Cant "universally zero init" (={0}) due to GCC bug 53119 */
696  clib_memset (&src, 0, sizeof src);
697  clib_memset (&dst, 0, sizeof dst);
698
699  /* Get a line of input. */
700  if (!unformat_user (input, unformat_line_input, line_input))
701    return 0;
702
703  while (unformat_check_input (line_input) != UNFORMAT_END_OF_INPUT)
704    {
705      if (unformat (line_input, "del"))
706	{
707	  is_add = 0;
708	}
709      else if (unformat (line_input, "src %U",
710			 unformat_ip4_address, &src.ip4))
711	{
712	  src_set = 1;
713	  ipv4_set = 1;
714	}
715      else if (unformat (line_input, "dst %U",
716			 unformat_ip4_address, &dst.ip4))
717	{
718	  dst_set = 1;
719	  ipv4_set = 1;
720	}
721      else if (unformat (line_input, "src %U",
722			 unformat_ip6_address, &src.ip6))
723	{
724	  src_set = 1;
725	  ipv6_set = 1;
726	}
727      else if (unformat (line_input, "dst %U",
728			 unformat_ip6_address, &dst.ip6))
729	{
730	  dst_set = 1;
731	  ipv6_set = 1;
732	}
733      else if (unformat (line_input, "group %U %U",
734			 unformat_ip4_address, &dst.ip4,
735			 unformat_vnet_sw_interface,
736			 vnet_get_main (), &mcast_sw_if_index))
737	{
738	  grp_set = dst_set = 1;
739	  ipv4_set = 1;
740	}
741      else if (unformat (line_input, "group %U %U",
742			 unformat_ip6_address, &dst.ip6,
743			 unformat_vnet_sw_interface,
744			 vnet_get_main (), &mcast_sw_if_index))
745	{
746	  grp_set = dst_set = 1;
747	  ipv6_set = 1;
748	}
749      else if (unformat (line_input, "encap-vrf-id %d", &tmp))
750	{
751	  encap_fib_index = fib_table_find (fib_ip_proto (ipv6_set), tmp);
752	  if (encap_fib_index == ~0)
753	    {
754	      error =
755		clib_error_return (0, "nonexistent encap-vrf-id %d", tmp);
756	      goto done;
757	    }
758	}
759      else if (unformat (line_input, "decap-next %U", unformat_decap_next,
760			 &decap_next_index, ipv4_set))
761	;
762      else if (unformat (line_input, "teid %d", &teid))
763	;
764      else
765	{
766	  error = clib_error_return (0, "parse error: '%U'",
767				     format_unformat_error, line_input);
768	  goto done;
769	}
770    }
771
772  if (src_set == 0)
773    {
774      error = clib_error_return (0, "tunnel src address not specified");
775      goto done;
776    }
777
778  if (dst_set == 0)
779    {
780      error = clib_error_return (0, "tunnel dst address not specified");
781      goto done;
782    }
783
784  if (grp_set && !ip46_address_is_multicast (&dst))
785    {
786      error = clib_error_return (0, "tunnel group address not multicast");
787      goto done;
788    }
789
790  if (grp_set == 0 && ip46_address_is_multicast (&dst))
791    {
792      error = clib_error_return (0, "dst address must be unicast");
793      goto done;
794    }
795
796  if (grp_set && mcast_sw_if_index == ~0)
797    {
798      error = clib_error_return (0, "tunnel nonexistent multicast device");
799      goto done;
800    }
801
802  if (ipv4_set && ipv6_set)
803    {
804      error = clib_error_return (0, "both IPv4 and IPv6 addresses specified");
805      goto done;
806    }
807
808  if (ip46_address_cmp (&src, &dst) == 0)
809    {
810      error = clib_error_return (0, "src and dst addresses are identical");
811      goto done;
812    }
813
814  if (decap_next_index == ~0)
815    {
816      error = clib_error_return (0, "next node not found");
817      goto done;
818    }
819
820  clib_memset (a, 0, sizeof (*a));
821
822  a->is_add = is_add;
823
824#define _(x) a->x = x;
825  foreach_copy_field;
826#undef _
827
828  rv = vnet_gtpu_add_del_tunnel (a, &tunnel_sw_if_index);
829
830  switch (rv)
831    {
832    case 0:
833      if (is_add)
834	vlib_cli_output (vm, "%U\n", format_vnet_sw_if_index_name,
835			 vnet_get_main (), tunnel_sw_if_index);
836      break;
837
838    case VNET_API_ERROR_TUNNEL_EXIST:
839      error = clib_error_return (0, "tunnel already exists...");
840      goto done;
841
842    case VNET_API_ERROR_NO_SUCH_ENTRY:
843      error = clib_error_return (0, "tunnel does not exist...");
844      goto done;
845
846    default:
847      error = clib_error_return
848	(0, "vnet_gtpu_add_del_tunnel returned %d", rv);
849      goto done;
850    }
851
852done:
853  unformat_free (line_input);
854
855  return error;
856}
857
858/*?
859 * Add or delete a GTPU Tunnel.
860 *
861 * GTPU provides the features needed to allow L2 bridge domains (BDs)
862 * to span multiple servers. This is done by building an L2 overlay on
863 * top of an L3 network underlay using GTPU tunnels.
864 *
865 * This makes it possible for servers to be co-located in the same data
866 * center or be separated geographically as long as they are reachable
867 * through the underlay L3 network.
868 *
869 * You can refer to this kind of L2 overlay bridge domain as a GTPU
870 * (Virtual eXtensible VLAN) segment.
871 *
872 * @cliexpar
873 * Example of how to create a GTPU Tunnel:
874 * @cliexcmd{create gtpu tunnel src 10.0.3.1 dst 10.0.3.3 teid 13 encap-vrf-id 7}
875 * Example of how to delete a GTPU Tunnel:
876 * @cliexcmd{create gtpu tunnel src 10.0.3.1 dst 10.0.3.3 teid 13 del}
877 ?*/
878/* *INDENT-OFF* */
879VLIB_CLI_COMMAND (create_gtpu_tunnel_command, static) = {
880  .path = "create gtpu tunnel",
881  .short_help =
882  "create gtpu tunnel src <local-vtep-addr>"
883  " {dst <remote-vtep-addr>|group <mcast-vtep-addr> <intf-name>} teid <nn>"
884  " [encap-vrf-id <nn>] [decap-next [l2|ip4|ip6|node <name>]] [del]",
885  .function = gtpu_add_del_tunnel_command_fn,
886};
887/* *INDENT-ON* */
888
889static clib_error_t *
890show_gtpu_tunnel_command_fn (vlib_main_t * vm,
891			     unformat_input_t * input,
892			     vlib_cli_command_t * cmd)
893{
894  gtpu_main_t *gtm = &gtpu_main;
895  gtpu_tunnel_t *t;
896
897  if (pool_elts (gtm->tunnels) == 0)
898    vlib_cli_output (vm, "No gtpu tunnels configured...");
899
900  pool_foreach (t, gtm->tunnels, (
901				   {
902				   vlib_cli_output (vm, "%U",
903						    format_gtpu_tunnel, t);
904				   }
905		));
906
907  return 0;
908}
909
910/*?
911 * Display all the GTPU Tunnel entries.
912 *
913 * @cliexpar
914 * Example of how to display the GTPU Tunnel entries:
915 * @cliexstart{show gtpu tunnel}
916 * [0] src 10.0.3.1 dst 10.0.3.3 teid 13 encap_fib_index 0 sw_if_index 5 decap_next l2
917 * @cliexend
918 ?*/
919/* *INDENT-OFF* */
920VLIB_CLI_COMMAND (show_gtpu_tunnel_command, static) = {
921    .path = "show gtpu tunnel",
922    .short_help = "show gtpu tunnel",
923    .function = show_gtpu_tunnel_command_fn,
924};
925/* *INDENT-ON* */
926
927void
928vnet_int_gtpu_bypass_mode (u32 sw_if_index, u8 is_ip6, u8 is_enable)
929{
930  if (is_ip6)
931    vnet_feature_enable_disable ("ip6-unicast", "ip6-gtpu-bypass",
932				 sw_if_index, is_enable, 0, 0);
933  else
934    vnet_feature_enable_disable ("ip4-unicast", "ip4-gtpu-bypass",
935				 sw_if_index, is_enable, 0, 0);
936}
937
938static clib_error_t *
939set_ip_gtpu_bypass (u32 is_ip6,
940		    unformat_input_t * input, vlib_cli_command_t * cmd)
941{
942  unformat_input_t _line_input, *line_input = &_line_input;
943  vnet_main_t *vnm = vnet_get_main ();
944  clib_error_t *error = 0;
945  u32 sw_if_index, is_enable;
946
947  sw_if_index = ~0;
948  is_enable = 1;
949
950  if (!unformat_user (input, unformat_line_input, line_input))
951    return 0;
952
953  while (unformat_check_input (line_input) != UNFORMAT_END_OF_INPUT)
954    {
955      if (unformat_user
956	  (line_input, unformat_vnet_sw_interface, vnm, &sw_if_index))
957	;
958      else if (unformat (line_input, "del"))
959	is_enable = 0;
960      else
961	{
962	  error = unformat_parse_error (line_input);
963	  goto done;
964	}
965    }
966
967  if (~0 == sw_if_index)
968    {
969      error = clib_error_return (0, "unknown interface `%U'",
970				 format_unformat_error, line_input);
971      goto done;
972    }
973
974  vnet_int_gtpu_bypass_mode (sw_if_index, is_ip6, is_enable);
975
976done:
977  unformat_free (line_input);
978
979  return error;
980}
981
982static clib_error_t *
983set_ip4_gtpu_bypass (vlib_main_t * vm,
984		     unformat_input_t * input, vlib_cli_command_t * cmd)
985{
986  return set_ip_gtpu_bypass (0, input, cmd);
987}
988
989/*?
990 * This command adds the 'ip4-gtpu-bypass' graph node for a given interface.
991 * By adding the IPv4 gtpu-bypass graph node to an interface, the node checks
992 *  for and validate input gtpu packet and bypass ip4-lookup, ip4-local,
993 * ip4-udp-lookup nodes to speedup gtpu packet forwarding. This node will
994 * cause extra overhead to for non-gtpu packets which is kept at a minimum.
995 *
996 * @cliexpar
997 * @parblock
998 * Example of graph node before ip4-gtpu-bypass is enabled:
999 * @cliexstart{show vlib graph ip4-gtpu-bypass}
1000 *            Name                      Next                    Previous
1001 * ip4-gtpu-bypass                error-drop [0]
1002 *                                gtpu4-input [1]
1003 *                                 ip4-lookup [2]
1004 * @cliexend
1005 *
1006 * Example of how to enable ip4-gtpu-bypass on an interface:
1007 * @cliexcmd{set interface ip gtpu-bypass GigabitEthernet2/0/0}
1008 *
1009 * Example of graph node after ip4-gtpu-bypass is enabled:
1010 * @cliexstart{show vlib graph ip4-gtpu-bypass}
1011 *            Name                      Next                    Previous
1012 * ip4-gtpu-bypass                error-drop [0]               ip4-input
1013 *                                gtpu4-input [1]        ip4-input-no-checksum
1014 *                                 ip4-lookup [2]
1015 * @cliexend
1016 *
1017 * Example of how to display the feature enabled on an interface:
1018 * @cliexstart{show ip interface features GigabitEthernet2/0/0}
1019 * IP feature paths configured on GigabitEthernet2/0/0...
1020 * ...
1021 * ipv4 unicast:
1022 *   ip4-gtpu-bypass
1023 *   ip4-lookup
1024 * ...
1025 * @cliexend
1026 *
1027 * Example of how to disable ip4-gtpu-bypass on an interface:
1028 * @cliexcmd{set interface ip gtpu-bypass GigabitEthernet2/0/0 del}
1029 * @endparblock
1030?*/
1031/* *INDENT-OFF* */
1032VLIB_CLI_COMMAND (set_interface_ip_gtpu_bypass_command, static) = {
1033  .path = "set interface ip gtpu-bypass",
1034  .function = set_ip4_gtpu_bypass,
1035  .short_help = "set interface ip gtpu-bypass <interface> [del]",
1036};
1037/* *INDENT-ON* */
1038
1039static clib_error_t *
1040set_ip6_gtpu_bypass (vlib_main_t * vm,
1041		     unformat_input_t * input, vlib_cli_command_t * cmd)
1042{
1043  return set_ip_gtpu_bypass (1, input, cmd);
1044}
1045
1046/*?
1047 * This command adds the 'ip6-gtpu-bypass' graph node for a given interface.
1048 * By adding the IPv6 gtpu-bypass graph node to an interface, the node checks
1049 *  for and validate input gtpu packet and bypass ip6-lookup, ip6-local,
1050 * ip6-udp-lookup nodes to speedup gtpu packet forwarding. This node will
1051 * cause extra overhead to for non-gtpu packets which is kept at a minimum.
1052 *
1053 * @cliexpar
1054 * @parblock
1055 * Example of graph node before ip6-gtpu-bypass is enabled:
1056 * @cliexstart{show vlib graph ip6-gtpu-bypass}
1057 *            Name                      Next                    Previous
1058 * ip6-gtpu-bypass                error-drop [0]
1059 *                                gtpu6-input [1]
1060 *                                 ip6-lookup [2]
1061 * @cliexend
1062 *
1063 * Example of how to enable ip6-gtpu-bypass on an interface:
1064 * @cliexcmd{set interface ip6 gtpu-bypass GigabitEthernet2/0/0}
1065 *
1066 * Example of graph node after ip6-gtpu-bypass is enabled:
1067 * @cliexstart{show vlib graph ip6-gtpu-bypass}
1068 *            Name                      Next                    Previous
1069 * ip6-gtpu-bypass                error-drop [0]               ip6-input
1070 *                                gtpu6-input [1]        ip4-input-no-checksum
1071 *                                 ip6-lookup [2]
1072 * @cliexend
1073 *
1074 * Example of how to display the feature enabled on an interface:
1075 * @cliexstart{show ip interface features GigabitEthernet2/0/0}
1076 * IP feature paths configured on GigabitEthernet2/0/0...
1077 * ...
1078 * ipv6 unicast:
1079 *   ip6-gtpu-bypass
1080 *   ip6-lookup
1081 * ...
1082 * @cliexend
1083 *
1084 * Example of how to disable ip6-gtpu-bypass on an interface:
1085 * @cliexcmd{set interface ip6 gtpu-bypass GigabitEthernet2/0/0 del}
1086 * @endparblock
1087?*/
1088/* *INDENT-OFF* */
1089VLIB_CLI_COMMAND (set_interface_ip6_gtpu_bypass_command, static) = {
1090  .path = "set interface ip6 gtpu-bypass",
1091  .function = set_ip6_gtpu_bypass,
1092  .short_help = "set interface ip6 gtpu-bypass <interface> [del]",
1093};
1094/* *INDENT-ON* */
1095
1096clib_error_t *
1097gtpu_init (vlib_main_t * vm)
1098{
1099  gtpu_main_t *gtm = &gtpu_main;
1100
1101  gtm->vnet_main = vnet_get_main ();
1102  gtm->vlib_main = vm;
1103
1104  /* initialize the ip6 hash */
1105  gtm->gtpu6_tunnel_by_key = hash_create_mem (0,
1106					      sizeof (gtpu6_tunnel_key_t),
1107					      sizeof (uword));
1108  gtm->vtep6 = hash_create_mem (0, sizeof (ip6_address_t), sizeof (uword));
1109  gtm->mcast_shared = hash_create_mem (0,
1110				       sizeof (ip46_address_t),
1111				       sizeof (mcast_shared_t));
1112
1113  gtm->fib_node_type = fib_node_register_new_type (&gtpu_vft);
1114
1115  return 0;
1116}
1117
1118VLIB_INIT_FUNCTION (gtpu_init);
1119
1120/* *INDENT-OFF* */
1121VLIB_PLUGIN_REGISTER () = {
1122    .version = VPP_BUILD_VER,
1123    .description = "GPRS Tunnelling Protocol, User Data (GTPv1-U)",
1124};
1125/* *INDENT-ON* */
1126
1127/*
1128 * fd.io coding-style-patch-verification: ON
1129 *
1130 * Local Variables:
1131 * eval: (c-set-style "gnu")
1132 * End:
1133 */
1134