FD.io VPP  v21.01.1
Vector Packet Processing
ip4_sv_reass.c
Go to the documentation of this file.
1 /*
2  * Copyright (c) 2017 Cisco and/or its affiliates.
3  * Licensed under the Apache License, Version 2.0 (the "License");
4  * you may not use this file except in compliance with the License.
5  * You may obtain a copy of the License at:
6  *
7  * http://www.apache.org/licenses/LICENSE-2.0
8  *
9  * Unless required by applicable law or agreed to in writing, software
10  * distributed under the License is distributed on an "AS IS" BASIS,
11  * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12  * See the License for the specific language governing permissions and
13  * limitations under the License.
14  */
15 
16 /**
17  * @file
18  * @brief IPv4 Shallow Virtual Reassembly.
19  *
20  * This file contains the source code for IPv4 Shallow Virtual reassembly.
21  */
22 
23 #include <vppinfra/vec.h>
24 #include <vnet/vnet.h>
25 #include <vnet/ip/ip.h>
26 #include <vnet/ip/ip4_to_ip6.h>
27 #include <vppinfra/fifo.h>
28 #include <vppinfra/bihash_16_8.h>
30 
31 #define MSEC_PER_SEC 1000
32 #define IP4_SV_REASS_TIMEOUT_DEFAULT_MS 100
33 #define IP4_SV_REASS_EXPIRE_WALK_INTERVAL_DEFAULT_MS 10000 // 10 seconds default
34 #define IP4_SV_REASS_MAX_REASSEMBLIES_DEFAULT 1024
35 #define IP4_SV_REASS_MAX_REASSEMBLY_LENGTH_DEFAULT 3
36 #define IP4_SV_REASS_HT_LOAD_FACTOR (0.75)
37 
38 typedef enum
39 {
44 
45 typedef struct
46 {
47  union
48  {
49  struct
50  {
57  };
58  u64 as_u64[2];
59  };
61 
62 typedef union
63 {
64  struct
65  {
68  };
71 
72 typedef union
73 {
74  struct
75  {
78  };
81 
82 typedef struct
83 {
84  // hash table key
86  // time when last packet was received
88  // internal id of this reassembly
90  // trace operation counter
92  // minimum fragment length for this reassembly - used to estimate MTU
94  // buffer indexes of buffers in this reassembly in chronological order -
95  // including overlaps and duplicate fragments
97  // set to true when this reassembly is completed
99  // ip protocol
104  // l4 src port
106  // l4 dst port
109  // lru indexes
113 
114 typedef struct
115 {
120  // lru indexes
123 
125 
126 typedef struct
127 {
128  // IPv4 config
132  // maximum number of fragments in one reassembly
134  // maximum number of reassemblies
136 
137  // IPv4 runtime
138  clib_bihash_16_8_t hash;
139  // per-thread data
141 
142  // convenience
145 
146  // node index of ip4-drop node
149 
150  /** Worker handoff */
153 
154  // reference count for enabling/disabling feature - per interface
156 
157  // reference count for enabling/disabling feature - per interface
159 
161 
163 
164 #ifndef CLIB_MARCH_VARIANT
166 #endif /* CLIB_MARCH_VARIANT */
167 
168 typedef enum
169 {
175 
176 typedef enum
177 {
183 
184 typedef struct
185 {
186  ip4_sv_reass_trace_operation_e action;
193 
196 
197 static u8 *
198 format_ip4_sv_reass_trace (u8 * s, va_list * args)
199 {
200  CLIB_UNUSED (vlib_main_t * vm) = va_arg (*args, vlib_main_t *);
201  CLIB_UNUSED (vlib_node_t * node) = va_arg (*args, vlib_node_t *);
202  ip4_sv_reass_trace_t *t = va_arg (*args, ip4_sv_reass_trace_t *);
203  if (REASS_PASSTHROUGH != t->action)
204  {
205  s = format (s, "reass id: %u, op id: %u ", t->reass_id, t->op_id);
206  }
207  switch (t->action)
208  {
210  s = format (s, "[cached]");
211  break;
212  case REASS_FINISH:
213  s =
214  format (s, "[finish, ip proto=%u, src_port=%u, dst_port=%u]",
215  t->ip_proto, clib_net_to_host_u16 (t->l4_src_port),
216  clib_net_to_host_u16 (t->l4_dst_port));
217  break;
219  s =
220  format (s, "[forward, ip proto=%u, src_port=%u, dst_port=%u]",
221  t->ip_proto, clib_net_to_host_u16 (t->l4_src_port),
222  clib_net_to_host_u16 (t->l4_dst_port));
223  break;
224  case REASS_PASSTHROUGH:
225  s = format (s, "[not-fragmented]");
226  break;
227  }
228  return s;
229 }
230 
231 static void
233  ip4_sv_reass_main_t * rm, ip4_sv_reass_t * reass,
234  u32 bi, ip4_sv_reass_trace_operation_e action,
235  u32 ip_proto, u16 l4_src_port, u16 l4_dst_port)
236 {
237  vlib_buffer_t *b = vlib_get_buffer (vm, bi);
240  {
241  // this buffer's trace is gone
242  b->flags &= ~VLIB_BUFFER_IS_TRACED;
243  return;
244  }
245  ip4_sv_reass_trace_t *t = vlib_add_trace (vm, node, b, sizeof (t[0]));
246  if (reass)
247  {
248  t->reass_id = reass->id;
249  t->op_id = reass->trace_op_counter;
250  ++reass->trace_op_counter;
251  }
252  t->action = action;
253  t->ip_proto = ip_proto;
254  t->l4_src_port = l4_src_port;
255  t->l4_dst_port = l4_dst_port;
256 #if 0
257  static u8 *s = NULL;
258  s = format (s, "%U", format_ip4_sv_reass_trace, NULL, NULL, t);
259  printf ("%.*s\n", vec_len (s), s);
260  fflush (stdout);
261  vec_reset_length (s);
262 #endif
263 }
264 
265 
266 always_inline void
269 {
271  kv.key[0] = reass->key.as_u64[0];
272  kv.key[1] = reass->key.as_u64[1];
273  clib_bihash_add_del_16_8 (&rm->hash, &kv, 0);
274  vlib_buffer_free (vm, reass->cached_buffers,
275  vec_len (reass->cached_buffers));
276  vec_free (reass->cached_buffers);
277  reass->cached_buffers = NULL;
278  if (~0 != reass->lru_prev)
279  {
280  ip4_sv_reass_t *lru_prev =
281  pool_elt_at_index (rt->pool, reass->lru_prev);
282  lru_prev->lru_next = reass->lru_next;
283  }
284  if (~0 != reass->lru_next)
285  {
286  ip4_sv_reass_t *lru_next =
287  pool_elt_at_index (rt->pool, reass->lru_next);
288  lru_next->lru_prev = reass->lru_prev;
289  }
290  if (rt->lru_first == reass - rt->pool)
291  {
292  rt->lru_first = reass->lru_next;
293  }
294  if (rt->lru_last == reass - rt->pool)
295  {
296  rt->lru_last = reass->lru_prev;
297  }
298  pool_put (rt->pool, reass);
299  --rt->reass_n;
300 }
301 
302 always_inline void
304 {
305  reass->cached_buffers = NULL;
306  reass->is_complete = false;
307 }
308 
312  ip4_sv_reass_kv_t * kv, u8 * do_handoff)
313 {
314  ip4_sv_reass_t *reass = NULL;
315  f64 now = vlib_time_now (vm);
316 
317  if (!clib_bihash_search_16_8 (&rm->hash, &kv->kv, &kv->kv))
318  {
319  if (vm->thread_index != kv->v.thread_index)
320  {
321  *do_handoff = 1;
322  return NULL;
323  }
324  reass = pool_elt_at_index (rt->pool, kv->v.reass_index);
325 
326  if (now > reass->last_heard + rm->timeout)
327  {
328  ip4_sv_reass_free (vm, rm, rt, reass);
329  reass = NULL;
330  }
331  }
332 
333  if (reass)
334  {
335  reass->last_heard = now;
336  return reass;
337  }
338 
339  if (rt->reass_n >= rm->max_reass_n && rm->max_reass_n)
340  {
341  reass = pool_elt_at_index (rt->pool, rt->lru_first);
342  ip4_sv_reass_free (vm, rm, rt, reass);
343  }
344 
345  pool_get (rt->pool, reass);
346  clib_memset (reass, 0, sizeof (*reass));
347  reass->id = ((u64) vm->thread_index * 1000000000) + rt->id_counter;
348  ++rt->id_counter;
349  ip4_sv_reass_init (reass);
350  ++rt->reass_n;
351  reass->lru_prev = reass->lru_next = ~0;
352 
353  if (~0 != rt->lru_last)
354  {
355  ip4_sv_reass_t *lru_last = pool_elt_at_index (rt->pool, rt->lru_last);
356  reass->lru_prev = rt->lru_last;
357  lru_last->lru_next = rt->lru_last = reass - rt->pool;
358  }
359 
360  if (~0 == rt->lru_first)
361  {
362  rt->lru_first = rt->lru_last = reass - rt->pool;
363  }
364 
365  reass->key.as_u64[0] = kv->kv.key[0];
366  reass->key.as_u64[1] = kv->kv.key[1];
367  kv->v.reass_index = (reass - rt->pool);
368  kv->v.thread_index = vm->thread_index;
369  reass->last_heard = now;
370 
371  if (clib_bihash_add_del_16_8 (&rm->hash, &kv->kv, 1))
372  {
373  ip4_sv_reass_free (vm, rm, rt, reass);
374  reass = NULL;
375  }
376 
377  return reass;
378 }
379 
380 always_inline ip4_sv_reass_rc_t
383  ip4_header_t * ip0, ip4_sv_reass_t * reass, u32 bi0)
384 {
385  vlib_buffer_t *b0 = vlib_get_buffer (vm, bi0);
386  ip4_sv_reass_rc_t rc = IP4_SV_REASS_RC_OK;
387  const u32 fragment_first = ip4_get_fragment_offset_bytes (ip0);
388  if (0 == fragment_first)
389  {
390  reass->ip_proto = ip0->protocol;
391  reass->l4_src_port = ip4_get_port (ip0, 1);
392  reass->l4_dst_port = ip4_get_port (ip0, 0);
393  if (!reass->l4_src_port || !reass->l4_dst_port)
395  if (IP_PROTOCOL_TCP == reass->ip_proto)
396  {
397  reass->icmp_type_or_tcp_flags = ((tcp_header_t *) (ip0 + 1))->flags;
398  reass->tcp_ack_number = ((tcp_header_t *) (ip0 + 1))->ack_number;
399  reass->tcp_seq_number = ((tcp_header_t *) (ip0 + 1))->seq_number;
400  }
401  else if (IP_PROTOCOL_ICMP == reass->ip_proto)
402  {
403  reass->icmp_type_or_tcp_flags =
404  ((icmp46_header_t *) (ip0 + 1))->type;
405  }
406  reass->is_complete = true;
407  vlib_buffer_t *b0 = vlib_get_buffer (vm, bi0);
408  if (PREDICT_FALSE (b0->flags & VLIB_BUFFER_IS_TRACED))
409  {
410  ip4_sv_reass_add_trace (vm, node, rm, reass, bi0, REASS_FINISH,
411  reass->ip_proto, reass->l4_src_port,
412  reass->l4_dst_port);
413  }
414  }
415  vec_add1 (reass->cached_buffers, bi0);
416  if (!reass->is_complete)
417  {
418  if (PREDICT_FALSE (b0->flags & VLIB_BUFFER_IS_TRACED))
419  {
420  ip4_sv_reass_add_trace (vm, node, rm, reass, bi0,
421  REASS_FRAGMENT_CACHE, ~0, ~0, ~0);
422  }
423  if (vec_len (reass->cached_buffers) > rm->max_reass_len)
424  {
426  }
427  }
428  return rc;
429 }
430 
433  vlib_frame_t * frame, bool is_feature,
434  bool is_output_feature, bool is_custom)
435 {
436  u32 *from = vlib_frame_vector_args (frame);
437  u32 n_left_from, n_left_to_next, *to_next, next_index;
440  clib_spinlock_lock (&rt->lock);
441 
442  n_left_from = frame->n_vectors;
443  next_index = node->cached_next_index;
444 
445  vlib_buffer_t *bufs[VLIB_FRAME_SIZE], **b;
446  vlib_get_buffers (vm, from, bufs, n_left_from);
447  u16 nexts[VLIB_FRAME_SIZE], *next = nexts;
448  b = bufs;
449 
450  /* optimistic case first - no fragments */
451  while (n_left_from >= 2)
452  {
453  vlib_buffer_t *b0, *b1;
454  u32 next0, next1;
455  b0 = *b;
456  b++;
457  b1 = *b;
458  b++;
459 
460  /* Prefetch next iteration. */
461  if (PREDICT_TRUE (n_left_from >= 4))
462  {
463  vlib_buffer_t *p2, *p3;
464 
465  p2 = *b;
466  p3 = *(b + 1);
467 
468  vlib_prefetch_buffer_header (p2, LOAD);
469  vlib_prefetch_buffer_header (p3, LOAD);
470 
473  }
474 
475  ip4_header_t *ip0 =
477  (is_output_feature ? 1 : 0) *
478  vnet_buffer (b0)->
479  ip.save_rewrite_length);
480  ip4_header_t *ip1 =
482  (is_output_feature ? 1 : 0) *
483  vnet_buffer (b1)->
484  ip.save_rewrite_length);
485  if (PREDICT_FALSE
488  {
489  // fragment found, go slow path
490  b -= 2;
491  if (b - bufs > 0)
492  {
493  vlib_buffer_enqueue_to_next (vm, node, from, (u16 *) nexts,
494  b - bufs);
495  }
496  goto slow_path;
497  }
498  if (is_feature)
499  {
500  vnet_feature_next (&next0, b0);
501  }
502  else
503  {
504  next0 = is_custom ? vnet_buffer (b0)->ip.reass.next_index :
506  }
507  vnet_buffer (b0)->ip.reass.is_non_first_fragment = 0;
508  vnet_buffer (b0)->ip.reass.ip_proto = ip0->protocol;
509  if (IP_PROTOCOL_TCP == ip0->protocol)
510  {
511  vnet_buffer (b0)->ip.reass.icmp_type_or_tcp_flags =
512  ((tcp_header_t *) (ip0 + 1))->flags;
513  vnet_buffer (b0)->ip.reass.tcp_ack_number =
514  ((tcp_header_t *) (ip0 + 1))->ack_number;
515  vnet_buffer (b0)->ip.reass.tcp_seq_number =
516  ((tcp_header_t *) (ip0 + 1))->seq_number;
517  }
518  else if (IP_PROTOCOL_ICMP == ip0->protocol)
519  {
520  vnet_buffer (b0)->ip.reass.icmp_type_or_tcp_flags =
521  ((icmp46_header_t *) (ip0 + 1))->type;
522  }
523  vnet_buffer (b0)->ip.reass.l4_src_port = ip4_get_port (ip0, 1);
524  vnet_buffer (b0)->ip.reass.l4_dst_port = ip4_get_port (ip0, 0);
525  if (PREDICT_FALSE (b0->flags & VLIB_BUFFER_IS_TRACED))
526  {
527  ip4_sv_reass_add_trace (vm, node, rm, NULL, from[(b - 2) - bufs],
529  vnet_buffer (b0)->ip.reass.ip_proto,
530  vnet_buffer (b0)->ip.reass.l4_src_port,
531  vnet_buffer (b0)->ip.reass.l4_dst_port);
532  }
533  if (is_feature)
534  {
535  vnet_feature_next (&next1, b1);
536  }
537  else
538  {
539  next1 = is_custom ? vnet_buffer (b1)->ip.reass.next_index :
541  }
542  vnet_buffer (b1)->ip.reass.is_non_first_fragment = 0;
543  vnet_buffer (b1)->ip.reass.ip_proto = ip1->protocol;
544  if (IP_PROTOCOL_TCP == ip1->protocol)
545  {
546  vnet_buffer (b1)->ip.reass.icmp_type_or_tcp_flags =
547  ((tcp_header_t *) (ip1 + 1))->flags;
548  vnet_buffer (b1)->ip.reass.tcp_ack_number =
549  ((tcp_header_t *) (ip1 + 1))->ack_number;
550  vnet_buffer (b1)->ip.reass.tcp_seq_number =
551  ((tcp_header_t *) (ip1 + 1))->seq_number;
552  }
553  else if (IP_PROTOCOL_ICMP == ip1->protocol)
554  {
555  vnet_buffer (b1)->ip.reass.icmp_type_or_tcp_flags =
556  ((icmp46_header_t *) (ip1 + 1))->type;
557  }
558  vnet_buffer (b1)->ip.reass.l4_src_port = ip4_get_port (ip1, 1);
559  vnet_buffer (b1)->ip.reass.l4_dst_port = ip4_get_port (ip1, 0);
560  if (PREDICT_FALSE (b1->flags & VLIB_BUFFER_IS_TRACED))
561  {
562  ip4_sv_reass_add_trace (vm, node, rm, NULL, from[(b - 1) - bufs],
564  vnet_buffer (b1)->ip.reass.ip_proto,
565  vnet_buffer (b1)->ip.reass.l4_src_port,
566  vnet_buffer (b1)->ip.reass.l4_dst_port);
567  }
568 
569  n_left_from -= 2;
570  next[0] = next0;
571  next[1] = next1;
572  next += 2;
573  }
574 
575  while (n_left_from > 0)
576  {
577  vlib_buffer_t *b0;
578  u32 next0;
579  b0 = *b;
580  b++;
581 
582  ip4_header_t *ip0 =
584  (is_output_feature ? 1 : 0) *
585  vnet_buffer (b0)->
586  ip.save_rewrite_length);
587  if (PREDICT_FALSE
589  {
590  // fragment found, go slow path
591  b -= 1;
592  if (b - bufs > 0)
593  {
594  vlib_buffer_enqueue_to_next (vm, node, from, (u16 *) nexts,
595  b - bufs);
596  }
597  goto slow_path;
598  }
599  if (is_feature)
600  {
601  vnet_feature_next (&next0, b0);
602  }
603  else
604  {
605  next0 =
606  is_custom ? vnet_buffer (b0)->ip.
607  reass.next_index : IP4_SV_REASSEMBLY_NEXT_INPUT;
608  }
609  vnet_buffer (b0)->ip.reass.is_non_first_fragment = 0;
610  vnet_buffer (b0)->ip.reass.ip_proto = ip0->protocol;
611  if (IP_PROTOCOL_TCP == ip0->protocol)
612  {
613  vnet_buffer (b0)->ip.reass.icmp_type_or_tcp_flags =
614  ((tcp_header_t *) (ip0 + 1))->flags;
615  vnet_buffer (b0)->ip.reass.tcp_ack_number =
616  ((tcp_header_t *) (ip0 + 1))->ack_number;
617  vnet_buffer (b0)->ip.reass.tcp_seq_number =
618  ((tcp_header_t *) (ip0 + 1))->seq_number;
619  }
620  else if (IP_PROTOCOL_ICMP == ip0->protocol)
621  {
622  vnet_buffer (b0)->ip.reass.icmp_type_or_tcp_flags =
623  ((icmp46_header_t *) (ip0 + 1))->type;
624  }
625  vnet_buffer (b0)->ip.reass.l4_src_port = ip4_get_port (ip0, 1);
626  vnet_buffer (b0)->ip.reass.l4_dst_port = ip4_get_port (ip0, 0);
627  if (PREDICT_FALSE (b0->flags & VLIB_BUFFER_IS_TRACED))
628  {
629  ip4_sv_reass_add_trace (vm, node, rm, NULL, from[(b - 1) - bufs],
631  vnet_buffer (b0)->ip.reass.ip_proto,
632  vnet_buffer (b0)->ip.reass.l4_src_port,
633  vnet_buffer (b0)->ip.reass.l4_dst_port);
634  }
635 
636  n_left_from -= 1;
637  next[0] = next0;
638  next += 1;
639  }
640 
641  vlib_buffer_enqueue_to_next (vm, node, from, (u16 *) nexts,
642  frame->n_vectors);
643 
644  goto done;
645 
646 slow_path:
647 
648  from += b - bufs;
649 
650  while (n_left_from > 0)
651  {
652  vlib_get_next_frame (vm, node, next_index, to_next, n_left_to_next);
653 
654  while (n_left_from > 0 && n_left_to_next > 0)
655  {
656  u32 bi0;
657  vlib_buffer_t *b0;
658  u32 next0;
659  u32 error0 = IP4_ERROR_NONE;
660 
661  bi0 = from[0];
662  b0 = vlib_get_buffer (vm, bi0);
663 
664  ip4_header_t *ip0 =
666  (is_output_feature ? 1 : 0) *
667  vnet_buffer (b0)->
668  ip.save_rewrite_length);
669  if (!ip4_get_fragment_more (ip0) && !ip4_get_fragment_offset (ip0))
670  {
671  // this is a regular packet - no fragmentation
672  if (is_custom)
673  {
674  next0 = vnet_buffer (b0)->ip.reass.next_index;
675  }
676  else
677  {
679  }
680  vnet_buffer (b0)->ip.reass.is_non_first_fragment = 0;
681  vnet_buffer (b0)->ip.reass.ip_proto = ip0->protocol;
682  if (IP_PROTOCOL_TCP == ip0->protocol)
683  {
684  vnet_buffer (b0)->ip.reass.icmp_type_or_tcp_flags =
685  ((tcp_header_t *) (ip0 + 1))->flags;
686  vnet_buffer (b0)->ip.reass.tcp_ack_number =
687  ((tcp_header_t *) (ip0 + 1))->ack_number;
688  vnet_buffer (b0)->ip.reass.tcp_seq_number =
689  ((tcp_header_t *) (ip0 + 1))->seq_number;
690  }
691  else if (IP_PROTOCOL_ICMP == ip0->protocol)
692  {
693  vnet_buffer (b0)->ip.reass.icmp_type_or_tcp_flags =
694  ((icmp46_header_t *) (ip0 + 1))->type;
695  }
696  vnet_buffer (b0)->ip.reass.l4_src_port = ip4_get_port (ip0, 1);
697  vnet_buffer (b0)->ip.reass.l4_dst_port = ip4_get_port (ip0, 0);
698  if (PREDICT_FALSE (b0->flags & VLIB_BUFFER_IS_TRACED))
699  {
700  ip4_sv_reass_add_trace (vm, node, rm, NULL, bi0,
702  vnet_buffer (b0)->ip.reass.ip_proto,
703  vnet_buffer (b0)->ip.
704  reass.l4_src_port,
705  vnet_buffer (b0)->ip.
706  reass.l4_dst_port);
707  }
708  goto packet_enqueue;
709  }
710  const u32 fragment_first = ip4_get_fragment_offset_bytes (ip0);
711  const u32 fragment_length =
712  clib_net_to_host_u16 (ip0->length) - ip4_header_bytes (ip0);
713  const u32 fragment_last = fragment_first + fragment_length - 1;
714  if (fragment_first > fragment_last || fragment_first + fragment_length > UINT16_MAX - 20 || (fragment_length < 8 && ip4_get_fragment_more (ip0))) // 8 is minimum frag length per RFC 791
715  {
717  error0 = IP4_ERROR_REASS_MALFORMED_PACKET;
718  b0->error = node->errors[error0];
719  goto packet_enqueue;
720  }
722  u8 do_handoff = 0;
723 
724  kv.k.as_u64[0] =
726  vnet_buffer (b0)->sw_if_index[VLIB_RX]) |
727  (u64) ip0->src_address.as_u32 << 32;
728  kv.k.as_u64[1] =
729  (u64) ip0->dst_address.
730  as_u32 | (u64) ip0->fragment_id << 32 | (u64) ip0->protocol << 48;
731 
732  ip4_sv_reass_t *reass =
733  ip4_sv_reass_find_or_create (vm, rm, rt, &kv, &do_handoff);
734 
735  if (PREDICT_FALSE (do_handoff))
736  {
738  vnet_buffer (b0)->ip.reass.owner_thread_index =
739  kv.v.thread_index;
740  goto packet_enqueue;
741  }
742 
743  if (!reass)
744  {
746  error0 = IP4_ERROR_REASS_LIMIT_REACHED;
747  b0->error = node->errors[error0];
748  goto packet_enqueue;
749  }
750 
751  if (reass->is_complete)
752  {
753  if (is_custom)
754  {
755  next0 = vnet_buffer (b0)->ip.reass.next_index;
756  }
757  else
758  {
760  }
761  vnet_buffer (b0)->ip.reass.is_non_first_fragment =
762  ! !fragment_first;
763  vnet_buffer (b0)->ip.reass.ip_proto = reass->ip_proto;
764  vnet_buffer (b0)->ip.reass.icmp_type_or_tcp_flags =
765  reass->icmp_type_or_tcp_flags;
766  vnet_buffer (b0)->ip.reass.tcp_ack_number =
767  reass->tcp_ack_number;
768  vnet_buffer (b0)->ip.reass.tcp_seq_number =
769  reass->tcp_seq_number;
770  vnet_buffer (b0)->ip.reass.l4_src_port = reass->l4_src_port;
771  vnet_buffer (b0)->ip.reass.l4_dst_port = reass->l4_dst_port;
772  if (PREDICT_FALSE (b0->flags & VLIB_BUFFER_IS_TRACED))
773  {
774  ip4_sv_reass_add_trace (vm, node, rm, reass, bi0,
776  reass->ip_proto,
777  reass->l4_src_port,
778  reass->l4_dst_port);
779  }
780  goto packet_enqueue;
781  }
782 
783  ip4_sv_reass_rc_t rc =
784  ip4_sv_reass_update (vm, node, rm, rt, ip0, reass, bi0);
785  switch (rc)
786  {
787  case IP4_SV_REASS_RC_OK:
788  /* nothing to do here */
789  break;
792  IP4_ERROR_REASS_FRAGMENT_CHAIN_TOO_LONG,
793  1);
794  ip4_sv_reass_free (vm, rm, rt, reass);
795  goto next_packet;
796  break;
799  IP4_ERROR_REASS_FRAGMENT_CHAIN_TOO_LONG,
800  1);
801  ip4_sv_reass_free (vm, rm, rt, reass);
802  goto next_packet;
803  break;
804  }
805  if (reass->is_complete)
806  {
807  u32 idx;
808  vec_foreach_index (idx, reass->cached_buffers)
809  {
810  u32 bi0 = vec_elt (reass->cached_buffers, idx);
811  vlib_buffer_t *b0 = vlib_get_buffer (vm, bi0);
812  ip0 =
814  (is_output_feature ? 1 : 0) *
815  vnet_buffer (b0)->
816  ip.save_rewrite_length);
818  if (is_feature)
819  {
820  vnet_feature_next (&next0, b0);
821  }
822  if (is_custom)
823  {
824  next0 = vnet_buffer (b0)->ip.reass.next_index;
825  }
826  if (0 == n_left_to_next)
827  {
828  vlib_put_next_frame (vm, node, next_index,
829  n_left_to_next);
830  vlib_get_next_frame (vm, node, next_index, to_next,
831  n_left_to_next);
832  }
833  to_next[0] = bi0;
834  to_next += 1;
835  n_left_to_next -= 1;
836  vnet_buffer (b0)->ip.reass.is_non_first_fragment =
837  ! !ip4_get_fragment_offset (ip0);
838  vnet_buffer (b0)->ip.reass.ip_proto = reass->ip_proto;
839  vnet_buffer (b0)->ip.reass.icmp_type_or_tcp_flags =
840  reass->icmp_type_or_tcp_flags;
841  vnet_buffer (b0)->ip.reass.tcp_ack_number =
842  reass->tcp_ack_number;
843  vnet_buffer (b0)->ip.reass.tcp_seq_number =
844  reass->tcp_seq_number;
845  vnet_buffer (b0)->ip.reass.l4_src_port = reass->l4_src_port;
846  vnet_buffer (b0)->ip.reass.l4_dst_port = reass->l4_dst_port;
847  if (PREDICT_FALSE (b0->flags & VLIB_BUFFER_IS_TRACED))
848  {
849  ip4_sv_reass_add_trace (vm, node, rm, reass, bi0,
851  reass->ip_proto,
852  reass->l4_src_port,
853  reass->l4_dst_port);
854  }
855  vlib_validate_buffer_enqueue_x1 (vm, node, next_index,
856  to_next, n_left_to_next, bi0,
857  next0);
858  }
859  _vec_len (reass->cached_buffers) = 0; // buffers are owned by frame now
860  }
861  goto next_packet;
862 
863  packet_enqueue:
864  to_next[0] = bi0;
865  to_next += 1;
866  n_left_to_next -= 1;
867  if (is_feature && IP4_ERROR_NONE == error0)
868  {
869  b0 = vlib_get_buffer (vm, bi0);
870  vnet_feature_next (&next0, b0);
871  }
872  vlib_validate_buffer_enqueue_x1 (vm, node, next_index,
873  to_next, n_left_to_next,
874  bi0, next0);
875 
876  next_packet:
877  from += 1;
878  n_left_from -= 1;
879  }
880 
881  vlib_put_next_frame (vm, node, next_index, n_left_to_next);
882  }
883 
884 done:
885  clib_spinlock_unlock (&rt->lock);
886  return frame->n_vectors;
887 }
888 
889 static char *ip4_sv_reass_error_strings[] = {
890 #define _(sym, string) string,
892 #undef _
893 };
894 
895 VLIB_NODE_FN (ip4_sv_reass_node) (vlib_main_t * vm,
898 {
899  return ip4_sv_reass_inline (vm, node, frame, false /* is_feature */ ,
900  false /* is_output_feature */ ,
901  false /* is_custom */ );
902 }
903 
904 /* *INDENT-OFF* */
905 VLIB_REGISTER_NODE (ip4_sv_reass_node) = {
906  .name = "ip4-sv-reassembly",
907  .vector_size = sizeof (u32),
908  .format_trace = format_ip4_sv_reass_trace,
909  .n_errors = ARRAY_LEN (ip4_sv_reass_error_strings),
910  .error_strings = ip4_sv_reass_error_strings,
911  .n_next_nodes = IP4_SV_REASSEMBLY_N_NEXT,
912  .next_nodes =
913  {
914  [IP4_SV_REASSEMBLY_NEXT_INPUT] = "ip4-input",
915  [IP4_SV_REASSEMBLY_NEXT_DROP] = "ip4-drop",
916  [IP4_SV_REASSEMBLY_NEXT_HANDOFF] = "ip4-sv-reassembly-handoff",
917 
918  },
919 };
920 /* *INDENT-ON* */
921 
922 VLIB_NODE_FN (ip4_sv_reass_node_feature) (vlib_main_t * vm,
925 {
926  return ip4_sv_reass_inline (vm, node, frame, true /* is_feature */ ,
927  false /* is_output_feature */ ,
928  false /* is_custom */ );
929 }
930 
931 /* *INDENT-OFF* */
932 VLIB_REGISTER_NODE (ip4_sv_reass_node_feature) = {
933  .name = "ip4-sv-reassembly-feature",
934  .vector_size = sizeof (u32),
935  .format_trace = format_ip4_sv_reass_trace,
936  .n_errors = ARRAY_LEN (ip4_sv_reass_error_strings),
937  .error_strings = ip4_sv_reass_error_strings,
938  .n_next_nodes = IP4_SV_REASSEMBLY_N_NEXT,
939  .next_nodes =
940  {
941  [IP4_SV_REASSEMBLY_NEXT_INPUT] = "ip4-input",
942  [IP4_SV_REASSEMBLY_NEXT_DROP] = "ip4-drop",
943  [IP4_SV_REASSEMBLY_NEXT_HANDOFF] = "ip4-sv-reass-feature-hoff",
944  },
945 };
946 /* *INDENT-ON* */
947 
948 /* *INDENT-OFF* */
949 VNET_FEATURE_INIT (ip4_sv_reass_feature) = {
950  .arc_name = "ip4-unicast",
951  .node_name = "ip4-sv-reassembly-feature",
952  .runs_before = VNET_FEATURES ("ip4-lookup"),
953  .runs_after = 0,
954 };
955 /* *INDENT-ON* */
956 
960 {
961  return ip4_sv_reass_inline (vm, node, frame, true /* is_feature */ ,
962  true /* is_output_feature */ ,
963  false /* is_custom */ );
964 }
965 
966 
967 /* *INDENT-OFF* */
969  .name = "ip4-sv-reassembly-output-feature",
970  .vector_size = sizeof (u32),
971  .format_trace = format_ip4_sv_reass_trace,
972  .n_errors = ARRAY_LEN (ip4_sv_reass_error_strings),
973  .error_strings = ip4_sv_reass_error_strings,
974  .n_next_nodes = IP4_SV_REASSEMBLY_N_NEXT,
975  .next_nodes =
976  {
977  [IP4_SV_REASSEMBLY_NEXT_INPUT] = "ip4-input",
978  [IP4_SV_REASSEMBLY_NEXT_DROP] = "ip4-drop",
979  [IP4_SV_REASSEMBLY_NEXT_HANDOFF] = "ip4-sv-reass-feature-hoff",
980  },
981 };
982 /* *INDENT-ON* */
983 
984 /* *INDENT-OFF* */
985 VNET_FEATURE_INIT (ip4_sv_reass_output_feature) = {
986  .arc_name = "ip4-output",
987  .node_name = "ip4-sv-reassembly-output-feature",
988  .runs_before = 0,
989  .runs_after = 0,
990 };
991 /* *INDENT-ON* */
992 
993 /* *INDENT-OFF* */
995  .name = "ip4-sv-reassembly-custom-next",
996  .vector_size = sizeof (u32),
997  .format_trace = format_ip4_sv_reass_trace,
998  .n_errors = ARRAY_LEN (ip4_sv_reass_error_strings),
999  .error_strings = ip4_sv_reass_error_strings,
1000  .n_next_nodes = IP4_SV_REASSEMBLY_N_NEXT,
1001  .next_nodes =
1002  {
1003  [IP4_SV_REASSEMBLY_NEXT_INPUT] = "ip4-input",
1004  [IP4_SV_REASSEMBLY_NEXT_DROP] = "ip4-drop",
1005  [IP4_SV_REASSEMBLY_NEXT_HANDOFF] = "ip4-sv-reassembly-handoff",
1006 
1007  },
1008 };
1009 /* *INDENT-ON* */
1010 
1013  vlib_frame_t * frame)
1014 {
1015  return ip4_sv_reass_inline (vm, node, frame, false /* is_feature */ ,
1016  false /* is_output_feature */ ,
1017  true /* is_custom */ );
1018 }
1019 
1020 #ifndef CLIB_MARCH_VARIANT
1023 {
1025  u32 nbuckets;
1026  u8 i;
1027 
1028  nbuckets = (u32) (rm->max_reass_n / IP4_SV_REASS_HT_LOAD_FACTOR);
1029 
1030  for (i = 0; i < 31; i++)
1031  if ((1 << i) >= nbuckets)
1032  break;
1033  nbuckets = 1 << i;
1034 
1035  return nbuckets;
1036 }
1037 #endif /* CLIB_MARCH_VARIANT */
1038 
1039 typedef enum
1040 {
1043 
1044 typedef struct
1045 {
1046  int failure;
1047  clib_bihash_16_8_t *new_hash;
1049 
1050 #ifndef CLIB_MARCH_VARIANT
1051 static int
1053 {
1054  ip4_rehash_cb_ctx *ctx = _ctx;
1055  if (clib_bihash_add_del_16_8 (ctx->new_hash, kv, 1))
1056  {
1057  ctx->failure = 1;
1058  }
1059  return (BIHASH_WALK_CONTINUE);
1060 }
1061 
1062 static void
1063 ip4_sv_reass_set_params (u32 timeout_ms, u32 max_reassemblies,
1064  u32 max_reassembly_length,
1065  u32 expire_walk_interval_ms)
1066 {
1067  ip4_sv_reass_main.timeout_ms = timeout_ms;
1068  ip4_sv_reass_main.timeout = (f64) timeout_ms / (f64) MSEC_PER_SEC;
1069  ip4_sv_reass_main.max_reass_n = max_reassemblies;
1070  ip4_sv_reass_main.max_reass_len = max_reassembly_length;
1071  ip4_sv_reass_main.expire_walk_interval_ms = expire_walk_interval_ms;
1072 }
1073 
1075 ip4_sv_reass_set (u32 timeout_ms, u32 max_reassemblies,
1076  u32 max_reassembly_length, u32 expire_walk_interval_ms)
1077 {
1078  u32 old_nbuckets = ip4_sv_reass_get_nbuckets ();
1079  ip4_sv_reass_set_params (timeout_ms, max_reassemblies,
1080  max_reassembly_length, expire_walk_interval_ms);
1081  vlib_process_signal_event (ip4_sv_reass_main.vlib_main,
1082  ip4_sv_reass_main.ip4_sv_reass_expire_node_idx,
1084  u32 new_nbuckets = ip4_sv_reass_get_nbuckets ();
1085  if (ip4_sv_reass_main.max_reass_n > 0 && new_nbuckets > old_nbuckets)
1086  {
1087  clib_bihash_16_8_t new_hash;
1088  clib_memset (&new_hash, 0, sizeof (new_hash));
1090  ctx.failure = 0;
1091  ctx.new_hash = &new_hash;
1092  clib_bihash_init_16_8 (&new_hash, "ip4-dr", new_nbuckets,
1093  new_nbuckets * 1024);
1094  clib_bihash_foreach_key_value_pair_16_8 (&ip4_sv_reass_main.hash,
1095  ip4_rehash_cb, &ctx);
1096  if (ctx.failure)
1097  {
1098  clib_bihash_free_16_8 (&new_hash);
1099  return -1;
1100  }
1101  else
1102  {
1103  clib_bihash_free_16_8 (&ip4_sv_reass_main.hash);
1104  clib_memcpy_fast (&ip4_sv_reass_main.hash, &new_hash,
1105  sizeof (ip4_sv_reass_main.hash));
1106  clib_bihash_copied (&ip4_sv_reass_main.hash, &new_hash);
1107  }
1108  }
1109  return 0;
1110 }
1111 
1113 ip4_sv_reass_get (u32 * timeout_ms, u32 * max_reassemblies,
1114  u32 * max_reassembly_length, u32 * expire_walk_interval_ms)
1115 {
1116  *timeout_ms = ip4_sv_reass_main.timeout_ms;
1117  *max_reassemblies = ip4_sv_reass_main.max_reass_n;
1118  *max_reassembly_length = ip4_sv_reass_main.max_reass_len;
1119  *expire_walk_interval_ms = ip4_sv_reass_main.expire_walk_interval_ms;
1120  return 0;
1121 }
1122 
1123 static clib_error_t *
1125 {
1127  clib_error_t *error = 0;
1128  u32 nbuckets;
1129  vlib_node_t *node;
1130 
1131  rm->vlib_main = vm;
1132  rm->vnet_main = vnet_get_main ();
1133 
1136  vec_foreach (rt, rm->per_thread_data)
1137  {
1138  clib_spinlock_init (&rt->lock);
1139  pool_alloc (rt->pool, rm->max_reass_n);
1140  rt->lru_first = rt->lru_last = ~0;
1141  }
1142 
1143  node = vlib_get_node_by_name (vm, (u8 *) "ip4-sv-reassembly-expire-walk");
1144  ASSERT (node);
1145  rm->ip4_sv_reass_expire_node_idx = node->index;
1146 
1151 
1152  nbuckets = ip4_sv_reass_get_nbuckets ();
1153  clib_bihash_init_16_8 (&rm->hash, "ip4-dr", nbuckets, nbuckets * 1024);
1154 
1155  node = vlib_get_node_by_name (vm, (u8 *) "ip4-drop");
1156  ASSERT (node);
1157  rm->ip4_drop_idx = node->index;
1158 
1159  rm->fq_index = vlib_frame_queue_main_init (ip4_sv_reass_node.index, 0);
1160  rm->fq_feature_index =
1161  vlib_frame_queue_main_init (ip4_sv_reass_node_feature.index, 0);
1162 
1163  rm->feature_use_refcount_per_intf = NULL;
1165 
1166  return error;
1167 }
1168 
1170 #endif /* CLIB_MARCH_VARIANT */
1171 
1172 static uword
1174  vlib_node_runtime_t * node, vlib_frame_t * f)
1175 {
1177  uword event_type, *event_data = 0;
1178 
1179  while (true)
1180  {
1182  (f64)
1184  (f64) MSEC_PER_SEC);
1185  event_type = vlib_process_get_events (vm, &event_data);
1186 
1187  switch (event_type)
1188  {
1189  case ~0: /* no events => timeout */
1190  /* nothing to do here */
1191  break;
1193  break;
1194  default:
1195  clib_warning ("BUG: event type 0x%wx", event_type);
1196  break;
1197  }
1198  f64 now = vlib_time_now (vm);
1199 
1200  ip4_sv_reass_t *reass;
1201  int *pool_indexes_to_free = NULL;
1202 
1203  uword thread_index = 0;
1204  int index;
1205  const uword nthreads = vlib_num_workers () + 1;
1206  for (thread_index = 0; thread_index < nthreads; ++thread_index)
1207  {
1208  ip4_sv_reass_per_thread_t *rt = &rm->per_thread_data[thread_index];
1209  clib_spinlock_lock (&rt->lock);
1210 
1211  vec_reset_length (pool_indexes_to_free);
1212  /* *INDENT-OFF* */
1213  pool_foreach_index (index, rt->pool) {
1214  reass = pool_elt_at_index (rt->pool, index);
1215  if (now > reass->last_heard + rm->timeout)
1216  {
1217  vec_add1 (pool_indexes_to_free, index);
1218  }
1219  }
1220  /* *INDENT-ON* */
1221  int *i;
1222  /* *INDENT-OFF* */
1223  vec_foreach (i, pool_indexes_to_free)
1224  {
1225  ip4_sv_reass_t *reass = pool_elt_at_index (rt->pool, i[0]);
1226  ip4_sv_reass_free (vm, rm, rt, reass);
1227  }
1228  /* *INDENT-ON* */
1229 
1230  clib_spinlock_unlock (&rt->lock);
1231  }
1232 
1233  vec_free (pool_indexes_to_free);
1234  if (event_data)
1235  {
1236  _vec_len (event_data) = 0;
1237  }
1238  }
1239 
1240  return 0;
1241 }
1242 
1243 /* *INDENT-OFF* */
1245  .function = ip4_sv_reass_walk_expired,
1246  .type = VLIB_NODE_TYPE_PROCESS,
1247  .name = "ip4-sv-reassembly-expire-walk",
1248  .format_trace = format_ip4_sv_reass_trace,
1249  .n_errors = ARRAY_LEN (ip4_sv_reass_error_strings),
1250  .error_strings = ip4_sv_reass_error_strings,
1251 
1252 };
1253 /* *INDENT-ON* */
1254 
1255 static u8 *
1256 format_ip4_sv_reass_key (u8 * s, va_list * args)
1257 {
1258  ip4_sv_reass_key_t *key = va_arg (*args, ip4_sv_reass_key_t *);
1259  s =
1260  format (s,
1261  "xx_id: %u, src: %U, dst: %U, frag_id: %u, proto: %u",
1263  &key->dst, clib_net_to_host_u16 (key->frag_id), key->proto);
1264  return s;
1265 }
1266 
1267 static u8 *
1268 format_ip4_sv_reass (u8 * s, va_list * args)
1269 {
1270  vlib_main_t *vm = va_arg (*args, vlib_main_t *);
1271  ip4_sv_reass_t *reass = va_arg (*args, ip4_sv_reass_t *);
1272 
1273  s = format (s, "ID: %lu, key: %U trace_op_counter: %u\n",
1274  reass->id, format_ip4_sv_reass_key, &reass->key,
1275  reass->trace_op_counter);
1276 
1277  vlib_buffer_t *b;
1278  u32 *bip;
1279  u32 counter = 0;
1280  vec_foreach (bip, reass->cached_buffers)
1281  {
1282  u32 bi = *bip;
1283  do
1284  {
1285  b = vlib_get_buffer (vm, bi);
1286  s = format (s, " #%03u: bi: %u, ", counter, bi);
1287  ++counter;
1288  bi = b->next_buffer;
1289  }
1290  while (b->flags & VLIB_BUFFER_NEXT_PRESENT);
1291  }
1292  return s;
1293 }
1294 
1295 static clib_error_t *
1297  unformat_input_t * input,
1299 {
1301 
1302  vlib_cli_output (vm, "---------------------");
1303  vlib_cli_output (vm, "IP4 reassembly status");
1304  vlib_cli_output (vm, "---------------------");
1305  bool details = false;
1306  if (unformat (input, "details"))
1307  {
1308  details = true;
1309  }
1310 
1311  u32 sum_reass_n = 0;
1312  ip4_sv_reass_t *reass;
1313  uword thread_index;
1314  const uword nthreads = vlib_num_workers () + 1;
1315  for (thread_index = 0; thread_index < nthreads; ++thread_index)
1316  {
1317  ip4_sv_reass_per_thread_t *rt = &rm->per_thread_data[thread_index];
1318  clib_spinlock_lock (&rt->lock);
1319  if (details)
1320  {
1321  /* *INDENT-OFF* */
1322  pool_foreach (reass, rt->pool) {
1323  vlib_cli_output (vm, "%U", format_ip4_sv_reass, vm, reass);
1324  }
1325  /* *INDENT-ON* */
1326  }
1327  sum_reass_n += rt->reass_n;
1328  clib_spinlock_unlock (&rt->lock);
1329  }
1330  vlib_cli_output (vm, "---------------------");
1331  vlib_cli_output (vm, "Current IP4 reassemblies count: %lu\n",
1332  (long unsigned) sum_reass_n);
1333  vlib_cli_output (vm,
1334  "Maximum configured concurrent shallow virtual IP4 reassemblies per worker-thread: %lu\n",
1335  (long unsigned) rm->max_reass_n);
1336  vlib_cli_output (vm,
1337  "Maximum configured shallow virtual IP4 reassembly timeout: %lums\n",
1338  (long unsigned) rm->timeout_ms);
1339  vlib_cli_output (vm,
1340  "Maximum configured shallow virtual IP4 reassembly expire walk interval: %lums\n",
1341  (long unsigned) rm->expire_walk_interval_ms);
1342  return 0;
1343 }
1344 
1345 /* *INDENT-OFF* */
1346 VLIB_CLI_COMMAND (show_ip4_sv_reass_cmd, static) = {
1347  .path = "show ip4-sv-reassembly",
1348  .short_help = "show ip4-sv-reassembly [details]",
1349  .function = show_ip4_reass,
1350 };
1351 /* *INDENT-ON* */
1352 
1353 #ifndef CLIB_MARCH_VARIANT
1356 {
1357  return ip4_sv_reass_enable_disable_with_refcnt (sw_if_index,
1358  enable_disable);
1359 }
1360 #endif /* CLIB_MARCH_VARIANT */
1361 
1362 
1363 #define foreach_ip4_sv_reass_handoff_error \
1364 _(CONGESTION_DROP, "congestion drop")
1365 
1366 
1367 typedef enum
1368 {
1369 #define _(sym,str) IP4_SV_REASSEMBLY_HANDOFF_ERROR_##sym,
1371 #undef _
1374 
1375 static char *ip4_sv_reass_handoff_error_strings[] = {
1376 #define _(sym,string) string,
1378 #undef _
1379 };
1380 
1381 typedef struct
1382 {
1385 
1386 static u8 *
1388 {
1389  CLIB_UNUSED (vlib_main_t * vm) = va_arg (*args, vlib_main_t *);
1390  CLIB_UNUSED (vlib_node_t * node) = va_arg (*args, vlib_node_t *);
1392  va_arg (*args, ip4_sv_reass_handoff_trace_t *);
1393 
1394  s =
1395  format (s, "ip4-sv-reassembly-handoff: next-worker %d",
1396  t->next_worker_index);
1397 
1398  return s;
1399 }
1400 
1403  vlib_node_runtime_t * node,
1404  vlib_frame_t * frame, bool is_feature)
1405 {
1407 
1408  vlib_buffer_t *bufs[VLIB_FRAME_SIZE], **b;
1409  u32 n_enq, n_left_from, *from;
1410  u16 thread_indices[VLIB_FRAME_SIZE], *ti;
1411  u32 fq_index;
1412 
1413  from = vlib_frame_vector_args (frame);
1414  n_left_from = frame->n_vectors;
1415  vlib_get_buffers (vm, from, bufs, n_left_from);
1416 
1417  b = bufs;
1418  ti = thread_indices;
1419 
1420  fq_index = (is_feature) ? rm->fq_feature_index : rm->fq_index;
1421 
1422  while (n_left_from > 0)
1423  {
1424  ti[0] = vnet_buffer (b[0])->ip.reass.owner_thread_index;
1425 
1426  if (PREDICT_FALSE
1427  ((node->flags & VLIB_NODE_FLAG_TRACE)
1428  && (b[0]->flags & VLIB_BUFFER_IS_TRACED)))
1429  {
1431  vlib_add_trace (vm, node, b[0], sizeof (*t));
1432  t->next_worker_index = ti[0];
1433  }
1434 
1435  n_left_from -= 1;
1436  ti += 1;
1437  b += 1;
1438  }
1439  n_enq =
1440  vlib_buffer_enqueue_to_thread (vm, fq_index, from, thread_indices,
1441  frame->n_vectors, 1);
1442 
1443  if (n_enq < frame->n_vectors)
1445  IP4_SV_REASSEMBLY_HANDOFF_ERROR_CONGESTION_DROP,
1446  frame->n_vectors - n_enq);
1447  return frame->n_vectors;
1448 }
1449 
1450 VLIB_NODE_FN (ip4_sv_reass_handoff_node) (vlib_main_t * vm,
1452  vlib_frame_t * frame)
1453 {
1454  return ip4_sv_reass_handoff_node_inline (vm, node, frame,
1455  false /* is_feature */ );
1456 }
1457 
1458 
1459 /* *INDENT-OFF* */
1460 VLIB_REGISTER_NODE (ip4_sv_reass_handoff_node) = {
1461  .name = "ip4-sv-reassembly-handoff",
1462  .vector_size = sizeof (u32),
1463  .n_errors = ARRAY_LEN(ip4_sv_reass_handoff_error_strings),
1464  .error_strings = ip4_sv_reass_handoff_error_strings,
1465  .format_trace = format_ip4_sv_reass_handoff_trace,
1466 
1467  .n_next_nodes = 1,
1468 
1469  .next_nodes = {
1470  [0] = "error-drop",
1471  },
1472 };
1473 /* *INDENT-ON* */
1474 
1475 
1476 /* *INDENT-OFF* */
1477 VLIB_NODE_FN (ip4_sv_reass_feature_handoff_node) (vlib_main_t * vm,
1479  node,
1480  vlib_frame_t * frame)
1481 {
1482  return ip4_sv_reass_handoff_node_inline (vm, node, frame,
1483  true /* is_feature */ );
1484 }
1485 /* *INDENT-ON* */
1486 
1487 
1488 /* *INDENT-OFF* */
1489 VLIB_REGISTER_NODE (ip4_sv_reass_feature_handoff_node) = {
1490  .name = "ip4-sv-reass-feature-hoff",
1491  .vector_size = sizeof (u32),
1492  .n_errors = ARRAY_LEN(ip4_sv_reass_handoff_error_strings),
1493  .error_strings = ip4_sv_reass_handoff_error_strings,
1494  .format_trace = format_ip4_sv_reass_handoff_trace,
1495 
1496  .n_next_nodes = 1,
1497 
1498  .next_nodes = {
1499  [0] = "error-drop",
1500  },
1501 };
1502 /* *INDENT-ON* */
1503 
1504 #ifndef CLIB_MARCH_VARIANT
1505 int
1507 {
1509  vec_validate (rm->feature_use_refcount_per_intf, sw_if_index);
1510  if (is_enable)
1511  {
1512  if (!rm->feature_use_refcount_per_intf[sw_if_index])
1513  {
1515  return vnet_feature_enable_disable ("ip4-unicast",
1516  "ip4-sv-reassembly-feature",
1517  sw_if_index, 1, 0, 0);
1518  }
1520  }
1521  else
1522  {
1523  if (rm->feature_use_refcount_per_intf[sw_if_index])
1525  if (!rm->feature_use_refcount_per_intf[sw_if_index])
1526  return vnet_feature_enable_disable ("ip4-unicast",
1527  "ip4-sv-reassembly-feature",
1528  sw_if_index, 0, 0, 0);
1529  }
1530  return 0;
1531 }
1532 
1533 uword
1535 {
1537  node_index);
1538 }
1539 
1540 int
1542  int is_enable)
1543 {
1546  if (is_enable)
1547  {
1548  if (!rm->output_feature_use_refcount_per_intf[sw_if_index])
1549  {
1551  return vnet_feature_enable_disable ("ip4-output",
1552  "ip4-sv-reassembly-output-feature",
1553  sw_if_index, 1, 0, 0);
1554  }
1556  }
1557  else
1558  {
1559  if (rm->output_feature_use_refcount_per_intf[sw_if_index])
1561  if (!rm->output_feature_use_refcount_per_intf[sw_if_index])
1562  return vnet_feature_enable_disable ("ip4-output",
1563  "ip4-sv-reassembly-output-feature",
1564  sw_if_index, 0, 0, 0);
1565  }
1566  return 0;
1567 }
1568 #endif
1569 
1570 /*
1571  * fd.io coding-style-patch-verification: ON
1572  *
1573  * Local Variables:
1574  * eval: (c-set-style "gnu")
1575  * End:
1576  */
#define vec_validate(V, I)
Make sure vector is long enough for given index (no header, unspecified alignment) ...
Definition: vec.h:509
ip4_sv_reass_t * pool
Definition: ip4_sv_reass.c:116
u32 flags
buffer flags: VLIB_BUFFER_FREE_LIST_INDEX_MASK: bits used to store free list index, VLIB_BUFFER_IS_TRACED: trace this buffer.
Definition: buffer.h:124
vnet_main_t * vnet_main
Definition: ip4_sv_reass.c:144
#define IP4_SV_REASS_TIMEOUT_DEFAULT_MS
Definition: ip4_sv_reass.c:32
#define vec_foreach_index(var, v)
Iterate over vector indices.
vnet_api_error_t
Definition: api_errno.h:162
ip4_sv_reass_rc_t
Definition: ip4_sv_reass.c:38
static void ip4_sv_reass_set_params(u32 timeout_ms, u32 max_reassemblies, u32 max_reassembly_length, u32 expire_walk_interval_ms)
#define IP4_SV_REASS_HT_LOAD_FACTOR
Definition: ip4_sv_reass.c:36
static_always_inline void clib_spinlock_unlock(clib_spinlock_t *p)
Definition: lock.h:121
#define pool_foreach_index(i, v)
Definition: pool.h:569
static_always_inline void clib_spinlock_lock(clib_spinlock_t *p)
Definition: lock.h:82
#define CLIB_UNUSED(x)
Definition: clib.h:87
clib_bihash_kv_16_8_t kv
Definition: ip4_sv_reass.c:79
static char * ip4_sv_reass_handoff_error_strings[]
u16 min_fragment_length
Definition: ip4_sv_reass.c:93
static f64 vlib_process_wait_for_event_or_clock(vlib_main_t *vm, f64 dt)
Suspend a cooperative multi-tasking thread Waits for an event, or for the indicated number of seconds...
Definition: node_funcs.h:751
static uword ip4_sv_reass_inline(vlib_main_t *vm, vlib_node_runtime_t *node, vlib_frame_t *frame, bool is_feature, bool is_output_feature, bool is_custom)
Definition: ip4_sv_reass.c:432
vl_api_wireguard_peer_flags_t flags
Definition: wireguard.api:105
int ip4_sv_reass_enable_disable_with_refcnt(u32 sw_if_index, int is_enable)
static void vlib_buffer_free(vlib_main_t *vm, u32 *buffers, u32 n_buffers)
Free buffers Frees the entire buffer chain for each buffer.
Definition: buffer_funcs.h:937
static u32 vlib_buffer_get_trace_index(vlib_buffer_t *b)
Extract the trace (pool) index from a trace handle.
Definition: buffer.h:392
ip4_address_t src_address
Definition: ip4_packet.h:125
static ip4_sv_reass_rc_t ip4_sv_reass_update(vlib_main_t *vm, vlib_node_runtime_t *node, ip4_sv_reass_main_t *rm, ip4_sv_reass_per_thread_t *rt, ip4_header_t *ip0, ip4_sv_reass_t *reass, u32 bi0)
Definition: ip4_sv_reass.c:381
#define IP4_SV_REASS_MAX_REASSEMBLIES_DEFAULT
Definition: ip4_sv_reass.c:34
vlib_node_registration_t ip4_sv_reass_custom_node
(constructor) VLIB_REGISTER_NODE (ip4_sv_reass_custom_node)
Definition: ip4_sv_reass.c:994
vnet_main_t * vnet_get_main(void)
Definition: misc.c:46
static u8 * format_ip4_sv_reass_key(u8 *s, va_list *args)
vlib_main_t * vlib_main
Definition: ip4_sv_reass.c:143
#define pool_foreach(VAR, POOL)
Iterate through pool.
Definition: pool.h:527
#define pool_alloc(P, N)
Allocate N more free elements to pool (unspecified alignment).
Definition: pool.h:360
u64 as_u64
Definition: bihash_doc.h:63
u32 * feature_use_refcount_per_intf
Definition: ip4_sv_reass.c:155
#define PREDICT_TRUE(x)
Definition: clib.h:122
unsigned long u64
Definition: types.h:89
vnet_api_error_t ip4_sv_reass_get(u32 *timeout_ms, u32 *max_reassemblies, u32 *max_reassembly_length, u32 *expire_walk_interval_ms)
get ip4 reassembly configuration
#define clib_memcpy_fast(a, b, c)
Definition: string.h:81
vlib_node_registration_t ip4_sv_reass_node_feature
(constructor) VLIB_REGISTER_NODE (ip4_sv_reass_node_feature)
Definition: ip4_sv_reass.c:932
ip_proto
Definition: ip_types.api:75
static void ip4_sv_reass_init(ip4_sv_reass_t *reass)
Definition: ip4_sv_reass.c:303
clib_memset(h->entries, 0, sizeof(h->entries[0]) *entries)
u32 index
Definition: node.h:280
static f64 vlib_time_now(vlib_main_t *vm)
Definition: main.h:334
static u32 ip4_sv_reass_get_nbuckets()
u32 * output_feature_use_refcount_per_intf
Definition: ip4_sv_reass.c:158
u32 vlib_frame_queue_main_init(u32 node_index, u32 frame_queue_nelts)
Definition: threads.c:1880
u32 thread_index
Definition: main.h:250
int ip4_sv_reass_output_enable_disable_with_refcnt(u32 sw_if_index, int is_enable)
static u8 * format_ip4_sv_reass_trace(u8 *s, va_list *args)
Definition: ip4_sv_reass.c:198
#define vec_add1(V, E)
Add 1 element to end of vector (unspecified alignment).
Definition: vec.h:592
#define IP4_SV_REASS_EXPIRE_WALK_INTERVAL_DEFAULT_MS
Definition: ip4_sv_reass.c:33
vlib_main_t * vm
Definition: in2out_ed.c:1580
u32 * fib_index_by_sw_if_index
Table index indexed by software interface.
Definition: ip4.h:123
vnet_api_error_t ip4_sv_reass_enable_disable(u32 sw_if_index, u8 enable_disable)
#define VLIB_NODE_FN(node)
Definition: node.h:203
vlib_error_t * errors
Vector of errors for this node.
Definition: node.h:470
#define pool_get(P, E)
Allocate an object E from a pool P (unspecified alignment).
Definition: pool.h:251
struct _tcp_header tcp_header_t
static void ip4_sv_reass_add_trace(vlib_main_t *vm, vlib_node_runtime_t *node, ip4_sv_reass_main_t *rm, ip4_sv_reass_t *reass, u32 bi, ip4_sv_reass_trace_operation_e action, u32 ip_proto, u16 l4_src_port, u16 l4_dst_port)
Definition: ip4_sv_reass.c:232
static uword vlib_node_add_next(vlib_main_t *vm, uword node, uword next_node)
Definition: node_funcs.h:1173
unsigned char u8
Definition: types.h:56
static int ip4_get_fragment_offset_bytes(const ip4_header_t *i)
Definition: ip4_packet.h:184
IPv4 to IPv6 translation.
#define vec_reset_length(v)
Reset vector length to zero NULL-pointer tolerant.
double f64
Definition: types.h:142
ip4_sv_reass_main_t ip4_sv_reass_main
Definition: ip4_sv_reass.c:165
ip4_address_t src
Definition: ip4_sv_reass.c:52
#define u8_ptr_add(ptr, index)
Definition: ip_types.h:56
format_function_t format_ip4_address
Definition: format.h:73
#define foreach_ip4_sv_reass_handoff_error
vlib_trace_header_t ** trace_buffer_pool
Definition: trace.h:86
static ip4_sv_reass_t * ip4_sv_reass_find_or_create(vlib_main_t *vm, ip4_sv_reass_main_t *rm, ip4_sv_reass_per_thread_t *rt, ip4_sv_reass_kv_t *kv, u8 *do_handoff)
Definition: ip4_sv_reass.c:310
static u32 slow_path(dslite_main_t *dm, dslite_session_key_t *in2out_key, dslite_session_t **sp, u32 next, u8 *error, u32 thread_index)
Definition: dslite_in2out.c:34
#define VLIB_INIT_FUNCTION(x)
Definition: init.h:173
static uword vlib_process_get_events(vlib_main_t *vm, uword **data_vector)
Return the first event type which has occurred and a vector of per-event data of that type...
Definition: node_funcs.h:579
static u16 ip4_get_fragment_more(const ip4_header_t *i)
Definition: ip4_packet.h:161
ip4_address_t dst_address
Definition: ip4_packet.h:125
description fragment has unexpected format
Definition: map.api:433
#define vlib_prefetch_buffer_header(b, type)
Prefetch buffer metadata.
Definition: buffer.h:207
clib_bihash_16_8_t * new_hash
const cJSON *const b
Definition: cJSON.h:255
ip4_sv_reass_trace_operation_e action
Definition: ip4_sv_reass.c:186
ip4_sv_reass_next_t
Definition: ip4_sv_reass.c:168
unsigned int u32
Definition: types.h:88
#define VLIB_FRAME_SIZE
Definition: node.h:378
ip4_sv_reass_handoff_error_t
static char * ip4_sv_reass_error_strings[]
Definition: ip4_sv_reass.c:889
static u16 ip4_get_port(ip4_header_t *ip, u8 sender)
Get TCP/UDP port number or ICMP id from IPv4 packet.
Definition: ip4_to_ip6.h:51
static void clib_spinlock_init(clib_spinlock_t *p)
Definition: lock.h:65
vl_api_fib_path_type_t type
Definition: fib_types.api:123
vlib_error_t error
Error code for buffers to be enqueued to error handler.
Definition: buffer.h:136
vlib_node_registration_t ip4_sv_reass_node
(constructor) VLIB_REGISTER_NODE (ip4_sv_reass_node)
Definition: ip4_sv_reass.c:905
vlib_node_registration_t ip4_sv_reass_expire_node
(constructor) VLIB_REGISTER_NODE (ip4_sv_reass_expire_node)
vnet_api_error_t ip4_sv_reass_set(u32 timeout_ms, u32 max_reassemblies, u32 max_reassembly_length, u32 expire_walk_interval_ms)
set ip4 reassembly configuration
Definition: cJSON.c:84
#define pool_elt_at_index(p, i)
Returns pointer to element at given index.
Definition: pool.h:546
vlib_node_t * vlib_get_node_by_name(vlib_main_t *vm, u8 *name)
Definition: node.c:45
static void vlib_process_signal_event(vlib_main_t *vm, uword node_index, uword type_opaque, uword data)
Definition: node_funcs.h:1015
#define IP4_SV_REASS_MAX_REASSEMBLY_LENGTH_DEFAULT
Definition: ip4_sv_reass.c:35
IPv4 shallow virtual reassembly.
long ctx[MAX_CONNS]
Definition: main.c:144
struct _unformat_input_t unformat_input_t
unsigned short u16
Definition: types.h:57
static void * vlib_buffer_get_current(vlib_buffer_t *b)
Get pointer to current data to process.
Definition: buffer.h:233
#define pool_put(P, E)
Free an object E in pool P.
Definition: pool.h:301
ip4_sv_reass_trace_operation_e
Definition: ip4_sv_reass.c:176
u32 fq_index
Worker handoff.
Definition: ip4_sv_reass.c:151
ip4_sv_reass_key_t k
Definition: ip4_sv_reass.c:76
#define PREDICT_FALSE(x)
Definition: clib.h:121
#define always_inline
Definition: ipsec.h:28
u32 node_index
Node index.
Definition: node.h:488
#define vlib_validate_buffer_enqueue_x1(vm, node, next_index, to_next, n_left_to_next, bi0, next0)
Finish enqueueing one buffer forward in the graph.
Definition: buffer_node.h:224
#define vlib_get_next_frame(vm, node, next_index, vectors, n_vectors_left)
Get pointer to next frame vector data by (vlib_node_runtime_t, next_index).
Definition: node_funcs.h:391
static void vlib_node_increment_counter(vlib_main_t *vm, u32 node_index, u32 counter_index, u64 increment)
Definition: node_funcs.h:1231
#define foreach_ip4_error
Definition: ip4_error.h:43
static uword ip4_sv_reass_handoff_node_inline(vlib_main_t *vm, vlib_node_runtime_t *node, vlib_frame_t *frame, bool is_feature)
#define VLIB_REGISTER_NODE(x,...)
Definition: node.h:170
u16 n_vectors
Definition: node.h:397
#define CLIB_PREFETCH(addr, size, type)
Definition: cache.h:80
clib_bihash_16_8_t hash
Definition: ip4_sv_reass.c:138
static_always_inline void vlib_buffer_enqueue_to_next(vlib_main_t *vm, vlib_node_runtime_t *node, u32 *buffers, u16 *nexts, uword count)
Definition: buffer_node.h:339
sll srl srl sll sra u16x4 i
Definition: vector_sse42.h:317
static u16 ip4_get_fragment_offset(const ip4_header_t *i)
Definition: ip4_packet.h:155
#define vec_free(V)
Free vector&#39;s memory (no header).
Definition: vec.h:380
static_always_inline void vnet_feature_next(u32 *next0, vlib_buffer_t *b0)
Definition: feature.h:322
ip4_sv_reass_event_t
#define clib_warning(format, args...)
Definition: error.h:59
u8 data[]
Packet data.
Definition: buffer.h:181
__clib_export void clib_bihash_copied(void *dst, void *src)
static clib_error_t * show_ip4_reass(vlib_main_t *vm, unformat_input_t *input, CLIB_UNUSED(vlib_cli_command_t *lmd))
#define pool_is_free_index(P, I)
Use free bitmap to query whether given index is free.
Definition: pool.h:298
#define ARRAY_LEN(x)
Definition: clib.h:67
void vlib_put_next_frame(vlib_main_t *vm, vlib_node_runtime_t *r, u32 next_index, u32 n_vectors_left)
Release pointer to next frame vector data.
Definition: main.c:483
vlib_main_t vlib_node_runtime_t * node
Definition: in2out_ed.c:1580
#define VLIB_CLI_COMMAND(x,...)
Definition: cli.h:158
VNET_FEATURE_INIT(ip4_sv_reass_feature)
static uword ip4_sv_reass_walk_expired(vlib_main_t *vm, vlib_node_runtime_t *node, vlib_frame_t *f)
u16 cached_next_index
Next frame index that vector arguments were last enqueued to last time this node ran.
Definition: node.h:511
#define ASSERT(truth)
void vlib_cli_output(vlib_main_t *vm, char *fmt,...)
Definition: cli.c:696
ip4_sv_reass_val_t v
Definition: ip4_sv_reass.c:77
vlib_trace_main_t trace_main
Definition: main.h:194
#define VNET_FEATURES(...)
Definition: feature.h:470
static vlib_main_t * vlib_get_main(void)
Definition: global_funcs.h:23
ip4_address_t dst
Definition: ip4_sv_reass.c:53
#define vec_elt(v, i)
Get vector value at index i.
typedef key
Definition: ipsec_types.api:86
struct _vlib_node_registration vlib_node_registration_t
vl_api_address_t ip
Definition: l2.api:501
vl_api_mac_event_action_t action
Definition: l2.api:181
#define vec_len(v)
Number of elements in vector (rvalue-only, NULL tolerant)
u32 next_buffer
Next buffer for this linked-list of buffers.
Definition: buffer.h:140
#define MSEC_PER_SEC
Definition: ip4_sv_reass.c:31
vlib_main_t vlib_node_runtime_t vlib_frame_t * frame
Definition: in2out_ed.c:1581
VLIB buffer representation.
Definition: buffer.h:102
u64 uword
Definition: types.h:112
u32 * cached_buffers
Definition: ip4_sv_reass.c:96
static void * vlib_frame_vector_args(vlib_frame_t *f)
Get pointer to frame vector data.
Definition: node_funcs.h:297
u32 index
Definition: flow_types.api:221
static_always_inline u32 vlib_buffer_enqueue_to_thread(vlib_main_t *vm, u32 frame_queue_index, u32 *buffer_indices, u16 *thread_indices, u32 n_packets, int drop_on_congestion)
Definition: buffer_node.h:494
ip4_sv_reass_per_thread_t * per_thread_data
Definition: ip4_sv_reass.c:140
#define vnet_buffer(b)
Definition: buffer.h:417
static u8 * format_ip4_sv_reass(u8 *s, va_list *args)
ip4_main_t ip4_main
Global ip4 main structure.
Definition: ip4_forward.c:1105
static u32 vlib_num_workers()
Definition: threads.h:377
#define vec_foreach(var, vec)
Vector iterator.
u16 flags
Copy of main node flags.
Definition: node.h:501
void * vlib_add_trace(vlib_main_t *vm, vlib_node_runtime_t *r, vlib_buffer_t *b, u32 n_data_bytes)
Definition: trace.c:634
static void ip4_sv_reass_free(vlib_main_t *vm, ip4_sv_reass_main_t *rm, ip4_sv_reass_per_thread_t *rt, ip4_sv_reass_t *reass)
Definition: ip4_sv_reass.c:267
static int ip4_rehash_cb(clib_bihash_kv_16_8_t *kv, void *_ctx)
static int ip4_header_bytes(const ip4_header_t *i)
Definition: ip4_packet.h:190
static_always_inline void vlib_get_buffers(vlib_main_t *vm, u32 *bi, vlib_buffer_t **b, int count)
Translate array of buffer indices into buffer pointers.
Definition: buffer_funcs.h:280
#define VLIB_NODE_FLAG_TRACE
Definition: node.h:302
vlib_node_registration_t ip4_sv_reass_node_output_feature
(constructor) VLIB_REGISTER_NODE (ip4_sv_reass_node_output_feature)
Definition: ip4_sv_reass.c:968
static u8 * format_ip4_sv_reass_handoff_trace(u8 *s, va_list *args)
#define CLIB_CACHE_LINE_BYTES
Definition: cache.h:59
uword ip4_sv_reass_custom_register_next_node(uword node_index)
static vlib_buffer_t * vlib_get_buffer(vlib_main_t *vm, u32 buffer_index)
Translate buffer index into buffer pointer.
Definition: buffer_funcs.h:85
ip4_sv_reass_key_t key
Definition: ip4_sv_reass.c:85
static clib_error_t * ip4_sv_reass_init_function(vlib_main_t *vm)
vl_api_interface_index_t sw_if_index
Definition: wireguard.api:34
uword unformat(unformat_input_t *i, const char *fmt,...)
Definition: unformat.c:978
Definition: defs.h:46
int vnet_feature_enable_disable(const char *arc_name, const char *node_name, u32 sw_if_index, int enable_disable, void *feature_config, u32 n_feature_config_bytes)
Definition: feature.c:303
CLIB vectors are ubiquitous dynamically resized arrays with by user defined "headers".