FD.io VPP  v20.09-64-g4f7b92f0a
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);
238  ip4_sv_reass_trace_t *t = vlib_add_trace (vm, node, b, sizeof (t[0]));
239  if (reass)
240  {
241  t->reass_id = reass->id;
242  t->op_id = reass->trace_op_counter;
243  ++reass->trace_op_counter;
244  }
245  t->action = action;
246  t->ip_proto = ip_proto;
247  t->l4_src_port = l4_src_port;
248  t->l4_dst_port = l4_dst_port;
249 #if 0
250  static u8 *s = NULL;
251  s = format (s, "%U", format_ip4_sv_reass_trace, NULL, NULL, t);
252  printf ("%.*s\n", vec_len (s), s);
253  fflush (stdout);
254  vec_reset_length (s);
255 #endif
256 }
257 
258 
259 always_inline void
262 {
264  kv.key[0] = reass->key.as_u64[0];
265  kv.key[1] = reass->key.as_u64[1];
266  clib_bihash_add_del_16_8 (&rm->hash, &kv, 0);
267  vlib_buffer_free (vm, reass->cached_buffers,
268  vec_len (reass->cached_buffers));
269  vec_free (reass->cached_buffers);
270  reass->cached_buffers = NULL;
271  if (~0 != reass->lru_prev)
272  {
273  ip4_sv_reass_t *lru_prev =
274  pool_elt_at_index (rt->pool, reass->lru_prev);
275  lru_prev->lru_next = reass->lru_next;
276  }
277  if (~0 != reass->lru_next)
278  {
279  ip4_sv_reass_t *lru_next =
280  pool_elt_at_index (rt->pool, reass->lru_next);
281  lru_next->lru_prev = reass->lru_prev;
282  }
283  if (rt->lru_first == reass - rt->pool)
284  {
285  rt->lru_first = reass->lru_next;
286  }
287  if (rt->lru_last == reass - rt->pool)
288  {
289  rt->lru_last = reass->lru_prev;
290  }
291  pool_put (rt->pool, reass);
292  --rt->reass_n;
293 }
294 
295 always_inline void
297 {
298  reass->cached_buffers = NULL;
299  reass->is_complete = false;
300 }
301 
305  ip4_sv_reass_kv_t * kv, u8 * do_handoff)
306 {
307  ip4_sv_reass_t *reass = NULL;
308  f64 now = vlib_time_now (vm);
309 
310  if (!clib_bihash_search_16_8 (&rm->hash, &kv->kv, &kv->kv))
311  {
312  if (vm->thread_index != kv->v.thread_index)
313  {
314  *do_handoff = 1;
315  return NULL;
316  }
317  reass = pool_elt_at_index (rt->pool, kv->v.reass_index);
318 
319  if (now > reass->last_heard + rm->timeout)
320  {
321  ip4_sv_reass_free (vm, rm, rt, reass);
322  reass = NULL;
323  }
324  }
325 
326  if (reass)
327  {
328  reass->last_heard = now;
329  return reass;
330  }
331 
332  if (rt->reass_n >= rm->max_reass_n && rm->max_reass_n)
333  {
334  reass = pool_elt_at_index (rt->pool, rt->lru_first);
335  ip4_sv_reass_free (vm, rm, rt, reass);
336  }
337 
338  pool_get (rt->pool, reass);
339  clib_memset (reass, 0, sizeof (*reass));
340  reass->id = ((u64) vm->thread_index * 1000000000) + rt->id_counter;
341  ++rt->id_counter;
342  ip4_sv_reass_init (reass);
343  ++rt->reass_n;
344  reass->lru_prev = reass->lru_next = ~0;
345 
346  if (~0 != rt->lru_last)
347  {
348  ip4_sv_reass_t *lru_last = pool_elt_at_index (rt->pool, rt->lru_last);
349  reass->lru_prev = rt->lru_last;
350  lru_last->lru_next = rt->lru_last = reass - rt->pool;
351  }
352 
353  if (~0 == rt->lru_first)
354  {
355  rt->lru_first = rt->lru_last = reass - rt->pool;
356  }
357 
358  reass->key.as_u64[0] = kv->kv.key[0];
359  reass->key.as_u64[1] = kv->kv.key[1];
360  kv->v.reass_index = (reass - rt->pool);
361  kv->v.thread_index = vm->thread_index;
362  reass->last_heard = now;
363 
364  if (clib_bihash_add_del_16_8 (&rm->hash, &kv->kv, 1))
365  {
366  ip4_sv_reass_free (vm, rm, rt, reass);
367  reass = NULL;
368  }
369 
370  return reass;
371 }
372 
373 always_inline ip4_sv_reass_rc_t
376  ip4_header_t * ip0, ip4_sv_reass_t * reass, u32 bi0)
377 {
378  vlib_buffer_t *b0 = vlib_get_buffer (vm, bi0);
379  ip4_sv_reass_rc_t rc = IP4_SV_REASS_RC_OK;
380  const u32 fragment_first = ip4_get_fragment_offset_bytes (ip0);
381  if (0 == fragment_first)
382  {
383  reass->ip_proto = ip0->protocol;
384  reass->l4_src_port = ip4_get_port (ip0, 1);
385  reass->l4_dst_port = ip4_get_port (ip0, 0);
386  if (!reass->l4_src_port || !reass->l4_dst_port)
388  if (IP_PROTOCOL_TCP == reass->ip_proto)
389  {
390  reass->icmp_type_or_tcp_flags = ((tcp_header_t *) (ip0 + 1))->flags;
391  reass->tcp_ack_number = ((tcp_header_t *) (ip0 + 1))->ack_number;
392  reass->tcp_seq_number = ((tcp_header_t *) (ip0 + 1))->seq_number;
393  }
394  else if (IP_PROTOCOL_ICMP == reass->ip_proto)
395  {
396  reass->icmp_type_or_tcp_flags =
397  ((icmp46_header_t *) (ip0 + 1))->type;
398  }
399  reass->is_complete = true;
400  vlib_buffer_t *b0 = vlib_get_buffer (vm, bi0);
401  if (PREDICT_FALSE (b0->flags & VLIB_BUFFER_IS_TRACED))
402  {
403  ip4_sv_reass_add_trace (vm, node, rm, reass, bi0, REASS_FINISH,
404  reass->ip_proto, reass->l4_src_port,
405  reass->l4_dst_port);
406  }
407  }
408  vec_add1 (reass->cached_buffers, bi0);
409  if (!reass->is_complete)
410  {
411  if (PREDICT_FALSE (b0->flags & VLIB_BUFFER_IS_TRACED))
412  {
413  ip4_sv_reass_add_trace (vm, node, rm, reass, bi0,
414  REASS_FRAGMENT_CACHE, ~0, ~0, ~0);
415  }
416  if (vec_len (reass->cached_buffers) > rm->max_reass_len)
417  {
419  }
420  }
421  return rc;
422 }
423 
426  vlib_frame_t * frame, bool is_feature,
427  bool is_output_feature, bool is_custom)
428 {
429  u32 *from = vlib_frame_vector_args (frame);
430  u32 n_left_from, n_left_to_next, *to_next, next_index;
433  clib_spinlock_lock (&rt->lock);
434 
435  n_left_from = frame->n_vectors;
436  next_index = node->cached_next_index;
437 
438  vlib_buffer_t *bufs[VLIB_FRAME_SIZE], **b;
439  vlib_get_buffers (vm, from, bufs, n_left_from);
440  u16 nexts[VLIB_FRAME_SIZE], *next = nexts;
441  b = bufs;
442 
443  /* optimistic case first - no fragments */
444  while (n_left_from >= 2)
445  {
446  vlib_buffer_t *b0, *b1;
447  u32 next0, next1;
448  b0 = *b;
449  b++;
450  b1 = *b;
451  b++;
452 
453  /* Prefetch next iteration. */
454  if (PREDICT_TRUE (n_left_from >= 4))
455  {
456  vlib_buffer_t *p2, *p3;
457 
458  p2 = *b;
459  p3 = *(b + 1);
460 
461  vlib_prefetch_buffer_header (p2, LOAD);
462  vlib_prefetch_buffer_header (p3, LOAD);
463 
466  }
467 
468  ip4_header_t *ip0 =
470  (is_output_feature ? 1 : 0) *
471  vnet_buffer (b0)->
472  ip.save_rewrite_length);
473  ip4_header_t *ip1 =
475  (is_output_feature ? 1 : 0) *
476  vnet_buffer (b1)->
477  ip.save_rewrite_length);
478  if (PREDICT_FALSE
481  {
482  // fragment found, go slow path
483  b -= 2;
484  if (b - bufs > 0)
485  {
486  vlib_buffer_enqueue_to_next (vm, node, from, (u16 *) nexts,
487  b - bufs);
488  }
489  goto slow_path;
490  }
491  if (is_feature)
492  {
493  vnet_feature_next (&next0, b0);
494  }
495  else
496  {
497  next0 = is_custom ? vnet_buffer (b0)->ip.reass.next_index :
499  }
500  vnet_buffer (b0)->ip.reass.is_non_first_fragment = 0;
501  vnet_buffer (b0)->ip.reass.ip_proto = ip0->protocol;
502  if (IP_PROTOCOL_TCP == ip0->protocol)
503  {
504  vnet_buffer (b0)->ip.reass.icmp_type_or_tcp_flags =
505  ((tcp_header_t *) (ip0 + 1))->flags;
506  vnet_buffer (b0)->ip.reass.tcp_ack_number =
507  ((tcp_header_t *) (ip0 + 1))->ack_number;
508  vnet_buffer (b0)->ip.reass.tcp_seq_number =
509  ((tcp_header_t *) (ip0 + 1))->seq_number;
510  }
511  else if (IP_PROTOCOL_ICMP == ip0->protocol)
512  {
513  vnet_buffer (b0)->ip.reass.icmp_type_or_tcp_flags =
514  ((icmp46_header_t *) (ip0 + 1))->type;
515  }
516  vnet_buffer (b0)->ip.reass.l4_src_port = ip4_get_port (ip0, 1);
517  vnet_buffer (b0)->ip.reass.l4_dst_port = ip4_get_port (ip0, 0);
518  if (PREDICT_FALSE (b0->flags & VLIB_BUFFER_IS_TRACED))
519  {
520  ip4_sv_reass_add_trace (vm, node, rm, NULL, from[(b - 2) - bufs],
522  vnet_buffer (b0)->ip.reass.ip_proto,
523  vnet_buffer (b0)->ip.reass.l4_src_port,
524  vnet_buffer (b0)->ip.reass.l4_dst_port);
525  }
526  if (is_feature)
527  {
528  vnet_feature_next (&next1, b1);
529  }
530  else
531  {
532  next1 = is_custom ? vnet_buffer (b1)->ip.reass.next_index :
534  }
535  vnet_buffer (b1)->ip.reass.is_non_first_fragment = 0;
536  vnet_buffer (b1)->ip.reass.ip_proto = ip1->protocol;
537  if (IP_PROTOCOL_TCP == ip1->protocol)
538  {
539  vnet_buffer (b1)->ip.reass.icmp_type_or_tcp_flags =
540  ((tcp_header_t *) (ip1 + 1))->flags;
541  vnet_buffer (b1)->ip.reass.tcp_ack_number =
542  ((tcp_header_t *) (ip1 + 1))->ack_number;
543  vnet_buffer (b1)->ip.reass.tcp_seq_number =
544  ((tcp_header_t *) (ip1 + 1))->seq_number;
545  }
546  else if (IP_PROTOCOL_ICMP == ip1->protocol)
547  {
548  vnet_buffer (b1)->ip.reass.icmp_type_or_tcp_flags =
549  ((icmp46_header_t *) (ip1 + 1))->type;
550  }
551  vnet_buffer (b1)->ip.reass.l4_src_port = ip4_get_port (ip1, 1);
552  vnet_buffer (b1)->ip.reass.l4_dst_port = ip4_get_port (ip1, 0);
553  if (PREDICT_FALSE (b1->flags & VLIB_BUFFER_IS_TRACED))
554  {
555  ip4_sv_reass_add_trace (vm, node, rm, NULL, from[(b - 1) - bufs],
557  vnet_buffer (b1)->ip.reass.ip_proto,
558  vnet_buffer (b1)->ip.reass.l4_src_port,
559  vnet_buffer (b1)->ip.reass.l4_dst_port);
560  }
561 
562  n_left_from -= 2;
563  next[0] = next0;
564  next[1] = next1;
565  next += 2;
566  }
567 
568  while (n_left_from > 0)
569  {
570  vlib_buffer_t *b0;
571  u32 next0;
572  b0 = *b;
573  b++;
574 
575  ip4_header_t *ip0 =
577  (is_output_feature ? 1 : 0) *
578  vnet_buffer (b0)->
579  ip.save_rewrite_length);
580  if (PREDICT_FALSE
582  {
583  // fragment found, go slow path
584  b -= 1;
585  if (b - bufs > 0)
586  {
587  vlib_buffer_enqueue_to_next (vm, node, from, (u16 *) nexts,
588  b - bufs);
589  }
590  goto slow_path;
591  }
592  if (is_feature)
593  {
594  vnet_feature_next (&next0, b0);
595  }
596  else
597  {
598  next0 =
599  is_custom ? vnet_buffer (b0)->ip.
600  reass.next_index : IP4_SV_REASSEMBLY_NEXT_INPUT;
601  }
602  vnet_buffer (b0)->ip.reass.is_non_first_fragment = 0;
603  vnet_buffer (b0)->ip.reass.ip_proto = ip0->protocol;
604  if (IP_PROTOCOL_TCP == ip0->protocol)
605  {
606  vnet_buffer (b0)->ip.reass.icmp_type_or_tcp_flags =
607  ((tcp_header_t *) (ip0 + 1))->flags;
608  vnet_buffer (b0)->ip.reass.tcp_ack_number =
609  ((tcp_header_t *) (ip0 + 1))->ack_number;
610  vnet_buffer (b0)->ip.reass.tcp_seq_number =
611  ((tcp_header_t *) (ip0 + 1))->seq_number;
612  }
613  else if (IP_PROTOCOL_ICMP == ip0->protocol)
614  {
615  vnet_buffer (b0)->ip.reass.icmp_type_or_tcp_flags =
616  ((icmp46_header_t *) (ip0 + 1))->type;
617  }
618  vnet_buffer (b0)->ip.reass.l4_src_port = ip4_get_port (ip0, 1);
619  vnet_buffer (b0)->ip.reass.l4_dst_port = ip4_get_port (ip0, 0);
620  if (PREDICT_FALSE (b0->flags & VLIB_BUFFER_IS_TRACED))
621  {
622  ip4_sv_reass_add_trace (vm, node, rm, NULL, from[(b - 1) - bufs],
624  vnet_buffer (b0)->ip.reass.ip_proto,
625  vnet_buffer (b0)->ip.reass.l4_src_port,
626  vnet_buffer (b0)->ip.reass.l4_dst_port);
627  }
628 
629  n_left_from -= 1;
630  next[0] = next0;
631  next += 1;
632  }
633 
634  vlib_buffer_enqueue_to_next (vm, node, from, (u16 *) nexts,
635  frame->n_vectors);
636 
637  goto done;
638 
639 slow_path:
640 
641  from += b - bufs;
642 
643  while (n_left_from > 0)
644  {
645  vlib_get_next_frame (vm, node, next_index, to_next, n_left_to_next);
646 
647  while (n_left_from > 0 && n_left_to_next > 0)
648  {
649  u32 bi0;
650  vlib_buffer_t *b0;
651  u32 next0;
652  u32 error0 = IP4_ERROR_NONE;
653 
654  bi0 = from[0];
655  b0 = vlib_get_buffer (vm, bi0);
656 
657  ip4_header_t *ip0 =
659  (is_output_feature ? 1 : 0) *
660  vnet_buffer (b0)->
661  ip.save_rewrite_length);
662  if (!ip4_get_fragment_more (ip0) && !ip4_get_fragment_offset (ip0))
663  {
664  // this is a regular packet - no fragmentation
665  if (is_custom)
666  {
667  next0 = vnet_buffer (b0)->ip.reass.next_index;
668  }
669  else
670  {
672  }
673  vnet_buffer (b0)->ip.reass.is_non_first_fragment = 0;
674  vnet_buffer (b0)->ip.reass.ip_proto = ip0->protocol;
675  if (IP_PROTOCOL_TCP == ip0->protocol)
676  {
677  vnet_buffer (b0)->ip.reass.icmp_type_or_tcp_flags =
678  ((tcp_header_t *) (ip0 + 1))->flags;
679  vnet_buffer (b0)->ip.reass.tcp_ack_number =
680  ((tcp_header_t *) (ip0 + 1))->ack_number;
681  vnet_buffer (b0)->ip.reass.tcp_seq_number =
682  ((tcp_header_t *) (ip0 + 1))->seq_number;
683  }
684  else if (IP_PROTOCOL_ICMP == ip0->protocol)
685  {
686  vnet_buffer (b0)->ip.reass.icmp_type_or_tcp_flags =
687  ((icmp46_header_t *) (ip0 + 1))->type;
688  }
689  vnet_buffer (b0)->ip.reass.l4_src_port = ip4_get_port (ip0, 1);
690  vnet_buffer (b0)->ip.reass.l4_dst_port = ip4_get_port (ip0, 0);
691  if (PREDICT_FALSE (b0->flags & VLIB_BUFFER_IS_TRACED))
692  {
693  ip4_sv_reass_add_trace (vm, node, rm, NULL, bi0,
695  vnet_buffer (b0)->ip.reass.ip_proto,
696  vnet_buffer (b0)->ip.
697  reass.l4_src_port,
698  vnet_buffer (b0)->ip.
699  reass.l4_dst_port);
700  }
701  goto packet_enqueue;
702  }
703  const u32 fragment_first = ip4_get_fragment_offset_bytes (ip0);
704  const u32 fragment_length =
705  clib_net_to_host_u16 (ip0->length) - ip4_header_bytes (ip0);
706  const u32 fragment_last = fragment_first + fragment_length - 1;
707  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
708  {
710  error0 = IP4_ERROR_REASS_MALFORMED_PACKET;
711  b0->error = node->errors[error0];
712  goto packet_enqueue;
713  }
715  u8 do_handoff = 0;
716 
717  kv.k.as_u64[0] =
719  vnet_buffer (b0)->sw_if_index[VLIB_RX]) |
720  (u64) ip0->src_address.as_u32 << 32;
721  kv.k.as_u64[1] =
722  (u64) ip0->dst_address.
723  as_u32 | (u64) ip0->fragment_id << 32 | (u64) ip0->protocol << 48;
724 
725  ip4_sv_reass_t *reass =
726  ip4_sv_reass_find_or_create (vm, rm, rt, &kv, &do_handoff);
727 
728  if (PREDICT_FALSE (do_handoff))
729  {
731  vnet_buffer (b0)->ip.reass.owner_thread_index =
732  kv.v.thread_index;
733  goto packet_enqueue;
734  }
735 
736  if (!reass)
737  {
739  error0 = IP4_ERROR_REASS_LIMIT_REACHED;
740  b0->error = node->errors[error0];
741  goto packet_enqueue;
742  }
743 
744  if (reass->is_complete)
745  {
746  if (is_custom)
747  {
748  next0 = vnet_buffer (b0)->ip.reass.next_index;
749  }
750  else
751  {
753  }
754  vnet_buffer (b0)->ip.reass.is_non_first_fragment =
755  ! !fragment_first;
756  vnet_buffer (b0)->ip.reass.ip_proto = reass->ip_proto;
757  vnet_buffer (b0)->ip.reass.icmp_type_or_tcp_flags =
758  reass->icmp_type_or_tcp_flags;
759  vnet_buffer (b0)->ip.reass.tcp_ack_number =
760  reass->tcp_ack_number;
761  vnet_buffer (b0)->ip.reass.tcp_seq_number =
762  reass->tcp_seq_number;
763  vnet_buffer (b0)->ip.reass.l4_src_port = reass->l4_src_port;
764  vnet_buffer (b0)->ip.reass.l4_dst_port = reass->l4_dst_port;
765  if (PREDICT_FALSE (b0->flags & VLIB_BUFFER_IS_TRACED))
766  {
767  ip4_sv_reass_add_trace (vm, node, rm, reass, bi0,
769  reass->ip_proto,
770  reass->l4_src_port,
771  reass->l4_dst_port);
772  }
773  goto packet_enqueue;
774  }
775 
776  ip4_sv_reass_rc_t rc =
777  ip4_sv_reass_update (vm, node, rm, rt, ip0, reass, bi0);
778  switch (rc)
779  {
780  case IP4_SV_REASS_RC_OK:
781  /* nothing to do here */
782  break;
785  IP4_ERROR_REASS_FRAGMENT_CHAIN_TOO_LONG,
786  1);
787  ip4_sv_reass_free (vm, rm, rt, reass);
788  goto next_packet;
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;
797  }
798  if (reass->is_complete)
799  {
800  u32 idx;
801  vec_foreach_index (idx, reass->cached_buffers)
802  {
803  u32 bi0 = vec_elt (reass->cached_buffers, idx);
804  vlib_buffer_t *b0 = vlib_get_buffer (vm, bi0);
805  ip0 =
807  (is_output_feature ? 1 : 0) *
808  vnet_buffer (b0)->
809  ip.save_rewrite_length);
811  if (is_feature)
812  {
813  vnet_feature_next (&next0, b0);
814  }
815  if (is_custom)
816  {
817  next0 = vnet_buffer (b0)->ip.reass.next_index;
818  }
819  if (0 == n_left_to_next)
820  {
821  vlib_put_next_frame (vm, node, next_index,
822  n_left_to_next);
823  vlib_get_next_frame (vm, node, next_index, to_next,
824  n_left_to_next);
825  }
826  to_next[0] = bi0;
827  to_next += 1;
828  n_left_to_next -= 1;
829  vnet_buffer (b0)->ip.reass.is_non_first_fragment =
830  ! !ip4_get_fragment_offset (ip0);
831  vnet_buffer (b0)->ip.reass.ip_proto = reass->ip_proto;
832  vnet_buffer (b0)->ip.reass.icmp_type_or_tcp_flags =
833  reass->icmp_type_or_tcp_flags;
834  vnet_buffer (b0)->ip.reass.tcp_ack_number =
835  reass->tcp_ack_number;
836  vnet_buffer (b0)->ip.reass.tcp_seq_number =
837  reass->tcp_seq_number;
838  vnet_buffer (b0)->ip.reass.l4_src_port = reass->l4_src_port;
839  vnet_buffer (b0)->ip.reass.l4_dst_port = reass->l4_dst_port;
840  if (PREDICT_FALSE (b0->flags & VLIB_BUFFER_IS_TRACED))
841  {
842  ip4_sv_reass_add_trace (vm, node, rm, reass, bi0,
844  reass->ip_proto,
845  reass->l4_src_port,
846  reass->l4_dst_port);
847  }
848  vlib_validate_buffer_enqueue_x1 (vm, node, next_index,
849  to_next, n_left_to_next, bi0,
850  next0);
851  }
852  _vec_len (reass->cached_buffers) = 0; // buffers are owned by frame now
853  }
854  goto next_packet;
855 
856  packet_enqueue:
857  to_next[0] = bi0;
858  to_next += 1;
859  n_left_to_next -= 1;
860  if (is_feature && IP4_ERROR_NONE == error0)
861  {
862  b0 = vlib_get_buffer (vm, bi0);
863  vnet_feature_next (&next0, b0);
864  }
865  vlib_validate_buffer_enqueue_x1 (vm, node, next_index,
866  to_next, n_left_to_next,
867  bi0, next0);
868 
869  next_packet:
870  from += 1;
871  n_left_from -= 1;
872  }
873 
874  vlib_put_next_frame (vm, node, next_index, n_left_to_next);
875  }
876 
877 done:
878  clib_spinlock_unlock (&rt->lock);
879  return frame->n_vectors;
880 }
881 
882 static char *ip4_sv_reass_error_strings[] = {
883 #define _(sym, string) string,
885 #undef _
886 };
887 
888 VLIB_NODE_FN (ip4_sv_reass_node) (vlib_main_t * vm,
891 {
892  return ip4_sv_reass_inline (vm, node, frame, false /* is_feature */ ,
893  false /* is_output_feature */ ,
894  false /* is_custom */ );
895 }
896 
897 /* *INDENT-OFF* */
898 VLIB_REGISTER_NODE (ip4_sv_reass_node) = {
899  .name = "ip4-sv-reassembly",
900  .vector_size = sizeof (u32),
901  .format_trace = format_ip4_sv_reass_trace,
902  .n_errors = ARRAY_LEN (ip4_sv_reass_error_strings),
903  .error_strings = ip4_sv_reass_error_strings,
904  .n_next_nodes = IP4_SV_REASSEMBLY_N_NEXT,
905  .next_nodes =
906  {
907  [IP4_SV_REASSEMBLY_NEXT_INPUT] = "ip4-input",
908  [IP4_SV_REASSEMBLY_NEXT_DROP] = "ip4-drop",
909  [IP4_SV_REASSEMBLY_NEXT_HANDOFF] = "ip4-sv-reassembly-handoff",
910 
911  },
912 };
913 /* *INDENT-ON* */
914 
915 VLIB_NODE_FN (ip4_sv_reass_node_feature) (vlib_main_t * vm,
918 {
919  return ip4_sv_reass_inline (vm, node, frame, true /* is_feature */ ,
920  false /* is_output_feature */ ,
921  false /* is_custom */ );
922 }
923 
924 /* *INDENT-OFF* */
925 VLIB_REGISTER_NODE (ip4_sv_reass_node_feature) = {
926  .name = "ip4-sv-reassembly-feature",
927  .vector_size = sizeof (u32),
928  .format_trace = format_ip4_sv_reass_trace,
929  .n_errors = ARRAY_LEN (ip4_sv_reass_error_strings),
930  .error_strings = ip4_sv_reass_error_strings,
931  .n_next_nodes = IP4_SV_REASSEMBLY_N_NEXT,
932  .next_nodes =
933  {
934  [IP4_SV_REASSEMBLY_NEXT_INPUT] = "ip4-input",
935  [IP4_SV_REASSEMBLY_NEXT_DROP] = "ip4-drop",
936  [IP4_SV_REASSEMBLY_NEXT_HANDOFF] = "ip4-sv-reass-feature-hoff",
937  },
938 };
939 /* *INDENT-ON* */
940 
941 /* *INDENT-OFF* */
942 VNET_FEATURE_INIT (ip4_sv_reass_feature) = {
943  .arc_name = "ip4-unicast",
944  .node_name = "ip4-sv-reassembly-feature",
945  .runs_before = VNET_FEATURES ("ip4-lookup"),
946  .runs_after = 0,
947 };
948 /* *INDENT-ON* */
949 
953 {
954  return ip4_sv_reass_inline (vm, node, frame, true /* is_feature */ ,
955  true /* is_output_feature */ ,
956  false /* is_custom */ );
957 }
958 
959 
960 /* *INDENT-OFF* */
962  .name = "ip4-sv-reassembly-output-feature",
963  .vector_size = sizeof (u32),
964  .format_trace = format_ip4_sv_reass_trace,
965  .n_errors = ARRAY_LEN (ip4_sv_reass_error_strings),
966  .error_strings = ip4_sv_reass_error_strings,
967  .n_next_nodes = IP4_SV_REASSEMBLY_N_NEXT,
968  .next_nodes =
969  {
970  [IP4_SV_REASSEMBLY_NEXT_INPUT] = "ip4-input",
971  [IP4_SV_REASSEMBLY_NEXT_DROP] = "ip4-drop",
972  [IP4_SV_REASSEMBLY_NEXT_HANDOFF] = "ip4-sv-reass-feature-hoff",
973  },
974 };
975 /* *INDENT-ON* */
976 
977 /* *INDENT-OFF* */
978 VNET_FEATURE_INIT (ip4_sv_reass_output_feature) = {
979  .arc_name = "ip4-output",
980  .node_name = "ip4-sv-reassembly-output-feature",
981  .runs_before = 0,
982  .runs_after = 0,
983 };
984 /* *INDENT-ON* */
985 
986 /* *INDENT-OFF* */
988  .name = "ip4-sv-reassembly-custom-next",
989  .vector_size = sizeof (u32),
990  .format_trace = format_ip4_sv_reass_trace,
991  .n_errors = ARRAY_LEN (ip4_sv_reass_error_strings),
992  .error_strings = ip4_sv_reass_error_strings,
993  .n_next_nodes = IP4_SV_REASSEMBLY_N_NEXT,
994  .next_nodes =
995  {
996  [IP4_SV_REASSEMBLY_NEXT_INPUT] = "ip4-input",
997  [IP4_SV_REASSEMBLY_NEXT_DROP] = "ip4-drop",
998  [IP4_SV_REASSEMBLY_NEXT_HANDOFF] = "ip4-sv-reassembly-handoff",
999 
1000  },
1001 };
1002 /* *INDENT-ON* */
1003 
1006  vlib_frame_t * frame)
1007 {
1008  return ip4_sv_reass_inline (vm, node, frame, false /* is_feature */ ,
1009  false /* is_output_feature */ ,
1010  true /* is_custom */ );
1011 }
1012 
1013 #ifndef CLIB_MARCH_VARIANT
1016 {
1018  u32 nbuckets;
1019  u8 i;
1020 
1021  nbuckets = (u32) (rm->max_reass_n / IP4_SV_REASS_HT_LOAD_FACTOR);
1022 
1023  for (i = 0; i < 31; i++)
1024  if ((1 << i) >= nbuckets)
1025  break;
1026  nbuckets = 1 << i;
1027 
1028  return nbuckets;
1029 }
1030 #endif /* CLIB_MARCH_VARIANT */
1031 
1032 typedef enum
1033 {
1036 
1037 typedef struct
1038 {
1039  int failure;
1040  clib_bihash_16_8_t *new_hash;
1042 
1043 #ifndef CLIB_MARCH_VARIANT
1044 static int
1046 {
1047  ip4_rehash_cb_ctx *ctx = _ctx;
1048  if (clib_bihash_add_del_16_8 (ctx->new_hash, kv, 1))
1049  {
1050  ctx->failure = 1;
1051  }
1052  return (BIHASH_WALK_CONTINUE);
1053 }
1054 
1055 static void
1056 ip4_sv_reass_set_params (u32 timeout_ms, u32 max_reassemblies,
1057  u32 max_reassembly_length,
1058  u32 expire_walk_interval_ms)
1059 {
1060  ip4_sv_reass_main.timeout_ms = timeout_ms;
1061  ip4_sv_reass_main.timeout = (f64) timeout_ms / (f64) MSEC_PER_SEC;
1062  ip4_sv_reass_main.max_reass_n = max_reassemblies;
1063  ip4_sv_reass_main.max_reass_len = max_reassembly_length;
1064  ip4_sv_reass_main.expire_walk_interval_ms = expire_walk_interval_ms;
1065 }
1066 
1068 ip4_sv_reass_set (u32 timeout_ms, u32 max_reassemblies,
1069  u32 max_reassembly_length, u32 expire_walk_interval_ms)
1070 {
1071  u32 old_nbuckets = ip4_sv_reass_get_nbuckets ();
1072  ip4_sv_reass_set_params (timeout_ms, max_reassemblies,
1073  max_reassembly_length, expire_walk_interval_ms);
1074  vlib_process_signal_event (ip4_sv_reass_main.vlib_main,
1075  ip4_sv_reass_main.ip4_sv_reass_expire_node_idx,
1077  u32 new_nbuckets = ip4_sv_reass_get_nbuckets ();
1078  if (ip4_sv_reass_main.max_reass_n > 0 && new_nbuckets > old_nbuckets)
1079  {
1080  clib_bihash_16_8_t new_hash;
1081  clib_memset (&new_hash, 0, sizeof (new_hash));
1083  ctx.failure = 0;
1084  ctx.new_hash = &new_hash;
1085  clib_bihash_init_16_8 (&new_hash, "ip4-dr", new_nbuckets,
1086  new_nbuckets * 1024);
1087  clib_bihash_foreach_key_value_pair_16_8 (&ip4_sv_reass_main.hash,
1088  ip4_rehash_cb, &ctx);
1089  if (ctx.failure)
1090  {
1091  clib_bihash_free_16_8 (&new_hash);
1092  return -1;
1093  }
1094  else
1095  {
1096  clib_bihash_free_16_8 (&ip4_sv_reass_main.hash);
1097  clib_memcpy_fast (&ip4_sv_reass_main.hash, &new_hash,
1098  sizeof (ip4_sv_reass_main.hash));
1099  clib_bihash_copied (&ip4_sv_reass_main.hash, &new_hash);
1100  }
1101  }
1102  return 0;
1103 }
1104 
1106 ip4_sv_reass_get (u32 * timeout_ms, u32 * max_reassemblies,
1107  u32 * max_reassembly_length, u32 * expire_walk_interval_ms)
1108 {
1109  *timeout_ms = ip4_sv_reass_main.timeout_ms;
1110  *max_reassemblies = ip4_sv_reass_main.max_reass_n;
1111  *max_reassembly_length = ip4_sv_reass_main.max_reass_len;
1112  *expire_walk_interval_ms = ip4_sv_reass_main.expire_walk_interval_ms;
1113  return 0;
1114 }
1115 
1116 static clib_error_t *
1118 {
1120  clib_error_t *error = 0;
1121  u32 nbuckets;
1122  vlib_node_t *node;
1123 
1124  rm->vlib_main = vm;
1125  rm->vnet_main = vnet_get_main ();
1126 
1129  vec_foreach (rt, rm->per_thread_data)
1130  {
1131  clib_spinlock_init (&rt->lock);
1132  pool_alloc (rt->pool, rm->max_reass_n);
1133  rt->lru_first = rt->lru_last = ~0;
1134  }
1135 
1136  node = vlib_get_node_by_name (vm, (u8 *) "ip4-sv-reassembly-expire-walk");
1137  ASSERT (node);
1138  rm->ip4_sv_reass_expire_node_idx = node->index;
1139 
1144 
1145  nbuckets = ip4_sv_reass_get_nbuckets ();
1146  clib_bihash_init_16_8 (&rm->hash, "ip4-dr", nbuckets, nbuckets * 1024);
1147 
1148  node = vlib_get_node_by_name (vm, (u8 *) "ip4-drop");
1149  ASSERT (node);
1150  rm->ip4_drop_idx = node->index;
1151 
1152  rm->fq_index = vlib_frame_queue_main_init (ip4_sv_reass_node.index, 0);
1153  rm->fq_feature_index =
1154  vlib_frame_queue_main_init (ip4_sv_reass_node_feature.index, 0);
1155 
1156  rm->feature_use_refcount_per_intf = NULL;
1158 
1159  return error;
1160 }
1161 
1163 #endif /* CLIB_MARCH_VARIANT */
1164 
1165 static uword
1167  vlib_node_runtime_t * node, vlib_frame_t * f)
1168 {
1170  uword event_type, *event_data = 0;
1171 
1172  while (true)
1173  {
1175  (f64)
1177  (f64) MSEC_PER_SEC);
1178  event_type = vlib_process_get_events (vm, &event_data);
1179 
1180  switch (event_type)
1181  {
1182  case ~0: /* no events => timeout */
1183  /* nothing to do here */
1184  break;
1186  break;
1187  default:
1188  clib_warning ("BUG: event type 0x%wx", event_type);
1189  break;
1190  }
1191  f64 now = vlib_time_now (vm);
1192 
1193  ip4_sv_reass_t *reass;
1194  int *pool_indexes_to_free = NULL;
1195 
1196  uword thread_index = 0;
1197  int index;
1198  const uword nthreads = vlib_num_workers () + 1;
1199  for (thread_index = 0; thread_index < nthreads; ++thread_index)
1200  {
1201  ip4_sv_reass_per_thread_t *rt = &rm->per_thread_data[thread_index];
1202  clib_spinlock_lock (&rt->lock);
1203 
1204  vec_reset_length (pool_indexes_to_free);
1205  /* *INDENT-OFF* */
1206  pool_foreach_index (index, rt->pool, ({
1207  reass = pool_elt_at_index (rt->pool, index);
1208  if (now > reass->last_heard + rm->timeout)
1209  {
1210  vec_add1 (pool_indexes_to_free, index);
1211  }
1212  }));
1213  /* *INDENT-ON* */
1214  int *i;
1215  /* *INDENT-OFF* */
1216  vec_foreach (i, pool_indexes_to_free)
1217  {
1218  ip4_sv_reass_t *reass = pool_elt_at_index (rt->pool, i[0]);
1219  ip4_sv_reass_free (vm, rm, rt, reass);
1220  }
1221  /* *INDENT-ON* */
1222 
1223  clib_spinlock_unlock (&rt->lock);
1224  }
1225 
1226  vec_free (pool_indexes_to_free);
1227  if (event_data)
1228  {
1229  _vec_len (event_data) = 0;
1230  }
1231  }
1232 
1233  return 0;
1234 }
1235 
1236 /* *INDENT-OFF* */
1238  .function = ip4_sv_reass_walk_expired,
1239  .type = VLIB_NODE_TYPE_PROCESS,
1240  .name = "ip4-sv-reassembly-expire-walk",
1241  .format_trace = format_ip4_sv_reass_trace,
1242  .n_errors = ARRAY_LEN (ip4_sv_reass_error_strings),
1243  .error_strings = ip4_sv_reass_error_strings,
1244 
1245 };
1246 /* *INDENT-ON* */
1247 
1248 static u8 *
1249 format_ip4_sv_reass_key (u8 * s, va_list * args)
1250 {
1251  ip4_sv_reass_key_t *key = va_arg (*args, ip4_sv_reass_key_t *);
1252  s =
1253  format (s,
1254  "xx_id: %u, src: %U, dst: %U, frag_id: %u, proto: %u",
1256  &key->dst, clib_net_to_host_u16 (key->frag_id), key->proto);
1257  return s;
1258 }
1259 
1260 static u8 *
1261 format_ip4_sv_reass (u8 * s, va_list * args)
1262 {
1263  vlib_main_t *vm = va_arg (*args, vlib_main_t *);
1264  ip4_sv_reass_t *reass = va_arg (*args, ip4_sv_reass_t *);
1265 
1266  s = format (s, "ID: %lu, key: %U trace_op_counter: %u\n",
1267  reass->id, format_ip4_sv_reass_key, &reass->key,
1268  reass->trace_op_counter);
1269 
1270  vlib_buffer_t *b;
1271  u32 *bip;
1272  u32 counter = 0;
1273  vec_foreach (bip, reass->cached_buffers)
1274  {
1275  u32 bi = *bip;
1276  do
1277  {
1278  b = vlib_get_buffer (vm, bi);
1279  s = format (s, " #%03u: bi: %u, ", counter, bi);
1280  ++counter;
1281  bi = b->next_buffer;
1282  }
1283  while (b->flags & VLIB_BUFFER_NEXT_PRESENT);
1284  }
1285  return s;
1286 }
1287 
1288 static clib_error_t *
1290  unformat_input_t * input,
1292 {
1294 
1295  vlib_cli_output (vm, "---------------------");
1296  vlib_cli_output (vm, "IP4 reassembly status");
1297  vlib_cli_output (vm, "---------------------");
1298  bool details = false;
1299  if (unformat (input, "details"))
1300  {
1301  details = true;
1302  }
1303 
1304  u32 sum_reass_n = 0;
1305  ip4_sv_reass_t *reass;
1306  uword thread_index;
1307  const uword nthreads = vlib_num_workers () + 1;
1308  for (thread_index = 0; thread_index < nthreads; ++thread_index)
1309  {
1310  ip4_sv_reass_per_thread_t *rt = &rm->per_thread_data[thread_index];
1311  clib_spinlock_lock (&rt->lock);
1312  if (details)
1313  {
1314  /* *INDENT-OFF* */
1315  pool_foreach (reass, rt->pool, {
1316  vlib_cli_output (vm, "%U", format_ip4_sv_reass, vm, reass);
1317  });
1318  /* *INDENT-ON* */
1319  }
1320  sum_reass_n += rt->reass_n;
1321  clib_spinlock_unlock (&rt->lock);
1322  }
1323  vlib_cli_output (vm, "---------------------");
1324  vlib_cli_output (vm, "Current IP4 reassemblies count: %lu\n",
1325  (long unsigned) sum_reass_n);
1326  vlib_cli_output (vm,
1327  "Maximum configured concurrent shallow virtual IP4 reassemblies per worker-thread: %lu\n",
1328  (long unsigned) rm->max_reass_n);
1329  vlib_cli_output (vm,
1330  "Maximum configured shallow virtual IP4 reassembly timeout: %lums\n",
1331  (long unsigned) rm->timeout_ms);
1332  vlib_cli_output (vm,
1333  "Maximum configured shallow virtual IP4 reassembly expire walk interval: %lums\n",
1334  (long unsigned) rm->expire_walk_interval_ms);
1335  return 0;
1336 }
1337 
1338 /* *INDENT-OFF* */
1340  .path = "show ip4-sv-reassembly",
1341  .short_help = "show ip4-sv-reassembly [details]",
1342  .function = show_ip4_reass,
1343 };
1344 /* *INDENT-ON* */
1345 
1346 #ifndef CLIB_MARCH_VARIANT
1349 {
1350  return ip4_sv_reass_enable_disable_with_refcnt (sw_if_index,
1351  enable_disable);
1352 }
1353 #endif /* CLIB_MARCH_VARIANT */
1354 
1355 
1356 #define foreach_ip4_sv_reass_handoff_error \
1357 _(CONGESTION_DROP, "congestion drop")
1358 
1359 
1360 typedef enum
1361 {
1362 #define _(sym,str) IP4_SV_REASSEMBLY_HANDOFF_ERROR_##sym,
1364 #undef _
1367 
1369 #define _(sym,string) string,
1371 #undef _
1372 };
1373 
1374 typedef struct
1375 {
1378 
1379 static u8 *
1381 {
1382  CLIB_UNUSED (vlib_main_t * vm) = va_arg (*args, vlib_main_t *);
1383  CLIB_UNUSED (vlib_node_t * node) = va_arg (*args, vlib_node_t *);
1385  va_arg (*args, ip4_sv_reass_handoff_trace_t *);
1386 
1387  s =
1388  format (s, "ip4-sv-reassembly-handoff: next-worker %d",
1389  t->next_worker_index);
1390 
1391  return s;
1392 }
1393 
1397  vlib_frame_t * frame, bool is_feature)
1398 {
1400 
1401  vlib_buffer_t *bufs[VLIB_FRAME_SIZE], **b;
1402  u32 n_enq, n_left_from, *from;
1403  u16 thread_indices[VLIB_FRAME_SIZE], *ti;
1404  u32 fq_index;
1405 
1406  from = vlib_frame_vector_args (frame);
1407  n_left_from = frame->n_vectors;
1408  vlib_get_buffers (vm, from, bufs, n_left_from);
1409 
1410  b = bufs;
1411  ti = thread_indices;
1412 
1413  fq_index = (is_feature) ? rm->fq_feature_index : rm->fq_index;
1414 
1415  while (n_left_from > 0)
1416  {
1417  ti[0] = vnet_buffer (b[0])->ip.reass.owner_thread_index;
1418 
1419  if (PREDICT_FALSE
1420  ((node->flags & VLIB_NODE_FLAG_TRACE)
1421  && (b[0]->flags & VLIB_BUFFER_IS_TRACED)))
1422  {
1424  vlib_add_trace (vm, node, b[0], sizeof (*t));
1425  t->next_worker_index = ti[0];
1426  }
1427 
1428  n_left_from -= 1;
1429  ti += 1;
1430  b += 1;
1431  }
1432  n_enq =
1433  vlib_buffer_enqueue_to_thread (vm, fq_index, from, thread_indices,
1434  frame->n_vectors, 1);
1435 
1436  if (n_enq < frame->n_vectors)
1438  IP4_SV_REASSEMBLY_HANDOFF_ERROR_CONGESTION_DROP,
1439  frame->n_vectors - n_enq);
1440  return frame->n_vectors;
1441 }
1442 
1445  vlib_frame_t * frame)
1446 {
1447  return ip4_sv_reass_handoff_node_inline (vm, node, frame,
1448  false /* is_feature */ );
1449 }
1450 
1451 
1452 /* *INDENT-OFF* */
1454  .name = "ip4-sv-reassembly-handoff",
1455  .vector_size = sizeof (u32),
1456  .n_errors = ARRAY_LEN(ip4_sv_reass_handoff_error_strings),
1457  .error_strings = ip4_sv_reass_handoff_error_strings,
1458  .format_trace = format_ip4_sv_reass_handoff_trace,
1459 
1460  .n_next_nodes = 1,
1461 
1462  .next_nodes = {
1463  [0] = "error-drop",
1464  },
1465 };
1466 /* *INDENT-ON* */
1467 
1468 
1469 /* *INDENT-OFF* */
1472  node,
1473  vlib_frame_t * frame)
1474 {
1475  return ip4_sv_reass_handoff_node_inline (vm, node, frame,
1476  true /* is_feature */ );
1477 }
1478 /* *INDENT-ON* */
1479 
1480 
1481 /* *INDENT-OFF* */
1483  .name = "ip4-sv-reass-feature-hoff",
1484  .vector_size = sizeof (u32),
1485  .n_errors = ARRAY_LEN(ip4_sv_reass_handoff_error_strings),
1486  .error_strings = ip4_sv_reass_handoff_error_strings,
1487  .format_trace = format_ip4_sv_reass_handoff_trace,
1488 
1489  .n_next_nodes = 1,
1490 
1491  .next_nodes = {
1492  [0] = "error-drop",
1493  },
1494 };
1495 /* *INDENT-ON* */
1496 
1497 #ifndef CLIB_MARCH_VARIANT
1498 int
1500 {
1502  vec_validate (rm->feature_use_refcount_per_intf, sw_if_index);
1503  if (is_enable)
1504  {
1505  if (!rm->feature_use_refcount_per_intf[sw_if_index])
1506  {
1508  return vnet_feature_enable_disable ("ip4-unicast",
1509  "ip4-sv-reassembly-feature",
1510  sw_if_index, 1, 0, 0);
1511  }
1513  }
1514  else
1515  {
1516  if (rm->feature_use_refcount_per_intf[sw_if_index])
1518  if (!rm->feature_use_refcount_per_intf[sw_if_index])
1519  return vnet_feature_enable_disable ("ip4-unicast",
1520  "ip4-sv-reassembly-feature",
1521  sw_if_index, 0, 0, 0);
1522  }
1523  return 0;
1524 }
1525 
1526 uword
1528 {
1530  node_index);
1531 }
1532 
1533 int
1535  int is_enable)
1536 {
1539  if (is_enable)
1540  {
1541  if (!rm->output_feature_use_refcount_per_intf[sw_if_index])
1542  {
1544  return vnet_feature_enable_disable ("ip4-output",
1545  "ip4-sv-reassembly-output-feature",
1546  sw_if_index, 1, 0, 0);
1547  }
1549  }
1550  else
1551  {
1552  if (rm->output_feature_use_refcount_per_intf[sw_if_index])
1554  if (!rm->output_feature_use_refcount_per_intf[sw_if_index])
1555  return vnet_feature_enable_disable ("ip4-output",
1556  "ip4-sv-reassembly-output-feature",
1557  sw_if_index, 0, 0, 0);
1558  }
1559  return 0;
1560 }
1561 #endif
1562 
1563 /*
1564  * fd.io coding-style-patch-verification: ON
1565  *
1566  * Local Variables:
1567  * eval: (c-set-style "gnu")
1568  * End:
1569  */
#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:119
static vlib_cli_command_t show_ip4_sv_reass_cmd
(constructor) VLIB_CLI_COMMAND (show_ip4_sv_reass_cmd)
static_always_inline void clib_spinlock_lock(clib_spinlock_t *p)
Definition: lock.h:80
#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:425
vl_api_wireguard_peer_flags_t flags
Definition: wireguard.api:103
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
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:374
#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:987
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_alloc(P, N)
Allocate N more free elements to pool (unspecified alignment).
Definition: pool.h:361
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:121
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:925
ip_proto
Definition: ip_types.api:64
static void ip4_sv_reass_init(ip4_sv_reass_t *reass)
Definition: ip4_sv_reass.c:296
clib_memset(h->entries, 0, sizeof(h->entries[0]) *entries)
u32 index
Definition: node.h:279
static f64 vlib_time_now(vlib_main_t *vm)
Definition: main.h:333
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:1873
u32 thread_index
Definition: main.h:249
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:1582
u32 * fib_index_by_sw_if_index
Table index indexed by software interface.
Definition: ip4.h:122
u8 * format(u8 *s, const char *fmt,...)
Definition: format.c:424
vnet_api_error_t ip4_sv_reass_enable_disable(u32 sw_if_index, u8 enable_disable)
#define VLIB_NODE_FN(node)
Definition: node.h:202
vlib_error_t * errors
Vector of errors for this node.
Definition: node.h:469
#define pool_get(P, E)
Allocate an object E from a pool P (unspecified alignment).
Definition: pool.h:252
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:43
format_function_t format_ip4_address
Definition: format.h:73
#define foreach_ip4_sv_reass_handoff_error
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:303
#define pool_foreach(VAR, POOL, BODY)
Iterate through pool.
Definition: pool.h:513
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
vlib_node_registration_t ip4_sv_reass_feature_handoff_node
(constructor) VLIB_REGISTER_NODE (ip4_sv_reass_feature_handoff_node)
#define vlib_prefetch_buffer_header(b, type)
Prefetch buffer metadata.
Definition: buffer.h:203
clib_bihash_16_8_t * new_hash
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:377
ip4_sv_reass_handoff_error_t
static char * ip4_sv_reass_error_strings[]
Definition: ip4_sv_reass.c:882
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:63
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:898
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
#define pool_elt_at_index(p, i)
Returns pointer to element at given index.
Definition: pool.h:534
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:229
#define pool_put(P, E)
Free an object E in pool P.
Definition: pool.h:302
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:120
#define always_inline
Definition: ipsec.h:28
u32 node_index
Node index.
Definition: node.h:487
#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:169
u16 n_vectors
Definition: node.h:396
#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
static clib_error_t * show_ip4_reass(vlib_main_t *vm, unformat_input_t *input, CLIB_UNUSED(vlib_cli_command_t *lmd))
#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:1582
#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:510
#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
#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.
void clib_bihash_copied(void *dst, void *src)
typedef key
Definition: ipsec_types.api:85
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:1583
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:1144
vlib_node_registration_t ip4_sv_reass_handoff_node
(constructor) VLIB_REGISTER_NODE (ip4_sv_reass_handoff_node)
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:500
void * vlib_add_trace(vlib_main_t *vm, vlib_node_runtime_t *r, vlib_buffer_t *b, u32 n_data_bytes)
Definition: trace.c:577
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:260
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
#define pool_foreach_index(i, v, body)
Iterate pool by index.
Definition: pool.h:558
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:301
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:961
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:33
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".