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 IPv6 Full Reassembly.
19 *
20 * This file contains the source code for IPv6 full reassembly.
21 */
22
23#include <vppinfra/vec.h>
24#include <vnet/vnet.h>
25#include <vnet/ip/ip.h>
26#include <vppinfra/bihash_48_8.h>
27#include <vnet/ip/reass/ip6_full_reass.h>
28
29#define MSEC_PER_SEC 1000
30#define IP6_FULL_REASS_TIMEOUT_DEFAULT_MS 100
31#define IP6_FULL_REASS_EXPIRE_WALK_INTERVAL_DEFAULT_MS 10000	// 10 seconds default
32#define IP6_FULL_REASS_MAX_REASSEMBLIES_DEFAULT 1024
33#define IP6_FULL_REASS_MAX_REASSEMBLY_LENGTH_DEFAULT 3
34#define IP6_FULL_REASS_HT_LOAD_FACTOR (0.75)
35
36typedef enum
37{
38  IP6_FULL_REASS_RC_OK,
39  IP6_FULL_REASS_RC_INTERNAL_ERROR,
40  IP6_FULL_REASS_RC_TOO_MANY_FRAGMENTS,
41  IP6_FULL_REASS_RC_NO_BUF,
42  IP6_FULL_REASS_RC_HANDOFF,
43} ip6_full_reass_rc_t;
44
45typedef struct
46{
47  union
48  {
49    struct
50    {
51      ip6_address_t src;
52      ip6_address_t dst;
53      u32 xx_id;
54      u32 frag_id;
55      u8 unused[7];
56      u8 proto;
57    };
58    u64 as_u64[6];
59  };
60} ip6_full_reass_key_t;
61
62typedef union
63{
64  struct
65  {
66    u32 reass_index;
67    u32 memory_owner_thread_index;
68  };
69  u64 as_u64;
70} ip6_full_reass_val_t;
71
72typedef union
73{
74  struct
75  {
76    ip6_full_reass_key_t k;
77    ip6_full_reass_val_t v;
78  };
79  clib_bihash_kv_48_8_t kv;
80} ip6_full_reass_kv_t;
81
82
83always_inline u32
84ip6_full_reass_buffer_get_data_offset (vlib_buffer_t * b)
85{
86  vnet_buffer_opaque_t *vnb = vnet_buffer (b);
87  return vnb->ip.reass.range_first - vnb->ip.reass.fragment_first;
88}
89
90always_inline u16
91ip6_full_reass_buffer_get_data_len (vlib_buffer_t * b)
92{
93  vnet_buffer_opaque_t *vnb = vnet_buffer (b);
94  return clib_min (vnb->ip.reass.range_last, vnb->ip.reass.fragment_last) -
95    (vnb->ip.reass.fragment_first +
96     ip6_full_reass_buffer_get_data_offset (b)) + 1;
97}
98
99typedef struct
100{
101  // hash table key
102  ip6_full_reass_key_t key;
103  // time when last packet was received
104  f64 last_heard;
105  // internal id of this reassembly
106  u64 id;
107  // buffer index of first buffer in this reassembly context
108  u32 first_bi;
109  // last octet of packet, ~0 until fragment without more_fragments arrives
110  u32 last_packet_octet;
111  // length of data collected so far
112  u32 data_len;
113  // trace operation counter
114  u32 trace_op_counter;
115  // next index - used by custom apps (~0 if not set)
116  u32 next_index;
117  // error next index - used by custom apps (~0 if not set)
118  u32 error_next_index;
119  // minimum fragment length for this reassembly - used to estimate MTU
120  u16 min_fragment_length;
121  // number of fragments for this reassembly
122  u32 fragments_n;
123  // thread owning memory for this context (whose pool contains this ctx)
124  u32 memory_owner_thread_index;
125  // thread which received fragment with offset 0 and which sends out the
126  // completed reassembly
127  u32 sendout_thread_index;
128} ip6_full_reass_t;
129
130typedef struct
131{
132  ip6_full_reass_t *pool;
133  u32 reass_n;
134  u32 id_counter;
135  clib_spinlock_t lock;
136} ip6_full_reass_per_thread_t;
137
138typedef struct
139{
140  // IPv6 config
141  u32 timeout_ms;
142  f64 timeout;
143  u32 expire_walk_interval_ms;
144  // maximum number of fragments in one reassembly
145  u32 max_reass_len;
146  // maximum number of reassemblies
147  u32 max_reass_n;
148
149  // IPv6 runtime
150  clib_bihash_48_8_t hash;
151
152  // per-thread data
153  ip6_full_reass_per_thread_t *per_thread_data;
154
155  // convenience
156  vlib_main_t *vlib_main;
157
158  // node index of ip6-drop node
159  u32 ip6_drop_idx;
160  u32 ip6_icmp_error_idx;
161  u32 ip6_full_reass_expire_node_idx;
162
163  /** Worker handoff */
164  u32 fq_index;
165  u32 fq_feature_index;
166
167  // reference count for enabling/disabling feature - per interface
168  u32 *feature_use_refcount_per_intf;
169} ip6_full_reass_main_t;
170
171extern ip6_full_reass_main_t ip6_full_reass_main;
172
173#ifndef CLIB_MARCH_VARIANT
174ip6_full_reass_main_t ip6_full_reass_main;
175#endif /* CLIB_MARCH_VARIANT */
176
177typedef enum
178{
179  IP6_FULL_REASSEMBLY_NEXT_INPUT,
180  IP6_FULL_REASSEMBLY_NEXT_DROP,
181  IP6_FULL_REASSEMBLY_NEXT_ICMP_ERROR,
182  IP6_FULL_REASSEMBLY_NEXT_HANDOFF,
183  IP6_FULL_REASSEMBLY_N_NEXT,
184} ip6_full_reass_next_t;
185
186typedef enum
187{
188  RANGE_NEW,
189  RANGE_OVERLAP,
190  ICMP_ERROR_RT_EXCEEDED,
191  ICMP_ERROR_FL_TOO_BIG,
192  ICMP_ERROR_FL_NOT_MULT_8,
193  FINALIZE,
194  HANDOFF,
195} ip6_full_reass_trace_operation_e;
196
197typedef struct
198{
199  u16 range_first;
200  u16 range_last;
201  u32 range_bi;
202  i32 data_offset;
203  u32 data_len;
204  u32 first_bi;
205} ip6_full_reass_range_trace_t;
206
207typedef struct
208{
209  ip6_full_reass_trace_operation_e action;
210  u32 reass_id;
211  ip6_full_reass_range_trace_t trace_range;
212  u32 op_id;
213  u32 fragment_first;
214  u32 fragment_last;
215  u32 total_data_len;
216  u32 thread_id;
217  u32 thread_id_to;
218  bool is_after_handoff;
219  ip6_header_t ip6_header;
220  ip6_frag_hdr_t ip6_frag_header;
221} ip6_full_reass_trace_t;
222
223static void
224ip6_full_reass_trace_details (vlib_main_t * vm, u32 bi,
225			      ip6_full_reass_range_trace_t * trace)
226{
227  vlib_buffer_t *b = vlib_get_buffer (vm, bi);
228  vnet_buffer_opaque_t *vnb = vnet_buffer (b);
229  trace->range_first = vnb->ip.reass.range_first;
230  trace->range_last = vnb->ip.reass.range_last;
231  trace->data_offset = ip6_full_reass_buffer_get_data_offset (b);
232  trace->data_len = ip6_full_reass_buffer_get_data_len (b);
233  trace->range_bi = bi;
234}
235
236static u8 *
237format_ip6_full_reass_range_trace (u8 * s, va_list * args)
238{
239  ip6_full_reass_range_trace_t *trace =
240    va_arg (*args, ip6_full_reass_range_trace_t *);
241  s =
242    format (s, "range: [%u, %u], off %d, len %u, bi %u", trace->range_first,
243	    trace->range_last, trace->data_offset, trace->data_len,
244	    trace->range_bi);
245  return s;
246}
247
248static u8 *
249format_ip6_full_reass_trace (u8 * s, va_list * args)
250{
251  CLIB_UNUSED (vlib_main_t * vm) = va_arg (*args, vlib_main_t *);
252  CLIB_UNUSED (vlib_node_t * node) = va_arg (*args, vlib_node_t *);
253  ip6_full_reass_trace_t *t = va_arg (*args, ip6_full_reass_trace_t *);
254  u32 indent = 0;
255  if (~0 != t->reass_id)
256    {
257      if (t->is_after_handoff)
258	{
259	  s =
260	    format (s, "%U\n", format_ip6_header, &t->ip6_header,
261		    sizeof (t->ip6_header));
262	  s =
263	    format (s, "  %U\n", format_ip6_frag_hdr, &t->ip6_frag_header,
264		    sizeof (t->ip6_frag_header));
265	  indent = 2;
266	}
267      s =
268	format (s, "%Ureass id: %u, op id: %u, ", format_white_space, indent,
269		t->reass_id, t->op_id);
270      indent = format_get_indent (s);
271      s = format (s, "first bi: %u, data len: %u, ip/fragment[%u, %u]",
272		  t->trace_range.first_bi, t->total_data_len,
273		  t->fragment_first, t->fragment_last);
274    }
275  switch (t->action)
276    {
277    case RANGE_NEW:
278      s = format (s, "\n%Unew %U", format_white_space, indent,
279		  format_ip6_full_reass_range_trace, &t->trace_range);
280      break;
281    case RANGE_OVERLAP:
282      s = format (s, "\n%Uoverlap %U", format_white_space, indent,
283		  format_ip6_full_reass_range_trace, &t->trace_range);
284      break;
285    case ICMP_ERROR_FL_TOO_BIG:
286      s = format (s, "\n%Uicmp-error - frag_len > 65535 %U",
287		  format_white_space, indent,
288		  format_ip6_full_reass_range_trace, &t->trace_range);
289      break;
290    case ICMP_ERROR_FL_NOT_MULT_8:
291      s = format (s, "\n%Uicmp-error - frag_len mod 8 != 0 %U",
292		  format_white_space, indent,
293		  format_ip6_full_reass_range_trace, &t->trace_range);
294      break;
295    case ICMP_ERROR_RT_EXCEEDED:
296      s = format (s, "\n%Uicmp-error - reassembly time exceeded",
297		  format_white_space, indent);
298      break;
299    case FINALIZE:
300      s = format (s, "\n%Ufinalize reassembly", format_white_space, indent);
301      break;
302    case HANDOFF:
303      s =
304	format (s, "handoff from thread #%u to thread #%u", t->thread_id,
305		t->thread_id_to);
306      break;
307    }
308  return s;
309}
310
311static void
312ip6_full_reass_add_trace (vlib_main_t * vm, vlib_node_runtime_t * node,
313			  ip6_full_reass_main_t * rm,
314			  ip6_full_reass_t * reass, u32 bi,
315			  ip6_frag_hdr_t * ip6_frag_header,
316			  ip6_full_reass_trace_operation_e action,
317			  u32 thread_id_to)
318{
319  vlib_buffer_t *b = vlib_get_buffer (vm, bi);
320  vnet_buffer_opaque_t *vnb = vnet_buffer (b);
321  bool is_after_handoff = false;
322  if (vlib_buffer_get_trace_thread (b) != vm->thread_index)
323    {
324      is_after_handoff = true;
325    }
326  ip6_full_reass_trace_t *t = vlib_add_trace (vm, node, b, sizeof (t[0]));
327  t->is_after_handoff = is_after_handoff;
328  if (t->is_after_handoff)
329    {
330      clib_memcpy (&t->ip6_header, vlib_buffer_get_current (b),
331		   clib_min (sizeof (t->ip6_header), b->current_length));
332      if (ip6_frag_header)
333	{
334	  clib_memcpy (&t->ip6_frag_header, ip6_frag_header,
335		       sizeof (t->ip6_frag_header));
336	}
337      else
338	{
339	  clib_memset (&t->ip6_frag_header, 0, sizeof (t->ip6_frag_header));
340	}
341    }
342  if (reass)
343    {
344      t->reass_id = reass->id;
345      t->op_id = reass->trace_op_counter;
346      t->trace_range.first_bi = reass->first_bi;
347      t->total_data_len = reass->data_len;
348      ++reass->trace_op_counter;
349    }
350  else
351    {
352      t->reass_id = ~0;
353    }
354  t->action = action;
355  t->thread_id = vm->thread_index;
356  t->thread_id_to = thread_id_to;
357  ip6_full_reass_trace_details (vm, bi, &t->trace_range);
358  t->fragment_first = vnb->ip.reass.fragment_first;
359  t->fragment_last = vnb->ip.reass.fragment_last;
360#if 0
361  static u8 *s = NULL;
362  s = format (s, "%U", format_ip6_full_reass_trace, NULL, NULL, t);
363  printf ("%.*s\n", vec_len (s), s);
364  fflush (stdout);
365  vec_reset_length (s);
366#endif
367}
368
369always_inline void
370ip6_full_reass_free_ctx (ip6_full_reass_per_thread_t * rt,
371			 ip6_full_reass_t * reass)
372{
373  pool_put (rt->pool, reass);
374  --rt->reass_n;
375}
376
377always_inline void
378ip6_full_reass_free (ip6_full_reass_main_t * rm,
379		     ip6_full_reass_per_thread_t * rt,
380		     ip6_full_reass_t * reass)
381{
382  clib_bihash_kv_48_8_t kv;
383  kv.key[0] = reass->key.as_u64[0];
384  kv.key[1] = reass->key.as_u64[1];
385  kv.key[2] = reass->key.as_u64[2];
386  kv.key[3] = reass->key.as_u64[3];
387  kv.key[4] = reass->key.as_u64[4];
388  kv.key[5] = reass->key.as_u64[5];
389  clib_bihash_add_del_48_8 (&rm->hash, &kv, 0);
390  ip6_full_reass_free_ctx (rt, reass);
391}
392
393always_inline void
394ip6_full_reass_drop_all (vlib_main_t * vm, vlib_node_runtime_t * node,
395			 ip6_full_reass_main_t * rm, ip6_full_reass_t * reass)
396{
397  u32 range_bi = reass->first_bi;
398  vlib_buffer_t *range_b;
399  vnet_buffer_opaque_t *range_vnb;
400  u32 *to_free = NULL;
401  while (~0 != range_bi)
402    {
403      range_b = vlib_get_buffer (vm, range_bi);
404      range_vnb = vnet_buffer (range_b);
405      u32 bi = range_bi;
406      while (~0 != bi)
407	{
408	  vec_add1 (to_free, bi);
409	  vlib_buffer_t *b = vlib_get_buffer (vm, bi);
410	  if (b->flags & VLIB_BUFFER_NEXT_PRESENT)
411	    {
412	      bi = b->next_buffer;
413	      b->flags &= ~VLIB_BUFFER_NEXT_PRESENT;
414	    }
415	  else
416	    {
417	      bi = ~0;
418	    }
419	}
420      range_bi = range_vnb->ip.reass.next_range_bi;
421    }
422  /* send to next_error_index */
423  if (~0 != reass->error_next_index)
424    {
425      u32 n_left_to_next, *to_next, next_index;
426
427      next_index = reass->error_next_index;
428      u32 bi = ~0;
429
430      while (vec_len (to_free) > 0)
431	{
432	  vlib_get_next_frame (vm, node, next_index, to_next, n_left_to_next);
433
434	  while (vec_len (to_free) > 0 && n_left_to_next > 0)
435	    {
436	      bi = vec_pop (to_free);
437
438	      if (~0 != bi)
439		{
440		  to_next[0] = bi;
441		  to_next += 1;
442		  n_left_to_next -= 1;
443		}
444	    }
445	  vlib_put_next_frame (vm, node, next_index, n_left_to_next);
446	}
447    }
448  else
449    {
450      vlib_buffer_free (vm, to_free, vec_len (to_free));
451    }
452  vec_free (to_free);
453}
454
455always_inline void
456ip6_full_reass_on_timeout (vlib_main_t * vm, vlib_node_runtime_t * node,
457			   ip6_full_reass_main_t * rm,
458			   ip6_full_reass_t * reass, u32 * icmp_bi)
459{
460  if (~0 == reass->first_bi)
461    {
462      return;
463    }
464  if (~0 == reass->next_index)	// custom apps don't want icmp
465    {
466      vlib_buffer_t *b = vlib_get_buffer (vm, reass->first_bi);
467      if (0 == vnet_buffer (b)->ip.reass.fragment_first)
468	{
469	  *icmp_bi = reass->first_bi;
470	  if (PREDICT_FALSE (b->flags & VLIB_BUFFER_IS_TRACED))
471	    {
472	      ip6_full_reass_add_trace (vm, node, rm, reass, reass->first_bi,
473					NULL, ICMP_ERROR_RT_EXCEEDED, ~0);
474	    }
475	  // fragment with offset zero received - send icmp message back
476	  if (b->flags & VLIB_BUFFER_NEXT_PRESENT)
477	    {
478	      // separate first buffer from chain and steer it towards icmp node
479	      b->flags &= ~VLIB_BUFFER_NEXT_PRESENT;
480	      reass->first_bi = b->next_buffer;
481	    }
482	  else
483	    {
484	      reass->first_bi = vnet_buffer (b)->ip.reass.next_range_bi;
485	    }
486	  icmp6_error_set_vnet_buffer (b, ICMP6_time_exceeded,
487				       ICMP6_time_exceeded_fragment_reassembly_time_exceeded,
488				       0);
489	}
490    }
491  ip6_full_reass_drop_all (vm, node, rm, reass);
492}
493
494always_inline ip6_full_reass_t *
495ip6_full_reass_find_or_create (vlib_main_t * vm, vlib_node_runtime_t * node,
496			       ip6_full_reass_main_t * rm,
497			       ip6_full_reass_per_thread_t * rt,
498			       ip6_full_reass_kv_t * kv, u32 * icmp_bi,
499			       u8 * do_handoff)
500{
501  ip6_full_reass_t *reass;
502  f64 now;
503
504again:
505
506  reass = NULL;
507  now = vlib_time_now (vm);
508
509  if (!clib_bihash_search_48_8 (&rm->hash, &kv->kv, &kv->kv))
510    {
511      if (vm->thread_index != kv->v.memory_owner_thread_index)
512	{
513	  *do_handoff = 1;
514	  return NULL;
515	}
516
517      reass =
518	pool_elt_at_index (rm->per_thread_data
519			   [kv->v.memory_owner_thread_index].pool,
520			   kv->v.reass_index);
521
522      if (now > reass->last_heard + rm->timeout)
523	{
524	  ip6_full_reass_on_timeout (vm, node, rm, reass, icmp_bi);
525	  ip6_full_reass_free (rm, rt, reass);
526	  reass = NULL;
527	}
528    }
529
530  if (reass)
531    {
532      reass->last_heard = now;
533      return reass;
534    }
535
536  if (rt->reass_n >= rm->max_reass_n)
537    {
538      reass = NULL;
539      return reass;
540    }
541  else
542    {
543      pool_get (rt->pool, reass);
544      clib_memset (reass, 0, sizeof (*reass));
545      reass->id = ((u64) vm->thread_index * 1000000000) + rt->id_counter;
546      ++rt->id_counter;
547      reass->first_bi = ~0;
548      reass->last_packet_octet = ~0;
549      reass->data_len = 0;
550      reass->next_index = ~0;
551      reass->error_next_index = ~0;
552      ++rt->reass_n;
553    }
554
555  reass->key.as_u64[0] = kv->kv.key[0];
556  reass->key.as_u64[1] = kv->kv.key[1];
557  reass->key.as_u64[2] = kv->kv.key[2];
558  reass->key.as_u64[3] = kv->kv.key[3];
559  reass->key.as_u64[4] = kv->kv.key[4];
560  reass->key.as_u64[5] = kv->kv.key[5];
561  kv->v.reass_index = (reass - rt->pool);
562  kv->v.memory_owner_thread_index = vm->thread_index;
563  reass->last_heard = now;
564
565  int rv = clib_bihash_add_del_48_8 (&rm->hash, &kv->kv, 2);
566  if (rv)
567    {
568      ip6_full_reass_free (rm, rt, reass);
569      reass = NULL;
570      // if other worker created a context already work with the other copy
571      if (-2 == rv)
572	goto again;
573    }
574
575  return reass;
576}
577
578always_inline ip6_full_reass_rc_t
579ip6_full_reass_finalize (vlib_main_t * vm, vlib_node_runtime_t * node,
580			 ip6_full_reass_main_t * rm,
581			 ip6_full_reass_per_thread_t * rt,
582			 ip6_full_reass_t * reass, u32 * bi0, u32 * next0,
583			 u32 * error0, bool is_custom_app)
584{
585  *bi0 = reass->first_bi;
586  *error0 = IP6_ERROR_NONE;
587  ip6_frag_hdr_t *frag_hdr;
588  vlib_buffer_t *last_b = NULL;
589  u32 sub_chain_bi = reass->first_bi;
590  u32 total_length = 0;
591  u32 buf_cnt = 0;
592  u32 dropped_cnt = 0;
593  u32 *vec_drop_compress = NULL;
594  ip6_full_reass_rc_t rv = IP6_FULL_REASS_RC_OK;
595  do
596    {
597      u32 tmp_bi = sub_chain_bi;
598      vlib_buffer_t *tmp = vlib_get_buffer (vm, tmp_bi);
599      vnet_buffer_opaque_t *vnb = vnet_buffer (tmp);
600      if (!(vnb->ip.reass.range_first >= vnb->ip.reass.fragment_first) &&
601	  !(vnb->ip.reass.range_last > vnb->ip.reass.fragment_first))
602	{
603	  rv = IP6_FULL_REASS_RC_INTERNAL_ERROR;
604	  goto free_buffers_and_return;
605	}
606
607      u32 data_len = ip6_full_reass_buffer_get_data_len (tmp);
608      u32 trim_front = vnet_buffer (tmp)->ip.reass.ip6_frag_hdr_offset +
609	sizeof (*frag_hdr) + ip6_full_reass_buffer_get_data_offset (tmp);
610      u32 trim_end =
611	vlib_buffer_length_in_chain (vm, tmp) - trim_front - data_len;
612      if (tmp_bi == reass->first_bi)
613	{
614	  /* first buffer - keep ip6 header */
615	  if (0 != ip6_full_reass_buffer_get_data_offset (tmp))
616	    {
617	      rv = IP6_FULL_REASS_RC_INTERNAL_ERROR;
618	      goto free_buffers_and_return;
619	    }
620	  trim_front = 0;
621	  trim_end = vlib_buffer_length_in_chain (vm, tmp) - data_len -
622	    (vnet_buffer (tmp)->ip.reass.ip6_frag_hdr_offset +
623	     sizeof (*frag_hdr));
624	  if (!(vlib_buffer_length_in_chain (vm, tmp) - trim_end > 0))
625	    {
626	      rv = IP6_FULL_REASS_RC_INTERNAL_ERROR;
627	      goto free_buffers_and_return;
628	    }
629	}
630      u32 keep_data =
631	vlib_buffer_length_in_chain (vm, tmp) - trim_front - trim_end;
632      while (1)
633	{
634	  ++buf_cnt;
635	  if (trim_front)
636	    {
637	      if (trim_front > tmp->current_length)
638		{
639		  /* drop whole buffer */
640		  vec_add1 (vec_drop_compress, tmp_bi);
641		  trim_front -= tmp->current_length;
642		  if (!(tmp->flags & VLIB_BUFFER_NEXT_PRESENT))
643		    {
644		      rv = IP6_FULL_REASS_RC_INTERNAL_ERROR;
645		      goto free_buffers_and_return;
646		    }
647		  tmp->flags &= ~VLIB_BUFFER_NEXT_PRESENT;
648		  tmp_bi = tmp->next_buffer;
649		  tmp = vlib_get_buffer (vm, tmp_bi);
650		  continue;
651		}
652	      else
653		{
654		  vlib_buffer_advance (tmp, trim_front);
655		  trim_front = 0;
656		}
657	    }
658	  if (keep_data)
659	    {
660	      if (last_b)
661		{
662		  last_b->flags |= VLIB_BUFFER_NEXT_PRESENT;
663		  last_b->next_buffer = tmp_bi;
664		}
665	      last_b = tmp;
666	      if (keep_data <= tmp->current_length)
667		{
668		  tmp->current_length = keep_data;
669		  keep_data = 0;
670		}
671	      else
672		{
673		  keep_data -= tmp->current_length;
674		  if (!(tmp->flags & VLIB_BUFFER_NEXT_PRESENT))
675		    {
676		      rv = IP6_FULL_REASS_RC_INTERNAL_ERROR;
677		      goto free_buffers_and_return;
678		    }
679		}
680	      total_length += tmp->current_length;
681	    }
682	  else
683	    {
684	      vec_add1 (vec_drop_compress, tmp_bi);
685	      if (reass->first_bi == tmp_bi)
686		{
687		  rv = IP6_FULL_REASS_RC_INTERNAL_ERROR;
688		  goto free_buffers_and_return;
689		}
690	      ++dropped_cnt;
691	    }
692	  if (tmp->flags & VLIB_BUFFER_NEXT_PRESENT)
693	    {
694	      tmp_bi = tmp->next_buffer;
695	      tmp = vlib_get_buffer (vm, tmp->next_buffer);
696	    }
697	  else
698	    {
699	      break;
700	    }
701	}
702      sub_chain_bi =
703	vnet_buffer (vlib_get_buffer (vm, sub_chain_bi))->ip.
704	reass.next_range_bi;
705    }
706  while (~0 != sub_chain_bi);
707
708  if (!last_b)
709    {
710      rv = IP6_FULL_REASS_RC_INTERNAL_ERROR;
711      goto free_buffers_and_return;
712    }
713  last_b->flags &= ~VLIB_BUFFER_NEXT_PRESENT;
714  vlib_buffer_t *first_b = vlib_get_buffer (vm, reass->first_bi);
715  if (total_length < first_b->current_length)
716    {
717      rv = IP6_FULL_REASS_RC_INTERNAL_ERROR;
718      goto free_buffers_and_return;
719    }
720  total_length -= first_b->current_length;
721  first_b->flags |= VLIB_BUFFER_TOTAL_LENGTH_VALID;
722  first_b->total_length_not_including_first_buffer = total_length;
723  // drop fragment header
724  vnet_buffer_opaque_t *first_b_vnb = vnet_buffer (first_b);
725  ip6_header_t *ip = vlib_buffer_get_current (first_b);
726  u16 ip6_frag_hdr_offset = first_b_vnb->ip.reass.ip6_frag_hdr_offset;
727  ip6_ext_header_t *prev_hdr;
728  frag_hdr =
729    ip6_ext_header_find (vm, first_b, ip, IP_PROTOCOL_IPV6_FRAGMENTATION,
730			 &prev_hdr);
731  if (prev_hdr)
732    {
733      prev_hdr->next_hdr = frag_hdr->next_hdr;
734    }
735  else
736    {
737      ip->protocol = frag_hdr->next_hdr;
738    }
739  if (!((u8 *) frag_hdr - (u8 *) ip == ip6_frag_hdr_offset))
740    {
741      rv = IP6_FULL_REASS_RC_INTERNAL_ERROR;
742      goto free_buffers_and_return;
743    }
744  memmove (frag_hdr, (u8 *) frag_hdr + sizeof (*frag_hdr),
745	   first_b->current_length - ip6_frag_hdr_offset -
746	   sizeof (ip6_frag_hdr_t));
747  first_b->current_length -= sizeof (*frag_hdr);
748  ip->payload_length =
749    clib_host_to_net_u16 (total_length + first_b->current_length -
750			  sizeof (*ip));
751  if (!vlib_buffer_chain_linearize (vm, first_b))
752    {
753      rv = IP6_FULL_REASS_RC_NO_BUF;
754      goto free_buffers_and_return;
755    }
756  first_b->flags &= ~VLIB_BUFFER_EXT_HDR_VALID;
757  if (PREDICT_FALSE (first_b->flags & VLIB_BUFFER_IS_TRACED))
758    {
759      ip6_full_reass_add_trace (vm, node, rm, reass, reass->first_bi, NULL,
760				FINALIZE, ~0);
761#if 0
762      // following code does a hexdump of packet fragments to stdout ...
763      do
764	{
765	  u32 bi = reass->first_bi;
766	  u8 *s = NULL;
767	  while (~0 != bi)
768	    {
769	      vlib_buffer_t *b = vlib_get_buffer (vm, bi);
770	      s = format (s, "%u: %U\n", bi, format_hexdump,
771			  vlib_buffer_get_current (b), b->current_length);
772	      if (b->flags & VLIB_BUFFER_NEXT_PRESENT)
773		{
774		  bi = b->next_buffer;
775		}
776	      else
777		{
778		  break;
779		}
780	    }
781	  printf ("%.*s\n", vec_len (s), s);
782	  fflush (stdout);
783	  vec_free (s);
784	}
785      while (0);
786#endif
787    }
788  if (!is_custom_app)
789    {
790      *next0 = IP6_FULL_REASSEMBLY_NEXT_INPUT;
791    }
792  else
793    {
794      *next0 = reass->next_index;
795    }
796  vnet_buffer (first_b)->ip.reass.estimated_mtu = reass->min_fragment_length;
797  ip6_full_reass_free (rm, rt, reass);
798  reass = NULL;
799free_buffers_and_return:
800  vlib_buffer_free (vm, vec_drop_compress, vec_len (vec_drop_compress));
801  vec_free (vec_drop_compress);
802  return rv;
803}
804
805always_inline void
806ip6_full_reass_insert_range_in_chain (vlib_main_t * vm,
807				      ip6_full_reass_main_t * rm,
808				      ip6_full_reass_per_thread_t * rt,
809				      ip6_full_reass_t * reass,
810				      u32 prev_range_bi, u32 new_next_bi)
811{
812
813  vlib_buffer_t *new_next_b = vlib_get_buffer (vm, new_next_bi);
814  vnet_buffer_opaque_t *new_next_vnb = vnet_buffer (new_next_b);
815  if (~0 != prev_range_bi)
816    {
817      vlib_buffer_t *prev_b = vlib_get_buffer (vm, prev_range_bi);
818      vnet_buffer_opaque_t *prev_vnb = vnet_buffer (prev_b);
819      new_next_vnb->ip.reass.next_range_bi = prev_vnb->ip.reass.next_range_bi;
820      prev_vnb->ip.reass.next_range_bi = new_next_bi;
821    }
822  else
823    {
824      if (~0 != reass->first_bi)
825	{
826	  new_next_vnb->ip.reass.next_range_bi = reass->first_bi;
827	}
828      reass->first_bi = new_next_bi;
829    }
830  reass->data_len += ip6_full_reass_buffer_get_data_len (new_next_b);
831}
832
833always_inline ip6_full_reass_rc_t
834ip6_full_reass_update (vlib_main_t * vm, vlib_node_runtime_t * node,
835		       ip6_full_reass_main_t * rm,
836		       ip6_full_reass_per_thread_t * rt,
837		       ip6_full_reass_t * reass, u32 * bi0, u32 * next0,
838		       u32 * error0, ip6_frag_hdr_t * frag_hdr,
839		       bool is_custom_app, u32 * handoff_thread_idx)
840{
841  int consumed = 0;
842  vlib_buffer_t *fb = vlib_get_buffer (vm, *bi0);
843  vnet_buffer_opaque_t *fvnb = vnet_buffer (fb);
844  if (is_custom_app)
845    {
846      reass->next_index = fvnb->ip.reass.next_index;	// store next_index before it's overwritten
847      reass->error_next_index = fvnb->ip.reass.error_next_index;	// store error_next_index before it is overwritten
848    }
849
850  fvnb->ip.reass.ip6_frag_hdr_offset =
851    (u8 *) frag_hdr - (u8 *) vlib_buffer_get_current (fb);
852  ip6_header_t *fip = vlib_buffer_get_current (fb);
853  if (fb->current_length < sizeof (*fip) ||
854      fvnb->ip.reass.ip6_frag_hdr_offset == 0 ||
855      fvnb->ip.reass.ip6_frag_hdr_offset >= fb->current_length)
856    {
857      return IP6_FULL_REASS_RC_INTERNAL_ERROR;
858    }
859
860  u32 fragment_first = fvnb->ip.reass.fragment_first =
861    ip6_frag_hdr_offset_bytes (frag_hdr);
862  u32 fragment_length =
863    vlib_buffer_length_in_chain (vm, fb) -
864    (fvnb->ip.reass.ip6_frag_hdr_offset + sizeof (*frag_hdr));
865  u32 fragment_last = fvnb->ip.reass.fragment_last =
866    fragment_first + fragment_length - 1;
867  int more_fragments = ip6_frag_hdr_more (frag_hdr);
868  u32 candidate_range_bi = reass->first_bi;
869  u32 prev_range_bi = ~0;
870  fvnb->ip.reass.range_first = fragment_first;
871  fvnb->ip.reass.range_last = fragment_last;
872  fvnb->ip.reass.next_range_bi = ~0;
873  if (!more_fragments)
874    {
875      reass->last_packet_octet = fragment_last;
876    }
877  if (~0 == reass->first_bi)
878    {
879      // starting a new reassembly
880      ip6_full_reass_insert_range_in_chain (vm, rm, rt, reass, prev_range_bi,
881					    *bi0);
882      reass->min_fragment_length = clib_net_to_host_u16 (fip->payload_length);
883      consumed = 1;
884      reass->fragments_n = 1;
885      goto check_if_done_maybe;
886    }
887  reass->min_fragment_length =
888    clib_min (clib_net_to_host_u16 (fip->payload_length),
889	      fvnb->ip.reass.estimated_mtu);
890  while (~0 != candidate_range_bi)
891    {
892      vlib_buffer_t *candidate_b = vlib_get_buffer (vm, candidate_range_bi);
893      vnet_buffer_opaque_t *candidate_vnb = vnet_buffer (candidate_b);
894      if (fragment_first > candidate_vnb->ip.reass.range_last)
895	{
896	  // this fragments starts after candidate range
897	  prev_range_bi = candidate_range_bi;
898	  candidate_range_bi = candidate_vnb->ip.reass.next_range_bi;
899	  if (candidate_vnb->ip.reass.range_last < fragment_last &&
900	      ~0 == candidate_range_bi)
901	    {
902	      // special case - this fragment falls beyond all known ranges
903	      ip6_full_reass_insert_range_in_chain (vm, rm, rt, reass,
904						    prev_range_bi, *bi0);
905	      consumed = 1;
906	      break;
907	    }
908	  continue;
909	}
910      if (fragment_last < candidate_vnb->ip.reass.range_first)
911	{
912	  // this fragment ends before candidate range without any overlap
913	  ip6_full_reass_insert_range_in_chain (vm, rm, rt, reass,
914						prev_range_bi, *bi0);
915	  consumed = 1;
916	}
917      else if (fragment_first == candidate_vnb->ip.reass.range_first &&
918	       fragment_last == candidate_vnb->ip.reass.range_last)
919	{
920	  // duplicate fragment - ignore
921	}
922      else
923	{
924	  // overlapping fragment - not allowed by RFC 8200
925	  if (PREDICT_FALSE (fb->flags & VLIB_BUFFER_IS_TRACED))
926	    {
927	      ip6_full_reass_add_trace (vm, node, rm, reass, *bi0, frag_hdr,
928					RANGE_OVERLAP, ~0);
929	    }
930	  ip6_full_reass_drop_all (vm, node, rm, reass);
931	  ip6_full_reass_free (rm, rt, reass);
932	  *next0 = IP6_FULL_REASSEMBLY_NEXT_DROP;
933	  *error0 = IP6_ERROR_REASS_OVERLAPPING_FRAGMENT;
934	  return IP6_FULL_REASS_RC_OK;
935	}
936      break;
937    }
938  ++reass->fragments_n;
939check_if_done_maybe:
940  if (consumed)
941    {
942      if (PREDICT_FALSE (fb->flags & VLIB_BUFFER_IS_TRACED))
943	{
944	  ip6_full_reass_add_trace (vm, node, rm, reass, *bi0, frag_hdr,
945				    RANGE_NEW, ~0);
946	}
947    }
948  if (~0 != reass->last_packet_octet &&
949      reass->data_len == reass->last_packet_octet + 1)
950    {
951      *handoff_thread_idx = reass->sendout_thread_index;
952      int handoff =
953	reass->memory_owner_thread_index != reass->sendout_thread_index;
954      ip6_full_reass_rc_t rc =
955	ip6_full_reass_finalize (vm, node, rm, rt, reass, bi0, next0, error0,
956				 is_custom_app);
957      if (IP6_FULL_REASS_RC_OK == rc && handoff)
958	{
959	  return IP6_FULL_REASS_RC_HANDOFF;
960	}
961      return rc;
962    }
963  else
964    {
965      if (consumed)
966	{
967	  *bi0 = ~0;
968	  if (reass->fragments_n > rm->max_reass_len)
969	    {
970	      return IP6_FULL_REASS_RC_TOO_MANY_FRAGMENTS;
971	    }
972	}
973      else
974	{
975	  *next0 = IP6_FULL_REASSEMBLY_NEXT_DROP;
976	  *error0 = IP6_ERROR_REASS_DUPLICATE_FRAGMENT;
977	}
978    }
979  return IP6_FULL_REASS_RC_OK;
980}
981
982always_inline bool
983ip6_full_reass_verify_upper_layer_present (vlib_node_runtime_t * node,
984					   vlib_buffer_t * b,
985					   ip6_frag_hdr_t * frag_hdr)
986{
987  ip6_ext_header_t *tmp = (ip6_ext_header_t *) frag_hdr;
988  while (ip6_ext_hdr (tmp->next_hdr))
989    {
990      tmp = ip6_ext_next_header (tmp);
991    }
992  if (IP_PROTOCOL_IP6_NONXT == tmp->next_hdr)
993    {
994      icmp6_error_set_vnet_buffer (b, ICMP6_parameter_problem,
995				   ICMP6_parameter_problem_first_fragment_has_incomplete_header_chain,
996				   0);
997      b->error = node->errors[IP6_ERROR_REASS_MISSING_UPPER];
998
999      return false;
1000    }
1001  return true;
1002}
1003
1004always_inline bool
1005ip6_full_reass_verify_fragment_multiple_8 (vlib_main_t * vm,
1006					   vlib_node_runtime_t * node,
1007					   vlib_buffer_t * b,
1008					   ip6_frag_hdr_t * frag_hdr)
1009{
1010  vnet_buffer_opaque_t *vnb = vnet_buffer (b);
1011  ip6_header_t *ip = vlib_buffer_get_current (b);
1012  int more_fragments = ip6_frag_hdr_more (frag_hdr);
1013  u32 fragment_length =
1014    vlib_buffer_length_in_chain (vm, b) -
1015    (vnb->ip.reass.ip6_frag_hdr_offset + sizeof (*frag_hdr));
1016  if (more_fragments && 0 != fragment_length % 8)
1017    {
1018      icmp6_error_set_vnet_buffer (b, ICMP6_parameter_problem,
1019				   ICMP6_parameter_problem_erroneous_header_field,
1020				   (u8 *) & ip->payload_length - (u8 *) ip);
1021      return false;
1022    }
1023  return true;
1024}
1025
1026always_inline bool
1027ip6_full_reass_verify_packet_size_lt_64k (vlib_main_t * vm,
1028					  vlib_node_runtime_t * node,
1029					  vlib_buffer_t * b,
1030					  ip6_frag_hdr_t * frag_hdr)
1031{
1032  vnet_buffer_opaque_t *vnb = vnet_buffer (b);
1033  u32 fragment_first = ip6_frag_hdr_offset_bytes (frag_hdr);
1034  u32 fragment_length =
1035    vlib_buffer_length_in_chain (vm, b) -
1036    (vnb->ip.reass.ip6_frag_hdr_offset + sizeof (*frag_hdr));
1037  if (fragment_first + fragment_length > 65535)
1038    {
1039      ip6_header_t *ip0 = vlib_buffer_get_current (b);
1040      icmp6_error_set_vnet_buffer (b, ICMP6_parameter_problem,
1041				   ICMP6_parameter_problem_erroneous_header_field,
1042				   (u8 *) & frag_hdr->fragment_offset_and_more
1043				   - (u8 *) ip0);
1044      return false;
1045    }
1046  return true;
1047}
1048
1049always_inline uword
1050ip6_full_reassembly_inline (vlib_main_t * vm,
1051			    vlib_node_runtime_t * node,
1052			    vlib_frame_t * frame, bool is_feature,
1053			    bool is_custom_app)
1054{
1055  u32 *from = vlib_frame_vector_args (frame);
1056  u32 n_left_from, n_left_to_next, *to_next, next_index;
1057  ip6_full_reass_main_t *rm = &ip6_full_reass_main;
1058  ip6_full_reass_per_thread_t *rt = &rm->per_thread_data[vm->thread_index];
1059  clib_spinlock_lock (&rt->lock);
1060
1061  n_left_from = frame->n_vectors;
1062  next_index = node->cached_next_index;
1063  while (n_left_from > 0)
1064    {
1065      vlib_get_next_frame (vm, node, next_index, to_next, n_left_to_next);
1066
1067      while (n_left_from > 0 && n_left_to_next > 0)
1068	{
1069	  u32 bi0;
1070	  vlib_buffer_t *b0;
1071	  u32 next0 = IP6_FULL_REASSEMBLY_NEXT_DROP;
1072	  u32 error0 = IP6_ERROR_NONE;
1073	  u32 icmp_bi = ~0;
1074
1075	  bi0 = from[0];
1076	  b0 = vlib_get_buffer (vm, bi0);
1077
1078	  ip6_header_t *ip0 = vlib_buffer_get_current (b0);
1079	  ip6_frag_hdr_t *frag_hdr = NULL;
1080	  ip6_ext_header_t *prev_hdr;
1081	  if (ip6_ext_hdr (ip0->protocol))
1082	    {
1083	      frag_hdr =
1084		ip6_ext_header_find (vm, b0, ip0,
1085				     IP_PROTOCOL_IPV6_FRAGMENTATION,
1086				     &prev_hdr);
1087	    }
1088	  if (!frag_hdr)
1089	    {
1090	      // this is a regular packet - no fragmentation
1091	      next0 = IP6_FULL_REASSEMBLY_NEXT_INPUT;
1092	      goto skip_reass;
1093	    }
1094	  vnet_buffer (b0)->ip.reass.ip6_frag_hdr_offset =
1095	    (u8 *) frag_hdr - (u8 *) ip0;
1096
1097	  if (0 == ip6_frag_hdr_offset (frag_hdr))
1098	    {
1099	      // first fragment - verify upper-layer is present
1100	      if (!ip6_full_reass_verify_upper_layer_present
1101		  (node, b0, frag_hdr))
1102		{
1103		  next0 = IP6_FULL_REASSEMBLY_NEXT_ICMP_ERROR;
1104		  goto skip_reass;
1105		}
1106	    }
1107	  if (!ip6_full_reass_verify_fragment_multiple_8
1108	      (vm, node, b0, frag_hdr)
1109	      || !ip6_full_reass_verify_packet_size_lt_64k (vm, node, b0,
1110							    frag_hdr))
1111	    {
1112	      next0 = IP6_FULL_REASSEMBLY_NEXT_ICMP_ERROR;
1113	      goto skip_reass;
1114	    }
1115	  ip6_full_reass_kv_t kv;
1116	  u8 do_handoff = 0;
1117
1118	  kv.k.as_u64[0] = ip0->src_address.as_u64[0];
1119	  kv.k.as_u64[1] = ip0->src_address.as_u64[1];
1120	  kv.k.as_u64[2] = ip0->dst_address.as_u64[0];
1121	  kv.k.as_u64[3] = ip0->dst_address.as_u64[1];
1122	  kv.k.as_u64[4] =
1123	    ((u64) vec_elt (ip6_main.fib_index_by_sw_if_index,
1124			    vnet_buffer (b0)->sw_if_index[VLIB_RX])) << 32 |
1125	    (u64) frag_hdr->identification;
1126	  kv.k.as_u64[5] = ip0->protocol;
1127
1128	  ip6_full_reass_t *reass =
1129	    ip6_full_reass_find_or_create (vm, node, rm, rt, &kv, &icmp_bi,
1130					   &do_handoff);
1131
1132	  if (reass)
1133	    {
1134	      const u32 fragment_first = ip6_frag_hdr_offset (frag_hdr);
1135	      if (0 == fragment_first)
1136		{
1137		  reass->sendout_thread_index = vm->thread_index;
1138		}
1139	    }
1140	  if (PREDICT_FALSE (do_handoff))
1141	    {
1142	      next0 = IP6_FULL_REASSEMBLY_NEXT_HANDOFF;
1143	      vnet_buffer (b0)->ip.reass.owner_thread_index =
1144		kv.v.memory_owner_thread_index;
1145	    }
1146	  else if (reass)
1147	    {
1148	      u32 handoff_thread_idx;
1149	      switch (ip6_full_reass_update
1150		      (vm, node, rm, rt, reass, &bi0, &next0, &error0,
1151		       frag_hdr, is_custom_app, &handoff_thread_idx))
1152		{
1153		case IP6_FULL_REASS_RC_OK:
1154		  /* nothing to do here */
1155		  break;
1156		case IP6_FULL_REASS_RC_HANDOFF:
1157		  next0 = IP6_FULL_REASSEMBLY_NEXT_HANDOFF;
1158		  b0 = vlib_get_buffer (vm, bi0);
1159		  vnet_buffer (b0)->ip.reass.owner_thread_index =
1160		    handoff_thread_idx;
1161		  break;
1162		case IP6_FULL_REASS_RC_TOO_MANY_FRAGMENTS:
1163		  vlib_node_increment_counter (vm, node->node_index,
1164					       IP6_ERROR_REASS_FRAGMENT_CHAIN_TOO_LONG,
1165					       1);
1166		  ip6_full_reass_drop_all (vm, node, rm, reass);
1167		  ip6_full_reass_free (rm, rt, reass);
1168		  goto next_packet;
1169		  break;
1170		case IP6_FULL_REASS_RC_NO_BUF:
1171		  vlib_node_increment_counter (vm, node->node_index,
1172					       IP6_ERROR_REASS_NO_BUF, 1);
1173		  ip6_full_reass_drop_all (vm, node, rm, reass);
1174		  ip6_full_reass_free (rm, rt, reass);
1175		  goto next_packet;
1176		  break;
1177		case IP6_FULL_REASS_RC_INTERNAL_ERROR:
1178		  vlib_node_increment_counter (vm, node->node_index,
1179					       IP6_ERROR_REASS_INTERNAL_ERROR,
1180					       1);
1181		  ip6_full_reass_drop_all (vm, node, rm, reass);
1182		  ip6_full_reass_free (rm, rt, reass);
1183		  goto next_packet;
1184		  break;
1185		}
1186	    }
1187	  else
1188	    {
1189	      if (is_feature)
1190		{
1191		  next0 = IP6_FULL_REASSEMBLY_NEXT_DROP;
1192		}
1193	      else
1194		{
1195		  vnet_buffer_opaque_t *fvnb = vnet_buffer (b0);
1196		  next0 = fvnb->ip.reass.error_next_index;
1197		}
1198	      error0 = IP6_ERROR_REASS_LIMIT_REACHED;
1199	    }
1200
1201	  if (~0 != bi0)
1202	    {
1203	    skip_reass:
1204	      to_next[0] = bi0;
1205	      to_next += 1;
1206	      n_left_to_next -= 1;
1207
1208	      /* bi0 might have been updated by reass_finalize, reload */
1209	      b0 = vlib_get_buffer (vm, bi0);
1210	      if (IP6_ERROR_NONE != error0)
1211		{
1212		  b0->error = node->errors[error0];
1213		}
1214
1215	      if (next0 == IP6_FULL_REASSEMBLY_NEXT_HANDOFF)
1216		{
1217		  if (PREDICT_FALSE (b0->flags & VLIB_BUFFER_IS_TRACED))
1218		    {
1219		      ip6_full_reass_add_trace (vm, node, rm, NULL, bi0,
1220						frag_hdr, HANDOFF,
1221						vnet_buffer (b0)->ip.
1222						reass.owner_thread_index);
1223		    }
1224		}
1225	      else if (is_feature && IP6_ERROR_NONE == error0)
1226		{
1227		  vnet_feature_next (&next0, b0);
1228		}
1229	      vlib_validate_buffer_enqueue_x1 (vm, node, next_index, to_next,
1230					       n_left_to_next, bi0, next0);
1231	    }
1232
1233	  if (~0 != icmp_bi)
1234	    {
1235	      next0 = IP6_FULL_REASSEMBLY_NEXT_ICMP_ERROR;
1236	      to_next[0] = icmp_bi;
1237	      to_next += 1;
1238	      n_left_to_next -= 1;
1239	      vlib_validate_buffer_enqueue_x1 (vm, node, next_index, to_next,
1240					       n_left_to_next, icmp_bi,
1241					       next0);
1242	    }
1243	next_packet:
1244	  from += 1;
1245	  n_left_from -= 1;
1246	}
1247
1248      vlib_put_next_frame (vm, node, next_index, n_left_to_next);
1249    }
1250
1251  clib_spinlock_unlock (&rt->lock);
1252  return frame->n_vectors;
1253}
1254
1255static char *ip6_full_reassembly_error_strings[] = {
1256#define _(sym, string) string,
1257  foreach_ip6_error
1258#undef _
1259};
1260
1261VLIB_NODE_FN (ip6_full_reass_node) (vlib_main_t * vm,
1262				    vlib_node_runtime_t * node,
1263				    vlib_frame_t * frame)
1264{
1265  return ip6_full_reassembly_inline (vm, node, frame, false /* is_feature */ ,
1266				     false /* is_custom_app */ );
1267}
1268
1269/* *INDENT-OFF* */
1270VLIB_REGISTER_NODE (ip6_full_reass_node) = {
1271    .name = "ip6-full-reassembly",
1272    .vector_size = sizeof (u32),
1273    .format_trace = format_ip6_full_reass_trace,
1274    .n_errors = ARRAY_LEN (ip6_full_reassembly_error_strings),
1275    .error_strings = ip6_full_reassembly_error_strings,
1276    .n_next_nodes = IP6_FULL_REASSEMBLY_N_NEXT,
1277    .next_nodes =
1278        {
1279                [IP6_FULL_REASSEMBLY_NEXT_INPUT] = "ip6-input",
1280                [IP6_FULL_REASSEMBLY_NEXT_DROP] = "ip6-drop",
1281                [IP6_FULL_REASSEMBLY_NEXT_ICMP_ERROR] = "ip6-icmp-error",
1282                [IP6_FULL_REASSEMBLY_NEXT_HANDOFF] = "ip6-full-reassembly-handoff",
1283        },
1284};
1285/* *INDENT-ON* */
1286
1287VLIB_NODE_FN (ip6_full_reass_node_feature) (vlib_main_t * vm,
1288					    vlib_node_runtime_t * node,
1289					    vlib_frame_t * frame)
1290{
1291  return ip6_full_reassembly_inline (vm, node, frame, true /* is_feature */ ,
1292				     false /* is_custom_app */ );
1293}
1294
1295/* *INDENT-OFF* */
1296VLIB_REGISTER_NODE (ip6_full_reass_node_feature) = {
1297    .name = "ip6-full-reassembly-feature",
1298    .vector_size = sizeof (u32),
1299    .format_trace = format_ip6_full_reass_trace,
1300    .n_errors = ARRAY_LEN (ip6_full_reassembly_error_strings),
1301    .error_strings = ip6_full_reassembly_error_strings,
1302    .n_next_nodes = IP6_FULL_REASSEMBLY_N_NEXT,
1303    .next_nodes =
1304        {
1305                [IP6_FULL_REASSEMBLY_NEXT_INPUT] = "ip6-input",
1306                [IP6_FULL_REASSEMBLY_NEXT_DROP] = "ip6-drop",
1307                [IP6_FULL_REASSEMBLY_NEXT_ICMP_ERROR] = "ip6-icmp-error",
1308                [IP6_FULL_REASSEMBLY_NEXT_HANDOFF] = "ip6-full-reass-feature-hoff",
1309        },
1310};
1311/* *INDENT-ON* */
1312
1313/* *INDENT-OFF* */
1314VNET_FEATURE_INIT (ip6_full_reassembly_feature, static) = {
1315    .arc_name = "ip6-unicast",
1316    .node_name = "ip6-full-reassembly-feature",
1317    .runs_before = VNET_FEATURES ("ip6-lookup",
1318                                  "ipsec6-input-feature"),
1319    .runs_after = 0,
1320};
1321/* *INDENT-ON* */
1322
1323#ifndef CLIB_MARCH_VARIANT
1324static u32
1325ip6_full_reass_get_nbuckets ()
1326{
1327  ip6_full_reass_main_t *rm = &ip6_full_reass_main;
1328  u32 nbuckets;
1329  u8 i;
1330
1331  nbuckets = (u32) (rm->max_reass_n / IP6_FULL_REASS_HT_LOAD_FACTOR);
1332
1333  for (i = 0; i < 31; i++)
1334    if ((1 << i) >= nbuckets)
1335      break;
1336  nbuckets = 1 << i;
1337
1338  return nbuckets;
1339}
1340#endif /* CLIB_MARCH_VARIANT */
1341
1342typedef enum
1343{
1344  IP6_EVENT_CONFIG_CHANGED = 1,
1345} ip6_full_reass_event_t;
1346
1347#ifndef CLIB_MARCH_VARIANT
1348typedef struct
1349{
1350  int failure;
1351  clib_bihash_48_8_t *new_hash;
1352} ip6_rehash_cb_ctx;
1353
1354static int
1355ip6_rehash_cb (clib_bihash_kv_48_8_t * kv, void *_ctx)
1356{
1357  ip6_rehash_cb_ctx *ctx = _ctx;
1358  if (clib_bihash_add_del_48_8 (ctx->new_hash, kv, 1))
1359    {
1360      ctx->failure = 1;
1361    }
1362  return (BIHASH_WALK_CONTINUE);
1363}
1364
1365static void
1366ip6_full_reass_set_params (u32 timeout_ms, u32 max_reassemblies,
1367			   u32 max_reassembly_length,
1368			   u32 expire_walk_interval_ms)
1369{
1370  ip6_full_reass_main.timeout_ms = timeout_ms;
1371  ip6_full_reass_main.timeout = (f64) timeout_ms / (f64) MSEC_PER_SEC;
1372  ip6_full_reass_main.max_reass_n = max_reassemblies;
1373  ip6_full_reass_main.max_reass_len = max_reassembly_length;
1374  ip6_full_reass_main.expire_walk_interval_ms = expire_walk_interval_ms;
1375}
1376
1377vnet_api_error_t
1378ip6_full_reass_set (u32 timeout_ms, u32 max_reassemblies,
1379		    u32 max_reassembly_length, u32 expire_walk_interval_ms)
1380{
1381  u32 old_nbuckets = ip6_full_reass_get_nbuckets ();
1382  ip6_full_reass_set_params (timeout_ms, max_reassemblies,
1383			     max_reassembly_length, expire_walk_interval_ms);
1384  vlib_process_signal_event (ip6_full_reass_main.vlib_main,
1385			     ip6_full_reass_main.ip6_full_reass_expire_node_idx,
1386			     IP6_EVENT_CONFIG_CHANGED, 0);
1387  u32 new_nbuckets = ip6_full_reass_get_nbuckets ();
1388  if (ip6_full_reass_main.max_reass_n > 0 && new_nbuckets > old_nbuckets)
1389    {
1390      clib_bihash_48_8_t new_hash;
1391      clib_memset (&new_hash, 0, sizeof (new_hash));
1392      ip6_rehash_cb_ctx ctx;
1393      ctx.failure = 0;
1394      ctx.new_hash = &new_hash;
1395      clib_bihash_init_48_8 (&new_hash, "ip6-full-reass", new_nbuckets,
1396			     new_nbuckets * 1024);
1397      clib_bihash_foreach_key_value_pair_48_8 (&ip6_full_reass_main.hash,
1398					       ip6_rehash_cb, &ctx);
1399      if (ctx.failure)
1400	{
1401	  clib_bihash_free_48_8 (&new_hash);
1402	  return -1;
1403	}
1404      else
1405	{
1406	  clib_bihash_free_48_8 (&ip6_full_reass_main.hash);
1407	  clib_memcpy_fast (&ip6_full_reass_main.hash, &new_hash,
1408			    sizeof (ip6_full_reass_main.hash));
1409	  clib_bihash_copied (&ip6_full_reass_main.hash, &new_hash);
1410	}
1411    }
1412  return 0;
1413}
1414
1415vnet_api_error_t
1416ip6_full_reass_get (u32 * timeout_ms, u32 * max_reassemblies,
1417		    u32 * max_reassembly_length,
1418		    u32 * expire_walk_interval_ms)
1419{
1420  *timeout_ms = ip6_full_reass_main.timeout_ms;
1421  *max_reassemblies = ip6_full_reass_main.max_reass_n;
1422  *max_reassembly_length = ip6_full_reass_main.max_reass_len;
1423  *expire_walk_interval_ms = ip6_full_reass_main.expire_walk_interval_ms;
1424  return 0;
1425}
1426
1427static clib_error_t *
1428ip6_full_reass_init_function (vlib_main_t * vm)
1429{
1430  ip6_full_reass_main_t *rm = &ip6_full_reass_main;
1431  clib_error_t *error = 0;
1432  u32 nbuckets;
1433  vlib_node_t *node;
1434
1435  rm->vlib_main = vm;
1436
1437  vec_validate (rm->per_thread_data, vlib_num_workers ());
1438  ip6_full_reass_per_thread_t *rt;
1439  vec_foreach (rt, rm->per_thread_data)
1440  {
1441    clib_spinlock_init (&rt->lock);
1442    pool_alloc (rt->pool, rm->max_reass_n);
1443  }
1444
1445  node = vlib_get_node_by_name (vm, (u8 *) "ip6-full-reassembly-expire-walk");
1446  ASSERT (node);
1447  rm->ip6_full_reass_expire_node_idx = node->index;
1448
1449  ip6_full_reass_set_params (IP6_FULL_REASS_TIMEOUT_DEFAULT_MS,
1450			     IP6_FULL_REASS_MAX_REASSEMBLIES_DEFAULT,
1451			     IP6_FULL_REASS_MAX_REASSEMBLY_LENGTH_DEFAULT,
1452			     IP6_FULL_REASS_EXPIRE_WALK_INTERVAL_DEFAULT_MS);
1453
1454  nbuckets = ip6_full_reass_get_nbuckets ();
1455  clib_bihash_init_48_8 (&rm->hash, "ip6-full-reass", nbuckets,
1456			 nbuckets * 1024);
1457
1458  node = vlib_get_node_by_name (vm, (u8 *) "ip6-drop");
1459  ASSERT (node);
1460  rm->ip6_drop_idx = node->index;
1461  node = vlib_get_node_by_name (vm, (u8 *) "ip6-icmp-error");
1462  ASSERT (node);
1463  rm->ip6_icmp_error_idx = node->index;
1464
1465  if ((error = vlib_call_init_function (vm, ip_main_init)))
1466    return error;
1467  ip6_register_protocol (IP_PROTOCOL_IPV6_FRAGMENTATION,
1468			 ip6_full_reass_node.index);
1469
1470  rm->fq_index = vlib_frame_queue_main_init (ip6_full_reass_node.index, 0);
1471  rm->fq_feature_index =
1472    vlib_frame_queue_main_init (ip6_full_reass_node_feature.index, 0);
1473
1474  rm->feature_use_refcount_per_intf = NULL;
1475  return error;
1476}
1477
1478VLIB_INIT_FUNCTION (ip6_full_reass_init_function);
1479#endif /* CLIB_MARCH_VARIANT */
1480
1481static uword
1482ip6_full_reass_walk_expired (vlib_main_t * vm,
1483			     vlib_node_runtime_t * node, vlib_frame_t * f)
1484{
1485  ip6_full_reass_main_t *rm = &ip6_full_reass_main;
1486  uword event_type, *event_data = 0;
1487
1488  while (true)
1489    {
1490      vlib_process_wait_for_event_or_clock (vm,
1491					    (f64) rm->expire_walk_interval_ms
1492					    / (f64) MSEC_PER_SEC);
1493      event_type = vlib_process_get_events (vm, &event_data);
1494
1495      switch (event_type)
1496	{
1497	case ~0:		/* no events => timeout */
1498	  /* nothing to do here */
1499	  break;
1500	case IP6_EVENT_CONFIG_CHANGED:
1501	  break;
1502	default:
1503	  clib_warning ("BUG: event type 0x%wx", event_type);
1504	  break;
1505	}
1506      f64 now = vlib_time_now (vm);
1507
1508      ip6_full_reass_t *reass;
1509      int *pool_indexes_to_free = NULL;
1510
1511      uword thread_index = 0;
1512      int index;
1513      const uword nthreads = vlib_num_workers () + 1;
1514      u32 *vec_icmp_bi = NULL;
1515      for (thread_index = 0; thread_index < nthreads; ++thread_index)
1516	{
1517	  ip6_full_reass_per_thread_t *rt =
1518	    &rm->per_thread_data[thread_index];
1519	  clib_spinlock_lock (&rt->lock);
1520
1521	  vec_reset_length (pool_indexes_to_free);
1522          /* *INDENT-OFF* */
1523          pool_foreach_index (index, rt->pool, ({
1524                                reass = pool_elt_at_index (rt->pool, index);
1525                                if (now > reass->last_heard + rm->timeout)
1526                                  {
1527                                    vec_add1 (pool_indexes_to_free, index);
1528                                  }
1529                              }));
1530          /* *INDENT-ON* */
1531	  int *i;
1532          /* *INDENT-OFF* */
1533          vec_foreach (i, pool_indexes_to_free)
1534          {
1535            ip6_full_reass_t *reass = pool_elt_at_index (rt->pool, i[0]);
1536            u32 icmp_bi = ~0;
1537            ip6_full_reass_on_timeout (vm, node, rm, reass, &icmp_bi);
1538            if (~0 != icmp_bi)
1539              vec_add1 (vec_icmp_bi, icmp_bi);
1540
1541            ip6_full_reass_free (rm, rt, reass);
1542          }
1543          /* *INDENT-ON* */
1544
1545	  clib_spinlock_unlock (&rt->lock);
1546	}
1547
1548      while (vec_len (vec_icmp_bi) > 0)
1549	{
1550	  vlib_frame_t *f =
1551	    vlib_get_frame_to_node (vm, rm->ip6_icmp_error_idx);
1552	  u32 *to_next = vlib_frame_vector_args (f);
1553	  u32 n_left_to_next = VLIB_FRAME_SIZE - f->n_vectors;
1554	  int trace_frame = 0;
1555	  while (vec_len (vec_icmp_bi) > 0 && n_left_to_next > 0)
1556	    {
1557	      u32 bi = vec_pop (vec_icmp_bi);
1558	      vlib_buffer_t *b = vlib_get_buffer (vm, bi);
1559	      if (PREDICT_FALSE (b->flags & VLIB_BUFFER_IS_TRACED))
1560		trace_frame = 1;
1561	      b->error = node->errors[IP6_ERROR_REASS_TIMEOUT];
1562	      to_next[0] = bi;
1563	      ++f->n_vectors;
1564	      to_next += 1;
1565	      n_left_to_next -= 1;
1566	    }
1567	  f->frame_flags |= (trace_frame * VLIB_FRAME_TRACE);
1568	  vlib_put_frame_to_node (vm, rm->ip6_icmp_error_idx, f);
1569	}
1570
1571      vec_free (pool_indexes_to_free);
1572      vec_free (vec_icmp_bi);
1573      if (event_data)
1574	{
1575	  _vec_len (event_data) = 0;
1576	}
1577    }
1578
1579  return 0;
1580}
1581
1582/* *INDENT-OFF* */
1583VLIB_REGISTER_NODE (ip6_full_reass_expire_node) = {
1584    .function = ip6_full_reass_walk_expired,
1585    .format_trace = format_ip6_full_reass_trace,
1586    .type = VLIB_NODE_TYPE_PROCESS,
1587    .name = "ip6-full-reassembly-expire-walk",
1588
1589    .n_errors = ARRAY_LEN (ip6_full_reassembly_error_strings),
1590    .error_strings = ip6_full_reassembly_error_strings,
1591
1592};
1593/* *INDENT-ON* */
1594
1595static u8 *
1596format_ip6_full_reass_key (u8 * s, va_list * args)
1597{
1598  ip6_full_reass_key_t *key = va_arg (*args, ip6_full_reass_key_t *);
1599  s = format (s, "xx_id: %u, src: %U, dst: %U, frag_id: %u, proto: %u",
1600	      key->xx_id, format_ip6_address, &key->src, format_ip6_address,
1601	      &key->dst, clib_net_to_host_u16 (key->frag_id), key->proto);
1602  return s;
1603}
1604
1605static u8 *
1606format_ip6_full_reass (u8 * s, va_list * args)
1607{
1608  vlib_main_t *vm = va_arg (*args, vlib_main_t *);
1609  ip6_full_reass_t *reass = va_arg (*args, ip6_full_reass_t *);
1610
1611  s = format (s, "ID: %lu, key: %U\n  first_bi: %u, data_len: %u, "
1612	      "last_packet_octet: %u, trace_op_counter: %u\n",
1613	      reass->id, format_ip6_full_reass_key, &reass->key,
1614	      reass->first_bi, reass->data_len, reass->last_packet_octet,
1615	      reass->trace_op_counter);
1616  u32 bi = reass->first_bi;
1617  u32 counter = 0;
1618  while (~0 != bi)
1619    {
1620      vlib_buffer_t *b = vlib_get_buffer (vm, bi);
1621      vnet_buffer_opaque_t *vnb = vnet_buffer (b);
1622      s = format (s, "  #%03u: range: [%u, %u], bi: %u, off: %d, len: %u, "
1623		  "fragment[%u, %u]\n",
1624		  counter, vnb->ip.reass.range_first,
1625		  vnb->ip.reass.range_last, bi,
1626		  ip6_full_reass_buffer_get_data_offset (b),
1627		  ip6_full_reass_buffer_get_data_len (b),
1628		  vnb->ip.reass.fragment_first, vnb->ip.reass.fragment_last);
1629      if (b->flags & VLIB_BUFFER_NEXT_PRESENT)
1630	{
1631	  bi = b->next_buffer;
1632	}
1633      else
1634	{
1635	  bi = ~0;
1636	}
1637    }
1638  return s;
1639}
1640
1641static clib_error_t *
1642show_ip6_full_reass (vlib_main_t * vm, unformat_input_t * input,
1643		     CLIB_UNUSED (vlib_cli_command_t * lmd))
1644{
1645  ip6_full_reass_main_t *rm = &ip6_full_reass_main;
1646
1647  vlib_cli_output (vm, "---------------------");
1648  vlib_cli_output (vm, "IP6 reassembly status");
1649  vlib_cli_output (vm, "---------------------");
1650  bool details = false;
1651  if (unformat (input, "details"))
1652    {
1653      details = true;
1654    }
1655
1656  u32 sum_reass_n = 0;
1657  u64 sum_buffers_n = 0;
1658  ip6_full_reass_t *reass;
1659  uword thread_index;
1660  const uword nthreads = vlib_num_workers () + 1;
1661  for (thread_index = 0; thread_index < nthreads; ++thread_index)
1662    {
1663      ip6_full_reass_per_thread_t *rt = &rm->per_thread_data[thread_index];
1664      clib_spinlock_lock (&rt->lock);
1665      if (details)
1666	{
1667          /* *INDENT-OFF* */
1668          pool_foreach (reass, rt->pool, {
1669            vlib_cli_output (vm, "%U", format_ip6_full_reass, vm, reass);
1670          });
1671          /* *INDENT-ON* */
1672	}
1673      sum_reass_n += rt->reass_n;
1674      clib_spinlock_unlock (&rt->lock);
1675    }
1676  vlib_cli_output (vm, "---------------------");
1677  vlib_cli_output (vm, "Current IP6 reassemblies count: %lu\n",
1678		   (long unsigned) sum_reass_n);
1679  vlib_cli_output (vm,
1680		   "Maximum configured concurrent full IP6 reassemblies per worker-thread: %lu\n",
1681		   (long unsigned) rm->max_reass_n);
1682  vlib_cli_output (vm,
1683		   "Maximum configured full IP6 reassembly timeout: %lums\n",
1684		   (long unsigned) rm->timeout_ms);
1685  vlib_cli_output (vm,
1686		   "Maximum configured full IP6 reassembly expire walk interval: %lums\n",
1687		   (long unsigned) rm->expire_walk_interval_ms);
1688  vlib_cli_output (vm, "Buffers in use: %lu\n",
1689		   (long unsigned) sum_buffers_n);
1690  return 0;
1691}
1692
1693/* *INDENT-OFF* */
1694VLIB_CLI_COMMAND (show_ip6_full_reassembly_cmd, static) = {
1695    .path = "show ip6-full-reassembly",
1696    .short_help = "show ip6-full-reassembly [details]",
1697    .function = show_ip6_full_reass,
1698};
1699/* *INDENT-ON* */
1700
1701#ifndef CLIB_MARCH_VARIANT
1702vnet_api_error_t
1703ip6_full_reass_enable_disable (u32 sw_if_index, u8 enable_disable)
1704{
1705  return vnet_feature_enable_disable ("ip6-unicast",
1706				      "ip6-full-reassembly-feature",
1707				      sw_if_index, enable_disable, 0, 0);
1708}
1709#endif /* CLIB_MARCH_VARIANT */
1710
1711#define foreach_ip6_full_reassembly_handoff_error                       \
1712_(CONGESTION_DROP, "congestion drop")
1713
1714
1715typedef enum
1716{
1717#define _(sym,str) IP6_FULL_REASSEMBLY_HANDOFF_ERROR_##sym,
1718  foreach_ip6_full_reassembly_handoff_error
1719#undef _
1720    IP6_FULL_REASSEMBLY_HANDOFF_N_ERROR,
1721} ip6_full_reassembly_handoff_error_t;
1722
1723static char *ip6_full_reassembly_handoff_error_strings[] = {
1724#define _(sym,string) string,
1725  foreach_ip6_full_reassembly_handoff_error
1726#undef _
1727};
1728
1729typedef struct
1730{
1731  u32 next_worker_index;
1732} ip6_full_reassembly_handoff_trace_t;
1733
1734static u8 *
1735format_ip6_full_reassembly_handoff_trace (u8 * s, va_list * args)
1736{
1737  CLIB_UNUSED (vlib_main_t * vm) = va_arg (*args, vlib_main_t *);
1738  CLIB_UNUSED (vlib_node_t * node) = va_arg (*args, vlib_node_t *);
1739  ip6_full_reassembly_handoff_trace_t *t =
1740    va_arg (*args, ip6_full_reassembly_handoff_trace_t *);
1741
1742  s =
1743    format (s, "ip6-full-reassembly-handoff: next-worker %d",
1744	    t->next_worker_index);
1745
1746  return s;
1747}
1748
1749always_inline uword
1750ip6_full_reassembly_handoff_inline (vlib_main_t * vm,
1751				    vlib_node_runtime_t * node,
1752				    vlib_frame_t * frame, bool is_feature)
1753{
1754  ip6_full_reass_main_t *rm = &ip6_full_reass_main;
1755
1756  vlib_buffer_t *bufs[VLIB_FRAME_SIZE], **b;
1757  u32 n_enq, n_left_from, *from;
1758  u16 thread_indices[VLIB_FRAME_SIZE], *ti;
1759  u32 fq_index;
1760
1761  from = vlib_frame_vector_args (frame);
1762  n_left_from = frame->n_vectors;
1763  vlib_get_buffers (vm, from, bufs, n_left_from);
1764
1765  b = bufs;
1766  ti = thread_indices;
1767
1768  fq_index = (is_feature) ? rm->fq_feature_index : rm->fq_index;
1769
1770  while (n_left_from > 0)
1771    {
1772      ti[0] = vnet_buffer (b[0])->ip.reass.owner_thread_index;
1773
1774      if (PREDICT_FALSE
1775	  ((node->flags & VLIB_NODE_FLAG_TRACE)
1776	   && (b[0]->flags & VLIB_BUFFER_IS_TRACED)))
1777	{
1778	  ip6_full_reassembly_handoff_trace_t *t =
1779	    vlib_add_trace (vm, node, b[0], sizeof (*t));
1780	  t->next_worker_index = ti[0];
1781	}
1782
1783      n_left_from -= 1;
1784      ti += 1;
1785      b += 1;
1786    }
1787  n_enq =
1788    vlib_buffer_enqueue_to_thread (vm, fq_index, from, thread_indices,
1789				   frame->n_vectors, 1);
1790
1791  if (n_enq < frame->n_vectors)
1792    vlib_node_increment_counter (vm, node->node_index,
1793				 IP6_FULL_REASSEMBLY_HANDOFF_ERROR_CONGESTION_DROP,
1794				 frame->n_vectors - n_enq);
1795  return frame->n_vectors;
1796}
1797
1798VLIB_NODE_FN (ip6_full_reassembly_handoff_node) (vlib_main_t * vm,
1799						 vlib_node_runtime_t * node,
1800						 vlib_frame_t * frame)
1801{
1802  return ip6_full_reassembly_handoff_inline (vm, node, frame,
1803					     false /* is_feature */ );
1804}
1805
1806/* *INDENT-OFF* */
1807VLIB_REGISTER_NODE (ip6_full_reassembly_handoff_node) = {
1808  .name = "ip6-full-reassembly-handoff",
1809  .vector_size = sizeof (u32),
1810  .n_errors = ARRAY_LEN(ip6_full_reassembly_handoff_error_strings),
1811  .error_strings = ip6_full_reassembly_handoff_error_strings,
1812  .format_trace = format_ip6_full_reassembly_handoff_trace,
1813
1814  .n_next_nodes = 1,
1815
1816  .next_nodes = {
1817    [0] = "error-drop",
1818  },
1819};
1820
1821
1822VLIB_NODE_FN (ip6_full_reassembly_feature_handoff_node) (vlib_main_t * vm,
1823                               vlib_node_runtime_t * node, vlib_frame_t * frame)
1824{
1825  return ip6_full_reassembly_handoff_inline (vm, node, frame, true /* is_feature */ );
1826}
1827
1828
1829/* *INDENT-OFF* */
1830VLIB_REGISTER_NODE (ip6_full_reassembly_feature_handoff_node) = {
1831  .name = "ip6-full-reass-feature-hoff",
1832  .vector_size = sizeof (u32),
1833  .n_errors = ARRAY_LEN(ip6_full_reassembly_handoff_error_strings),
1834  .error_strings = ip6_full_reassembly_handoff_error_strings,
1835  .format_trace = format_ip6_full_reassembly_handoff_trace,
1836
1837  .n_next_nodes = 1,
1838
1839  .next_nodes = {
1840    [0] = "error-drop",
1841  },
1842};
1843/* *INDENT-ON* */
1844
1845#ifndef CLIB_MARCH_VARIANT
1846int
1847ip6_full_reass_enable_disable_with_refcnt (u32 sw_if_index, int is_enable)
1848{
1849  ip6_full_reass_main_t *rm = &ip6_full_reass_main;
1850  vec_validate (rm->feature_use_refcount_per_intf, sw_if_index);
1851  if (is_enable)
1852    {
1853      if (!rm->feature_use_refcount_per_intf[sw_if_index])
1854	{
1855	  ++rm->feature_use_refcount_per_intf[sw_if_index];
1856	  return vnet_feature_enable_disable ("ip6-unicast",
1857					      "ip6-full-reassembly-feature",
1858					      sw_if_index, 1, 0, 0);
1859	}
1860      ++rm->feature_use_refcount_per_intf[sw_if_index];
1861    }
1862  else
1863    {
1864      --rm->feature_use_refcount_per_intf[sw_if_index];
1865      if (!rm->feature_use_refcount_per_intf[sw_if_index])
1866	return vnet_feature_enable_disable ("ip6-unicast",
1867					    "ip6-full-reassembly-feature",
1868					    sw_if_index, 0, 0, 0);
1869    }
1870  return -1;
1871}
1872#endif
1873
1874/*
1875 * fd.io coding-style-patch-verification: ON
1876 *
1877 * Local Variables:
1878 * eval: (c-set-style "gnu")
1879 * End:
1880 */
1881