ip6_full_reass.c revision cf7803d2
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
510      (&rm->hash, (clib_bihash_kv_48_8_t *) kv, (clib_bihash_kv_48_8_t *) kv))
511    {
512      reass =
513	pool_elt_at_index (rm->per_thread_data
514			   [kv->v.memory_owner_thread_index].pool,
515			   kv->v.reass_index);
516      if (vm->thread_index != kv->v.memory_owner_thread_index)
517	{
518	  *do_handoff = 1;
519	  return reass;
520	}
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] = ((clib_bihash_kv_48_8_t *) kv)->key[0];
556  reass->key.as_u64[1] = ((clib_bihash_kv_48_8_t *) kv)->key[1];
557  reass->key.as_u64[2] = ((clib_bihash_kv_48_8_t *) kv)->key[2];
558  reass->key.as_u64[3] = ((clib_bihash_kv_48_8_t *) kv)->key[3];
559  reass->key.as_u64[4] = ((clib_bihash_kv_48_8_t *) kv)->key[4];
560  reass->key.as_u64[5] = ((clib_bihash_kv_48_8_t *) 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 =
566    clib_bihash_add_del_48_8 (&rm->hash, (clib_bihash_kv_48_8_t *) kv, 2);
567  if (rv)
568    {
569      ip6_full_reass_free (rm, rt, reass);
570      reass = NULL;
571      // if other worker created a context already work with the other copy
572      if (-2 == rv)
573	goto again;
574    }
575
576  return reass;
577}
578
579always_inline ip6_full_reass_rc_t
580ip6_full_reass_finalize (vlib_main_t * vm, vlib_node_runtime_t * node,
581			 ip6_full_reass_main_t * rm,
582			 ip6_full_reass_per_thread_t * rt,
583			 ip6_full_reass_t * reass, u32 * bi0, u32 * next0,
584			 u32 * error0, bool is_custom_app)
585{
586  *bi0 = reass->first_bi;
587  *error0 = IP6_ERROR_NONE;
588  ip6_frag_hdr_t *frag_hdr;
589  vlib_buffer_t *last_b = NULL;
590  u32 sub_chain_bi = reass->first_bi;
591  u32 total_length = 0;
592  u32 buf_cnt = 0;
593  u32 dropped_cnt = 0;
594  u32 *vec_drop_compress = NULL;
595  ip6_full_reass_rc_t rv = IP6_FULL_REASS_RC_OK;
596  do
597    {
598      u32 tmp_bi = sub_chain_bi;
599      vlib_buffer_t *tmp = vlib_get_buffer (vm, tmp_bi);
600      vnet_buffer_opaque_t *vnb = vnet_buffer (tmp);
601      if (!(vnb->ip.reass.range_first >= vnb->ip.reass.fragment_first) &&
602	  !(vnb->ip.reass.range_last > vnb->ip.reass.fragment_first))
603	{
604	  rv = IP6_FULL_REASS_RC_INTERNAL_ERROR;
605	  goto free_buffers_and_return;
606	}
607
608      u32 data_len = ip6_full_reass_buffer_get_data_len (tmp);
609      u32 trim_front = vnet_buffer (tmp)->ip.reass.ip6_frag_hdr_offset +
610	sizeof (*frag_hdr) + ip6_full_reass_buffer_get_data_offset (tmp);
611      u32 trim_end =
612	vlib_buffer_length_in_chain (vm, tmp) - trim_front - data_len;
613      if (tmp_bi == reass->first_bi)
614	{
615	  /* first buffer - keep ip6 header */
616	  if (0 != ip6_full_reass_buffer_get_data_offset (tmp))
617	    {
618	      rv = IP6_FULL_REASS_RC_INTERNAL_ERROR;
619	      goto free_buffers_and_return;
620	    }
621	  trim_front = 0;
622	  trim_end = vlib_buffer_length_in_chain (vm, tmp) - data_len -
623	    (vnet_buffer (tmp)->ip.reass.ip6_frag_hdr_offset +
624	     sizeof (*frag_hdr));
625	  if (!(vlib_buffer_length_in_chain (vm, tmp) - trim_end > 0))
626	    {
627	      rv = IP6_FULL_REASS_RC_INTERNAL_ERROR;
628	      goto free_buffers_and_return;
629	    }
630	}
631      u32 keep_data =
632	vlib_buffer_length_in_chain (vm, tmp) - trim_front - trim_end;
633      while (1)
634	{
635	  ++buf_cnt;
636	  if (trim_front)
637	    {
638	      if (trim_front > tmp->current_length)
639		{
640		  /* drop whole buffer */
641		  vec_add1 (vec_drop_compress, tmp_bi);
642		  trim_front -= tmp->current_length;
643		  if (!(tmp->flags & VLIB_BUFFER_NEXT_PRESENT))
644		    {
645		      rv = IP6_FULL_REASS_RC_INTERNAL_ERROR;
646		      goto free_buffers_and_return;
647		    }
648		  tmp->flags &= ~VLIB_BUFFER_NEXT_PRESENT;
649		  tmp_bi = tmp->next_buffer;
650		  tmp = vlib_get_buffer (vm, tmp_bi);
651		  continue;
652		}
653	      else
654		{
655		  vlib_buffer_advance (tmp, trim_front);
656		  trim_front = 0;
657		}
658	    }
659	  if (keep_data)
660	    {
661	      if (last_b)
662		{
663		  last_b->flags |= VLIB_BUFFER_NEXT_PRESENT;
664		  last_b->next_buffer = tmp_bi;
665		}
666	      last_b = tmp;
667	      if (keep_data <= tmp->current_length)
668		{
669		  tmp->current_length = keep_data;
670		  keep_data = 0;
671		}
672	      else
673		{
674		  keep_data -= tmp->current_length;
675		  if (!(tmp->flags & VLIB_BUFFER_NEXT_PRESENT))
676		    {
677		      rv = IP6_FULL_REASS_RC_INTERNAL_ERROR;
678		      goto free_buffers_and_return;
679		    }
680		}
681	      total_length += tmp->current_length;
682	    }
683	  else
684	    {
685	      vec_add1 (vec_drop_compress, tmp_bi);
686	      if (reass->first_bi == tmp_bi)
687		{
688		  rv = IP6_FULL_REASS_RC_INTERNAL_ERROR;
689		  goto free_buffers_and_return;
690		}
691	      ++dropped_cnt;
692	    }
693	  if (tmp->flags & VLIB_BUFFER_NEXT_PRESENT)
694	    {
695	      tmp_bi = tmp->next_buffer;
696	      tmp = vlib_get_buffer (vm, tmp->next_buffer);
697	    }
698	  else
699	    {
700	      break;
701	    }
702	}
703      sub_chain_bi =
704	vnet_buffer (vlib_get_buffer (vm, sub_chain_bi))->ip.
705	reass.next_range_bi;
706    }
707  while (~0 != sub_chain_bi);
708
709  if (!last_b)
710    {
711      rv = IP6_FULL_REASS_RC_INTERNAL_ERROR;
712      goto free_buffers_and_return;
713    }
714  last_b->flags &= ~VLIB_BUFFER_NEXT_PRESENT;
715  vlib_buffer_t *first_b = vlib_get_buffer (vm, reass->first_bi);
716  if (total_length < first_b->current_length)
717    {
718      rv = IP6_FULL_REASS_RC_INTERNAL_ERROR;
719      goto free_buffers_and_return;
720    }
721  total_length -= first_b->current_length;
722  first_b->flags |= VLIB_BUFFER_TOTAL_LENGTH_VALID;
723  first_b->total_length_not_including_first_buffer = total_length;
724  // drop fragment header
725  vnet_buffer_opaque_t *first_b_vnb = vnet_buffer (first_b);
726  ip6_header_t *ip = vlib_buffer_get_current (first_b);
727  u16 ip6_frag_hdr_offset = first_b_vnb->ip.reass.ip6_frag_hdr_offset;
728  ip6_ext_header_t *prev_hdr;
729  frag_hdr =
730    ip6_ext_header_find (vm, first_b, ip, IP_PROTOCOL_IPV6_FRAGMENTATION,
731			 &prev_hdr);
732  if (prev_hdr)
733    {
734      prev_hdr->next_hdr = frag_hdr->next_hdr;
735    }
736  else
737    {
738      ip->protocol = frag_hdr->next_hdr;
739    }
740  if (!((u8 *) frag_hdr - (u8 *) ip == ip6_frag_hdr_offset))
741    {
742      rv = IP6_FULL_REASS_RC_INTERNAL_ERROR;
743      goto free_buffers_and_return;
744    }
745  memmove (frag_hdr, (u8 *) frag_hdr + sizeof (*frag_hdr),
746	   first_b->current_length - ip6_frag_hdr_offset -
747	   sizeof (ip6_frag_hdr_t));
748  first_b->current_length -= sizeof (*frag_hdr);
749  ip->payload_length =
750    clib_host_to_net_u16 (total_length + first_b->current_length -
751			  sizeof (*ip));
752  if (!vlib_buffer_chain_linearize (vm, first_b))
753    {
754      rv = IP6_FULL_REASS_RC_NO_BUF;
755      goto free_buffers_and_return;
756    }
757  first_b->flags &= ~VLIB_BUFFER_EXT_HDR_VALID;
758  if (PREDICT_FALSE (first_b->flags & VLIB_BUFFER_IS_TRACED))
759    {
760      ip6_full_reass_add_trace (vm, node, rm, reass, reass->first_bi, NULL,
761				FINALIZE, ~0);
762#if 0
763      // following code does a hexdump of packet fragments to stdout ...
764      do
765	{
766	  u32 bi = reass->first_bi;
767	  u8 *s = NULL;
768	  while (~0 != bi)
769	    {
770	      vlib_buffer_t *b = vlib_get_buffer (vm, bi);
771	      s = format (s, "%u: %U\n", bi, format_hexdump,
772			  vlib_buffer_get_current (b), b->current_length);
773	      if (b->flags & VLIB_BUFFER_NEXT_PRESENT)
774		{
775		  bi = b->next_buffer;
776		}
777	      else
778		{
779		  break;
780		}
781	    }
782	  printf ("%.*s\n", vec_len (s), s);
783	  fflush (stdout);
784	  vec_free (s);
785	}
786      while (0);
787#endif
788    }
789  if (!is_custom_app)
790    {
791      *next0 = IP6_FULL_REASSEMBLY_NEXT_INPUT;
792    }
793  else
794    {
795      *next0 = reass->next_index;
796    }
797  vnet_buffer (first_b)->ip.reass.estimated_mtu = reass->min_fragment_length;
798  ip6_full_reass_free (rm, rt, reass);
799  reass = NULL;
800free_buffers_and_return:
801  vlib_buffer_free (vm, vec_drop_compress, vec_len (vec_drop_compress));
802  vec_free (vec_drop_compress);
803  return rv;
804}
805
806always_inline void
807ip6_full_reass_insert_range_in_chain (vlib_main_t * vm,
808				      ip6_full_reass_main_t * rm,
809				      ip6_full_reass_per_thread_t * rt,
810				      ip6_full_reass_t * reass,
811				      u32 prev_range_bi, u32 new_next_bi)
812{
813
814  vlib_buffer_t *new_next_b = vlib_get_buffer (vm, new_next_bi);
815  vnet_buffer_opaque_t *new_next_vnb = vnet_buffer (new_next_b);
816  if (~0 != prev_range_bi)
817    {
818      vlib_buffer_t *prev_b = vlib_get_buffer (vm, prev_range_bi);
819      vnet_buffer_opaque_t *prev_vnb = vnet_buffer (prev_b);
820      new_next_vnb->ip.reass.next_range_bi = prev_vnb->ip.reass.next_range_bi;
821      prev_vnb->ip.reass.next_range_bi = new_next_bi;
822    }
823  else
824    {
825      if (~0 != reass->first_bi)
826	{
827	  new_next_vnb->ip.reass.next_range_bi = reass->first_bi;
828	}
829      reass->first_bi = new_next_bi;
830    }
831  reass->data_len += ip6_full_reass_buffer_get_data_len (new_next_b);
832}
833
834always_inline ip6_full_reass_rc_t
835ip6_full_reass_update (vlib_main_t * vm, vlib_node_runtime_t * node,
836		       ip6_full_reass_main_t * rm,
837		       ip6_full_reass_per_thread_t * rt,
838		       ip6_full_reass_t * reass, u32 * bi0, u32 * next0,
839		       u32 * error0, ip6_frag_hdr_t * frag_hdr,
840		       bool is_custom_app, u32 * handoff_thread_idx)
841{
842  int consumed = 0;
843  vlib_buffer_t *fb = vlib_get_buffer (vm, *bi0);
844  vnet_buffer_opaque_t *fvnb = vnet_buffer (fb);
845  if (is_custom_app)
846    {
847      reass->next_index = fvnb->ip.reass.next_index;	// store next_index before it's overwritten
848      reass->error_next_index = fvnb->ip.reass.error_next_index;	// store error_next_index before it is overwritten
849    }
850
851  fvnb->ip.reass.ip6_frag_hdr_offset =
852    (u8 *) frag_hdr - (u8 *) vlib_buffer_get_current (fb);
853  ip6_header_t *fip = vlib_buffer_get_current (fb);
854  if (fb->current_length < sizeof (*fip) ||
855      fvnb->ip.reass.ip6_frag_hdr_offset == 0 ||
856      fvnb->ip.reass.ip6_frag_hdr_offset >= fb->current_length)
857    {
858      return IP6_FULL_REASS_RC_INTERNAL_ERROR;
859    }
860
861  u32 fragment_first = fvnb->ip.reass.fragment_first =
862    ip6_frag_hdr_offset_bytes (frag_hdr);
863  u32 fragment_length =
864    vlib_buffer_length_in_chain (vm, fb) -
865    (fvnb->ip.reass.ip6_frag_hdr_offset + sizeof (*frag_hdr));
866  u32 fragment_last = fvnb->ip.reass.fragment_last =
867    fragment_first + fragment_length - 1;
868  int more_fragments = ip6_frag_hdr_more (frag_hdr);
869  u32 candidate_range_bi = reass->first_bi;
870  u32 prev_range_bi = ~0;
871  fvnb->ip.reass.range_first = fragment_first;
872  fvnb->ip.reass.range_last = fragment_last;
873  fvnb->ip.reass.next_range_bi = ~0;
874  if (!more_fragments)
875    {
876      reass->last_packet_octet = fragment_last;
877    }
878  if (~0 == reass->first_bi)
879    {
880      // starting a new reassembly
881      ip6_full_reass_insert_range_in_chain (vm, rm, rt, reass, prev_range_bi,
882					    *bi0);
883      reass->min_fragment_length = clib_net_to_host_u16 (fip->payload_length);
884      consumed = 1;
885      reass->fragments_n = 1;
886      goto check_if_done_maybe;
887    }
888  reass->min_fragment_length =
889    clib_min (clib_net_to_host_u16 (fip->payload_length),
890	      fvnb->ip.reass.estimated_mtu);
891  while (~0 != candidate_range_bi)
892    {
893      vlib_buffer_t *candidate_b = vlib_get_buffer (vm, candidate_range_bi);
894      vnet_buffer_opaque_t *candidate_vnb = vnet_buffer (candidate_b);
895      if (fragment_first > candidate_vnb->ip.reass.range_last)
896	{
897	  // this fragments starts after candidate range
898	  prev_range_bi = candidate_range_bi;
899	  candidate_range_bi = candidate_vnb->ip.reass.next_range_bi;
900	  if (candidate_vnb->ip.reass.range_last < fragment_last &&
901	      ~0 == candidate_range_bi)
902	    {
903	      // special case - this fragment falls beyond all known ranges
904	      ip6_full_reass_insert_range_in_chain (vm, rm, rt, reass,
905						    prev_range_bi, *bi0);
906	      consumed = 1;
907	      break;
908	    }
909	  continue;
910	}
911      if (fragment_last < candidate_vnb->ip.reass.range_first)
912	{
913	  // this fragment ends before candidate range without any overlap
914	  ip6_full_reass_insert_range_in_chain (vm, rm, rt, reass,
915						prev_range_bi, *bi0);
916	  consumed = 1;
917	}
918      else if (fragment_first == candidate_vnb->ip.reass.range_first &&
919	       fragment_last == candidate_vnb->ip.reass.range_last)
920	{
921	  // duplicate fragment - ignore
922	}
923      else
924	{
925	  // overlapping fragment - not allowed by RFC 8200
926	  if (PREDICT_FALSE (fb->flags & VLIB_BUFFER_IS_TRACED))
927	    {
928	      ip6_full_reass_add_trace (vm, node, rm, reass, *bi0, frag_hdr,
929					RANGE_OVERLAP, ~0);
930	    }
931	  ip6_full_reass_drop_all (vm, node, rm, reass);
932	  ip6_full_reass_free (rm, rt, reass);
933	  *next0 = IP6_FULL_REASSEMBLY_NEXT_DROP;
934	  *error0 = IP6_ERROR_REASS_OVERLAPPING_FRAGMENT;
935	  return IP6_FULL_REASS_RC_OK;
936	}
937      break;
938    }
939  ++reass->fragments_n;
940check_if_done_maybe:
941  if (consumed)
942    {
943      if (PREDICT_FALSE (fb->flags & VLIB_BUFFER_IS_TRACED))
944	{
945	  ip6_full_reass_add_trace (vm, node, rm, reass, *bi0, frag_hdr,
946				    RANGE_NEW, ~0);
947	}
948    }
949  if (~0 != reass->last_packet_octet &&
950      reass->data_len == reass->last_packet_octet + 1)
951    {
952      *handoff_thread_idx = reass->sendout_thread_index;
953      int handoff =
954	reass->memory_owner_thread_index != reass->sendout_thread_index;
955      ip6_full_reass_rc_t rc =
956	ip6_full_reass_finalize (vm, node, rm, rt, reass, bi0, next0, error0,
957				 is_custom_app);
958      if (IP6_FULL_REASS_RC_OK == rc && handoff)
959	{
960	  return IP6_FULL_REASS_RC_HANDOFF;
961	}
962      return rc;
963    }
964  else
965    {
966      if (consumed)
967	{
968	  *bi0 = ~0;
969	  if (reass->fragments_n > rm->max_reass_len)
970	    {
971	      return IP6_FULL_REASS_RC_TOO_MANY_FRAGMENTS;
972	    }
973	}
974      else
975	{
976	  *next0 = IP6_FULL_REASSEMBLY_NEXT_DROP;
977	  *error0 = IP6_ERROR_REASS_DUPLICATE_FRAGMENT;
978	}
979    }
980  return IP6_FULL_REASS_RC_OK;
981}
982
983always_inline bool
984ip6_full_reass_verify_upper_layer_present (vlib_node_runtime_t * node,
985					   vlib_buffer_t * b,
986					   ip6_frag_hdr_t * frag_hdr)
987{
988  ip6_ext_header_t *tmp = (ip6_ext_header_t *) frag_hdr;
989  while (ip6_ext_hdr (tmp->next_hdr))
990    {
991      tmp = ip6_ext_next_header (tmp);
992    }
993  if (IP_PROTOCOL_IP6_NONXT == tmp->next_hdr)
994    {
995      icmp6_error_set_vnet_buffer (b, ICMP6_parameter_problem,
996				   ICMP6_parameter_problem_first_fragment_has_incomplete_header_chain,
997				   0);
998      b->error = node->errors[IP6_ERROR_REASS_MISSING_UPPER];
999
1000      return false;
1001    }
1002  return true;
1003}
1004
1005always_inline bool
1006ip6_full_reass_verify_fragment_multiple_8 (vlib_main_t * vm,
1007					   vlib_node_runtime_t * node,
1008					   vlib_buffer_t * b,
1009					   ip6_frag_hdr_t * frag_hdr)
1010{
1011  vnet_buffer_opaque_t *vnb = vnet_buffer (b);
1012  ip6_header_t *ip = vlib_buffer_get_current (b);
1013  int more_fragments = ip6_frag_hdr_more (frag_hdr);
1014  u32 fragment_length =
1015    vlib_buffer_length_in_chain (vm, b) -
1016    (vnb->ip.reass.ip6_frag_hdr_offset + sizeof (*frag_hdr));
1017  if (more_fragments && 0 != fragment_length % 8)
1018    {
1019      icmp6_error_set_vnet_buffer (b, ICMP6_parameter_problem,
1020				   ICMP6_parameter_problem_erroneous_header_field,
1021				   (u8 *) & ip->payload_length - (u8 *) ip);
1022      return false;
1023    }
1024  return true;
1025}
1026
1027always_inline bool
1028ip6_full_reass_verify_packet_size_lt_64k (vlib_main_t * vm,
1029					  vlib_node_runtime_t * node,
1030					  vlib_buffer_t * b,
1031					  ip6_frag_hdr_t * frag_hdr)
1032{
1033  vnet_buffer_opaque_t *vnb = vnet_buffer (b);
1034  u32 fragment_first = ip6_frag_hdr_offset_bytes (frag_hdr);
1035  u32 fragment_length =
1036    vlib_buffer_length_in_chain (vm, b) -
1037    (vnb->ip.reass.ip6_frag_hdr_offset + sizeof (*frag_hdr));
1038  if (fragment_first + fragment_length > 65535)
1039    {
1040      ip6_header_t *ip0 = vlib_buffer_get_current (b);
1041      icmp6_error_set_vnet_buffer (b, ICMP6_parameter_problem,
1042				   ICMP6_parameter_problem_erroneous_header_field,
1043				   (u8 *) & frag_hdr->fragment_offset_and_more
1044				   - (u8 *) ip0);
1045      return false;
1046    }
1047  return true;
1048}
1049
1050always_inline uword
1051ip6_full_reassembly_inline (vlib_main_t * vm,
1052			    vlib_node_runtime_t * node,
1053			    vlib_frame_t * frame, bool is_feature,
1054			    bool is_custom_app)
1055{
1056  u32 *from = vlib_frame_vector_args (frame);
1057  u32 n_left_from, n_left_to_next, *to_next, next_index;
1058  ip6_full_reass_main_t *rm = &ip6_full_reass_main;
1059  ip6_full_reass_per_thread_t *rt = &rm->per_thread_data[vm->thread_index];
1060  clib_spinlock_lock (&rt->lock);
1061
1062  n_left_from = frame->n_vectors;
1063  next_index = node->cached_next_index;
1064  while (n_left_from > 0)
1065    {
1066      vlib_get_next_frame (vm, node, next_index, to_next, n_left_to_next);
1067
1068      while (n_left_from > 0 && n_left_to_next > 0)
1069	{
1070	  u32 bi0;
1071	  vlib_buffer_t *b0;
1072	  u32 next0 = IP6_FULL_REASSEMBLY_NEXT_DROP;
1073	  u32 error0 = IP6_ERROR_NONE;
1074	  u32 icmp_bi = ~0;
1075
1076	  bi0 = from[0];
1077	  b0 = vlib_get_buffer (vm, bi0);
1078
1079	  ip6_header_t *ip0 = vlib_buffer_get_current (b0);
1080	  ip6_frag_hdr_t *frag_hdr = NULL;
1081	  ip6_ext_header_t *prev_hdr;
1082	  if (ip6_ext_hdr (ip0->protocol))
1083	    {
1084	      frag_hdr =
1085		ip6_ext_header_find (vm, b0, ip0,
1086				     IP_PROTOCOL_IPV6_FRAGMENTATION,
1087				     &prev_hdr);
1088	    }
1089	  if (!frag_hdr)
1090	    {
1091	      // this is a regular packet - no fragmentation
1092	      next0 = IP6_FULL_REASSEMBLY_NEXT_INPUT;
1093	      goto skip_reass;
1094	    }
1095	  if (0 == ip6_frag_hdr_offset (frag_hdr))
1096	    {
1097	      // first fragment - verify upper-layer is present
1098	      if (!ip6_full_reass_verify_upper_layer_present
1099		  (node, b0, frag_hdr))
1100		{
1101		  next0 = IP6_FULL_REASSEMBLY_NEXT_ICMP_ERROR;
1102		  goto skip_reass;
1103		}
1104	    }
1105	  if (!ip6_full_reass_verify_fragment_multiple_8
1106	      (vm, node, b0, frag_hdr)
1107	      || !ip6_full_reass_verify_packet_size_lt_64k (vm, node, b0,
1108							    frag_hdr))
1109	    {
1110	      next0 = IP6_FULL_REASSEMBLY_NEXT_ICMP_ERROR;
1111	      goto skip_reass;
1112	    }
1113	  vnet_buffer (b0)->ip.reass.ip6_frag_hdr_offset =
1114	    (u8 *) frag_hdr - (u8 *) ip0;
1115
1116	  ip6_full_reass_kv_t kv;
1117	  u8 do_handoff = 0;
1118
1119	  kv.k.as_u64[0] = ip0->src_address.as_u64[0];
1120	  kv.k.as_u64[1] = ip0->src_address.as_u64[1];
1121	  kv.k.as_u64[2] = ip0->dst_address.as_u64[0];
1122	  kv.k.as_u64[3] = ip0->dst_address.as_u64[1];
1123	  kv.k.as_u64[4] =
1124	    ((u64) vec_elt (ip6_main.fib_index_by_sw_if_index,
1125			    vnet_buffer (b0)->sw_if_index[VLIB_RX])) << 32 |
1126	    (u64) frag_hdr->identification;
1127	  kv.k.as_u64[5] = ip0->protocol;
1128
1129	  ip6_full_reass_t *reass =
1130	    ip6_full_reass_find_or_create (vm, node, rm, rt, &kv, &icmp_bi,
1131					   &do_handoff);
1132
1133	  if (reass)
1134	    {
1135	      const u32 fragment_first = ip6_frag_hdr_offset (frag_hdr);
1136	      if (0 == fragment_first)
1137		{
1138		  reass->sendout_thread_index = vm->thread_index;
1139		}
1140	    }
1141	  if (PREDICT_FALSE (do_handoff))
1142	    {
1143	      next0 = IP6_FULL_REASSEMBLY_NEXT_HANDOFF;
1144	      vnet_buffer (b0)->ip.reass.owner_thread_index =
1145		kv.v.memory_owner_thread_index;
1146	    }
1147	  else if (reass)
1148	    {
1149	      u32 handoff_thread_idx;
1150	      switch (ip6_full_reass_update
1151		      (vm, node, rm, rt, reass, &bi0, &next0, &error0,
1152		       frag_hdr, is_custom_app, &handoff_thread_idx))
1153		{
1154		case IP6_FULL_REASS_RC_OK:
1155		  /* nothing to do here */
1156		  break;
1157		case IP6_FULL_REASS_RC_HANDOFF:
1158		  next0 = IP6_FULL_REASSEMBLY_NEXT_HANDOFF;
1159		  b0 = vlib_get_buffer (vm, bi0);
1160		  vnet_buffer (b0)->ip.reass.owner_thread_index =
1161		    handoff_thread_idx;
1162		  break;
1163		case IP6_FULL_REASS_RC_TOO_MANY_FRAGMENTS:
1164		  vlib_node_increment_counter (vm, node->node_index,
1165					       IP6_ERROR_REASS_FRAGMENT_CHAIN_TOO_LONG,
1166					       1);
1167		  ip6_full_reass_drop_all (vm, node, rm, reass);
1168		  ip6_full_reass_free (rm, rt, reass);
1169		  goto next_packet;
1170		  break;
1171		case IP6_FULL_REASS_RC_NO_BUF:
1172		  vlib_node_increment_counter (vm, node->node_index,
1173					       IP6_ERROR_REASS_NO_BUF, 1);
1174		  ip6_full_reass_drop_all (vm, node, rm, reass);
1175		  ip6_full_reass_free (rm, rt, reass);
1176		  goto next_packet;
1177		  break;
1178		case IP6_FULL_REASS_RC_INTERNAL_ERROR:
1179		  vlib_node_increment_counter (vm, node->node_index,
1180					       IP6_ERROR_REASS_INTERNAL_ERROR,
1181					       1);
1182		  ip6_full_reass_drop_all (vm, node, rm, reass);
1183		  ip6_full_reass_free (rm, rt, reass);
1184		  goto next_packet;
1185		  break;
1186		}
1187	    }
1188	  else
1189	    {
1190	      if (is_feature)
1191		{
1192		  next0 = IP6_FULL_REASSEMBLY_NEXT_DROP;
1193		}
1194	      else
1195		{
1196		  vnet_buffer_opaque_t *fvnb = vnet_buffer (b0);
1197		  next0 = fvnb->ip.reass.error_next_index;
1198		}
1199	      error0 = IP6_ERROR_REASS_LIMIT_REACHED;
1200	    }
1201
1202	  if (~0 != bi0)
1203	    {
1204	    skip_reass:
1205	      to_next[0] = bi0;
1206	      to_next += 1;
1207	      n_left_to_next -= 1;
1208
1209	      /* bi0 might have been updated by reass_finalize, reload */
1210	      b0 = vlib_get_buffer (vm, bi0);
1211	      b0->error = node->errors[error0];
1212
1213	      if (next0 == IP6_FULL_REASSEMBLY_NEXT_HANDOFF)
1214		{
1215		  if (PREDICT_FALSE (b0->flags & VLIB_BUFFER_IS_TRACED))
1216		    {
1217		      ip6_full_reass_add_trace (vm, node, rm, NULL, bi0,
1218						frag_hdr, HANDOFF,
1219						vnet_buffer (b0)->ip.
1220						reass.owner_thread_index);
1221		    }
1222		}
1223	      else if (is_feature && IP6_ERROR_NONE == error0)
1224		{
1225		  vnet_feature_next (&next0, b0);
1226		}
1227	      vlib_validate_buffer_enqueue_x1 (vm, node, next_index, to_next,
1228					       n_left_to_next, bi0, next0);
1229	    }
1230
1231	  if (~0 != icmp_bi)
1232	    {
1233	      next0 = IP6_FULL_REASSEMBLY_NEXT_ICMP_ERROR;
1234	      to_next[0] = icmp_bi;
1235	      to_next += 1;
1236	      n_left_to_next -= 1;
1237	      vlib_validate_buffer_enqueue_x1 (vm, node, next_index, to_next,
1238					       n_left_to_next, icmp_bi,
1239					       next0);
1240	    }
1241	next_packet:
1242	  from += 1;
1243	  n_left_from -= 1;
1244	}
1245
1246      vlib_put_next_frame (vm, node, next_index, n_left_to_next);
1247    }
1248
1249  clib_spinlock_unlock (&rt->lock);
1250  return frame->n_vectors;
1251}
1252
1253static char *ip6_full_reassembly_error_strings[] = {
1254#define _(sym, string) string,
1255  foreach_ip6_error
1256#undef _
1257};
1258
1259VLIB_NODE_FN (ip6_full_reass_node) (vlib_main_t * vm,
1260				    vlib_node_runtime_t * node,
1261				    vlib_frame_t * frame)
1262{
1263  return ip6_full_reassembly_inline (vm, node, frame, false /* is_feature */ ,
1264				     false /* is_custom_app */ );
1265}
1266
1267/* *INDENT-OFF* */
1268VLIB_REGISTER_NODE (ip6_full_reass_node) = {
1269    .name = "ip6-full-reassembly",
1270    .vector_size = sizeof (u32),
1271    .format_trace = format_ip6_full_reass_trace,
1272    .n_errors = ARRAY_LEN (ip6_full_reassembly_error_strings),
1273    .error_strings = ip6_full_reassembly_error_strings,
1274    .n_next_nodes = IP6_FULL_REASSEMBLY_N_NEXT,
1275    .next_nodes =
1276        {
1277                [IP6_FULL_REASSEMBLY_NEXT_INPUT] = "ip6-input",
1278                [IP6_FULL_REASSEMBLY_NEXT_DROP] = "ip6-drop",
1279                [IP6_FULL_REASSEMBLY_NEXT_ICMP_ERROR] = "ip6-icmp-error",
1280                [IP6_FULL_REASSEMBLY_NEXT_HANDOFF] = "ip6-full-reassembly-handoff",
1281        },
1282};
1283/* *INDENT-ON* */
1284
1285VLIB_NODE_FN (ip6_full_reass_node_feature) (vlib_main_t * vm,
1286					    vlib_node_runtime_t * node,
1287					    vlib_frame_t * frame)
1288{
1289  return ip6_full_reassembly_inline (vm, node, frame, true /* is_feature */ ,
1290				     false /* is_custom_app */ );
1291}
1292
1293/* *INDENT-OFF* */
1294VLIB_REGISTER_NODE (ip6_full_reass_node_feature) = {
1295    .name = "ip6-full-reassembly-feature",
1296    .vector_size = sizeof (u32),
1297    .format_trace = format_ip6_full_reass_trace,
1298    .n_errors = ARRAY_LEN (ip6_full_reassembly_error_strings),
1299    .error_strings = ip6_full_reassembly_error_strings,
1300    .n_next_nodes = IP6_FULL_REASSEMBLY_N_NEXT,
1301    .next_nodes =
1302        {
1303                [IP6_FULL_REASSEMBLY_NEXT_INPUT] = "ip6-input",
1304                [IP6_FULL_REASSEMBLY_NEXT_DROP] = "ip6-drop",
1305                [IP6_FULL_REASSEMBLY_NEXT_ICMP_ERROR] = "ip6-icmp-error",
1306                [IP6_FULL_REASSEMBLY_NEXT_HANDOFF] = "ip6-full-reass-feature-hoff",
1307        },
1308};
1309/* *INDENT-ON* */
1310
1311/* *INDENT-OFF* */
1312VNET_FEATURE_INIT (ip6_full_reassembly_feature, static) = {
1313    .arc_name = "ip6-unicast",
1314    .node_name = "ip6-full-reassembly-feature",
1315    .runs_before = VNET_FEATURES ("ip6-lookup",
1316                                  "ipsec6-input-feature"),
1317    .runs_after = 0,
1318};
1319/* *INDENT-ON* */
1320
1321#ifndef CLIB_MARCH_VARIANT
1322static u32
1323ip6_full_reass_get_nbuckets ()
1324{
1325  ip6_full_reass_main_t *rm = &ip6_full_reass_main;
1326  u32 nbuckets;
1327  u8 i;
1328
1329  nbuckets = (u32) (rm->max_reass_n / IP6_FULL_REASS_HT_LOAD_FACTOR);
1330
1331  for (i = 0; i < 31; i++)
1332    if ((1 << i) >= nbuckets)
1333      break;
1334  nbuckets = 1 << i;
1335
1336  return nbuckets;
1337}
1338#endif /* CLIB_MARCH_VARIANT */
1339
1340typedef enum
1341{
1342  IP6_EVENT_CONFIG_CHANGED = 1,
1343} ip6_full_reass_event_t;
1344
1345#ifndef CLIB_MARCH_VARIANT
1346typedef struct
1347{
1348  int failure;
1349  clib_bihash_48_8_t *new_hash;
1350} ip6_rehash_cb_ctx;
1351
1352static void
1353ip6_rehash_cb (clib_bihash_kv_48_8_t * kv, void *_ctx)
1354{
1355  ip6_rehash_cb_ctx *ctx = _ctx;
1356  if (clib_bihash_add_del_48_8 (ctx->new_hash, kv, 1))
1357    {
1358      ctx->failure = 1;
1359    }
1360}
1361
1362static void
1363ip6_full_reass_set_params (u32 timeout_ms, u32 max_reassemblies,
1364			   u32 max_reassembly_length,
1365			   u32 expire_walk_interval_ms)
1366{
1367  ip6_full_reass_main.timeout_ms = timeout_ms;
1368  ip6_full_reass_main.timeout = (f64) timeout_ms / (f64) MSEC_PER_SEC;
1369  ip6_full_reass_main.max_reass_n = max_reassemblies;
1370  ip6_full_reass_main.max_reass_len = max_reassembly_length;
1371  ip6_full_reass_main.expire_walk_interval_ms = expire_walk_interval_ms;
1372}
1373
1374vnet_api_error_t
1375ip6_full_reass_set (u32 timeout_ms, u32 max_reassemblies,
1376		    u32 max_reassembly_length, u32 expire_walk_interval_ms)
1377{
1378  u32 old_nbuckets = ip6_full_reass_get_nbuckets ();
1379  ip6_full_reass_set_params (timeout_ms, max_reassemblies,
1380			     max_reassembly_length, expire_walk_interval_ms);
1381  vlib_process_signal_event (ip6_full_reass_main.vlib_main,
1382			     ip6_full_reass_main.ip6_full_reass_expire_node_idx,
1383			     IP6_EVENT_CONFIG_CHANGED, 0);
1384  u32 new_nbuckets = ip6_full_reass_get_nbuckets ();
1385  if (ip6_full_reass_main.max_reass_n > 0 && new_nbuckets > old_nbuckets)
1386    {
1387      clib_bihash_48_8_t new_hash;
1388      clib_memset (&new_hash, 0, sizeof (new_hash));
1389      ip6_rehash_cb_ctx ctx;
1390      ctx.failure = 0;
1391      ctx.new_hash = &new_hash;
1392      clib_bihash_init_48_8 (&new_hash, "ip6-full-reass", new_nbuckets,
1393			     new_nbuckets * 1024);
1394      clib_bihash_foreach_key_value_pair_48_8 (&ip6_full_reass_main.hash,
1395					       ip6_rehash_cb, &ctx);
1396      if (ctx.failure)
1397	{
1398	  clib_bihash_free_48_8 (&new_hash);
1399	  return -1;
1400	}
1401      else
1402	{
1403	  clib_bihash_free_48_8 (&ip6_full_reass_main.hash);
1404	  clib_memcpy_fast (&ip6_full_reass_main.hash, &new_hash,
1405			    sizeof (ip6_full_reass_main.hash));
1406	  clib_bihash_copied (&ip6_full_reass_main.hash, &new_hash);
1407	}
1408    }
1409  return 0;
1410}
1411
1412vnet_api_error_t
1413ip6_full_reass_get (u32 * timeout_ms, u32 * max_reassemblies,
1414		    u32 * max_reassembly_length,
1415		    u32 * expire_walk_interval_ms)
1416{
1417  *timeout_ms = ip6_full_reass_main.timeout_ms;
1418  *max_reassemblies = ip6_full_reass_main.max_reass_n;
1419  *max_reassembly_length = ip6_full_reass_main.max_reass_len;
1420  *expire_walk_interval_ms = ip6_full_reass_main.expire_walk_interval_ms;
1421  return 0;
1422}
1423
1424static clib_error_t *
1425ip6_full_reass_init_function (vlib_main_t * vm)
1426{
1427  ip6_full_reass_main_t *rm = &ip6_full_reass_main;
1428  clib_error_t *error = 0;
1429  u32 nbuckets;
1430  vlib_node_t *node;
1431
1432  rm->vlib_main = vm;
1433
1434  vec_validate (rm->per_thread_data, vlib_num_workers ());
1435  ip6_full_reass_per_thread_t *rt;
1436  vec_foreach (rt, rm->per_thread_data)
1437  {
1438    clib_spinlock_init (&rt->lock);
1439    pool_alloc (rt->pool, rm->max_reass_n);
1440  }
1441
1442  node = vlib_get_node_by_name (vm, (u8 *) "ip6-full-reassembly-expire-walk");
1443  ASSERT (node);
1444  rm->ip6_full_reass_expire_node_idx = node->index;
1445
1446  ip6_full_reass_set_params (IP6_FULL_REASS_TIMEOUT_DEFAULT_MS,
1447			     IP6_FULL_REASS_MAX_REASSEMBLIES_DEFAULT,
1448			     IP6_FULL_REASS_MAX_REASSEMBLY_LENGTH_DEFAULT,
1449			     IP6_FULL_REASS_EXPIRE_WALK_INTERVAL_DEFAULT_MS);
1450
1451  nbuckets = ip6_full_reass_get_nbuckets ();
1452  clib_bihash_init_48_8 (&rm->hash, "ip6-full-reass", nbuckets,
1453			 nbuckets * 1024);
1454
1455  node = vlib_get_node_by_name (vm, (u8 *) "ip6-drop");
1456  ASSERT (node);
1457  rm->ip6_drop_idx = node->index;
1458  node = vlib_get_node_by_name (vm, (u8 *) "ip6-icmp-error");
1459  ASSERT (node);
1460  rm->ip6_icmp_error_idx = node->index;
1461
1462  if ((error = vlib_call_init_function (vm, ip_main_init)))
1463    return error;
1464  ip6_register_protocol (IP_PROTOCOL_IPV6_FRAGMENTATION,
1465			 ip6_full_reass_node.index);
1466
1467  rm->fq_index = vlib_frame_queue_main_init (ip6_full_reass_node.index, 0);
1468  rm->fq_feature_index =
1469    vlib_frame_queue_main_init (ip6_full_reass_node_feature.index, 0);
1470
1471  rm->feature_use_refcount_per_intf = NULL;
1472  return error;
1473}
1474
1475VLIB_INIT_FUNCTION (ip6_full_reass_init_function);
1476#endif /* CLIB_MARCH_VARIANT */
1477
1478static uword
1479ip6_full_reass_walk_expired (vlib_main_t * vm,
1480			     vlib_node_runtime_t * node, vlib_frame_t * f)
1481{
1482  ip6_full_reass_main_t *rm = &ip6_full_reass_main;
1483  uword event_type, *event_data = 0;
1484
1485  while (true)
1486    {
1487      vlib_process_wait_for_event_or_clock (vm,
1488					    (f64) rm->expire_walk_interval_ms
1489					    / (f64) MSEC_PER_SEC);
1490      event_type = vlib_process_get_events (vm, &event_data);
1491
1492      switch (event_type)
1493	{
1494	case ~0:		/* no events => timeout */
1495	  /* nothing to do here */
1496	  break;
1497	case IP6_EVENT_CONFIG_CHANGED:
1498	  break;
1499	default:
1500	  clib_warning ("BUG: event type 0x%wx", event_type);
1501	  break;
1502	}
1503      f64 now = vlib_time_now (vm);
1504
1505      ip6_full_reass_t *reass;
1506      int *pool_indexes_to_free = NULL;
1507
1508      uword thread_index = 0;
1509      int index;
1510      const uword nthreads = vlib_num_workers () + 1;
1511      u32 *vec_icmp_bi = NULL;
1512      for (thread_index = 0; thread_index < nthreads; ++thread_index)
1513	{
1514	  ip6_full_reass_per_thread_t *rt =
1515	    &rm->per_thread_data[thread_index];
1516	  clib_spinlock_lock (&rt->lock);
1517
1518	  vec_reset_length (pool_indexes_to_free);
1519          /* *INDENT-OFF* */
1520          pool_foreach_index (index, rt->pool, ({
1521                                reass = pool_elt_at_index (rt->pool, index);
1522                                if (now > reass->last_heard + rm->timeout)
1523                                  {
1524                                    vec_add1 (pool_indexes_to_free, index);
1525                                  }
1526                              }));
1527          /* *INDENT-ON* */
1528	  int *i;
1529          /* *INDENT-OFF* */
1530          vec_foreach (i, pool_indexes_to_free)
1531          {
1532            ip6_full_reass_t *reass = pool_elt_at_index (rt->pool, i[0]);
1533            u32 icmp_bi = ~0;
1534            ip6_full_reass_on_timeout (vm, node, rm, reass, &icmp_bi);
1535            if (~0 != icmp_bi)
1536              vec_add1 (vec_icmp_bi, icmp_bi);
1537
1538            ip6_full_reass_free (rm, rt, reass);
1539          }
1540          /* *INDENT-ON* */
1541
1542	  clib_spinlock_unlock (&rt->lock);
1543	}
1544
1545      while (vec_len (vec_icmp_bi) > 0)
1546	{
1547	  vlib_frame_t *f =
1548	    vlib_get_frame_to_node (vm, rm->ip6_icmp_error_idx);
1549	  u32 *to_next = vlib_frame_vector_args (f);
1550	  u32 n_left_to_next = VLIB_FRAME_SIZE - f->n_vectors;
1551	  int trace_frame = 0;
1552	  while (vec_len (vec_icmp_bi) > 0 && n_left_to_next > 0)
1553	    {
1554	      u32 bi = vec_pop (vec_icmp_bi);
1555	      vlib_buffer_t *b = vlib_get_buffer (vm, bi);
1556	      if (PREDICT_FALSE (b->flags & VLIB_BUFFER_IS_TRACED))
1557		trace_frame = 1;
1558	      b->error = node->errors[IP6_ERROR_REASS_TIMEOUT];
1559	      to_next[0] = bi;
1560	      ++f->n_vectors;
1561	      to_next += 1;
1562	      n_left_to_next -= 1;
1563	    }
1564	  f->frame_flags |= (trace_frame * VLIB_FRAME_TRACE);
1565	  vlib_put_frame_to_node (vm, rm->ip6_icmp_error_idx, f);
1566	}
1567
1568      vec_free (pool_indexes_to_free);
1569      vec_free (vec_icmp_bi);
1570      if (event_data)
1571	{
1572	  _vec_len (event_data) = 0;
1573	}
1574    }
1575
1576  return 0;
1577}
1578
1579/* *INDENT-OFF* */
1580VLIB_REGISTER_NODE (ip6_full_reass_expire_node) = {
1581    .function = ip6_full_reass_walk_expired,
1582    .format_trace = format_ip6_full_reass_trace,
1583    .type = VLIB_NODE_TYPE_PROCESS,
1584    .name = "ip6-full-reassembly-expire-walk",
1585
1586    .n_errors = ARRAY_LEN (ip6_full_reassembly_error_strings),
1587    .error_strings = ip6_full_reassembly_error_strings,
1588
1589};
1590/* *INDENT-ON* */
1591
1592static u8 *
1593format_ip6_full_reass_key (u8 * s, va_list * args)
1594{
1595  ip6_full_reass_key_t *key = va_arg (*args, ip6_full_reass_key_t *);
1596  s = format (s, "xx_id: %u, src: %U, dst: %U, frag_id: %u, proto: %u",
1597	      key->xx_id, format_ip6_address, &key->src, format_ip6_address,
1598	      &key->dst, clib_net_to_host_u16 (key->frag_id), key->proto);
1599  return s;
1600}
1601
1602static u8 *
1603format_ip6_full_reass (u8 * s, va_list * args)
1604{
1605  vlib_main_t *vm = va_arg (*args, vlib_main_t *);
1606  ip6_full_reass_t *reass = va_arg (*args, ip6_full_reass_t *);
1607
1608  s = format (s, "ID: %lu, key: %U\n  first_bi: %u, data_len: %u, "
1609	      "last_packet_octet: %u, trace_op_counter: %u\n",
1610	      reass->id, format_ip6_full_reass_key, &reass->key,
1611	      reass->first_bi, reass->data_len, reass->last_packet_octet,
1612	      reass->trace_op_counter);
1613  u32 bi = reass->first_bi;
1614  u32 counter = 0;
1615  while (~0 != bi)
1616    {
1617      vlib_buffer_t *b = vlib_get_buffer (vm, bi);
1618      vnet_buffer_opaque_t *vnb = vnet_buffer (b);
1619      s = format (s, "  #%03u: range: [%u, %u], bi: %u, off: %d, len: %u, "
1620		  "fragment[%u, %u]\n",
1621		  counter, vnb->ip.reass.range_first,
1622		  vnb->ip.reass.range_last, bi,
1623		  ip6_full_reass_buffer_get_data_offset (b),
1624		  ip6_full_reass_buffer_get_data_len (b),
1625		  vnb->ip.reass.fragment_first, vnb->ip.reass.fragment_last);
1626      if (b->flags & VLIB_BUFFER_NEXT_PRESENT)
1627	{
1628	  bi = b->next_buffer;
1629	}
1630      else
1631	{
1632	  bi = ~0;
1633	}
1634    }
1635  return s;
1636}
1637
1638static clib_error_t *
1639show_ip6_full_reass (vlib_main_t * vm, unformat_input_t * input,
1640		     CLIB_UNUSED (vlib_cli_command_t * lmd))
1641{
1642  ip6_full_reass_main_t *rm = &ip6_full_reass_main;
1643
1644  vlib_cli_output (vm, "---------------------");
1645  vlib_cli_output (vm, "IP6 reassembly status");
1646  vlib_cli_output (vm, "---------------------");
1647  bool details = false;
1648  if (unformat (input, "details"))
1649    {
1650      details = true;
1651    }
1652
1653  u32 sum_reass_n = 0;
1654  u64 sum_buffers_n = 0;
1655  ip6_full_reass_t *reass;
1656  uword thread_index;
1657  const uword nthreads = vlib_num_workers () + 1;
1658  for (thread_index = 0; thread_index < nthreads; ++thread_index)
1659    {
1660      ip6_full_reass_per_thread_t *rt = &rm->per_thread_data[thread_index];
1661      clib_spinlock_lock (&rt->lock);
1662      if (details)
1663	{
1664          /* *INDENT-OFF* */
1665          pool_foreach (reass, rt->pool, {
1666            vlib_cli_output (vm, "%U", format_ip6_full_reass, vm, reass);
1667          });
1668          /* *INDENT-ON* */
1669	}
1670      sum_reass_n += rt->reass_n;
1671      clib_spinlock_unlock (&rt->lock);
1672    }
1673  vlib_cli_output (vm, "---------------------");
1674  vlib_cli_output (vm, "Current IP6 reassemblies count: %lu\n",
1675		   (long unsigned) sum_reass_n);
1676  vlib_cli_output (vm, "Maximum configured concurrent IP6 reassemblies per "
1677		   "worker-thread: %lu\n", (long unsigned) rm->max_reass_n);
1678  vlib_cli_output (vm, "Buffers in use: %lu\n",
1679		   (long unsigned) sum_buffers_n);
1680  return 0;
1681}
1682
1683/* *INDENT-OFF* */
1684VLIB_CLI_COMMAND (show_ip6_full_reassembly_cmd, static) = {
1685    .path = "show ip6-full-reassembly",
1686    .short_help = "show ip6-full-reassembly [details]",
1687    .function = show_ip6_full_reass,
1688};
1689/* *INDENT-ON* */
1690
1691#ifndef CLIB_MARCH_VARIANT
1692vnet_api_error_t
1693ip6_full_reass_enable_disable (u32 sw_if_index, u8 enable_disable)
1694{
1695  return vnet_feature_enable_disable ("ip6-unicast",
1696				      "ip6-full-reassembly-feature",
1697				      sw_if_index, enable_disable, 0, 0);
1698}
1699#endif /* CLIB_MARCH_VARIANT */
1700
1701#define foreach_ip6_full_reassembly_handoff_error                       \
1702_(CONGESTION_DROP, "congestion drop")
1703
1704
1705typedef enum
1706{
1707#define _(sym,str) IP6_FULL_REASSEMBLY_HANDOFF_ERROR_##sym,
1708  foreach_ip6_full_reassembly_handoff_error
1709#undef _
1710    IP6_FULL_REASSEMBLY_HANDOFF_N_ERROR,
1711} ip6_full_reassembly_handoff_error_t;
1712
1713static char *ip6_full_reassembly_handoff_error_strings[] = {
1714#define _(sym,string) string,
1715  foreach_ip6_full_reassembly_handoff_error
1716#undef _
1717};
1718
1719typedef struct
1720{
1721  u32 next_worker_index;
1722} ip6_full_reassembly_handoff_trace_t;
1723
1724static u8 *
1725format_ip6_full_reassembly_handoff_trace (u8 * s, va_list * args)
1726{
1727  CLIB_UNUSED (vlib_main_t * vm) = va_arg (*args, vlib_main_t *);
1728  CLIB_UNUSED (vlib_node_t * node) = va_arg (*args, vlib_node_t *);
1729  ip6_full_reassembly_handoff_trace_t *t =
1730    va_arg (*args, ip6_full_reassembly_handoff_trace_t *);
1731
1732  s =
1733    format (s, "ip6-full-reassembly-handoff: next-worker %d",
1734	    t->next_worker_index);
1735
1736  return s;
1737}
1738
1739always_inline uword
1740ip6_full_reassembly_handoff_inline (vlib_main_t * vm,
1741				    vlib_node_runtime_t * node,
1742				    vlib_frame_t * frame, bool is_feature)
1743{
1744  ip6_full_reass_main_t *rm = &ip6_full_reass_main;
1745
1746  vlib_buffer_t *bufs[VLIB_FRAME_SIZE], **b;
1747  u32 n_enq, n_left_from, *from;
1748  u16 thread_indices[VLIB_FRAME_SIZE], *ti;
1749  u32 fq_index;
1750
1751  from = vlib_frame_vector_args (frame);
1752  n_left_from = frame->n_vectors;
1753  vlib_get_buffers (vm, from, bufs, n_left_from);
1754
1755  b = bufs;
1756  ti = thread_indices;
1757
1758  fq_index = (is_feature) ? rm->fq_feature_index : rm->fq_index;
1759
1760  while (n_left_from > 0)
1761    {
1762      ti[0] = vnet_buffer (b[0])->ip.reass.owner_thread_index;
1763
1764      if (PREDICT_FALSE
1765	  ((node->flags & VLIB_NODE_FLAG_TRACE)
1766	   && (b[0]->flags & VLIB_BUFFER_IS_TRACED)))
1767	{
1768	  ip6_full_reassembly_handoff_trace_t *t =
1769	    vlib_add_trace (vm, node, b[0], sizeof (*t));
1770	  t->next_worker_index = ti[0];
1771	}
1772
1773      n_left_from -= 1;
1774      ti += 1;
1775      b += 1;
1776    }
1777  n_enq =
1778    vlib_buffer_enqueue_to_thread (vm, fq_index, from, thread_indices,
1779				   frame->n_vectors, 1);
1780
1781  if (n_enq < frame->n_vectors)
1782    vlib_node_increment_counter (vm, node->node_index,
1783				 IP6_FULL_REASSEMBLY_HANDOFF_ERROR_CONGESTION_DROP,
1784				 frame->n_vectors - n_enq);
1785  return frame->n_vectors;
1786}
1787
1788VLIB_NODE_FN (ip6_full_reassembly_handoff_node) (vlib_main_t * vm,
1789						 vlib_node_runtime_t * node,
1790						 vlib_frame_t * frame)
1791{
1792  return ip6_full_reassembly_handoff_inline (vm, node, frame,
1793					     false /* is_feature */ );
1794}
1795
1796/* *INDENT-OFF* */
1797VLIB_REGISTER_NODE (ip6_full_reassembly_handoff_node) = {
1798  .name = "ip6-full-reassembly-handoff",
1799  .vector_size = sizeof (u32),
1800  .n_errors = ARRAY_LEN(ip6_full_reassembly_handoff_error_strings),
1801  .error_strings = ip6_full_reassembly_handoff_error_strings,
1802  .format_trace = format_ip6_full_reassembly_handoff_trace,
1803
1804  .n_next_nodes = 1,
1805
1806  .next_nodes = {
1807    [0] = "error-drop",
1808  },
1809};
1810
1811
1812VLIB_NODE_FN (ip6_full_reassembly_feature_handoff_node) (vlib_main_t * vm,
1813                               vlib_node_runtime_t * node, vlib_frame_t * frame)
1814{
1815  return ip6_full_reassembly_handoff_inline (vm, node, frame, true /* is_feature */ );
1816}
1817
1818
1819/* *INDENT-OFF* */
1820VLIB_REGISTER_NODE (ip6_full_reassembly_feature_handoff_node) = {
1821  .name = "ip6-full-reass-feature-hoff",
1822  .vector_size = sizeof (u32),
1823  .n_errors = ARRAY_LEN(ip6_full_reassembly_handoff_error_strings),
1824  .error_strings = ip6_full_reassembly_handoff_error_strings,
1825  .format_trace = format_ip6_full_reassembly_handoff_trace,
1826
1827  .n_next_nodes = 1,
1828
1829  .next_nodes = {
1830    [0] = "error-drop",
1831  },
1832};
1833/* *INDENT-ON* */
1834
1835#ifndef CLIB_MARCH_VARIANT
1836int
1837ip6_full_reass_enable_disable_with_refcnt (u32 sw_if_index, int is_enable)
1838{
1839  ip6_full_reass_main_t *rm = &ip6_full_reass_main;
1840  vec_validate (rm->feature_use_refcount_per_intf, sw_if_index);
1841  if (is_enable)
1842    {
1843      if (!rm->feature_use_refcount_per_intf[sw_if_index])
1844	{
1845	  ++rm->feature_use_refcount_per_intf[sw_if_index];
1846	  return vnet_feature_enable_disable ("ip6-unicast",
1847					      "ip6-full-reassembly-feature",
1848					      sw_if_index, 1, 0, 0);
1849	}
1850      ++rm->feature_use_refcount_per_intf[sw_if_index];
1851    }
1852  else
1853    {
1854      --rm->feature_use_refcount_per_intf[sw_if_index];
1855      if (!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, 0, 0, 0);
1859    }
1860  return -1;
1861}
1862#endif
1863
1864/*
1865 * fd.io coding-style-patch-verification: ON
1866 *
1867 * Local Variables:
1868 * eval: (c-set-style "gnu")
1869 * End:
1870 */
1871