FD.io VPP  v17.04-9-g99c0734
Vector Packet Processing
lb.c
Go to the documentation of this file.
1 /*
2  * Copyright (c) 2016 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 #include <lb/lb.h>
17 #include <vnet/plugin/plugin.h>
18 #include <vpp/app/version.h>
19 #include <vnet/api_errno.h>
20 
21 //GC runs at most once every so many seconds
22 #define LB_GARBAGE_RUN 60
23 
24 //After so many seconds. It is assumed that inter-core race condition will not occur.
25 #define LB_CONCURRENCY_TIMEOUT 10
26 
28 
29 #define lb_get_writer_lock() do {} while(__sync_lock_test_and_set (lb_main.writer_lock, 1))
30 #define lb_put_writer_lock() lb_main.writer_lock[0] = 0
31 
32 static void lb_as_stack (lb_as_t *as);
33 
34 
35 const static char * const lb_dpo_gre4_ip4[] = { "lb4-gre4" , NULL };
36 const static char * const lb_dpo_gre4_ip6[] = { "lb6-gre4" , NULL };
37 const static char* const * const lb_dpo_gre4_nodes[DPO_PROTO_NUM] =
38  {
41  };
42 
43 const static char * const lb_dpo_gre6_ip4[] = { "lb4-gre6" , NULL };
44 const static char * const lb_dpo_gre6_ip6[] = { "lb6-gre6" , NULL };
45 const static char* const * const lb_dpo_gre6_nodes[DPO_PROTO_NUM] =
46  {
49  };
50 
52 {
53  return (u32) (vlib_time_now(vm) + 10000);
54 }
55 
56 u8 *format_lb_main (u8 * s, va_list * args)
57 {
59  lb_main_t *lbm = &lb_main;
60  s = format(s, "lb_main");
61  s = format(s, " ip4-src-address: %U \n", format_ip4_address, &lbm->ip4_src_address);
62  s = format(s, " ip6-src-address: %U \n", format_ip6_address, &lbm->ip6_src_address);
63  s = format(s, " #vips: %u\n", pool_elts(lbm->vips));
64  s = format(s, " #ass: %u\n", pool_elts(lbm->ass) - 1);
65 
66  u32 cpu_index;
67  for(cpu_index = 0; cpu_index < tm->n_vlib_mains; cpu_index++ ) {
68  lb_hash_t *h = lbm->per_cpu[cpu_index].sticky_ht;
69  if (h) {
70  s = format(s, "core %d\n", cpu_index);
71  s = format(s, " timeout: %ds\n", h->timeout);
72  s = format(s, " usage: %d / %d\n", lb_hash_elts(h, lb_hash_time_now(vlib_get_main())), lb_hash_size(h));
73  }
74  }
75 
76  return s;
77 }
78 
79 static char *lb_vip_type_strings[] = {
80  [LB_VIP_TYPE_IP6_GRE6] = "ip6-gre6",
81  [LB_VIP_TYPE_IP6_GRE4] = "ip6-gre4",
82  [LB_VIP_TYPE_IP4_GRE6] = "ip4-gre6",
83  [LB_VIP_TYPE_IP4_GRE4] = "ip4-gre4",
84 };
85 
86 u8 *format_lb_vip_type (u8 * s, va_list * args)
87 {
88  lb_vip_type_t vipt = va_arg (*args, lb_vip_type_t);
89  u32 i;
90  for (i=0; i<LB_VIP_N_TYPES; i++)
91  if (vipt == i)
92  return format(s, lb_vip_type_strings[i]);
93  return format(s, "_WRONG_TYPE_");
94 }
95 
96 uword unformat_lb_vip_type (unformat_input_t * input, va_list * args)
97 {
98  lb_vip_type_t *vipt = va_arg (*args, lb_vip_type_t *);
99  u32 i;
100  for (i=0; i<LB_VIP_N_TYPES; i++)
101  if (unformat(input, lb_vip_type_strings[i])) {
102  *vipt = i;
103  return 1;
104  }
105  return 0;
106 }
107 
108 u8 *format_lb_vip (u8 * s, va_list * args)
109 {
110  lb_vip_t *vip = va_arg (*args, lb_vip_t *);
111  return format(s, "%U %U new_size:%u #as:%u%s",
112  format_lb_vip_type, vip->type,
114  vip->new_flow_table_mask + 1,
115  pool_elts(vip->as_indexes),
116  (vip->flags & LB_VIP_FLAGS_USED)?"":" removed");
117 }
118 
119 u8 *format_lb_as (u8 * s, va_list * args)
120 {
121  lb_as_t *as = va_arg (*args, lb_as_t *);
122  return format(s, "%U %s", format_ip46_address,
123  &as->address, IP46_TYPE_ANY,
124  (as->flags & LB_AS_FLAGS_USED)?"used":"removed");
125 }
126 
127 u8 *format_lb_vip_detailed (u8 * s, va_list * args)
128 {
129  lb_main_t *lbm = &lb_main;
130  lb_vip_t *vip = va_arg (*args, lb_vip_t *);
131  uword indent = format_get_indent (s);
132 
133  s = format(s, "%U %U [%u] %U%s\n"
134  "%U new_size:%u\n",
135  format_white_space, indent,
136  format_lb_vip_type, vip->type,
137  vip - lbm->vips, format_ip46_prefix, &vip->prefix, vip->plen, IP46_TYPE_ANY,
138  (vip->flags & LB_VIP_FLAGS_USED)?"":" removed",
139  format_white_space, indent,
140  vip->new_flow_table_mask + 1);
141 
142  //Print counters
143  s = format(s, "%U counters:\n",
144  format_white_space, indent);
145  u32 i;
146  for (i=0; i<LB_N_VIP_COUNTERS; i++)
147  s = format(s, "%U %s: %d\n",
148  format_white_space, indent,
149  lbm->vip_counters[i].name,
150  vlib_get_simple_counter(&lbm->vip_counters[i], vip - lbm->vips));
151 
152 
153  s = format(s, "%U #as:%u\n",
154  format_white_space, indent,
155  pool_elts(vip->as_indexes));
156 
157  //Let's count the buckets for each AS
158  u32 *count = 0;
159  vec_validate(count, pool_len(lbm->ass)); //Possibly big alloc for not much...
160  lb_new_flow_entry_t *nfe;
161  vec_foreach(nfe, vip->new_flow_table)
162  count[nfe->as_index]++;
163 
164  lb_as_t *as;
165  u32 *as_index;
166  pool_foreach(as_index, vip->as_indexes, {
167  as = &lbm->ass[*as_index];
168  s = format(s, "%U %U %d buckets %d flows dpo:%u %s\n",
169  format_white_space, indent,
170  format_ip46_address, &as->address, IP46_TYPE_ANY,
171  count[as - lbm->ass],
172  vlib_refcount_get(&lbm->as_refcount, as - lbm->ass),
173  as->dpo.dpoi_index,
174  (as->flags & LB_AS_FLAGS_USED)?"used":" removed");
175  });
176 
177  vec_free(count);
178 
179  /*
180  s = format(s, "%U new flows table:\n", format_white_space, indent);
181  lb_new_flow_entry_t *nfe;
182  vec_foreach(nfe, vip->new_flow_table) {
183  s = format(s, "%U %d: %d\n", format_white_space, indent, nfe - vip->new_flow_table, nfe->as_index);
184  }
185  */
186  return s;
187 }
188 
189 typedef struct {
194 
195 static int lb_pseudorand_compare(void *a, void *b)
196 {
197  lb_as_t *asa, *asb;
198  lb_main_t *lbm = &lb_main;
199  asa = &lbm->ass[((lb_pseudorand_t *)a)->as_index];
200  asb = &lbm->ass[((lb_pseudorand_t *)b)->as_index];
201  return memcmp(&asa->address, &asb->address, sizeof(asb->address));
202 }
203 
205 {
206  lb_main_t *lbm = &lb_main;
207  ASSERT (lbm->writer_lock[0]);
208 
209  u32 now = (u32) vlib_time_now(vlib_get_main());
211  return;
212 
213  vip->last_garbage_collection = now;
214  lb_as_t *as;
215  u32 *as_index;
216  pool_foreach(as_index, vip->as_indexes, {
217  as = &lbm->ass[*as_index];
218  if (!(as->flags & LB_AS_FLAGS_USED) && //Not used
219  clib_u32_loop_gt(now, as->last_used + LB_CONCURRENCY_TIMEOUT) && //Not recently used
220  (vlib_refcount_get(&lbm->as_refcount, as - lbm->ass) == 0))
221  { //Not referenced
222  fib_entry_child_remove(as->next_hop_fib_entry_index,
223  as->next_hop_child_index);
224  fib_table_entry_delete_index(as->next_hop_fib_entry_index,
225  FIB_SOURCE_RR);
226  as->next_hop_fib_entry_index = FIB_NODE_INDEX_INVALID;
227 
228  pool_put(vip->as_indexes, as_index);
229  pool_put(lbm->ass, as);
230  }
231  });
232 }
233 
235 {
236  lb_main_t *lbm = &lb_main;
238  lb_vip_t *vip;
239  u32 *to_be_removed_vips = 0, *i;
240  pool_foreach(vip, lbm->vips, {
241  lb_vip_garbage_collection(vip);
242 
243  if (!(vip->flags & LB_VIP_FLAGS_USED) &&
244  (pool_elts(vip->as_indexes) == 0)) {
245  vec_add1(to_be_removed_vips, vip - lbm->vips);
246  }
247  });
248 
249  vec_foreach(i, to_be_removed_vips) {
250  vip = &lbm->vips[*i];
251  pool_put(lbm->vips, vip);
252  pool_free(vip->as_indexes);
253  }
254 
255  vec_free(to_be_removed_vips);
257 }
258 
260 {
261  lb_main_t *lbm = &lb_main;
262  lb_new_flow_entry_t *old_table;
263  u32 i, *as_index;
264  lb_new_flow_entry_t *new_flow_table = 0;
265  lb_as_t *as;
266  lb_pseudorand_t *pr, *sort_arr = 0;
267  u32 count;
268 
269  ASSERT (lbm->writer_lock[0]); //We must have the lock
270 
271  //Check if some AS is configured or not
272  i = 0;
273  pool_foreach(as_index, vip->as_indexes, {
274  as = &lbm->ass[*as_index];
275  if (as->flags & LB_AS_FLAGS_USED) { //Not used anymore
276  i = 1;
277  goto out; //Not sure 'break' works in this macro-loop
278  }
279  });
280 
281 out:
282  if (i == 0) {
283  //Only the default. i.e. no AS
284  vec_validate(new_flow_table, vip->new_flow_table_mask);
285  for (i=0; i<vec_len(new_flow_table); i++)
286  new_flow_table[i].as_index = 0;
287 
288  goto finished;
289  }
290 
291  //First, let's sort the ASs
292  sort_arr = 0;
293  vec_alloc(sort_arr, pool_elts(vip->as_indexes));
294 
295  i = 0;
296  pool_foreach(as_index, vip->as_indexes, {
297  as = &lbm->ass[*as_index];
298  if (!(as->flags & LB_AS_FLAGS_USED)) //Not used anymore
299  continue;
300 
301  sort_arr[i].as_index = as - lbm->ass;
302  i++;
303  });
304  _vec_len(sort_arr) = i;
305 
307 
308  //Now let's pseudo-randomly generate permutations
309  vec_foreach(pr, sort_arr) {
310  lb_as_t *as = &lbm->ass[pr->as_index];
311 
312  u64 seed = clib_xxhash(as->address.as_u64[0] ^
313  as->address.as_u64[1]);
314  /* We have 2^n buckets.
315  * skip must be prime with 2^n.
316  * So skip must be odd.
317  * MagLev actually state that M should be prime,
318  * but this has a big computation cost (% operation).
319  * Using 2^n is more better (& operation).
320  */
321  pr->skip = ((seed & 0xffffffff) | 1) & vip->new_flow_table_mask;
322  pr->last = (seed >> 32) & vip->new_flow_table_mask;
323  }
324 
325  //Let's create a new flow table
326  vec_validate(new_flow_table, vip->new_flow_table_mask);
327  for (i=0; i<vec_len(new_flow_table); i++)
328  new_flow_table[i].as_index = ~0;
329 
330  u32 done = 0;
331  while (1) {
332  vec_foreach(pr, sort_arr) {
333  while (1) {
334  u32 last = pr->last;
335  pr->last = (pr->last + pr->skip) & vip->new_flow_table_mask;
336  if (new_flow_table[last].as_index == ~0) {
337  new_flow_table[last].as_index = pr->as_index;
338  break;
339  }
340  }
341  done++;
342  if (done == vec_len(new_flow_table))
343  goto finished;
344  }
345  }
346 
347  vec_free(sort_arr);
348 
349 finished:
350 
351 //Count number of changed entries
352  count = 0;
353  for (i=0; i<vec_len(new_flow_table); i++)
354  if (vip->new_flow_table == 0 ||
355  new_flow_table[i].as_index != vip->new_flow_table[i].as_index)
356  count++;
357 
358  old_table = vip->new_flow_table;
359  vip->new_flow_table = new_flow_table;
360  vec_free(old_table);
361 }
362 
363 int lb_conf(ip4_address_t *ip4_address, ip6_address_t *ip6_address,
364  u32 per_cpu_sticky_buckets, u32 flow_timeout)
365 {
366  lb_main_t *lbm = &lb_main;
367 
368  if (!is_pow2(per_cpu_sticky_buckets))
369  return VNET_API_ERROR_INVALID_MEMORY_SIZE;
370 
371  lb_get_writer_lock(); //Not exactly necessary but just a reminder that it exists for my future self
372  lbm->ip4_src_address = *ip4_address;
373  lbm->ip6_src_address = *ip6_address;
374  lbm->per_cpu_sticky_buckets = per_cpu_sticky_buckets;
375  lbm->flow_timeout = flow_timeout;
377  return 0;
378 }
379 
380 static
381 int lb_vip_find_index_with_lock(ip46_address_t *prefix, u8 plen, u32 *vip_index)
382 {
383  lb_main_t *lbm = &lb_main;
384  lb_vip_t *vip;
385  ASSERT (lbm->writer_lock[0]); //This must be called with the lock owned
386  ip46_prefix_normalize(prefix, plen);
387  pool_foreach(vip, lbm->vips, {
388  if ((vip->flags & LB_AS_FLAGS_USED) &&
389  vip->plen == plen &&
390  vip->prefix.as_u64[0] == prefix->as_u64[0] &&
391  vip->prefix.as_u64[1] == prefix->as_u64[1]) {
392  *vip_index = vip - lbm->vips;
393  return 0;
394  }
395  });
396  return VNET_API_ERROR_NO_SUCH_ENTRY;
397 }
398 
399 int lb_vip_find_index(ip46_address_t *prefix, u8 plen, u32 *vip_index)
400 {
401  int ret;
403  ret = lb_vip_find_index_with_lock(prefix, plen, vip_index);
405  return ret;
406 }
407 
408 static int lb_as_find_index_vip(lb_vip_t *vip, ip46_address_t *address, u32 *as_index)
409 {
410  lb_main_t *lbm = &lb_main;
411  ASSERT (lbm->writer_lock[0]); //This must be called with the lock owned
412  lb_as_t *as;
413  u32 *asi;
414  pool_foreach(asi, vip->as_indexes, {
415  as = &lbm->ass[*asi];
416  if (as->vip_index == (vip - lbm->vips) &&
417  as->address.as_u64[0] == address->as_u64[0] &&
418  as->address.as_u64[1] == address->as_u64[1]) {
419  *as_index = as - lbm->ass;
420  return 0;
421  }
422  });
423  return -1;
424 }
425 
426 int lb_vip_add_ass(u32 vip_index, ip46_address_t *addresses, u32 n)
427 {
428  lb_main_t *lbm = &lb_main;
430  lb_vip_t *vip;
431  if (!(vip = lb_vip_get_by_index(vip_index))) {
433  return VNET_API_ERROR_NO_SUCH_ENTRY;
434  }
435 
437  u32 *to_be_added = 0;
438  u32 *to_be_updated = 0;
439  u32 i;
440  u32 *ip;
441 
442  //Sanity check
443  while (n--) {
444 
445  if (!lb_as_find_index_vip(vip, &addresses[n], &i)) {
446  if (lbm->ass[i].flags & LB_AS_FLAGS_USED) {
447  vec_free(to_be_added);
448  vec_free(to_be_updated);
450  return VNET_API_ERROR_VALUE_EXIST;
451  }
452  vec_add1(to_be_updated, i);
453  goto next;
454  }
455 
456  if (ip46_address_type(&addresses[n]) != type) {
457  vec_free(to_be_added);
458  vec_free(to_be_updated);
460  return VNET_API_ERROR_INVALID_ADDRESS_FAMILY;
461  }
462 
463  if (n) {
464  u32 n2 = n;
465  while(n2--) //Check for duplicates
466  if (addresses[n2].as_u64[0] == addresses[n].as_u64[0] &&
467  addresses[n2].as_u64[1] == addresses[n].as_u64[1])
468  goto next;
469  }
470 
471  vec_add1(to_be_added, n);
472 
473 next:
474  continue;
475  }
476 
477  //Update reused ASs
478  vec_foreach(ip, to_be_updated) {
479  lbm->ass[*ip].flags = LB_AS_FLAGS_USED;
480  }
481  vec_free(to_be_updated);
482 
483  //Create those who have to be created
484  vec_foreach(ip, to_be_added) {
485  lb_as_t *as;
486  u32 *as_index;
487  pool_get(lbm->ass, as);
488  as->address = addresses[*ip];
489  as->flags = LB_AS_FLAGS_USED;
490  as->vip_index = vip_index;
491  pool_get(vip->as_indexes, as_index);
492  *as_index = as - lbm->ass;
493 
494  /*
495  * become a child of the FIB entry
496  * so we are informed when its forwarding changes
497  */
498  fib_prefix_t nh = {};
499  if (lb_vip_is_gre4(vip)) {
500  nh.fp_addr.ip4 = as->address.ip4;
501  nh.fp_len = 32;
503  } else {
504  nh.fp_addr.ip6 = as->address.ip6;
505  nh.fp_len = 128;
507  }
508 
511  &nh,
516  lbm->fib_node_type,
517  as - lbm->ass);
518 
519  lb_as_stack(as);
520  }
521  vec_free(to_be_added);
522 
523  //Recompute flows
525 
526  //Garbage collection maybe
528 
530  return 0;
531 }
532 
533 int lb_vip_del_ass_withlock(u32 vip_index, ip46_address_t *addresses, u32 n)
534 {
535  lb_main_t *lbm = &lb_main;
536  u32 now = (u32) vlib_time_now(vlib_get_main());
537  u32 *ip = 0;
538 
539  lb_vip_t *vip;
540  if (!(vip = lb_vip_get_by_index(vip_index))) {
541  return VNET_API_ERROR_NO_SUCH_ENTRY;
542  }
543 
544  u32 *indexes = NULL;
545  while (n--) {
546  u32 i;
547  if (lb_as_find_index_vip(vip, &addresses[n], &i)) {
548  vec_free(indexes);
549  return VNET_API_ERROR_NO_SUCH_ENTRY;
550  }
551 
552  if (n) { //Check for duplicates
553  u32 n2 = n - 1;
554  while(n2--) {
555  if (addresses[n2].as_u64[0] == addresses[n].as_u64[0] &&
556  addresses[n2].as_u64[1] == addresses[n].as_u64[1])
557  goto next;
558  }
559  }
560 
561  vec_add1(indexes, i);
562 next:
563  continue;
564  }
565 
566  //Garbage collection maybe
568 
569  if (indexes != NULL) {
570  vec_foreach(ip, indexes) {
571  lbm->ass[*ip].flags &= ~LB_AS_FLAGS_USED;
572  lbm->ass[*ip].last_used = now;
573  }
574 
575  //Recompute flows
577  }
578 
579  vec_free(indexes);
580  return 0;
581 }
582 
583 int lb_vip_del_ass(u32 vip_index, ip46_address_t *addresses, u32 n)
584 {
586  int ret = lb_vip_del_ass_withlock(vip_index, addresses, n);
588  return ret;
589 }
590 
591 /**
592  * Add the VIP adjacency to the ip4 or ip6 fib
593  */
594 static void lb_vip_add_adjacency(lb_main_t *lbm, lb_vip_t *vip)
595 {
596  dpo_proto_t proto = 0;
597  dpo_id_t dpo = DPO_INVALID;
598  fib_prefix_t pfx = {};
599  if (lb_vip_is_ip4(vip)) {
600  pfx.fp_addr.ip4 = vip->prefix.ip4;
601  pfx.fp_len = vip->plen - 96;
603  proto = DPO_PROTO_IP4;
604  } else {
605  pfx.fp_addr.ip6 = vip->prefix.ip6;
606  pfx.fp_len = vip->plen;
608  proto = DPO_PROTO_IP6;
609  }
610  dpo_set(&dpo, lb_vip_is_gre4(vip)?lbm->dpo_gre4_type:lbm->dpo_gre6_type,
611  proto, vip - lbm->vips);
613  &pfx,
616  &dpo);
617  dpo_reset(&dpo);
618 }
619 
620 /**
621  * Deletes the adjacency associated with the VIP
622  */
623 static void lb_vip_del_adjacency(lb_main_t *lbm, lb_vip_t *vip)
624 {
625  fib_prefix_t pfx = {};
626  if (lb_vip_is_ip4(vip)) {
627  pfx.fp_addr.ip4 = vip->prefix.ip4;
628  pfx.fp_len = vip->plen - 96;
630  } else {
631  pfx.fp_addr.ip6 = vip->prefix.ip6;
632  pfx.fp_len = vip->plen;
634  }
636 }
637 
638 int lb_vip_add(ip46_address_t *prefix, u8 plen, lb_vip_type_t type, u32 new_length, u32 *vip_index)
639 {
640  lb_main_t *lbm = &lb_main;
641  lb_vip_t *vip;
643  ip46_prefix_normalize(prefix, plen);
644 
645  if (!lb_vip_find_index_with_lock(prefix, plen, vip_index)) {
647  return VNET_API_ERROR_VALUE_EXIST;
648  }
649 
650  if (!is_pow2(new_length)) {
652  return VNET_API_ERROR_INVALID_MEMORY_SIZE;
653  }
654 
655  if (ip46_prefix_is_ip4(prefix, plen) &&
656  (type != LB_VIP_TYPE_IP4_GRE4) &&
657  (type != LB_VIP_TYPE_IP4_GRE6))
658  return VNET_API_ERROR_INVALID_ADDRESS_FAMILY;
659 
660 
661  //Allocate
662  pool_get(lbm->vips, vip);
663 
664  //Init
665  vip->prefix = *prefix;
666  vip->plen = plen;
668  vip->type = type;
669  vip->flags = LB_VIP_FLAGS_USED;
670  vip->as_indexes = 0;
671 
672  //Validate counters
673  u32 i;
674  for (i = 0; i < LB_N_VIP_COUNTERS; i++) {
675  vlib_validate_simple_counter(&lbm->vip_counters[i], vip - lbm->vips);
676  vlib_zero_simple_counter(&lbm->vip_counters[i], vip - lbm->vips);
677  }
678 
679  //Configure new flow table
680  vip->new_flow_table_mask = new_length - 1;
681  vip->new_flow_table = 0;
682 
683  //Create a new flow hash table full of the default entry
685 
686  //Create adjacency to direct traffic
687  lb_vip_add_adjacency(lbm, vip);
688 
689  //Return result
690  *vip_index = vip - lbm->vips;
691 
693  return 0;
694 }
695 
696 int lb_vip_del(u32 vip_index)
697 {
698  lb_main_t *lbm = &lb_main;
699  lb_vip_t *vip;
701  if (!(vip = lb_vip_get_by_index(vip_index))) {
703  return VNET_API_ERROR_NO_SUCH_ENTRY;
704  }
705 
706  //FIXME: This operation is actually not working
707  //We will need to remove state before performing this.
708 
709  {
710  //Remove all ASs
711  ip46_address_t *ass = 0;
712  lb_as_t *as;
713  u32 *as_index;
714  pool_foreach(as_index, vip->as_indexes, {
715  as = &lbm->ass[*as_index];
716  vec_add1(ass, as->address);
717  });
718  if (vec_len(ass))
719  lb_vip_del_ass_withlock(vip_index, ass, vec_len(ass));
720  vec_free(ass);
721  }
722 
723  //Delete adjacency
724  lb_vip_del_adjacency(lbm, vip);
725 
726  //Set the VIP as unused
727  vip->flags &= ~LB_VIP_FLAGS_USED;
728 
730  return 0;
731 }
732 
733 /* *INDENT-OFF* */
735  .version = VPP_BUILD_VER,
736  .description = "Load Balancer",
737 };
738 /* *INDENT-ON* */
739 
740 u8 *format_lb_dpo (u8 * s, va_list * va)
741 {
742  index_t index = va_arg (*va, index_t);
743  CLIB_UNUSED(u32 indent) = va_arg (*va, u32);
744  lb_main_t *lbm = &lb_main;
745  lb_vip_t *vip = pool_elt_at_index (lbm->vips, index);
746  return format (s, "%U", format_lb_vip, vip);
747 }
748 
749 static void lb_dpo_lock (dpo_id_t *dpo) {}
750 static void lb_dpo_unlock (dpo_id_t *dpo) {}
751 
752 static fib_node_t *
754 {
755  lb_main_t *lbm = &lb_main;
756  lb_as_t *as = pool_elt_at_index (lbm->ass, index);
757  return (&as->fib_node);
758 }
759 
760 static void
762 {
763 }
764 
765 static lb_as_t *
767 {
768  return ((lb_as_t*)(((char*)node) -
769  STRUCT_OFFSET_OF(lb_as_t, fib_node)));
770 }
771 
772 static void
774 {
775  lb_main_t *lbm = &lb_main;
776  lb_vip_t *vip = &lbm->vips[as->vip_index];
779  &as->dpo,
782 }
783 
787 {
790 }
791 
792 clib_error_t *
794 {
796  lb_main_t *lbm = &lb_main;
797  lb_as_t *default_as;
798  fib_node_vft_t lb_fib_node_vft = {
800  .fnv_last_lock = lb_fib_node_last_lock_gone,
801  .fnv_back_walk = lb_fib_node_back_walk_notify,
802  };
803  dpo_vft_t lb_vft = {
804  .dv_lock = lb_dpo_lock,
805  .dv_unlock = lb_dpo_unlock,
806  .dv_format = format_lb_dpo,
807  };
808 
809  lbm->vips = 0;
810  lbm->per_cpu = 0;
811  vec_validate(lbm->per_cpu, tm->n_vlib_mains - 1);
813  lbm->writer_lock[0] = 0;
816  lbm->ip4_src_address.as_u32 = 0xffffffff;
817  lbm->ip6_src_address.as_u64[0] = 0xffffffffffffffffL;
818  lbm->ip6_src_address.as_u64[1] = 0xffffffffffffffffL;
821  lbm->fib_node_type = fib_node_register_new_type(&lb_fib_node_vft);
822 
823  //Init AS reference counters
825 
826  //Allocate and init default AS.
827  lbm->ass = 0;
828  pool_get(lbm->ass, default_as);
829  default_as->flags = 0;
830  default_as->dpo.dpoi_next_node = LB_NEXT_DROP;
831  default_as->vip_index = ~0;
832  default_as->address.ip6.as_u64[0] = 0xffffffffffffffffL;
833  default_as->address.ip6.as_u64[1] = 0xffffffffffffffffL;
834 
835 #define _(a,b,c) lbm->vip_counters[c].name = b;
837 #undef _
838  return NULL;
839 }
840 
u32 skip
Definition: lb.c:192
#define vec_validate(V, I)
Make sure vector is long enough for given index (no header, unspecified alignment) ...
Definition: vec.h:436
fib_protocol_t fp_proto
protocol type
Definition: fib_types.h:169
dpo_lock_fn_t dv_lock
A reference counting lock function.
Definition: dpo.h:335
u32 lb_hash_time_now(vlib_main_t *vm)
Definition: lb.c:51
Recursive resolution source.
Definition: fib_entry.h:109
static int lb_pseudorand_compare(void *a, void *b)
Definition: lb.c:195
Each VIP is configured with a set of application server.
Definition: lb.h:55
sll srl srl sll sra u16x4 i
Definition: vector_sse2.h:343
#define LB_GARBAGE_RUN
Definition: lb.c:22
#define CLIB_UNUSED(x)
Definition: clib.h:79
A virtual function table regisitered for a DPO type.
Definition: dpo.h:330
ip46_type_t
Definition: format.h:63
a
Definition: bitmap.h:516
u32 last
Definition: lb.c:191
u32 per_cpu_sticky_buckets
Number of buckets in the per-cpu sticky hash table.
Definition: lb.h:272
clib_error_t * lb_init(vlib_main_t *vm)
Definition: lb.c:793
u64 as_u64
Definition: bihash_doc.h:63
u32 fib_entry_child_add(fib_node_index_t fib_entry_index, fib_node_type_t child_type, fib_node_index_t child_index)
Definition: fib_entry.c:472
static void lb_fib_node_last_lock_gone(fib_node_t *node)
Definition: lb.c:761
u64 as_u64[2]
Definition: ip6_packet.h:51
static void lb_vip_update_new_flow_table(lb_vip_t *vip)
Definition: lb.c:259
static int lb_as_find_index_vip(lb_vip_t *vip, ip46_address_t *address, u32 *as_index)
Definition: lb.c:408
#define NULL
Definition: clib.h:55
static f64 vlib_time_now(vlib_main_t *vm)
Definition: main.h:185
enum fib_node_back_walk_rc_t_ fib_node_back_walk_rc_t
Return code from a back walk function.
const dpo_id_t * fib_entry_contribute_ip_forwarding(fib_node_index_t fib_entry_index)
Definition: fib_entry.c:432
static void lb_vip_del_adjacency(lb_main_t *lbm, lb_vip_t *vip)
Deletes the adjacency associated with the VIP.
Definition: lb.c:623
int lb_conf(ip4_address_t *ip4_address, ip6_address_t *ip6_address, u32 per_cpu_sticky_buckets, u32 flow_timeout)
Fix global load-balancer parameters.
Definition: lb.c:363
#define lb_get_writer_lock()
Definition: lb.c:29
u8 * format_ip46_prefix(u8 *s, va_list *args)
Definition: util.c:54
int lb_vip_add_ass(u32 vip_index, ip46_address_t *addresses, u32 n)
Definition: lb.c:426
u32 index_t
A Data-Path Object is an object that represents actions that are applied to packets are they are swit...
Definition: dpo.h:41
#define vec_add1(V, E)
Add 1 element to end of vector (unspecified alignment).
Definition: vec.h:522
ip46_address_t prefix
A Virtual IP represents a given service delivered by a set of application servers.
Definition: lb.h:184
#define clib_u32_loop_gt(a, b)
32 bits integer comparison for running values.
Definition: util.h:38
static u64 clib_xxhash(u64 key)
Definition: xxhash.h:58
static heap_elt_t * last(heap_header_t *h)
Definition: heap.c:53
static_always_inline void vlib_refcount_init(vlib_refcount_t *r)
Definition: refcount.h:60
static void lb_dpo_lock(dpo_id_t *dpo)
Definition: lb.c:749
format_function_t format_ip46_address
Definition: format.h:61
#define STRUCT_OFFSET_OF(t, f)
Definition: clib.h:62
u8 * format(u8 *s, const char *fmt,...)
Definition: format.c:418
static void lb_as_stack(lb_as_t *as)
Definition: lb.c:773
#define lb_vip_get_by_index(index)
Definition: lb.h:322
#define pool_get(P, E)
Allocate an object E from a pool P (unspecified alignment).
Definition: pool.h:200
u32 vip_index
ASs are indexed by address and VIP Index.
Definition: lb.h:74
#define vec_alloc(V, N)
Allocate space for N more elements (no header, unspecified alignment)
Definition: vec.h:279
static const char *const *const lb_dpo_gre6_nodes[DPO_PROTO_NUM]
Definition: lb.c:45
lb_hash_t * sticky_ht
Each CPU has its own sticky flow hash table.
Definition: lb.h:225
#define pool_len(p)
Number of elements in pool vector.
Definition: pool.h:121
fib_node_type_t fib_node_register_new_type(const fib_node_vft_t *vft)
Create a new FIB node type and Register the function table for it.
Definition: fib_node.c:78
#define LB_VIP_FLAGS_USED
Definition: lb.h:206
#define ip46_address_type(ip46)
Definition: util.h:26
ip46_address_t address
Destination address used to tunnel traffic towards that application server.
Definition: lb.h:67
int lb_vip_del_ass_withlock(u32 vip_index, ip46_address_t *addresses, u32 n)
Definition: lb.c:533
u32 timeout
Definition: lbhash.h:59
static counter_t vlib_get_simple_counter(vlib_simple_counter_main_t *cm, u32 index)
Get the value of a simple counter Scrapes the entire set of per-thread counters.
Definition: counter.h:97
format_function_t format_ip4_address
Definition: format.h:79
#define LB_AS_FLAGS_USED
Definition: lb.h:82
#define pool_foreach(VAR, POOL, BODY)
Iterate through pool.
Definition: pool.h:376
#define VLIB_INIT_FUNCTION(x)
Definition: init.h:111
static lb_as_t * lb_as_from_fib_node(fib_node_t *node)
Definition: lb.c:766
void fib_table_entry_special_remove(u32 fib_index, const fib_prefix_t *prefix, fib_source_t source)
Remove a &#39;special&#39; entry from the FIB.
Definition: fib_table.c:388
VLIB_PLUGIN_REGISTER()
static uword format_get_indent(u8 *s)
Definition: format.h:72
u8 * format_white_space(u8 *s, va_list *va)
Definition: std-formats.c:113
#define LB_DEFAULT_PER_CPU_STICKY_BUCKETS
lb-plugin implements a MagLev-like load balancer.
Definition: lb.h:43
lb_main_t lb_main
Definition: lb.c:27
static const char *const lb_dpo_gre4_ip4[]
Definition: lb.c:35
u32 flow_timeout
Flow timeout in seconds.
Definition: lb.h:277
A high priority source a plugin can use.
Definition: fib_entry.h:54
Definition: lb.h:228
fib_node_type_t fib_node_type
Node type for registering to fib changes.
Definition: lb.h:293
dpo_type_t dpo_gre4_type
DPO used to send packet from IP4/6 lookup to LB node.
Definition: lb.h:287
Aggregrate type for a prefix.
Definition: fib_types.h:160
static void lb_vip_add_adjacency(lb_main_t *lbm, lb_vip_t *vip)
Add the VIP adjacency to the ip4 or ip6 fib.
Definition: lb.c:594
vlib_refcount_t as_refcount
Each AS has an associated reference counter.
Definition: lb.h:247
unsigned long u64
Definition: types.h:89
static void lb_vip_garbage_collection(lb_vip_t *vip)
Definition: lb.c:204
u8 * format_lb_main(u8 *s, va_list *args)
Definition: lb.c:56
u8 * format_lb_vip(u8 *s, va_list *args)
Definition: lb.c:108
enum dpo_proto_t_ dpo_proto_t
Data path protocol.
u16 fp_len
The mask length.
Definition: fib_types.h:164
#define lb_vip_is_ip4(vip)
Definition: lb.h:215
lb_vip_t * vips
Pool of all Virtual IPs.
Definition: lb.h:232
dpo_type_t dpo_register_new_type(const dpo_vft_t *vft, const char *const *const *nodes)
Create and register a new DPO type.
Definition: dpo.c:261
u32 last_used
Rotating timestamp of when LB_AS_FLAGS_USED flag was last set.
Definition: lb.h:93
ip4_address_t ip4_src_address
Source address used for IPv4 encapsulated traffic.
Definition: lb.h:267
Definition: fib_entry.h:227
char * name
The counter collection&#39;s name.
Definition: counter.h:65
u8 plen
The VIP prefix length.
Definition: lb.h:190
The identity of a DPO is a combination of its type and its instance number/index of objects of that t...
Definition: dpo.h:146
Definition: fib_entry.h:231
#define lb_vip_is_gre4(vip)
Definition: lb.h:216
static const char *const lb_dpo_gre6_ip4[]
Definition: lb.c:43
#define pool_elt_at_index(p, i)
Returns pointer to element at given index.
Definition: pool.h:397
ip46_address_t fp_addr
The address type is not deriveable from the fp_addr member.
Definition: fib_types.h:183
int lb_vip_del(u32 vip_index)
Definition: lb.c:696
u8 * format_lb_vip_type(u8 *s, va_list *args)
Definition: lb.c:86
struct _unformat_input_t unformat_input_t
#define pool_put(P, E)
Free an object E in pool P.
Definition: pool.h:241
uword unformat_lb_vip_type(unformat_input_t *input, va_list *args)
Definition: lb.c:96
#define LB_DEFAULT_FLOW_TIMEOUT
Definition: lb.h:44
void ip46_prefix_normalize(ip46_address_t *prefix, u8 plen)
Definition: util.c:18
static const char *const *const lb_dpo_gre4_nodes[DPO_PROTO_NUM]
Definition: lb.c:37
An node in the FIB graph.
Definition: fib_node.h:277
Definition: lb.h:114
fib_node_t fib_node
Registration to FIB event.
Definition: lb.h:59
static const char *const lb_dpo_gre6_ip6[]
Definition: lb.c:44
static const dpo_vft_t lb_vft
Definition: load_balance.c:775
#define ip46_prefix_is_ip4(ip46, len)
Definition: util.h:27
#define pool_free(p)
Free a pool.
Definition: pool.h:290
fib_node_index_t fib_table_entry_special_add(u32 fib_index, const fib_prefix_t *prefix, fib_source_t source, fib_entry_flag_t flags)
Add a &#39;special&#39; entry to the FIB.
Definition: fib_table.c:369
int lb_vip_add(ip46_address_t *prefix, u8 plen, lb_vip_type_t type, u32 new_length, u32 *vip_index)
Definition: lb.c:638
format_function_t format_ip6_address
Definition: format.h:95
vlib_main_t * vm
Definition: buffer.c:276
vec_header_t h
Definition: buffer.c:275
#define vec_free(V)
Free vector&#39;s memory (no header).
Definition: vec.h:340
volatile u32 * writer_lock
Definition: lb.h:300
#define lb_foreach_vip_counter
Definition: lb.h:118
fib_node_get_t fnv_get
Definition: fib_node.h:265
u32 fib_node_index_t
A typedef of a node index.
Definition: fib_types.h:28
u32 as_index
Definition: lb.h:115
static fib_node_back_walk_rc_t lb_fib_node_back_walk_notify(fib_node_t *node, fib_node_back_walk_ctx_t *ctx)
Definition: lb.c:785
int lb_vip_del_ass(u32 vip_index, ip46_address_t *addresses, u32 n)
Definition: lb.c:583
void dpo_set(dpo_id_t *dpo, dpo_type_t type, dpo_proto_t proto, index_t index)
Set/create a DPO ID The DPO will be locked.
Definition: dpo.c:157
dpo_type_t dpo_gre6_type
Definition: lb.h:288
u32 last_garbage_collection
Last time garbage collection was run to free the ASs.
Definition: lb.h:173
lb_as_t * ass
Pool of ASs.
Definition: lb.h:240
lb_vip_type_t type
The type of traffic for this.
Definition: lb.h:196
Context passed between object during a back walk.
Definition: fib_node.h:190
fib_node_index_t fib_table_entry_special_dpo_add(u32 fib_index, const fib_prefix_t *prefix, fib_source_t source, fib_entry_flag_t flags, const dpo_id_t *dpo)
Add a &#39;special&#39; entry to the FIB that links to the DPO passed A special entry is an entry that the FI...
Definition: fib_table.c:288
void vlib_validate_simple_counter(vlib_simple_counter_main_t *cm, u32 index)
validate a simple counter
Definition: counter.c:78
#define ASSERT(truth)
unsigned int u32
Definition: types.h:88
lb_vip_type_t
The load balancer supports IPv4 and IPv6 traffic and GRE4 and GRE6 encap.
Definition: lb.h:135
int lb_vip_find_index(ip46_address_t *prefix, u8 plen, u32 *vip_index)
Definition: lb.c:399
u8 * format_lb_as(u8 *s, va_list *args)
Definition: lb.c:119
u32 new_flow_table_mask
New flows table length - 1 (length MUST be a power of 2)
Definition: lb.h:168
static void vlib_zero_simple_counter(vlib_simple_counter_main_t *cm, u32 index)
Clear a simple counter Clears the set of per-thread u16 counters, and the u64 counter.
Definition: counter.h:123
lb_per_cpu_t * per_cpu
Some global data is per-cpu.
Definition: lb.h:252
static void lb_dpo_unlock(dpo_id_t *dpo)
Definition: lb.c:750
static vlib_main_t * vlib_get_main(void)
Definition: global_funcs.h:23
static uword is_pow2(uword x)
Definition: clib.h:272
u32 as_index
Definition: lb.c:190
u64 uword
Definition: types.h:112
vlib_simple_counter_main_t vip_counters[LB_N_VIP_COUNTERS]
Per VIP counter.
Definition: lb.h:282
static const char *const lb_dpo_gre4_ip6[]
Definition: lb.c:36
#define DPO_PROTO_NUM
Definition: dpo.h:73
ip6_address_t ip6_src_address
Source address used in IPv6 encapsulated traffic.
Definition: lb.h:262
u8 * format_lb_vip_detailed(u8 *s, va_list *args)
Definition: lb.c:127
#define vec_len(v)
Number of elements in vector (rvalue-only, NULL tolerant)
unsigned char u8
Definition: types.h:56
static fib_node_t * lb_fib_node_get_node(fib_node_index_t index)
Definition: lb.c:753
#define vec_sort_with_function(vec, f)
Sort a vector using the supplied element comparison function.
Definition: vec.h:960
#define DPO_INVALID
An initialiser for DPOs declared on the stack.
Definition: dpo.h:173
void lb_garbage_collection()
Definition: lb.c:234
static int lb_vip_find_index_with_lock(ip46_address_t *prefix, u8 plen, u32 *vip_index)
Definition: lb.c:381
u32 next_hop_child_index
The child index on the FIB entry.
Definition: lb.h:103
A FIB graph nodes virtual function table.
Definition: fib_node.h:264
static void * clib_mem_alloc_aligned(uword size, uword align)
Definition: mem.h:117
static vlib_thread_main_t * vlib_get_thread_main()
Definition: global_funcs.h:32
void dpo_reset(dpo_id_t *dpo)
reset a DPO ID The DPO will be unlocked.
Definition: dpo.c:194
#define vec_foreach(var, vec)
Vector iterator.
dpo_id_t dpo
The next DPO in the graph to follow.
Definition: lb.h:108
u8 flags
Some per-AS flags.
Definition: lb.h:80
u16 dpoi_next_node
The next VLIB node to follow.
Definition: dpo.h:158
lb_new_flow_entry_t * new_flow_table
Vector mapping (flow-hash & new_connect_table_mask) to AS index.
Definition: lb.h:162
u8 flags
Flags related to this VIP.
Definition: lb.h:205
#define CLIB_CACHE_LINE_BYTES
Definition: cache.h:67
static char * lb_vip_type_strings[]
Definition: lb.c:79
u8 * format_lb_dpo(u8 *s, va_list *va)
Definition: lb.c:740
Load balancing service is provided per VIP.
Definition: lb.h:154
u32 * as_indexes
Pool of AS indexes used for this VIP.
Definition: lb.h:212
#define lb_hash_size(h)
Definition: lbhash.h:64
uword unformat(unformat_input_t *i, const char *fmt,...)
Definition: unformat.c:971
void dpo_stack(dpo_type_t child_type, dpo_proto_t child_proto, dpo_id_t *dpo, const dpo_id_t *parent)
Stack one DPO object on another, and thus establish a child-parent relationship.
Definition: dpo.c:401
#define lb_put_writer_lock()
Definition: lb.c:30
fib_node_index_t next_hop_fib_entry_index
The FIB entry index for the next-hop.
Definition: lb.h:98
static_always_inline u32 lb_hash_elts(lb_hash_t *h, u32 time_now)
Definition: lbhash.h:205
static uword pool_elts(void *v)
Number of active elements in a pool.
Definition: pool.h:109