]> git.ipfire.org Git - thirdparty/bird.git/blob - nest/rt-table.c
Nest: Keep multipath next hops sorted
[thirdparty/bird.git] / nest / rt-table.c
1 /*
2 * BIRD -- Routing Tables
3 *
4 * (c) 1998--2000 Martin Mares <mj@ucw.cz>
5 *
6 * Can be freely distributed and used under the terms of the GNU GPL.
7 */
8
9 /**
10 * DOC: Routing tables
11 *
12 * Routing tables are probably the most important structures BIRD uses. They
13 * hold all the information about known networks, the associated routes and
14 * their attributes.
15 *
16 * There are multiple routing tables (a primary one together with any
17 * number of secondary ones if requested by the configuration). Each table
18 * is basically a FIB containing entries describing the individual
19 * destination networks. For each network (represented by structure &net),
20 * there is a one-way linked list of route entries (&rte), the first entry
21 * on the list being the best one (i.e., the one we currently use
22 * for routing), the order of the other ones is undetermined.
23 *
24 * The &rte contains information specific to the route (preference, protocol
25 * metrics, time of last modification etc.) and a pointer to a &rta structure
26 * (see the route attribute module for a precise explanation) holding the
27 * remaining route attributes which are expected to be shared by multiple
28 * routes in order to conserve memory.
29 */
30
31 #undef LOCAL_DEBUG
32
33 #include "nest/bird.h"
34 #include "nest/route.h"
35 #include "nest/protocol.h"
36 #include "nest/cli.h"
37 #include "nest/iface.h"
38 #include "lib/resource.h"
39 #include "lib/event.h"
40 #include "lib/string.h"
41 #include "conf/conf.h"
42 #include "filter/filter.h"
43 #include "lib/string.h"
44 #include "lib/alloca.h"
45
46 pool *rt_table_pool;
47
48 static slab *rte_slab;
49 static linpool *rte_update_pool;
50
51 static list routing_tables;
52
53 static byte *rt_format_via(rte *e);
54 static void rt_free_hostcache(rtable *tab);
55 static void rt_notify_hostcache(rtable *tab, net *net);
56 static void rt_update_hostcache(rtable *tab);
57 static void rt_next_hop_update(rtable *tab);
58 static inline int rt_prune_table(rtable *tab);
59 static inline void rt_schedule_gc(rtable *tab);
60 static inline void rt_schedule_prune(rtable *tab);
61
62
63 static inline struct ea_list *
64 make_tmp_attrs(struct rte *rt, struct linpool *pool)
65 {
66 struct ea_list *(*mta)(struct rte *rt, struct linpool *pool);
67 mta = rt->attrs->src->proto->make_tmp_attrs;
68 return mta ? mta(rt, rte_update_pool) : NULL;
69 }
70
71 /* Like fib_route(), but skips empty net entries */
72 static net *
73 net_route(rtable *tab, ip_addr a, int len)
74 {
75 ip_addr a0;
76 net *n;
77
78 while (len >= 0)
79 {
80 a0 = ipa_and(a, ipa_mkmask(len));
81 n = fib_find(&tab->fib, &a0, len);
82 if (n && rte_is_valid(n->routes))
83 return n;
84 len--;
85 }
86 return NULL;
87 }
88
89 static void
90 rte_init(struct fib_node *N)
91 {
92 net *n = (net *) N;
93
94 N->flags = 0;
95 n->routes = NULL;
96 }
97
98 /**
99 * rte_find - find a route
100 * @net: network node
101 * @src: route source
102 *
103 * The rte_find() function returns a route for destination @net
104 * which is from route source @src.
105 */
106 rte *
107 rte_find(net *net, struct rte_src *src)
108 {
109 rte *e = net->routes;
110
111 while (e && e->attrs->src != src)
112 e = e->next;
113 return e;
114 }
115
116 /**
117 * rte_get_temp - get a temporary &rte
118 * @a: attributes to assign to the new route (a &rta; in case it's
119 * un-cached, rte_update() will create a cached copy automatically)
120 *
121 * Create a temporary &rte and bind it with the attributes @a.
122 * Also set route preference to the default preference set for
123 * the protocol.
124 */
125 rte *
126 rte_get_temp(rta *a)
127 {
128 rte *e = sl_alloc(rte_slab);
129
130 e->attrs = a;
131 e->flags = 0;
132 e->pref = a->src->proto->preference;
133 return e;
134 }
135
136 rte *
137 rte_do_cow(rte *r)
138 {
139 rte *e = sl_alloc(rte_slab);
140
141 memcpy(e, r, sizeof(rte));
142 e->attrs = rta_clone(r->attrs);
143 e->flags = 0;
144 return e;
145 }
146
147 /**
148 * rte_cow_rta - get a private writable copy of &rte with writable &rta
149 * @r: a route entry to be copied
150 * @lp: a linpool from which to allocate &rta
151 *
152 * rte_cow_rta() takes a &rte and prepares it and associated &rta for
153 * modification. There are three possibilities: First, both &rte and &rta are
154 * private copies, in that case they are returned unchanged. Second, &rte is
155 * private copy, but &rta is cached, in that case &rta is duplicated using
156 * rta_do_cow(). Third, both &rte is shared and &rta is cached, in that case
157 * both structures are duplicated by rte_do_cow() and rta_do_cow().
158 *
159 * Note that in the second case, cached &rta loses one reference, while private
160 * copy created by rta_do_cow() is a shallow copy sharing indirect data (eattrs,
161 * nexthops, ...) with it. To work properly, original shared &rta should have
162 * another reference during the life of created private copy.
163 *
164 * Result: a pointer to the new writable &rte with writable &rta.
165 */
166 rte *
167 rte_cow_rta(rte *r, linpool *lp)
168 {
169 if (!rta_is_cached(r->attrs))
170 return r;
171
172 rte *e = rte_cow(r);
173 rta *a = rta_do_cow(r->attrs, lp);
174 rta_free(e->attrs);
175 e->attrs = a;
176 return e;
177 }
178
179 static int /* Actually better or at least as good as */
180 rte_better(rte *new, rte *old)
181 {
182 int (*better)(rte *, rte *);
183
184 if (!rte_is_valid(old))
185 return 1;
186 if (!rte_is_valid(new))
187 return 0;
188
189 if (new->pref > old->pref)
190 return 1;
191 if (new->pref < old->pref)
192 return 0;
193 if (new->attrs->src->proto->proto != old->attrs->src->proto->proto)
194 {
195 /*
196 * If the user has configured protocol preferences, so that two different protocols
197 * have the same preference, try to break the tie by comparing addresses. Not too
198 * useful, but keeps the ordering of routes unambiguous.
199 */
200 return new->attrs->src->proto->proto > old->attrs->src->proto->proto;
201 }
202 if (better = new->attrs->src->proto->rte_better)
203 return better(new, old);
204 return 0;
205 }
206
207 static int
208 rte_mergable(rte *pri, rte *sec)
209 {
210 int (*mergable)(rte *, rte *);
211
212 if (!rte_is_valid(pri) || !rte_is_valid(sec))
213 return 0;
214
215 if (pri->pref != sec->pref)
216 return 0;
217
218 if (pri->attrs->src->proto->proto != sec->attrs->src->proto->proto)
219 return 0;
220
221 if (mergable = pri->attrs->src->proto->rte_mergable)
222 return mergable(pri, sec);
223
224 return 0;
225 }
226
227 static void
228 rte_trace(struct proto *p, rte *e, int dir, char *msg)
229 {
230 log(L_TRACE "%s %c %s %I/%d %s", p->name, dir, msg, e->net->n.prefix, e->net->n.pxlen, rt_format_via(e));
231 }
232
233 static inline void
234 rte_trace_in(uint flag, struct proto *p, rte *e, char *msg)
235 {
236 if (p->debug & flag)
237 rte_trace(p, e, '>', msg);
238 }
239
240 static inline void
241 rte_trace_out(uint flag, struct proto *p, rte *e, char *msg)
242 {
243 if (p->debug & flag)
244 rte_trace(p, e, '<', msg);
245 }
246
247 static rte *
248 export_filter(struct announce_hook *ah, rte *rt0, rte **rt_free, ea_list **tmpa, int silent)
249 {
250 struct proto *p = ah->proto;
251 struct filter *filter = ah->out_filter;
252 struct proto_stats *stats = ah->stats;
253 ea_list *tmpb = NULL;
254 rte *rt;
255 int v;
256
257 rt = rt0;
258 *rt_free = NULL;
259
260 if (!tmpa)
261 tmpa = &tmpb;
262
263 *tmpa = make_tmp_attrs(rt, rte_update_pool);
264
265 v = p->import_control ? p->import_control(p, &rt, tmpa, rte_update_pool) : 0;
266 if (v < 0)
267 {
268 if (silent)
269 goto reject;
270
271 stats->exp_updates_rejected++;
272 if (v == RIC_REJECT)
273 rte_trace_out(D_FILTERS, p, rt, "rejected by protocol");
274 goto reject;
275 }
276 if (v > 0)
277 {
278 if (!silent)
279 rte_trace_out(D_FILTERS, p, rt, "forced accept by protocol");
280 goto accept;
281 }
282
283 v = filter && ((filter == FILTER_REJECT) ||
284 (f_run(filter, &rt, tmpa, rte_update_pool, FF_FORCE_TMPATTR) > F_ACCEPT));
285 if (v)
286 {
287 if (silent)
288 goto reject;
289
290 stats->exp_updates_filtered++;
291 rte_trace_out(D_FILTERS, p, rt, "filtered out");
292 goto reject;
293 }
294
295 accept:
296 if (rt != rt0)
297 *rt_free = rt;
298 return rt;
299
300 reject:
301 /* Discard temporary rte */
302 if (rt != rt0)
303 rte_free(rt);
304 return NULL;
305 }
306
307 static void
308 do_rt_notify(struct announce_hook *ah, net *net, rte *new, rte *old, ea_list *tmpa, int refeed)
309 {
310 struct proto *p = ah->proto;
311 struct proto_stats *stats = ah->stats;
312
313
314 /*
315 * First, apply export limit.
316 *
317 * Export route limits has several problems. Because exp_routes
318 * counter is reset before refeed, we don't really know whether
319 * limit is breached and whether the update is new or not. Therefore
320 * the number of really exported routes may exceed the limit
321 * temporarily (routes exported before and new routes in refeed).
322 *
323 * Minor advantage is that if the limit is decreased and refeed is
324 * requested, the number of exported routes really decrease.
325 *
326 * Second problem is that with export limits, we don't know whether
327 * old was really exported (it might be blocked by limit). When a
328 * withdraw is exported, we announce it even when the previous
329 * update was blocked. This is not a big issue, but the same problem
330 * is in updating exp_routes counter. Therefore, to be consistent in
331 * increases and decreases of exp_routes, we count exported routes
332 * regardless of blocking by limits.
333 *
334 * Similar problem is in handling updates - when a new route is
335 * received and blocking is active, the route would be blocked, but
336 * when an update for the route will be received later, the update
337 * would be propagated (as old != NULL). Therefore, we have to block
338 * also non-new updates (contrary to import blocking).
339 */
340
341 struct proto_limit *l = ah->out_limit;
342 if (l && new)
343 {
344 if ((!old || refeed) && (stats->exp_routes >= l->limit))
345 proto_notify_limit(ah, l, PLD_OUT, stats->exp_routes);
346
347 if (l->state == PLS_BLOCKED)
348 {
349 stats->exp_routes++; /* see note above */
350 stats->exp_updates_rejected++;
351 rte_trace_out(D_FILTERS, p, new, "rejected [limit]");
352 new = NULL;
353
354 if (!old)
355 return;
356 }
357 }
358
359
360 if (new)
361 stats->exp_updates_accepted++;
362 else
363 stats->exp_withdraws_accepted++;
364
365 /* Hack: We do not decrease exp_routes during refeed, we instead
366 reset exp_routes at the start of refeed. */
367 if (new)
368 stats->exp_routes++;
369 if (old && !refeed)
370 stats->exp_routes--;
371
372 if (p->debug & D_ROUTES)
373 {
374 if (new && old)
375 rte_trace_out(D_ROUTES, p, new, "replaced");
376 else if (new)
377 rte_trace_out(D_ROUTES, p, new, "added");
378 else if (old)
379 rte_trace_out(D_ROUTES, p, old, "removed");
380 }
381 if (!new)
382 p->rt_notify(p, ah->table, net, NULL, old, NULL);
383 else if (tmpa)
384 {
385 ea_list *t = tmpa;
386 while (t->next)
387 t = t->next;
388 t->next = new->attrs->eattrs;
389 p->rt_notify(p, ah->table, net, new, old, tmpa);
390 t->next = NULL;
391 }
392 else
393 p->rt_notify(p, ah->table, net, new, old, new->attrs->eattrs);
394 }
395
396 static void
397 rt_notify_basic(struct announce_hook *ah, net *net, rte *new0, rte *old0, int refeed)
398 {
399 struct proto *p = ah->proto;
400 struct proto_stats *stats = ah->stats;
401
402 rte *new = new0;
403 rte *old = old0;
404 rte *new_free = NULL;
405 rte *old_free = NULL;
406 ea_list *tmpa = NULL;
407
408 if (new)
409 stats->exp_updates_received++;
410 else
411 stats->exp_withdraws_received++;
412
413 /*
414 * This is a tricky part - we don't know whether route 'old' was
415 * exported to protocol 'p' or was filtered by the export filter.
416 * We try to run the export filter to know this to have a correct
417 * value in 'old' argument of rte_update (and proper filter value)
418 *
419 * FIXME - this is broken because 'configure soft' may change
420 * filters but keep routes. Refeed is expected to be called after
421 * change of the filters and with old == new, therefore we do not
422 * even try to run the filter on an old route, This may lead to
423 * 'spurious withdraws' but ensure that there are no 'missing
424 * withdraws'.
425 *
426 * This is not completely safe as there is a window between
427 * reconfiguration and the end of refeed - if a newly filtered
428 * route disappears during this period, proper withdraw is not
429 * sent (because old would be also filtered) and the route is
430 * not refeeded (because it disappeared before that).
431 */
432
433 if (new)
434 new = export_filter(ah, new, &new_free, &tmpa, 0);
435
436 if (old && !refeed)
437 old = export_filter(ah, old, &old_free, NULL, 1);
438
439 if (!new && !old)
440 {
441 /*
442 * As mentioned above, 'old' value may be incorrect in some race conditions.
443 * We generally ignore it with the exception of withdraw to pipe protocol.
444 * In that case we rather propagate unfiltered withdraws regardless of
445 * export filters to ensure that when a protocol is flushed, its routes are
446 * removed from all tables. Possible spurious unfiltered withdraws are not
447 * problem here as they are ignored if there is no corresponding route at
448 * the other end of the pipe. We directly call rt_notify() hook instead of
449 * do_rt_notify() to avoid logging and stat counters.
450 */
451
452 #ifdef CONFIG_PIPE
453 if ((p->proto == &proto_pipe) && !new0 && (p != old0->sender->proto))
454 p->rt_notify(p, ah->table, net, NULL, old0, NULL);
455 #endif
456
457 return;
458 }
459
460 do_rt_notify(ah, net, new, old, tmpa, refeed);
461
462 /* Discard temporary rte's */
463 if (new_free)
464 rte_free(new_free);
465 if (old_free)
466 rte_free(old_free);
467 }
468
469 static void
470 rt_notify_accepted(struct announce_hook *ah, net *net, rte *new_changed, rte *old_changed, rte *before_old, int feed)
471 {
472 // struct proto *p = ah->proto;
473 struct proto_stats *stats = ah->stats;
474
475 rte *r;
476 rte *new_best = NULL;
477 rte *old_best = NULL;
478 rte *new_free = NULL;
479 rte *old_free = NULL;
480 ea_list *tmpa = NULL;
481
482 /* Used to track whether we met old_changed position. If before_old is NULL
483 old_changed was the first and we met it implicitly before current best route. */
484 int old_meet = old_changed && !before_old;
485
486 /* Note that before_old is either NULL or valid (not rejected) route.
487 If old_changed is valid, before_old have to be too. If old changed route
488 was not valid, caller must use NULL for both old_changed and before_old. */
489
490 if (new_changed)
491 stats->exp_updates_received++;
492 else
493 stats->exp_withdraws_received++;
494
495 /* First, find the new_best route - first accepted by filters */
496 for (r=net->routes; rte_is_valid(r); r=r->next)
497 {
498 if (new_best = export_filter(ah, r, &new_free, &tmpa, 0))
499 break;
500
501 /* Note if we walked around the position of old_changed route */
502 if (r == before_old)
503 old_meet = 1;
504 }
505
506 /*
507 * Second, handle the feed case. That means we do not care for
508 * old_best. It is NULL for feed, and the new_best for refeed.
509 * For refeed, there is a hack similar to one in rt_notify_basic()
510 * to ensure withdraws in case of changed filters
511 */
512 if (feed)
513 {
514 if (feed == 2) /* refeed */
515 old_best = new_best ? new_best :
516 (rte_is_valid(net->routes) ? net->routes : NULL);
517 else
518 old_best = NULL;
519
520 if (!new_best && !old_best)
521 return;
522
523 goto found;
524 }
525
526 /*
527 * Now, we find the old_best route. Generally, it is the same as the
528 * new_best, unless new_best is the same as new_changed or
529 * old_changed is accepted before new_best.
530 *
531 * There are four cases:
532 *
533 * - We would find and accept old_changed before new_best, therefore
534 * old_changed is old_best. In remaining cases we suppose this
535 * is not true.
536 *
537 * - We found no new_best, therefore there is also no old_best and
538 * we ignore this withdraw.
539 *
540 * - We found new_best different than new_changed, therefore
541 * old_best is the same as new_best and we ignore this update.
542 *
543 * - We found new_best the same as new_changed, therefore it cannot
544 * be old_best and we have to continue search for old_best.
545 */
546
547 /* First case */
548 if (old_meet)
549 if (old_best = export_filter(ah, old_changed, &old_free, NULL, 1))
550 goto found;
551
552 /* Second case */
553 if (!new_best)
554 return;
555
556 /* Third case, we use r instead of new_best, because export_filter() could change it */
557 if (r != new_changed)
558 {
559 if (new_free)
560 rte_free(new_free);
561 return;
562 }
563
564 /* Fourth case */
565 for (r=r->next; rte_is_valid(r); r=r->next)
566 {
567 if (old_best = export_filter(ah, r, &old_free, NULL, 1))
568 goto found;
569
570 if (r == before_old)
571 if (old_best = export_filter(ah, old_changed, &old_free, NULL, 1))
572 goto found;
573 }
574
575 /* Implicitly, old_best is NULL and new_best is non-NULL */
576
577 found:
578 do_rt_notify(ah, net, new_best, old_best, tmpa, (feed == 2));
579
580 /* Discard temporary rte's */
581 if (new_free)
582 rte_free(new_free);
583 if (old_free)
584 rte_free(old_free);
585 }
586
587
588 static struct mpnh *
589 mpnh_merge_rta(struct mpnh *nhs, rta *a, int max)
590 {
591 struct mpnh nh = { .gw = a->gw, .iface = a->iface };
592 struct mpnh *nh2 = (a->dest == RTD_MULTIPATH) ? a->nexthops : &nh;
593 return mpnh_merge(nhs, nh2, 1, 0, max, rte_update_pool);
594 }
595
596 rte *
597 rt_export_merged(struct announce_hook *ah, net *net, rte **rt_free, ea_list **tmpa, int silent)
598 {
599 // struct proto *p = ah->proto;
600 struct mpnh *nhs = NULL;
601 rte *best0, *best, *rt0, *rt, *tmp;
602
603 best0 = net->routes;
604 *rt_free = NULL;
605
606 if (!rte_is_valid(best0))
607 return NULL;
608
609 best = export_filter(ah, best0, rt_free, tmpa, silent);
610
611 if (!best || !rte_is_reachable(best))
612 return best;
613
614 for (rt0 = best0->next; rt0; rt0 = rt0->next)
615 {
616 if (!rte_mergable(best0, rt0))
617 continue;
618
619 rt = export_filter(ah, rt0, &tmp, NULL, 1);
620
621 if (!rt)
622 continue;
623
624 if (rte_is_reachable(rt))
625 nhs = mpnh_merge_rta(nhs, rt->attrs, ah->proto->merge_limit);
626
627 if (tmp)
628 rte_free(tmp);
629 }
630
631 if (nhs)
632 {
633 nhs = mpnh_merge_rta(nhs, best->attrs, ah->proto->merge_limit);
634
635 if (nhs->next)
636 {
637 best = rte_cow_rta(best, rte_update_pool);
638 best->attrs->dest = RTD_MULTIPATH;
639 best->attrs->nexthops = nhs;
640 }
641 }
642
643 if (best != best0)
644 *rt_free = best;
645
646 return best;
647 }
648
649
650 static void
651 rt_notify_merged(struct announce_hook *ah, net *net, rte *new_changed, rte *old_changed,
652 rte *new_best, rte*old_best, int refeed)
653 {
654 // struct proto *p = ah->proto;
655
656 rte *new_best_free = NULL;
657 rte *old_best_free = NULL;
658 rte *new_changed_free = NULL;
659 rte *old_changed_free = NULL;
660 ea_list *tmpa = NULL;
661
662 /* We assume that all rte arguments are either NULL or rte_is_valid() */
663
664 /* This check should be done by the caller */
665 if (!new_best && !old_best)
666 return;
667
668 /* Check whether the change is relevant to the merged route */
669 if ((new_best == old_best) && !refeed)
670 {
671 new_changed = rte_mergable(new_best, new_changed) ?
672 export_filter(ah, new_changed, &new_changed_free, NULL, 1) : NULL;
673
674 old_changed = rte_mergable(old_best, old_changed) ?
675 export_filter(ah, old_changed, &old_changed_free, NULL, 1) : NULL;
676
677 if (!new_changed && !old_changed)
678 return;
679 }
680
681 if (new_best)
682 ah->stats->exp_updates_received++;
683 else
684 ah->stats->exp_withdraws_received++;
685
686 /* Prepare new merged route */
687 if (new_best)
688 new_best = rt_export_merged(ah, net, &new_best_free, &tmpa, 0);
689
690 /* Prepare old merged route (without proper merged next hops) */
691 /* There are some issues with running filter on old route - see rt_notify_basic() */
692 if (old_best && !refeed)
693 old_best = export_filter(ah, old_best, &old_best_free, NULL, 1);
694
695 if (new_best || old_best)
696 do_rt_notify(ah, net, new_best, old_best, tmpa, refeed);
697
698 /* Discard temporary rte's */
699 if (new_best_free)
700 rte_free(new_best_free);
701 if (old_best_free)
702 rte_free(old_best_free);
703 if (new_changed_free)
704 rte_free(new_changed_free);
705 if (old_changed_free)
706 rte_free(old_changed_free);
707 }
708
709
710 /**
711 * rte_announce - announce a routing table change
712 * @tab: table the route has been added to
713 * @type: type of route announcement (RA_OPTIMAL or RA_ANY)
714 * @net: network in question
715 * @new: the new route to be announced
716 * @old: the previous route for the same network
717 * @new_best: the new best route for the same network
718 * @old_best: the previous best route for the same network
719 * @before_old: The previous route before @old for the same network.
720 * If @before_old is NULL @old was the first.
721 *
722 * This function gets a routing table update and announces it
723 * to all protocols that acccepts given type of route announcement
724 * and are connected to the same table by their announcement hooks.
725 *
726 * Route announcement of type %RA_OPTIMAL si generated when optimal
727 * route (in routing table @tab) changes. In that case @old stores the
728 * old optimal route.
729 *
730 * Route announcement of type %RA_ANY si generated when any route (in
731 * routing table @tab) changes In that case @old stores the old route
732 * from the same protocol.
733 *
734 * For each appropriate protocol, we first call its import_control()
735 * hook which performs basic checks on the route (each protocol has a
736 * right to veto or force accept of the route before any filter is
737 * asked) and adds default values of attributes specific to the new
738 * protocol (metrics, tags etc.). Then it consults the protocol's
739 * export filter and if it accepts the route, the rt_notify() hook of
740 * the protocol gets called.
741 */
742 static void
743 rte_announce(rtable *tab, unsigned type, net *net, rte *new, rte *old,
744 rte *new_best, rte *old_best, rte *before_old)
745 {
746 if (!rte_is_valid(new))
747 new = NULL;
748
749 if (!rte_is_valid(old))
750 old = before_old = NULL;
751
752 if (!rte_is_valid(new_best))
753 new_best = NULL;
754
755 if (!rte_is_valid(old_best))
756 old_best = NULL;
757
758 if (!old && !new)
759 return;
760
761 if (type == RA_OPTIMAL)
762 {
763 if (new)
764 new->attrs->src->proto->stats.pref_routes++;
765 if (old)
766 old->attrs->src->proto->stats.pref_routes--;
767
768 if (tab->hostcache)
769 rt_notify_hostcache(tab, net);
770 }
771
772 struct announce_hook *a;
773 WALK_LIST(a, tab->hooks)
774 {
775 ASSERT(a->proto->export_state != ES_DOWN);
776 if (a->proto->accept_ra_types == type)
777 if (type == RA_ACCEPTED)
778 rt_notify_accepted(a, net, new, old, before_old, 0);
779 else if (type == RA_MERGED)
780 rt_notify_merged(a, net, new, old, new_best, old_best, 0);
781 else
782 rt_notify_basic(a, net, new, old, 0);
783 }
784 }
785
786 static inline int
787 rte_validate(rte *e)
788 {
789 int c;
790 net *n = e->net;
791
792 if ((n->n.pxlen > BITS_PER_IP_ADDRESS) || !ip_is_prefix(n->n.prefix,n->n.pxlen))
793 {
794 log(L_WARN "Ignoring bogus prefix %I/%d received via %s",
795 n->n.prefix, n->n.pxlen, e->sender->proto->name);
796 return 0;
797 }
798
799 c = ipa_classify_net(n->n.prefix);
800 if ((c < 0) || !(c & IADDR_HOST) || ((c & IADDR_SCOPE_MASK) <= SCOPE_LINK))
801 {
802 log(L_WARN "Ignoring bogus route %I/%d received via %s",
803 n->n.prefix, n->n.pxlen, e->sender->proto->name);
804 return 0;
805 }
806
807 if ((e->attrs->dest == RTD_MULTIPATH) && !mpnh_is_sorted(e->attrs->nexthops))
808 {
809 log(L_WARN "Ignoring unsorted multipath route %I/%d received via %s",
810 n->n.prefix, n->n.pxlen, e->sender->proto->name);
811 return 0;
812 }
813
814 return 1;
815 }
816
817 /**
818 * rte_free - delete a &rte
819 * @e: &rte to be deleted
820 *
821 * rte_free() deletes the given &rte from the routing table it's linked to.
822 */
823 void
824 rte_free(rte *e)
825 {
826 if (rta_is_cached(e->attrs))
827 rta_free(e->attrs);
828 sl_free(rte_slab, e);
829 }
830
831 static inline void
832 rte_free_quick(rte *e)
833 {
834 rta_free(e->attrs);
835 sl_free(rte_slab, e);
836 }
837
838 static int
839 rte_same(rte *x, rte *y)
840 {
841 return
842 x->attrs == y->attrs &&
843 x->flags == y->flags &&
844 x->pflags == y->pflags &&
845 x->pref == y->pref &&
846 (!x->attrs->src->proto->rte_same || x->attrs->src->proto->rte_same(x, y));
847 }
848
849 static inline int rte_is_ok(rte *e) { return e && !rte_is_filtered(e); }
850
851 static void
852 rte_recalculate(struct announce_hook *ah, net *net, rte *new, struct rte_src *src)
853 {
854 struct proto *p = ah->proto;
855 struct rtable *table = ah->table;
856 struct proto_stats *stats = ah->stats;
857 static struct tbf rl_pipe = TBF_DEFAULT_LOG_LIMITS;
858 rte *before_old = NULL;
859 rte *old_best = net->routes;
860 rte *old = NULL;
861 rte **k;
862
863 k = &net->routes; /* Find and remove original route from the same protocol */
864 while (old = *k)
865 {
866 if (old->attrs->src == src)
867 {
868 /* If there is the same route in the routing table but from
869 * a different sender, then there are two paths from the
870 * source protocol to this routing table through transparent
871 * pipes, which is not allowed.
872 *
873 * We log that and ignore the route. If it is withdraw, we
874 * ignore it completely (there might be 'spurious withdraws',
875 * see FIXME in do_rte_announce())
876 */
877 if (old->sender->proto != p)
878 {
879 if (new)
880 {
881 log_rl(&rl_pipe, L_ERR "Pipe collision detected when sending %I/%d to table %s",
882 net->n.prefix, net->n.pxlen, table->name);
883 rte_free_quick(new);
884 }
885 return;
886 }
887
888 if (new && rte_same(old, new))
889 {
890 /* No changes, ignore the new route */
891
892 if (!rte_is_filtered(new))
893 {
894 stats->imp_updates_ignored++;
895 rte_trace_in(D_ROUTES, p, new, "ignored");
896 }
897
898 rte_free_quick(new);
899 return;
900 }
901 *k = old->next;
902 break;
903 }
904 k = &old->next;
905 before_old = old;
906 }
907
908 if (!old)
909 before_old = NULL;
910
911 if (!old && !new)
912 {
913 stats->imp_withdraws_ignored++;
914 return;
915 }
916
917 int new_ok = rte_is_ok(new);
918 int old_ok = rte_is_ok(old);
919
920 struct proto_limit *l = ah->rx_limit;
921 if (l && !old && new)
922 {
923 u32 all_routes = stats->imp_routes + stats->filt_routes;
924
925 if (all_routes >= l->limit)
926 proto_notify_limit(ah, l, PLD_RX, all_routes);
927
928 if (l->state == PLS_BLOCKED)
929 {
930 /* In receive limit the situation is simple, old is NULL so
931 we just free new and exit like nothing happened */
932
933 stats->imp_updates_ignored++;
934 rte_trace_in(D_FILTERS, p, new, "ignored [limit]");
935 rte_free_quick(new);
936 return;
937 }
938 }
939
940 l = ah->in_limit;
941 if (l && !old_ok && new_ok)
942 {
943 if (stats->imp_routes >= l->limit)
944 proto_notify_limit(ah, l, PLD_IN, stats->imp_routes);
945
946 if (l->state == PLS_BLOCKED)
947 {
948 /* In import limit the situation is more complicated. We
949 shouldn't just drop the route, we should handle it like
950 it was filtered. We also have to continue the route
951 processing if old or new is non-NULL, but we should exit
952 if both are NULL as this case is probably assumed to be
953 already handled. */
954
955 stats->imp_updates_ignored++;
956 rte_trace_in(D_FILTERS, p, new, "ignored [limit]");
957
958 if (ah->in_keep_filtered)
959 new->flags |= REF_FILTERED;
960 else
961 { rte_free_quick(new); new = NULL; }
962
963 /* Note that old && !new could be possible when
964 ah->in_keep_filtered changed in the recent past. */
965
966 if (!old && !new)
967 return;
968
969 new_ok = 0;
970 goto skip_stats1;
971 }
972 }
973
974 if (new_ok)
975 stats->imp_updates_accepted++;
976 else if (old_ok)
977 stats->imp_withdraws_accepted++;
978 else
979 stats->imp_withdraws_ignored++;
980
981 skip_stats1:
982
983 if (new)
984 rte_is_filtered(new) ? stats->filt_routes++ : stats->imp_routes++;
985 if (old)
986 rte_is_filtered(old) ? stats->filt_routes-- : stats->imp_routes--;
987
988 if (table->config->sorted)
989 {
990 /* If routes are sorted, just insert new route to appropriate position */
991 if (new)
992 {
993 if (before_old && !rte_better(new, before_old))
994 k = &before_old->next;
995 else
996 k = &net->routes;
997
998 for (; *k; k=&(*k)->next)
999 if (rte_better(new, *k))
1000 break;
1001
1002 new->next = *k;
1003 *k = new;
1004 }
1005 }
1006 else
1007 {
1008 /* If routes are not sorted, find the best route and move it on
1009 the first position. There are several optimized cases. */
1010
1011 if (src->proto->rte_recalculate && src->proto->rte_recalculate(table, net, new, old, old_best))
1012 goto do_recalculate;
1013
1014 if (new && rte_better(new, old_best))
1015 {
1016 /* The first case - the new route is cleary optimal,
1017 we link it at the first position */
1018
1019 new->next = net->routes;
1020 net->routes = new;
1021 }
1022 else if (old == old_best)
1023 {
1024 /* The second case - the old best route disappeared, we add the
1025 new route (if we have any) to the list (we don't care about
1026 position) and then we elect the new optimal route and relink
1027 that route at the first position and announce it. New optimal
1028 route might be NULL if there is no more routes */
1029
1030 do_recalculate:
1031 /* Add the new route to the list */
1032 if (new)
1033 {
1034 new->next = net->routes;
1035 net->routes = new;
1036 }
1037
1038 /* Find a new optimal route (if there is any) */
1039 if (net->routes)
1040 {
1041 rte **bp = &net->routes;
1042 for (k=&(*bp)->next; *k; k=&(*k)->next)
1043 if (rte_better(*k, *bp))
1044 bp = k;
1045
1046 /* And relink it */
1047 rte *best = *bp;
1048 *bp = best->next;
1049 best->next = net->routes;
1050 net->routes = best;
1051 }
1052 }
1053 else if (new)
1054 {
1055 /* The third case - the new route is not better than the old
1056 best route (therefore old_best != NULL) and the old best
1057 route was not removed (therefore old_best == net->routes).
1058 We just link the new route after the old best route. */
1059
1060 ASSERT(net->routes != NULL);
1061 new->next = net->routes->next;
1062 net->routes->next = new;
1063 }
1064 /* The fourth (empty) case - suboptimal route was removed, nothing to do */
1065 }
1066
1067 if (new)
1068 new->lastmod = now;
1069
1070 /* Log the route change */
1071 if (p->debug & D_ROUTES)
1072 {
1073 if (new_ok)
1074 rte_trace(p, new, '>', new == net->routes ? "added [best]" : "added");
1075 else if (old_ok)
1076 {
1077 if (old != old_best)
1078 rte_trace(p, old, '>', "removed");
1079 else if (rte_is_ok(net->routes))
1080 rte_trace(p, old, '>', "removed [replaced]");
1081 else
1082 rte_trace(p, old, '>', "removed [sole]");
1083 }
1084 }
1085
1086 /* Propagate the route change */
1087 rte_announce(table, RA_ANY, net, new, old, NULL, NULL, NULL);
1088 if (net->routes != old_best)
1089 rte_announce(table, RA_OPTIMAL, net, net->routes, old_best, NULL, NULL, NULL);
1090 if (table->config->sorted)
1091 rte_announce(table, RA_ACCEPTED, net, new, old, NULL, NULL, before_old);
1092 rte_announce(table, RA_MERGED, net, new, old, net->routes, old_best, NULL);
1093
1094 if (!net->routes &&
1095 (table->gc_counter++ >= table->config->gc_max_ops) &&
1096 (table->gc_time + table->config->gc_min_time <= now))
1097 rt_schedule_gc(table);
1098
1099 if (old_ok && p->rte_remove)
1100 p->rte_remove(net, old);
1101 if (new_ok && p->rte_insert)
1102 p->rte_insert(net, new);
1103
1104 if (old)
1105 rte_free_quick(old);
1106 }
1107
1108 static int rte_update_nest_cnt; /* Nesting counter to allow recursive updates */
1109
1110 static inline void
1111 rte_update_lock(void)
1112 {
1113 rte_update_nest_cnt++;
1114 }
1115
1116 static inline void
1117 rte_update_unlock(void)
1118 {
1119 if (!--rte_update_nest_cnt)
1120 lp_flush(rte_update_pool);
1121 }
1122
1123 static inline void
1124 rte_hide_dummy_routes(net *net, rte **dummy)
1125 {
1126 if (net->routes && net->routes->attrs->source == RTS_DUMMY)
1127 {
1128 *dummy = net->routes;
1129 net->routes = (*dummy)->next;
1130 }
1131 }
1132
1133 static inline void
1134 rte_unhide_dummy_routes(net *net, rte **dummy)
1135 {
1136 if (*dummy)
1137 {
1138 (*dummy)->next = net->routes;
1139 net->routes = *dummy;
1140 }
1141 }
1142
1143 /**
1144 * rte_update - enter a new update to a routing table
1145 * @table: table to be updated
1146 * @ah: pointer to table announce hook
1147 * @net: network node
1148 * @p: protocol submitting the update
1149 * @src: protocol originating the update
1150 * @new: a &rte representing the new route or %NULL for route removal.
1151 *
1152 * This function is called by the routing protocols whenever they discover
1153 * a new route or wish to update/remove an existing route. The right announcement
1154 * sequence is to build route attributes first (either un-cached with @aflags set
1155 * to zero or a cached one using rta_lookup(); in this case please note that
1156 * you need to increase the use count of the attributes yourself by calling
1157 * rta_clone()), call rte_get_temp() to obtain a temporary &rte, fill in all
1158 * the appropriate data and finally submit the new &rte by calling rte_update().
1159 *
1160 * @src specifies the protocol that originally created the route and the meaning
1161 * of protocol-dependent data of @new. If @new is not %NULL, @src have to be the
1162 * same value as @new->attrs->proto. @p specifies the protocol that called
1163 * rte_update(). In most cases it is the same protocol as @src. rte_update()
1164 * stores @p in @new->sender;
1165 *
1166 * When rte_update() gets any route, it automatically validates it (checks,
1167 * whether the network and next hop address are valid IP addresses and also
1168 * whether a normal routing protocol doesn't try to smuggle a host or link
1169 * scope route to the table), converts all protocol dependent attributes stored
1170 * in the &rte to temporary extended attributes, consults import filters of the
1171 * protocol to see if the route should be accepted and/or its attributes modified,
1172 * stores the temporary attributes back to the &rte.
1173 *
1174 * Now, having a "public" version of the route, we
1175 * automatically find any old route defined by the protocol @src
1176 * for network @n, replace it by the new one (or removing it if @new is %NULL),
1177 * recalculate the optimal route for this destination and finally broadcast
1178 * the change (if any) to all routing protocols by calling rte_announce().
1179 *
1180 * All memory used for attribute lists and other temporary allocations is taken
1181 * from a special linear pool @rte_update_pool and freed when rte_update()
1182 * finishes.
1183 */
1184
1185 void
1186 rte_update2(struct announce_hook *ah, net *net, rte *new, struct rte_src *src)
1187 {
1188 struct proto *p = ah->proto;
1189 struct proto_stats *stats = ah->stats;
1190 struct filter *filter = ah->in_filter;
1191 ea_list *tmpa = NULL;
1192 rte *dummy = NULL;
1193
1194 rte_update_lock();
1195 if (new)
1196 {
1197 new->sender = ah;
1198
1199 stats->imp_updates_received++;
1200 if (!rte_validate(new))
1201 {
1202 rte_trace_in(D_FILTERS, p, new, "invalid");
1203 stats->imp_updates_invalid++;
1204 goto drop;
1205 }
1206
1207 if (filter == FILTER_REJECT)
1208 {
1209 stats->imp_updates_filtered++;
1210 rte_trace_in(D_FILTERS, p, new, "filtered out");
1211
1212 if (! ah->in_keep_filtered)
1213 goto drop;
1214
1215 /* new is a private copy, i could modify it */
1216 new->flags |= REF_FILTERED;
1217 }
1218 else
1219 {
1220 tmpa = make_tmp_attrs(new, rte_update_pool);
1221 if (filter && (filter != FILTER_REJECT))
1222 {
1223 ea_list *old_tmpa = tmpa;
1224 int fr = f_run(filter, &new, &tmpa, rte_update_pool, 0);
1225 if (fr > F_ACCEPT)
1226 {
1227 stats->imp_updates_filtered++;
1228 rte_trace_in(D_FILTERS, p, new, "filtered out");
1229
1230 if (! ah->in_keep_filtered)
1231 goto drop;
1232
1233 new->flags |= REF_FILTERED;
1234 }
1235 if (tmpa != old_tmpa && src->proto->store_tmp_attrs)
1236 src->proto->store_tmp_attrs(new, tmpa);
1237 }
1238 }
1239 if (!rta_is_cached(new->attrs)) /* Need to copy attributes */
1240 new->attrs = rta_lookup(new->attrs);
1241 new->flags |= REF_COW;
1242 }
1243 else
1244 {
1245 stats->imp_withdraws_received++;
1246
1247 if (!net || !src)
1248 {
1249 stats->imp_withdraws_ignored++;
1250 rte_update_unlock();
1251 return;
1252 }
1253 }
1254
1255 recalc:
1256 rte_hide_dummy_routes(net, &dummy);
1257 rte_recalculate(ah, net, new, src);
1258 rte_unhide_dummy_routes(net, &dummy);
1259 rte_update_unlock();
1260 return;
1261
1262 drop:
1263 rte_free(new);
1264 new = NULL;
1265 goto recalc;
1266 }
1267
1268 /* Independent call to rte_announce(), used from next hop
1269 recalculation, outside of rte_update(). new must be non-NULL */
1270 static inline void
1271 rte_announce_i(rtable *tab, unsigned type, net *net, rte *new, rte *old,
1272 rte *new_best, rte *old_best)
1273 {
1274 rte_update_lock();
1275 rte_announce(tab, type, net, new, old, new_best, old_best, NULL);
1276 rte_update_unlock();
1277 }
1278
1279 void
1280 rte_discard(rtable *t, rte *old) /* Non-filtered route deletion, used during garbage collection */
1281 {
1282 rte_update_lock();
1283 rte_recalculate(old->sender, old->net, NULL, old->attrs->src);
1284 rte_update_unlock();
1285 }
1286
1287 /* Check rtable for best route to given net whether it would be exported do p */
1288 int
1289 rt_examine(rtable *t, ip_addr prefix, int pxlen, struct proto *p, struct filter *filter)
1290 {
1291 net *n = net_find(t, prefix, pxlen);
1292 rte *rt = n ? n->routes : NULL;
1293
1294 if (!rte_is_valid(rt))
1295 return 0;
1296
1297 rte_update_lock();
1298
1299 /* Rest is stripped down export_filter() */
1300 ea_list *tmpa = make_tmp_attrs(rt, rte_update_pool);
1301 int v = p->import_control ? p->import_control(p, &rt, &tmpa, rte_update_pool) : 0;
1302 if (v == RIC_PROCESS)
1303 v = (f_run(filter, &rt, &tmpa, rte_update_pool, FF_FORCE_TMPATTR) <= F_ACCEPT);
1304
1305 /* Discard temporary rte */
1306 if (rt != n->routes)
1307 rte_free(rt);
1308
1309 rte_update_unlock();
1310
1311 return v > 0;
1312 }
1313
1314
1315 /**
1316 * rt_refresh_begin - start a refresh cycle
1317 * @t: related routing table
1318 * @ah: related announce hook
1319 *
1320 * This function starts a refresh cycle for given routing table and announce
1321 * hook. The refresh cycle is a sequence where the protocol sends all its valid
1322 * routes to the routing table (by rte_update()). After that, all protocol
1323 * routes (more precisely routes with @ah as @sender) not sent during the
1324 * refresh cycle but still in the table from the past are pruned. This is
1325 * implemented by marking all related routes as stale by REF_STALE flag in
1326 * rt_refresh_begin(), then marking all related stale routes with REF_DISCARD
1327 * flag in rt_refresh_end() and then removing such routes in the prune loop.
1328 */
1329 void
1330 rt_refresh_begin(rtable *t, struct announce_hook *ah)
1331 {
1332 net *n;
1333 rte *e;
1334
1335 FIB_WALK(&t->fib, fn)
1336 {
1337 n = (net *) fn;
1338 for (e = n->routes; e; e = e->next)
1339 if (e->sender == ah)
1340 e->flags |= REF_STALE;
1341 }
1342 FIB_WALK_END;
1343 }
1344
1345 /**
1346 * rt_refresh_end - end a refresh cycle
1347 * @t: related routing table
1348 * @ah: related announce hook
1349 *
1350 * This function starts a refresh cycle for given routing table and announce
1351 * hook. See rt_refresh_begin() for description of refresh cycles.
1352 */
1353 void
1354 rt_refresh_end(rtable *t, struct announce_hook *ah)
1355 {
1356 int prune = 0;
1357 net *n;
1358 rte *e;
1359
1360 FIB_WALK(&t->fib, fn)
1361 {
1362 n = (net *) fn;
1363 for (e = n->routes; e; e = e->next)
1364 if ((e->sender == ah) && (e->flags & REF_STALE))
1365 {
1366 e->flags |= REF_DISCARD;
1367 prune = 1;
1368 }
1369 }
1370 FIB_WALK_END;
1371
1372 if (prune)
1373 rt_schedule_prune(t);
1374 }
1375
1376
1377 /**
1378 * rte_dump - dump a route
1379 * @e: &rte to be dumped
1380 *
1381 * This functions dumps contents of a &rte to debug output.
1382 */
1383 void
1384 rte_dump(rte *e)
1385 {
1386 net *n = e->net;
1387 debug("%-1I/%2d ", n->n.prefix, n->n.pxlen);
1388 debug("KF=%02x PF=%02x pref=%d lm=%d ", n->n.flags, e->pflags, e->pref, now-e->lastmod);
1389 rta_dump(e->attrs);
1390 if (e->attrs->src->proto->proto->dump_attrs)
1391 e->attrs->src->proto->proto->dump_attrs(e);
1392 debug("\n");
1393 }
1394
1395 /**
1396 * rt_dump - dump a routing table
1397 * @t: routing table to be dumped
1398 *
1399 * This function dumps contents of a given routing table to debug output.
1400 */
1401 void
1402 rt_dump(rtable *t)
1403 {
1404 rte *e;
1405 net *n;
1406 struct announce_hook *a;
1407
1408 debug("Dump of routing table <%s>\n", t->name);
1409 #ifdef DEBUGGING
1410 fib_check(&t->fib);
1411 #endif
1412 FIB_WALK(&t->fib, fn)
1413 {
1414 n = (net *) fn;
1415 for(e=n->routes; e; e=e->next)
1416 rte_dump(e);
1417 }
1418 FIB_WALK_END;
1419 WALK_LIST(a, t->hooks)
1420 debug("\tAnnounces routes to protocol %s\n", a->proto->name);
1421 debug("\n");
1422 }
1423
1424 /**
1425 * rt_dump_all - dump all routing tables
1426 *
1427 * This function dumps contents of all routing tables to debug output.
1428 */
1429 void
1430 rt_dump_all(void)
1431 {
1432 rtable *t;
1433
1434 WALK_LIST(t, routing_tables)
1435 rt_dump(t);
1436 }
1437
1438 static inline void
1439 rt_schedule_prune(rtable *tab)
1440 {
1441 rt_mark_for_prune(tab);
1442 ev_schedule(tab->rt_event);
1443 }
1444
1445 static inline void
1446 rt_schedule_gc(rtable *tab)
1447 {
1448 if (tab->gc_scheduled)
1449 return;
1450
1451 tab->gc_scheduled = 1;
1452 ev_schedule(tab->rt_event);
1453 }
1454
1455 static inline void
1456 rt_schedule_hcu(rtable *tab)
1457 {
1458 if (tab->hcu_scheduled)
1459 return;
1460
1461 tab->hcu_scheduled = 1;
1462 ev_schedule(tab->rt_event);
1463 }
1464
1465 static inline void
1466 rt_schedule_nhu(rtable *tab)
1467 {
1468 if (tab->nhu_state == 0)
1469 ev_schedule(tab->rt_event);
1470
1471 /* state change 0->1, 2->3 */
1472 tab->nhu_state |= 1;
1473 }
1474
1475
1476 static void
1477 rt_prune_nets(rtable *tab)
1478 {
1479 struct fib_iterator fit;
1480 int ncnt = 0, ndel = 0;
1481
1482 #ifdef DEBUGGING
1483 fib_check(&tab->fib);
1484 #endif
1485
1486 FIB_ITERATE_INIT(&fit, &tab->fib);
1487 again:
1488 FIB_ITERATE_START(&tab->fib, &fit, f)
1489 {
1490 net *n = (net *) f;
1491 ncnt++;
1492 if (!n->routes) /* Orphaned FIB entry */
1493 {
1494 FIB_ITERATE_PUT(&fit, f);
1495 fib_delete(&tab->fib, f);
1496 ndel++;
1497 goto again;
1498 }
1499 }
1500 FIB_ITERATE_END(f);
1501 DBG("Pruned %d of %d networks\n", ndel, ncnt);
1502
1503 tab->gc_counter = 0;
1504 tab->gc_time = now;
1505 tab->gc_scheduled = 0;
1506 }
1507
1508 static void
1509 rt_event(void *ptr)
1510 {
1511 rtable *tab = ptr;
1512
1513 if (tab->hcu_scheduled)
1514 rt_update_hostcache(tab);
1515
1516 if (tab->nhu_state)
1517 rt_next_hop_update(tab);
1518
1519 if (tab->prune_state)
1520 if (!rt_prune_table(tab))
1521 {
1522 /* Table prune unfinished */
1523 ev_schedule(tab->rt_event);
1524 return;
1525 }
1526
1527 if (tab->gc_scheduled)
1528 {
1529 rt_prune_nets(tab);
1530 rt_prune_sources(); // FIXME this should be moved to independent event
1531 }
1532 }
1533
1534 void
1535 rt_setup(pool *p, rtable *t, char *name, struct rtable_config *cf)
1536 {
1537 bzero(t, sizeof(*t));
1538 fib_init(&t->fib, p, sizeof(net), 0, rte_init);
1539 t->name = name;
1540 t->config = cf;
1541 init_list(&t->hooks);
1542 if (cf)
1543 {
1544 t->rt_event = ev_new(p);
1545 t->rt_event->hook = rt_event;
1546 t->rt_event->data = t;
1547 t->gc_time = now;
1548 }
1549 }
1550
1551 /**
1552 * rt_init - initialize routing tables
1553 *
1554 * This function is called during BIRD startup. It initializes the
1555 * routing table module.
1556 */
1557 void
1558 rt_init(void)
1559 {
1560 rta_init();
1561 rt_table_pool = rp_new(&root_pool, "Routing tables");
1562 rte_update_pool = lp_new(rt_table_pool, 4080);
1563 rte_slab = sl_new(rt_table_pool, sizeof(rte));
1564 init_list(&routing_tables);
1565 }
1566
1567
1568 static int
1569 rt_prune_step(rtable *tab, int *limit)
1570 {
1571 struct fib_iterator *fit = &tab->prune_fit;
1572
1573 DBG("Pruning route table %s\n", tab->name);
1574 #ifdef DEBUGGING
1575 fib_check(&tab->fib);
1576 #endif
1577
1578 if (tab->prune_state == RPS_NONE)
1579 return 1;
1580
1581 if (tab->prune_state == RPS_SCHEDULED)
1582 {
1583 FIB_ITERATE_INIT(fit, &tab->fib);
1584 tab->prune_state = RPS_RUNNING;
1585 }
1586
1587 again:
1588 FIB_ITERATE_START(&tab->fib, fit, fn)
1589 {
1590 net *n = (net *) fn;
1591 rte *e;
1592
1593 rescan:
1594 for (e=n->routes; e; e=e->next)
1595 if (e->sender->proto->flushing || (e->flags & REF_DISCARD))
1596 {
1597 if (*limit <= 0)
1598 {
1599 FIB_ITERATE_PUT(fit, fn);
1600 return 0;
1601 }
1602
1603 rte_discard(tab, e);
1604 (*limit)--;
1605
1606 goto rescan;
1607 }
1608 if (!n->routes) /* Orphaned FIB entry */
1609 {
1610 FIB_ITERATE_PUT(fit, fn);
1611 fib_delete(&tab->fib, fn);
1612 goto again;
1613 }
1614 }
1615 FIB_ITERATE_END(fn);
1616
1617 #ifdef DEBUGGING
1618 fib_check(&tab->fib);
1619 #endif
1620
1621 tab->prune_state = RPS_NONE;
1622 return 1;
1623 }
1624
1625 /**
1626 * rt_prune_table - prune a routing table
1627 * @tab: a routing table for pruning
1628 *
1629 * This function scans the routing table @tab and removes routes belonging to
1630 * flushing protocols, discarded routes and also stale network entries, in a
1631 * similar fashion like rt_prune_loop(). Returns 1 when all such routes are
1632 * pruned. Contrary to rt_prune_loop(), this function is not a part of the
1633 * protocol flushing loop, but it is called from rt_event() for just one routing
1634 * table.
1635 *
1636 * Note that rt_prune_table() and rt_prune_loop() share (for each table) the
1637 * prune state (@prune_state) and also the pruning iterator (@prune_fit).
1638 */
1639 static inline int
1640 rt_prune_table(rtable *tab)
1641 {
1642 int limit = 512;
1643 return rt_prune_step(tab, &limit);
1644 }
1645
1646 /**
1647 * rt_prune_loop - prune routing tables
1648 *
1649 * The prune loop scans routing tables and removes routes belonging to flushing
1650 * protocols, discarded routes and also stale network entries. Returns 1 when
1651 * all such routes are pruned. It is a part of the protocol flushing loop.
1652 */
1653 int
1654 rt_prune_loop(void)
1655 {
1656 int limit = 512;
1657 rtable *t;
1658
1659 WALK_LIST(t, routing_tables)
1660 if (! rt_prune_step(t, &limit))
1661 return 0;
1662
1663 return 1;
1664 }
1665
1666 void
1667 rt_preconfig(struct config *c)
1668 {
1669 struct symbol *s = cf_get_symbol("master");
1670
1671 init_list(&c->tables);
1672 c->master_rtc = rt_new_table(s);
1673 }
1674
1675
1676 /*
1677 * Some functions for handing internal next hop updates
1678 * triggered by rt_schedule_nhu().
1679 */
1680
1681 static inline int
1682 rta_next_hop_outdated(rta *a)
1683 {
1684 struct hostentry *he = a->hostentry;
1685
1686 if (!he)
1687 return 0;
1688
1689 if (!he->src)
1690 return a->dest != RTD_UNREACHABLE;
1691
1692 return (a->iface != he->src->iface) || !ipa_equal(a->gw, he->gw) ||
1693 (a->dest != he->dest) || (a->igp_metric != he->igp_metric) ||
1694 !mpnh_same(a->nexthops, he->src->nexthops);
1695 }
1696
1697 static inline void
1698 rta_apply_hostentry(rta *a, struct hostentry *he)
1699 {
1700 a->hostentry = he;
1701 a->iface = he->src ? he->src->iface : NULL;
1702 a->gw = he->gw;
1703 a->dest = he->dest;
1704 a->igp_metric = he->igp_metric;
1705 a->nexthops = he->src ? he->src->nexthops : NULL;
1706 }
1707
1708 static inline rte *
1709 rt_next_hop_update_rte(rtable *tab, rte *old)
1710 {
1711 rta a;
1712 memcpy(&a, old->attrs, sizeof(rta));
1713 rta_apply_hostentry(&a, old->attrs->hostentry);
1714 a.aflags = 0;
1715
1716 rte *e = sl_alloc(rte_slab);
1717 memcpy(e, old, sizeof(rte));
1718 e->attrs = rta_lookup(&a);
1719
1720 return e;
1721 }
1722
1723 static inline int
1724 rt_next_hop_update_net(rtable *tab, net *n)
1725 {
1726 rte **k, *e, *new, *old_best, **new_best;
1727 int count = 0;
1728 int free_old_best = 0;
1729
1730 old_best = n->routes;
1731 if (!old_best)
1732 return 0;
1733
1734 for (k = &n->routes; e = *k; k = &e->next)
1735 if (rta_next_hop_outdated(e->attrs))
1736 {
1737 new = rt_next_hop_update_rte(tab, e);
1738 *k = new;
1739
1740 rte_announce_i(tab, RA_ANY, n, new, e, NULL, NULL);
1741 rte_trace_in(D_ROUTES, new->sender->proto, new, "updated");
1742
1743 /* Call a pre-comparison hook */
1744 /* Not really an efficient way to compute this */
1745 if (e->attrs->src->proto->rte_recalculate)
1746 e->attrs->src->proto->rte_recalculate(tab, n, new, e, NULL);
1747
1748 if (e != old_best)
1749 rte_free_quick(e);
1750 else /* Freeing of the old best rte is postponed */
1751 free_old_best = 1;
1752
1753 e = new;
1754 count++;
1755 }
1756
1757 if (!count)
1758 return 0;
1759
1760 /* Find the new best route */
1761 new_best = NULL;
1762 for (k = &n->routes; e = *k; k = &e->next)
1763 {
1764 if (!new_best || rte_better(e, *new_best))
1765 new_best = k;
1766 }
1767
1768 /* Relink the new best route to the first position */
1769 new = *new_best;
1770 if (new != n->routes)
1771 {
1772 *new_best = new->next;
1773 new->next = n->routes;
1774 n->routes = new;
1775 }
1776
1777 /* Announce the new best route */
1778 if (new != old_best)
1779 {
1780 rte_announce_i(tab, RA_OPTIMAL, n, new, old_best, NULL, NULL);
1781 rte_trace_in(D_ROUTES, new->sender->proto, new, "updated [best]");
1782 }
1783
1784 /* FIXME: Better announcement of merged routes */
1785 rte_announce_i(tab, RA_MERGED, n, new, old_best, new, old_best);
1786
1787 if (free_old_best)
1788 rte_free_quick(old_best);
1789
1790 return count;
1791 }
1792
1793 static void
1794 rt_next_hop_update(rtable *tab)
1795 {
1796 struct fib_iterator *fit = &tab->nhu_fit;
1797 int max_feed = 32;
1798
1799 if (tab->nhu_state == 0)
1800 return;
1801
1802 if (tab->nhu_state == 1)
1803 {
1804 FIB_ITERATE_INIT(fit, &tab->fib);
1805 tab->nhu_state = 2;
1806 }
1807
1808 FIB_ITERATE_START(&tab->fib, fit, fn)
1809 {
1810 if (max_feed <= 0)
1811 {
1812 FIB_ITERATE_PUT(fit, fn);
1813 ev_schedule(tab->rt_event);
1814 return;
1815 }
1816 max_feed -= rt_next_hop_update_net(tab, (net *) fn);
1817 }
1818 FIB_ITERATE_END(fn);
1819
1820 /* state change 2->0, 3->1 */
1821 tab->nhu_state &= 1;
1822
1823 if (tab->nhu_state > 0)
1824 ev_schedule(tab->rt_event);
1825 }
1826
1827
1828 struct rtable_config *
1829 rt_new_table(struct symbol *s)
1830 {
1831 /* Hack that allows to 'redefine' the master table */
1832 if ((s->class == SYM_TABLE) && (s->def == new_config->master_rtc))
1833 return s->def;
1834
1835 struct rtable_config *c = cfg_allocz(sizeof(struct rtable_config));
1836
1837 cf_define_symbol(s, SYM_TABLE, c);
1838 c->name = s->name;
1839 add_tail(&new_config->tables, &c->n);
1840 c->gc_max_ops = 1000;
1841 c->gc_min_time = 5;
1842 return c;
1843 }
1844
1845 /**
1846 * rt_lock_table - lock a routing table
1847 * @r: routing table to be locked
1848 *
1849 * Lock a routing table, because it's in use by a protocol,
1850 * preventing it from being freed when it gets undefined in a new
1851 * configuration.
1852 */
1853 void
1854 rt_lock_table(rtable *r)
1855 {
1856 r->use_count++;
1857 }
1858
1859 /**
1860 * rt_unlock_table - unlock a routing table
1861 * @r: routing table to be unlocked
1862 *
1863 * Unlock a routing table formerly locked by rt_lock_table(),
1864 * that is decrease its use count and delete it if it's scheduled
1865 * for deletion by configuration changes.
1866 */
1867 void
1868 rt_unlock_table(rtable *r)
1869 {
1870 if (!--r->use_count && r->deleted)
1871 {
1872 struct config *conf = r->deleted;
1873 DBG("Deleting routing table %s\n", r->name);
1874 r->config->table = NULL;
1875 if (r->hostcache)
1876 rt_free_hostcache(r);
1877 rem_node(&r->n);
1878 fib_free(&r->fib);
1879 rfree(r->rt_event);
1880 mb_free(r);
1881 config_del_obstacle(conf);
1882 }
1883 }
1884
1885 /**
1886 * rt_commit - commit new routing table configuration
1887 * @new: new configuration
1888 * @old: original configuration or %NULL if it's boot time config
1889 *
1890 * Scan differences between @old and @new configuration and modify
1891 * the routing tables according to these changes. If @new defines a
1892 * previously unknown table, create it, if it omits a table existing
1893 * in @old, schedule it for deletion (it gets deleted when all protocols
1894 * disconnect from it by calling rt_unlock_table()), if it exists
1895 * in both configurations, leave it unchanged.
1896 */
1897 void
1898 rt_commit(struct config *new, struct config *old)
1899 {
1900 struct rtable_config *o, *r;
1901
1902 DBG("rt_commit:\n");
1903 if (old)
1904 {
1905 WALK_LIST(o, old->tables)
1906 {
1907 rtable *ot = o->table;
1908 if (!ot->deleted)
1909 {
1910 struct symbol *sym = cf_find_symbol(new, o->name);
1911 if (sym && sym->class == SYM_TABLE && !new->shutdown)
1912 {
1913 DBG("\t%s: same\n", o->name);
1914 r = sym->def;
1915 r->table = ot;
1916 ot->name = r->name;
1917 ot->config = r;
1918 if (o->sorted != r->sorted)
1919 log(L_WARN "Reconfiguration of rtable sorted flag not implemented");
1920 }
1921 else
1922 {
1923 DBG("\t%s: deleted\n", o->name);
1924 ot->deleted = old;
1925 config_add_obstacle(old);
1926 rt_lock_table(ot);
1927 rt_unlock_table(ot);
1928 }
1929 }
1930 }
1931 }
1932
1933 WALK_LIST(r, new->tables)
1934 if (!r->table)
1935 {
1936 rtable *t = mb_alloc(rt_table_pool, sizeof(struct rtable));
1937 DBG("\t%s: created\n", r->name);
1938 rt_setup(rt_table_pool, t, r->name, r);
1939 add_tail(&routing_tables, &t->n);
1940 r->table = t;
1941 }
1942 DBG("\tdone\n");
1943 }
1944
1945 static inline void
1946 do_feed_baby(struct proto *p, int type, struct announce_hook *h, net *n, rte *e)
1947 {
1948 rte_update_lock();
1949 if (type == RA_ACCEPTED)
1950 rt_notify_accepted(h, n, e, NULL, NULL, p->refeeding ? 2 : 1);
1951 else if (type == RA_MERGED)
1952 rt_notify_merged(h, n, NULL, NULL, e, p->refeeding ? e : NULL, p->refeeding);
1953 else
1954 rt_notify_basic(h, n, e, p->refeeding ? e : NULL, p->refeeding);
1955 rte_update_unlock();
1956 }
1957
1958 /**
1959 * rt_feed_baby - advertise routes to a new protocol
1960 * @p: protocol to be fed
1961 *
1962 * This function performs one pass of advertisement of routes to a newly
1963 * initialized protocol. It's called by the protocol code as long as it
1964 * has something to do. (We avoid transferring all the routes in single
1965 * pass in order not to monopolize CPU time.)
1966 */
1967 int
1968 rt_feed_baby(struct proto *p)
1969 {
1970 struct announce_hook *h;
1971 struct fib_iterator *fit;
1972 int max_feed = 256;
1973
1974 if (!p->feed_ahook) /* Need to initialize first */
1975 {
1976 if (!p->ahooks)
1977 return 1;
1978 DBG("Announcing routes to new protocol %s\n", p->name);
1979 p->feed_ahook = p->ahooks;
1980 fit = p->feed_iterator = mb_alloc(p->pool, sizeof(struct fib_iterator));
1981 goto next_hook;
1982 }
1983 fit = p->feed_iterator;
1984
1985 again:
1986 h = p->feed_ahook;
1987 FIB_ITERATE_START(&h->table->fib, fit, fn)
1988 {
1989 net *n = (net *) fn;
1990 rte *e = n->routes;
1991 if (max_feed <= 0)
1992 {
1993 FIB_ITERATE_PUT(fit, fn);
1994 return 0;
1995 }
1996
1997 /* XXXX perhaps we should change feed for RA_ACCEPTED to not use 'new' */
1998
1999 if ((p->accept_ra_types == RA_OPTIMAL) ||
2000 (p->accept_ra_types == RA_ACCEPTED) ||
2001 (p->accept_ra_types == RA_MERGED))
2002 if (rte_is_valid(e))
2003 {
2004 if (p->export_state != ES_FEEDING)
2005 return 1; /* In the meantime, the protocol fell down. */
2006
2007 do_feed_baby(p, p->accept_ra_types, h, n, e);
2008 max_feed--;
2009 }
2010
2011 if (p->accept_ra_types == RA_ANY)
2012 for(e = n->routes; e; e = e->next)
2013 {
2014 if (p->export_state != ES_FEEDING)
2015 return 1; /* In the meantime, the protocol fell down. */
2016
2017 if (!rte_is_valid(e))
2018 continue;
2019
2020 do_feed_baby(p, RA_ANY, h, n, e);
2021 max_feed--;
2022 }
2023 }
2024 FIB_ITERATE_END(fn);
2025 p->feed_ahook = h->next;
2026 if (!p->feed_ahook)
2027 {
2028 mb_free(p->feed_iterator);
2029 p->feed_iterator = NULL;
2030 return 1;
2031 }
2032
2033 next_hook:
2034 h = p->feed_ahook;
2035 FIB_ITERATE_INIT(fit, &h->table->fib);
2036 goto again;
2037 }
2038
2039 /**
2040 * rt_feed_baby_abort - abort protocol feeding
2041 * @p: protocol
2042 *
2043 * This function is called by the protocol code when the protocol
2044 * stops or ceases to exist before the last iteration of rt_feed_baby()
2045 * has finished.
2046 */
2047 void
2048 rt_feed_baby_abort(struct proto *p)
2049 {
2050 if (p->feed_ahook)
2051 {
2052 /* Unlink the iterator and exit */
2053 fit_get(&p->feed_ahook->table->fib, p->feed_iterator);
2054 p->feed_ahook = NULL;
2055 }
2056 }
2057
2058
2059 static inline unsigned
2060 ptr_hash(void *ptr)
2061 {
2062 uintptr_t p = (uintptr_t) ptr;
2063 return p ^ (p << 8) ^ (p >> 16);
2064 }
2065
2066 static inline unsigned
2067 hc_hash(ip_addr a, rtable *dep)
2068 {
2069 return (ipa_hash(a) ^ ptr_hash(dep)) & 0xffff;
2070 }
2071
2072 static inline void
2073 hc_insert(struct hostcache *hc, struct hostentry *he)
2074 {
2075 uint k = he->hash_key >> hc->hash_shift;
2076 he->next = hc->hash_table[k];
2077 hc->hash_table[k] = he;
2078 }
2079
2080 static inline void
2081 hc_remove(struct hostcache *hc, struct hostentry *he)
2082 {
2083 struct hostentry **hep;
2084 uint k = he->hash_key >> hc->hash_shift;
2085
2086 for (hep = &hc->hash_table[k]; *hep != he; hep = &(*hep)->next);
2087 *hep = he->next;
2088 }
2089
2090 #define HC_DEF_ORDER 10
2091 #define HC_HI_MARK *4
2092 #define HC_HI_STEP 2
2093 #define HC_HI_ORDER 16 /* Must be at most 16 */
2094 #define HC_LO_MARK /5
2095 #define HC_LO_STEP 2
2096 #define HC_LO_ORDER 10
2097
2098 static void
2099 hc_alloc_table(struct hostcache *hc, unsigned order)
2100 {
2101 unsigned hsize = 1 << order;
2102 hc->hash_order = order;
2103 hc->hash_shift = 16 - order;
2104 hc->hash_max = (order >= HC_HI_ORDER) ? ~0 : (hsize HC_HI_MARK);
2105 hc->hash_min = (order <= HC_LO_ORDER) ? 0 : (hsize HC_LO_MARK);
2106
2107 hc->hash_table = mb_allocz(rt_table_pool, hsize * sizeof(struct hostentry *));
2108 }
2109
2110 static void
2111 hc_resize(struct hostcache *hc, unsigned new_order)
2112 {
2113 unsigned old_size = 1 << hc->hash_order;
2114 struct hostentry **old_table = hc->hash_table;
2115 struct hostentry *he, *hen;
2116 int i;
2117
2118 hc_alloc_table(hc, new_order);
2119 for (i = 0; i < old_size; i++)
2120 for (he = old_table[i]; he != NULL; he=hen)
2121 {
2122 hen = he->next;
2123 hc_insert(hc, he);
2124 }
2125 mb_free(old_table);
2126 }
2127
2128 static struct hostentry *
2129 hc_new_hostentry(struct hostcache *hc, ip_addr a, ip_addr ll, rtable *dep, unsigned k)
2130 {
2131 struct hostentry *he = sl_alloc(hc->slab);
2132
2133 he->addr = a;
2134 he->link = ll;
2135 he->tab = dep;
2136 he->hash_key = k;
2137 he->uc = 0;
2138 he->src = NULL;
2139
2140 add_tail(&hc->hostentries, &he->ln);
2141 hc_insert(hc, he);
2142
2143 hc->hash_items++;
2144 if (hc->hash_items > hc->hash_max)
2145 hc_resize(hc, hc->hash_order + HC_HI_STEP);
2146
2147 return he;
2148 }
2149
2150 static void
2151 hc_delete_hostentry(struct hostcache *hc, struct hostentry *he)
2152 {
2153 rta_free(he->src);
2154
2155 rem_node(&he->ln);
2156 hc_remove(hc, he);
2157 sl_free(hc->slab, he);
2158
2159 hc->hash_items--;
2160 if (hc->hash_items < hc->hash_min)
2161 hc_resize(hc, hc->hash_order - HC_LO_STEP);
2162 }
2163
2164 static void
2165 rt_init_hostcache(rtable *tab)
2166 {
2167 struct hostcache *hc = mb_allocz(rt_table_pool, sizeof(struct hostcache));
2168 init_list(&hc->hostentries);
2169
2170 hc->hash_items = 0;
2171 hc_alloc_table(hc, HC_DEF_ORDER);
2172 hc->slab = sl_new(rt_table_pool, sizeof(struct hostentry));
2173
2174 hc->lp = lp_new(rt_table_pool, 1008);
2175 hc->trie = f_new_trie(hc->lp, sizeof(struct f_trie_node));
2176
2177 tab->hostcache = hc;
2178 }
2179
2180 static void
2181 rt_free_hostcache(rtable *tab)
2182 {
2183 struct hostcache *hc = tab->hostcache;
2184
2185 node *n;
2186 WALK_LIST(n, hc->hostentries)
2187 {
2188 struct hostentry *he = SKIP_BACK(struct hostentry, ln, n);
2189 rta_free(he->src);
2190
2191 if (he->uc)
2192 log(L_ERR "Hostcache is not empty in table %s", tab->name);
2193 }
2194
2195 rfree(hc->slab);
2196 rfree(hc->lp);
2197 mb_free(hc->hash_table);
2198 mb_free(hc);
2199 }
2200
2201 static void
2202 rt_notify_hostcache(rtable *tab, net *net)
2203 {
2204 struct hostcache *hc = tab->hostcache;
2205
2206 if (tab->hcu_scheduled)
2207 return;
2208
2209 if (trie_match_prefix(hc->trie, net->n.prefix, net->n.pxlen))
2210 rt_schedule_hcu(tab);
2211 }
2212
2213 static int
2214 if_local_addr(ip_addr a, struct iface *i)
2215 {
2216 struct ifa *b;
2217
2218 WALK_LIST(b, i->addrs)
2219 if (ipa_equal(a, b->ip))
2220 return 1;
2221
2222 return 0;
2223 }
2224
2225 static u32
2226 rt_get_igp_metric(rte *rt)
2227 {
2228 eattr *ea = ea_find(rt->attrs->eattrs, EA_GEN_IGP_METRIC);
2229
2230 if (ea)
2231 return ea->u.data;
2232
2233 rta *a = rt->attrs;
2234
2235 #ifdef CONFIG_OSPF
2236 if ((a->source == RTS_OSPF) ||
2237 (a->source == RTS_OSPF_IA) ||
2238 (a->source == RTS_OSPF_EXT1))
2239 return rt->u.ospf.metric1;
2240 #endif
2241
2242 #ifdef CONFIG_RIP
2243 if (a->source == RTS_RIP)
2244 return rt->u.rip.metric;
2245 #endif
2246
2247 /* Device routes */
2248 if ((a->dest != RTD_ROUTER) && (a->dest != RTD_MULTIPATH))
2249 return 0;
2250
2251 return IGP_METRIC_UNKNOWN;
2252 }
2253
2254 static int
2255 rt_update_hostentry(rtable *tab, struct hostentry *he)
2256 {
2257 rta *old_src = he->src;
2258 int pxlen = 0;
2259
2260 /* Reset the hostentry */
2261 he->src = NULL;
2262 he->gw = IPA_NONE;
2263 he->dest = RTD_UNREACHABLE;
2264 he->igp_metric = 0;
2265
2266 net *n = net_route(tab, he->addr, MAX_PREFIX_LENGTH);
2267 if (n)
2268 {
2269 rte *e = n->routes;
2270 rta *a = e->attrs;
2271 pxlen = n->n.pxlen;
2272
2273 if (a->hostentry)
2274 {
2275 /* Recursive route should not depend on another recursive route */
2276 log(L_WARN "Next hop address %I resolvable through recursive route for %I/%d",
2277 he->addr, n->n.prefix, pxlen);
2278 goto done;
2279 }
2280
2281 if (a->dest == RTD_DEVICE)
2282 {
2283 if (if_local_addr(he->addr, a->iface))
2284 {
2285 /* The host address is a local address, this is not valid */
2286 log(L_WARN "Next hop address %I is a local address of iface %s",
2287 he->addr, a->iface->name);
2288 goto done;
2289 }
2290
2291 /* The host is directly reachable, use link as a gateway */
2292 he->gw = he->link;
2293 he->dest = RTD_ROUTER;
2294 }
2295 else
2296 {
2297 /* The host is reachable through some route entry */
2298 he->gw = a->gw;
2299 he->dest = a->dest;
2300 }
2301
2302 he->src = rta_clone(a);
2303 he->igp_metric = rt_get_igp_metric(e);
2304 }
2305
2306 done:
2307 /* Add a prefix range to the trie */
2308 trie_add_prefix(tab->hostcache->trie, he->addr, MAX_PREFIX_LENGTH, pxlen, MAX_PREFIX_LENGTH);
2309
2310 rta_free(old_src);
2311 return old_src != he->src;
2312 }
2313
2314 static void
2315 rt_update_hostcache(rtable *tab)
2316 {
2317 struct hostcache *hc = tab->hostcache;
2318 struct hostentry *he;
2319 node *n, *x;
2320
2321 /* Reset the trie */
2322 lp_flush(hc->lp);
2323 hc->trie = f_new_trie(hc->lp, sizeof(struct f_trie_node));
2324
2325 WALK_LIST_DELSAFE(n, x, hc->hostentries)
2326 {
2327 he = SKIP_BACK(struct hostentry, ln, n);
2328 if (!he->uc)
2329 {
2330 hc_delete_hostentry(hc, he);
2331 continue;
2332 }
2333
2334 if (rt_update_hostentry(tab, he))
2335 rt_schedule_nhu(he->tab);
2336 }
2337
2338 tab->hcu_scheduled = 0;
2339 }
2340
2341 static struct hostentry *
2342 rt_get_hostentry(rtable *tab, ip_addr a, ip_addr ll, rtable *dep)
2343 {
2344 struct hostentry *he;
2345
2346 if (!tab->hostcache)
2347 rt_init_hostcache(tab);
2348
2349 uint k = hc_hash(a, dep);
2350 struct hostcache *hc = tab->hostcache;
2351 for (he = hc->hash_table[k >> hc->hash_shift]; he != NULL; he = he->next)
2352 if (ipa_equal(he->addr, a) && (he->tab == dep))
2353 return he;
2354
2355 he = hc_new_hostentry(hc, a, ll, dep, k);
2356 rt_update_hostentry(tab, he);
2357 return he;
2358 }
2359
2360 void
2361 rta_set_recursive_next_hop(rtable *dep, rta *a, rtable *tab, ip_addr *gw, ip_addr *ll)
2362 {
2363 rta_apply_hostentry(a, rt_get_hostentry(tab, *gw, *ll, dep));
2364 }
2365
2366
2367 /*
2368 * CLI commands
2369 */
2370
2371 static byte *
2372 rt_format_via(rte *e)
2373 {
2374 rta *a = e->attrs;
2375
2376 /* Max text length w/o IP addr and interface name is 16 */
2377 static byte via[STD_ADDRESS_P_LENGTH+sizeof(a->iface->name)+16];
2378
2379 switch (a->dest)
2380 {
2381 case RTD_ROUTER: bsprintf(via, "via %I on %s", a->gw, a->iface->name); break;
2382 case RTD_DEVICE: bsprintf(via, "dev %s", a->iface->name); break;
2383 case RTD_BLACKHOLE: bsprintf(via, "blackhole"); break;
2384 case RTD_UNREACHABLE: bsprintf(via, "unreachable"); break;
2385 case RTD_PROHIBIT: bsprintf(via, "prohibited"); break;
2386 case RTD_MULTIPATH: bsprintf(via, "multipath"); break;
2387 default: bsprintf(via, "???");
2388 }
2389 return via;
2390 }
2391
2392 static void
2393 rt_show_rte(struct cli *c, byte *ia, rte *e, struct rt_show_data *d, ea_list *tmpa)
2394 {
2395 byte from[STD_ADDRESS_P_LENGTH+8];
2396 byte tm[TM_DATETIME_BUFFER_SIZE], info[256];
2397 rta *a = e->attrs;
2398 int primary = (e->net->routes == e);
2399 int sync_error = (e->net->n.flags & KRF_SYNC_ERROR);
2400 void (*get_route_info)(struct rte *, byte *buf, struct ea_list *attrs);
2401 struct mpnh *nh;
2402
2403 tm_format_datetime(tm, &config->tf_route, e->lastmod);
2404 if (ipa_nonzero(a->from) && !ipa_equal(a->from, a->gw))
2405 bsprintf(from, " from %I", a->from);
2406 else
2407 from[0] = 0;
2408
2409 get_route_info = a->src->proto->proto->get_route_info;
2410 if (get_route_info || d->verbose)
2411 {
2412 /* Need to normalize the extended attributes */
2413 ea_list *t = tmpa;
2414 t = ea_append(t, a->eattrs);
2415 tmpa = alloca(ea_scan(t));
2416 ea_merge(t, tmpa);
2417 ea_sort(tmpa);
2418 }
2419 if (get_route_info)
2420 get_route_info(e, info, tmpa);
2421 else
2422 bsprintf(info, " (%d)", e->pref);
2423 cli_printf(c, -1007, "%-18s %s [%s %s%s]%s%s", ia, rt_format_via(e), a->src->proto->name,
2424 tm, from, primary ? (sync_error ? " !" : " *") : "", info);
2425 for (nh = a->nexthops; nh; nh = nh->next)
2426 cli_printf(c, -1007, "\tvia %I on %s weight %d", nh->gw, nh->iface->name, nh->weight + 1);
2427 if (d->verbose)
2428 rta_show(c, a, tmpa);
2429 }
2430
2431 static void
2432 rt_show_net(struct cli *c, net *n, struct rt_show_data *d)
2433 {
2434 rte *e, *ee;
2435 byte ia[STD_ADDRESS_P_LENGTH+8];
2436 struct ea_list *tmpa;
2437 struct announce_hook *a = NULL;
2438 int first = 1;
2439 int pass = 0;
2440
2441 bsprintf(ia, "%I/%d", n->n.prefix, n->n.pxlen);
2442
2443 if (d->export_mode)
2444 {
2445 if (! d->export_protocol->rt_notify)
2446 return;
2447
2448 a = proto_find_announce_hook(d->export_protocol, d->table);
2449 if (!a)
2450 return;
2451 }
2452
2453 for (e = n->routes; e; e = e->next)
2454 {
2455 if (rte_is_filtered(e) != d->filtered)
2456 continue;
2457
2458 d->rt_counter++;
2459 d->net_counter += first;
2460 first = 0;
2461
2462 if (pass)
2463 continue;
2464
2465 ee = e;
2466 rte_update_lock(); /* We use the update buffer for filtering */
2467 tmpa = make_tmp_attrs(e, rte_update_pool);
2468
2469 /* Special case for merged export */
2470 if ((d->export_mode == RSEM_EXPORT) && (d->export_protocol->accept_ra_types == RA_MERGED))
2471 {
2472 rte *rt_free;
2473 e = rt_export_merged(a, n, &rt_free, &tmpa, 1);
2474 pass = 1;
2475
2476 if (!e)
2477 { e = ee; goto skip; }
2478 }
2479 else if (d->export_mode)
2480 {
2481 struct proto *ep = d->export_protocol;
2482 int ic = ep->import_control ? ep->import_control(ep, &e, &tmpa, rte_update_pool) : 0;
2483
2484 if (ep->accept_ra_types == RA_OPTIMAL || ep->accept_ra_types == RA_MERGED)
2485 pass = 1;
2486
2487 if (ic < 0)
2488 goto skip;
2489
2490 if (d->export_mode > RSEM_PREEXPORT)
2491 {
2492 /*
2493 * FIXME - This shows what should be exported according to current
2494 * filters, but not what was really exported. 'configure soft'
2495 * command may change the export filter and do not update routes.
2496 */
2497 int do_export = (ic > 0) ||
2498 (f_run(a->out_filter, &e, &tmpa, rte_update_pool, FF_FORCE_TMPATTR) <= F_ACCEPT);
2499
2500 if (do_export != (d->export_mode == RSEM_EXPORT))
2501 goto skip;
2502
2503 if ((d->export_mode == RSEM_EXPORT) && (ep->accept_ra_types == RA_ACCEPTED))
2504 pass = 1;
2505 }
2506 }
2507
2508 if (d->show_protocol && (d->show_protocol != e->attrs->src->proto))
2509 goto skip;
2510
2511 if (f_run(d->filter, &e, &tmpa, rte_update_pool, FF_FORCE_TMPATTR) > F_ACCEPT)
2512 goto skip;
2513
2514 d->show_counter++;
2515 if (d->stats < 2)
2516 rt_show_rte(c, ia, e, d, tmpa);
2517 ia[0] = 0;
2518
2519 skip:
2520 if (e != ee)
2521 {
2522 rte_free(e);
2523 e = ee;
2524 }
2525 rte_update_unlock();
2526
2527 if (d->primary_only)
2528 break;
2529 }
2530 }
2531
2532 static void
2533 rt_show_cont(struct cli *c)
2534 {
2535 struct rt_show_data *d = c->rover;
2536 #ifdef DEBUGGING
2537 unsigned max = 4;
2538 #else
2539 unsigned max = 64;
2540 #endif
2541 struct fib *fib = &d->table->fib;
2542 struct fib_iterator *it = &d->fit;
2543
2544 FIB_ITERATE_START(fib, it, f)
2545 {
2546 net *n = (net *) f;
2547 if (d->running_on_config && d->running_on_config != config)
2548 {
2549 cli_printf(c, 8004, "Stopped due to reconfiguration");
2550 goto done;
2551 }
2552 if (d->export_protocol && (d->export_protocol->export_state == ES_DOWN))
2553 {
2554 cli_printf(c, 8005, "Protocol is down");
2555 goto done;
2556 }
2557 if (!max--)
2558 {
2559 FIB_ITERATE_PUT(it, f);
2560 return;
2561 }
2562 rt_show_net(c, n, d);
2563 }
2564 FIB_ITERATE_END(f);
2565 if (d->stats)
2566 cli_printf(c, 14, "%d of %d routes for %d networks", d->show_counter, d->rt_counter, d->net_counter);
2567 else
2568 cli_printf(c, 0, "");
2569 done:
2570 c->cont = c->cleanup = NULL;
2571 }
2572
2573 static void
2574 rt_show_cleanup(struct cli *c)
2575 {
2576 struct rt_show_data *d = c->rover;
2577
2578 /* Unlink the iterator */
2579 fit_get(&d->table->fib, &d->fit);
2580 }
2581
2582 void
2583 rt_show(struct rt_show_data *d)
2584 {
2585 net *n;
2586
2587 /* Default is either a master table or a table related to a respective protocol */
2588 if (!d->table && d->export_protocol) d->table = d->export_protocol->table;
2589 if (!d->table && d->show_protocol) d->table = d->show_protocol->table;
2590 if (!d->table) d->table = config->master_rtc->table;
2591
2592 /* Filtered routes are neither exported nor have sensible ordering */
2593 if (d->filtered && (d->export_mode || d->primary_only))
2594 cli_msg(0, "");
2595
2596 if (d->pxlen == 256)
2597 {
2598 FIB_ITERATE_INIT(&d->fit, &d->table->fib);
2599 this_cli->cont = rt_show_cont;
2600 this_cli->cleanup = rt_show_cleanup;
2601 this_cli->rover = d;
2602 }
2603 else
2604 {
2605 if (d->show_for)
2606 n = net_route(d->table, d->prefix, d->pxlen);
2607 else
2608 n = net_find(d->table, d->prefix, d->pxlen);
2609
2610 if (n)
2611 rt_show_net(this_cli, n, d);
2612
2613 if (d->rt_counter)
2614 cli_msg(0, "");
2615 else
2616 cli_msg(8001, "Network not in table");
2617 }
2618 }
2619
2620 /*
2621 * Documentation for functions declared inline in route.h
2622 */
2623 #if 0
2624
2625 /**
2626 * net_find - find a network entry
2627 * @tab: a routing table
2628 * @addr: address of the network
2629 * @len: length of the network prefix
2630 *
2631 * net_find() looks up the given network in routing table @tab and
2632 * returns a pointer to its &net entry or %NULL if no such network
2633 * exists.
2634 */
2635 static inline net *net_find(rtable *tab, ip_addr addr, unsigned len)
2636 { DUMMY; }
2637
2638 /**
2639 * net_get - obtain a network entry
2640 * @tab: a routing table
2641 * @addr: address of the network
2642 * @len: length of the network prefix
2643 *
2644 * net_get() looks up the given network in routing table @tab and
2645 * returns a pointer to its &net entry. If no such entry exists, it's
2646 * created.
2647 */
2648 static inline net *net_get(rtable *tab, ip_addr addr, unsigned len)
2649 { DUMMY; }
2650
2651 /**
2652 * rte_cow - copy a route for writing
2653 * @r: a route entry to be copied
2654 *
2655 * rte_cow() takes a &rte and prepares it for modification. The exact action
2656 * taken depends on the flags of the &rte -- if it's a temporary entry, it's
2657 * just returned unchanged, else a new temporary entry with the same contents
2658 * is created.
2659 *
2660 * The primary use of this function is inside the filter machinery -- when
2661 * a filter wants to modify &rte contents (to change the preference or to
2662 * attach another set of attributes), it must ensure that the &rte is not
2663 * shared with anyone else (and especially that it isn't stored in any routing
2664 * table).
2665 *
2666 * Result: a pointer to the new writable &rte.
2667 */
2668 static inline rte * rte_cow(rte *r)
2669 { DUMMY; }
2670
2671 #endif