xref: /openbsd/usr.sbin/ospfd/kroute.c (revision 905646f0)
1 /*	$OpenBSD: kroute.c,v 1.114 2020/08/20 03:09:28 jmatthew Exp $ */
2 
3 /*
4  * Copyright (c) 2004 Esben Norby <norby@openbsd.org>
5  * Copyright (c) 2003, 2004 Henning Brauer <henning@openbsd.org>
6  *
7  * Permission to use, copy, modify, and distribute this software for any
8  * purpose with or without fee is hereby granted, provided that the above
9  * copyright notice and this permission notice appear in all copies.
10  *
11  * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
12  * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
13  * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
14  * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
15  * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
16  * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
17  * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
18  */
19 
20 #include <sys/types.h>
21 #include <sys/socket.h>
22 #include <sys/sysctl.h>
23 #include <sys/tree.h>
24 #include <sys/uio.h>
25 #include <netinet/in.h>
26 #include <arpa/inet.h>
27 #include <net/if.h>
28 #include <net/if_dl.h>
29 #include <net/if_types.h>
30 #include <net/route.h>
31 #include <err.h>
32 #include <errno.h>
33 #include <fcntl.h>
34 #include <stdio.h>
35 #include <stdlib.h>
36 #include <string.h>
37 #include <unistd.h>
38 #include <limits.h>
39 
40 #include "ospfd.h"
41 #include "log.h"
42 
43 struct {
44 	u_int32_t		rtseq;
45 	pid_t			pid;
46 	int			fib_sync;
47 	int			fib_serial;
48 	u_int8_t		fib_prio;
49 	int			fd;
50 	struct event		ev;
51 	struct event		reload;
52 	u_int			rdomain;
53 #define KR_RELOAD_IDLE	0
54 #define KR_RELOAD_FETCH	1
55 #define KR_RELOAD_HOLD	2
56 	int			reload_state;
57 } kr_state;
58 
59 struct kroute_node {
60 	RB_ENTRY(kroute_node)	 entry;
61 	struct kroute_node	*next;
62 	struct kroute		 r;
63 	int			 serial;
64 };
65 
66 struct kif_node {
67 	RB_ENTRY(kif_node)	 entry;
68 	TAILQ_HEAD(, kif_addr)	 addrs;
69 	struct kif		 k;
70 };
71 
72 void	kr_redist_remove(struct kroute_node *, struct kroute_node *);
73 int	kr_redist_eval(struct kroute *, struct kroute *);
74 void	kr_redistribute(struct kroute_node *);
75 int	kroute_compare(struct kroute_node *, struct kroute_node *);
76 int	kif_compare(struct kif_node *, struct kif_node *);
77 int	kr_change_fib(struct kroute_node *, struct kroute *, int, int);
78 int	kr_delete_fib(struct kroute_node *);
79 
80 struct kroute_node	*kroute_find(in_addr_t, u_int8_t, u_int8_t);
81 struct kroute_node	*kroute_matchgw(struct kroute_node *, struct in_addr);
82 int			 kroute_insert(struct kroute_node *);
83 int			 kroute_remove(struct kroute_node *);
84 void			 kroute_clear(void);
85 
86 struct kif_node		*kif_find(u_short);
87 struct kif_node		*kif_insert(u_short);
88 int			 kif_remove(struct kif_node *);
89 struct kif		*kif_update(u_short, int, struct if_data *,
90 			    struct sockaddr_dl *);
91 int			 kif_validate(u_short);
92 
93 struct kroute_node	*kroute_match(in_addr_t);
94 
95 int		protect_lo(void);
96 u_int8_t	prefixlen_classful(in_addr_t);
97 void		get_rtaddrs(int, struct sockaddr *, struct sockaddr **);
98 void		if_change(u_short, int, struct if_data *, struct sockaddr_dl *);
99 void		if_newaddr(u_short, struct sockaddr_in *, struct sockaddr_in *,
100 		    struct sockaddr_in *);
101 void		if_deladdr(u_short, struct sockaddr_in *, struct sockaddr_in *,
102 		    struct sockaddr_in *);
103 void		if_announce(void *);
104 
105 int		send_rtmsg(int, int, struct kroute *);
106 int		dispatch_rtmsg(void);
107 int		fetchtable(void);
108 int		fetchifs(u_short);
109 int		rtmsg_process(char *, size_t);
110 void		kr_fib_reload_timer(int, short, void *);
111 void		kr_fib_reload_arm_timer(int);
112 
113 RB_HEAD(kroute_tree, kroute_node)	krt = RB_INITIALIZER(&krt);
114 RB_PROTOTYPE(kroute_tree, kroute_node, entry, kroute_compare)
115 RB_GENERATE(kroute_tree, kroute_node, entry, kroute_compare)
116 
117 RB_HEAD(kif_tree, kif_node)		kit = RB_INITIALIZER(&kit);
118 RB_PROTOTYPE(kif_tree, kif_node, entry, kif_compare)
119 RB_GENERATE(kif_tree, kif_node, entry, kif_compare)
120 
121 int
122 kif_init(void)
123 {
124 	if (fetchifs(0) == -1)
125 		return (-1);
126 
127 	return (0);
128 }
129 
130 int
131 kr_init(int fs, u_int rdomain, int redis_label_or_prefix, u_int8_t fib_prio)
132 {
133 	int		opt = 0, rcvbuf, default_rcvbuf;
134 	socklen_t	optlen;
135 	int		filter_prio = fib_prio;
136 	int		filter_flags = RTF_LLINFO | RTF_BROADCAST;
137 
138 	kr_state.fib_sync = fs;
139 	kr_state.rdomain = rdomain;
140 	kr_state.fib_prio = fib_prio;
141 
142 	if ((kr_state.fd = socket(AF_ROUTE,
143 	    SOCK_RAW | SOCK_CLOEXEC | SOCK_NONBLOCK, AF_INET)) == -1) {
144 		log_warn("kr_init: socket");
145 		return (-1);
146 	}
147 
148 	/* not interested in my own messages */
149 	if (setsockopt(kr_state.fd, SOL_SOCKET, SO_USELOOPBACK,
150 	    &opt, sizeof(opt)) == -1)
151 		log_warn("kr_init: setsockopt");	/* not fatal */
152 
153 	if (redis_label_or_prefix) {
154 		filter_prio = 0;
155 		log_info("%s: priority filter disabled", __func__);
156 	} else
157 		log_debug("%s: priority filter enabled", __func__);
158 
159 	if (setsockopt(kr_state.fd, AF_ROUTE, ROUTE_PRIOFILTER, &filter_prio,
160 	    sizeof(filter_prio)) == -1) {
161 		log_warn("%s: setsockopt AF_ROUTE ROUTE_PRIOFILTER", __func__);
162 		/* not fatal */
163 	}
164 	if (setsockopt(kr_state.fd, AF_ROUTE, ROUTE_FLAGFILTER, &filter_flags,
165 	    sizeof(filter_flags)) == -1) {
166 		log_warn("%s: setsockopt AF_ROUTE ROUTE_FLAGFILTER", __func__);
167 		/* not fatal */
168 	}
169 
170 	/* grow receive buffer, don't wanna miss messages */
171 	optlen = sizeof(default_rcvbuf);
172 	if (getsockopt(kr_state.fd, SOL_SOCKET, SO_RCVBUF,
173 	    &default_rcvbuf, &optlen) == -1)
174 		log_warn("kr_init getsockopt SOL_SOCKET SO_RCVBUF");
175 	else
176 		for (rcvbuf = MAX_RTSOCK_BUF;
177 		    rcvbuf > default_rcvbuf &&
178 		    setsockopt(kr_state.fd, SOL_SOCKET, SO_RCVBUF,
179 		    &rcvbuf, sizeof(rcvbuf)) == -1 && errno == ENOBUFS;
180 		    rcvbuf /= 2)
181 			;	/* nothing */
182 
183 	kr_state.pid = getpid();
184 	kr_state.rtseq = 1;
185 
186 	if (fetchtable() == -1)
187 		return (-1);
188 
189 	if (protect_lo() == -1)
190 		return (-1);
191 
192 	event_set(&kr_state.ev, kr_state.fd, EV_READ | EV_PERSIST,
193 	    kr_dispatch_msg, NULL);
194 	event_add(&kr_state.ev, NULL);
195 
196 	kr_state.reload_state = KR_RELOAD_IDLE;
197 	evtimer_set(&kr_state.reload, kr_fib_reload_timer, NULL);
198 
199 	return (0);
200 }
201 
202 int
203 kr_change_fib(struct kroute_node *kr, struct kroute *kroute, int krcount,
204     int action)
205 {
206 	int			 i;
207 	struct kroute_node	*kn, *nkn;
208 
209 	if (action == RTM_ADD) {
210 		/*
211 		 * First remove all stale multipath routes.
212 		 * This step must be skipped when the action is RTM_CHANGE
213 		 * because it is already a single path route that will be
214 		 * changed.
215 		 */
216 		for (kn = kr; kn != NULL; kn = nkn) {
217 			for (i = 0; i < krcount; i++) {
218 				if (kn->r.nexthop.s_addr ==
219 				    kroute[i].nexthop.s_addr)
220 					break;
221 			}
222 			nkn = kn->next;
223 			if (i == krcount) {
224 				/* stale route */
225 				if (kr_delete_fib(kn) == -1)
226 					log_warnx("kr_delete_fib failed");
227 				/*
228 				 * if head element was removed we need to adjust
229 				 * the head
230 				 */
231 				if (kr == kn)
232 					kr = nkn;
233 			}
234 		}
235 	}
236 
237 	/*
238 	 * now add or change the route
239 	 */
240 	for (i = 0; i < krcount; i++) {
241 		/* nexthop within 127/8 -> ignore silently */
242 		if ((kroute[i].nexthop.s_addr & htonl(IN_CLASSA_NET)) ==
243 		    htonl(INADDR_LOOPBACK & IN_CLASSA_NET))
244 			continue;
245 
246 		if (action == RTM_ADD && kr) {
247 			for (kn = kr; kn != NULL; kn = kn->next) {
248 				if (kn->r.nexthop.s_addr ==
249 				    kroute[i].nexthop.s_addr)
250 					break;
251 			}
252 
253 			if (kn != NULL)
254 				/* nexthop already present, skip it */
255 				continue;
256 		} else
257 			/* modify first entry */
258 			kn = kr;
259 
260 		/* send update */
261 		if (send_rtmsg(kr_state.fd, action, &kroute[i]) == -1)
262 			return (-1);
263 
264 		/* create new entry unless we are changing the first entry */
265 		if (action == RTM_ADD)
266 			if ((kn = calloc(1, sizeof(*kn))) == NULL)
267 				fatal(NULL);
268 
269 		kn->r.prefix.s_addr = kroute[i].prefix.s_addr;
270 		kn->r.prefixlen = kroute[i].prefixlen;
271 		kn->r.nexthop.s_addr = kroute[i].nexthop.s_addr;
272 		kn->r.flags = kroute[i].flags | F_OSPFD_INSERTED;
273 		kn->r.priority = kr_state.fib_prio;
274 		kn->r.ext_tag = kroute[i].ext_tag;
275 		rtlabel_unref(kn->r.rtlabel);	/* for RTM_CHANGE */
276 		kn->r.rtlabel = kroute[i].rtlabel;
277 
278 		if (action == RTM_ADD)
279 			if (kroute_insert(kn) == -1) {
280 				log_debug("kr_update_fib: cannot insert %s",
281 				    inet_ntoa(kn->r.nexthop));
282 				free(kn);
283 			}
284 		action = RTM_ADD;
285 	}
286 	return  (0);
287 }
288 
289 int
290 kr_change(struct kroute *kroute, int krcount)
291 {
292 	struct kroute_node	*kr;
293 	int			 action = RTM_ADD;
294 
295 	kroute->rtlabel = rtlabel_tag2id(kroute->ext_tag);
296 
297 	kr = kroute_find(kroute->prefix.s_addr, kroute->prefixlen,
298 	    kr_state.fib_prio);
299 	if (kr != NULL && kr->next == NULL && krcount == 1)
300 		/* single path OSPF route */
301 		action = RTM_CHANGE;
302 
303 	return (kr_change_fib(kr, kroute, krcount, action));
304 }
305 
306 int
307 kr_delete_fib(struct kroute_node *kr)
308 {
309 	if (kr->r.priority != kr_state.fib_prio)
310 		log_warn("kr_delete_fib: %s/%d has wrong priority %d",
311 		    inet_ntoa(kr->r.prefix), kr->r.prefixlen, kr->r.priority);
312 
313 	if (send_rtmsg(kr_state.fd, RTM_DELETE, &kr->r) == -1)
314 		return (-1);
315 
316 	if (kroute_remove(kr) == -1)
317 		return (-1);
318 
319 	return (0);
320 }
321 
322 int
323 kr_delete(struct kroute *kroute)
324 {
325 	struct kroute_node	*kr, *nkr;
326 
327 	if ((kr = kroute_find(kroute->prefix.s_addr, kroute->prefixlen,
328 	    kr_state.fib_prio)) == NULL)
329 		return (0);
330 
331 	while (kr != NULL) {
332 		nkr = kr->next;
333 		if (kr_delete_fib(kr) == -1)
334 			return (-1);
335 		kr = nkr;
336 	}
337 	return (0);
338 }
339 
340 void
341 kr_shutdown(void)
342 {
343 	kr_fib_decouple();
344 	kroute_clear();
345 	kif_clear();
346 }
347 
348 void
349 kr_fib_couple(void)
350 {
351 	struct kroute_node	*kr;
352 	struct kroute_node	*kn;
353 
354 	if (kr_state.fib_sync == 1)	/* already coupled */
355 		return;
356 
357 	kr_state.fib_sync = 1;
358 
359 	RB_FOREACH(kr, kroute_tree, &krt)
360 		if (kr->r.priority == kr_state.fib_prio)
361 			for (kn = kr; kn != NULL; kn = kn->next)
362 				send_rtmsg(kr_state.fd, RTM_ADD, &kn->r);
363 
364 	log_info("kernel routing table coupled");
365 }
366 
367 void
368 kr_fib_decouple(void)
369 {
370 	struct kroute_node	*kr;
371 	struct kroute_node	*kn;
372 
373 	if (kr_state.fib_sync == 0)	/* already decoupled */
374 		return;
375 
376 	RB_FOREACH(kr, kroute_tree, &krt)
377 		if (kr->r.priority == kr_state.fib_prio)
378 			for (kn = kr; kn != NULL; kn = kn->next)
379 				send_rtmsg(kr_state.fd, RTM_DELETE, &kn->r);
380 
381 	kr_state.fib_sync = 0;
382 
383 	log_info("kernel routing table decoupled");
384 }
385 
386 void
387 kr_fib_reload_timer(int fd, short event, void *bula)
388 {
389 	if (kr_state.reload_state == KR_RELOAD_FETCH) {
390 		kr_fib_reload();
391 		kr_state.reload_state = KR_RELOAD_HOLD;
392 		kr_fib_reload_arm_timer(KR_RELOAD_HOLD_TIMER);
393 	} else {
394 		kr_state.reload_state = KR_RELOAD_IDLE;
395 	}
396 }
397 
398 void
399 kr_fib_reload_arm_timer(int delay)
400 {
401 	struct timeval		 tv;
402 
403 	timerclear(&tv);
404 	tv.tv_sec = delay / 1000;
405 	tv.tv_usec = (delay % 1000) * 1000;
406 
407 	if (evtimer_add(&kr_state.reload, &tv) == -1)
408 		fatal("add_reload_timer");
409 }
410 
411 void
412 kr_fib_reload()
413 {
414 	struct kroute_node	*krn, *kr, *kn;
415 
416 	log_info("reloading interface list and routing table");
417 
418 	kr_state.fib_serial++;
419 
420 	if (fetchifs(0) == -1 || fetchtable() == -1)
421 		return;
422 
423 	for (kr = RB_MIN(kroute_tree, &krt); kr != NULL; kr = krn) {
424 		krn = RB_NEXT(kroute_tree, &krt, kr);
425 
426 		do {
427 			kn = kr->next;
428 
429 			if (kr->serial != kr_state.fib_serial) {
430 				if (kr->r.priority == kr_state.fib_prio) {
431 					kr->serial = kr_state.fib_serial;
432 					if (send_rtmsg(kr_state.fd,
433 					    RTM_ADD, &kr->r) != 0)
434 						break;
435 				} else
436 					kroute_remove(kr);
437 			}
438 
439 		} while ((kr = kn) != NULL);
440 	}
441 }
442 
443 void
444 kr_fib_update_prio(u_int8_t fib_prio)
445 {
446 	struct kroute_node      *kr;
447 
448 	RB_FOREACH(kr, kroute_tree, &krt)
449 		if ((kr->r.flags & F_OSPFD_INSERTED))
450 			kr->r.priority = fib_prio;
451 
452 	log_info("fib priority changed from %hhu to %hhu",
453 	    kr_state.fib_prio, fib_prio);
454 
455 	kr_state.fib_prio = fib_prio;
456  }
457 
458 /* ARGSUSED */
459 void
460 kr_dispatch_msg(int fd, short event, void *bula)
461 {
462 	/* XXX this is stupid */
463 	if (dispatch_rtmsg() == -1)
464 		event_loopexit(NULL);
465 }
466 
467 void
468 kr_show_route(struct imsg *imsg)
469 {
470 	struct kroute_node	*kr;
471 	struct kroute_node	*kn;
472 	int			 flags;
473 	struct in_addr		 addr;
474 
475 	switch (imsg->hdr.type) {
476 	case IMSG_CTL_KROUTE:
477 		if (imsg->hdr.len != IMSG_HEADER_SIZE + sizeof(flags)) {
478 			log_warnx("kr_show_route: wrong imsg len");
479 			return;
480 		}
481 		memcpy(&flags, imsg->data, sizeof(flags));
482 		RB_FOREACH(kr, kroute_tree, &krt)
483 			if (!flags || kr->r.flags & flags) {
484 				kn = kr;
485 				do {
486 					main_imsg_compose_ospfe(IMSG_CTL_KROUTE,
487 					    imsg->hdr.pid,
488 					    &kn->r, sizeof(kn->r));
489 				} while ((kn = kn->next) != NULL);
490 			}
491 		break;
492 	case IMSG_CTL_KROUTE_ADDR:
493 		if (imsg->hdr.len != IMSG_HEADER_SIZE +
494 		    sizeof(struct in_addr)) {
495 			log_warnx("kr_show_route: wrong imsg len");
496 			return;
497 		}
498 		memcpy(&addr, imsg->data, sizeof(addr));
499 		kr = NULL;
500 		kr = kroute_match(addr.s_addr);
501 		if (kr != NULL)
502 			main_imsg_compose_ospfe(IMSG_CTL_KROUTE, imsg->hdr.pid,
503 			    &kr->r, sizeof(kr->r));
504 		break;
505 	default:
506 		log_debug("kr_show_route: error handling imsg");
507 		break;
508 	}
509 
510 	main_imsg_compose_ospfe(IMSG_CTL_END, imsg->hdr.pid, NULL, 0);
511 }
512 
513 void
514 kr_ifinfo(char *ifname, pid_t pid)
515 {
516 	struct kif_node	*kif;
517 
518 	RB_FOREACH(kif, kif_tree, &kit)
519 		if (ifname == NULL || !strcmp(ifname, kif->k.ifname)) {
520 			main_imsg_compose_ospfe(IMSG_CTL_IFINFO,
521 			    pid, &kif->k, sizeof(kif->k));
522 		}
523 
524 	main_imsg_compose_ospfe(IMSG_CTL_END, pid, NULL, 0);
525 }
526 
527 void
528 kr_redist_remove(struct kroute_node *kh, struct kroute_node *kn)
529 {
530 	struct kroute	*kr;
531 
532 	/* was the route redistributed? */
533 	if ((kn->r.flags & F_REDISTRIBUTED) == 0)
534 		return;
535 
536 	/* remove redistributed flag */
537 	kn->r.flags &= ~F_REDISTRIBUTED;
538 	kr = &kn->r;
539 
540 	/* probably inform the RDE (check if no other path is redistributed) */
541 	for (kn = kh; kn; kn = kn->next)
542 		if (kn->r.flags & F_REDISTRIBUTED)
543 			break;
544 
545 	if (kn == NULL)
546 		main_imsg_compose_rde(IMSG_NETWORK_DEL, 0, kr,
547 		    sizeof(struct kroute));
548 }
549 
550 int
551 kr_redist_eval(struct kroute *kr, struct kroute *new_kr)
552 {
553 	u_int32_t	 a, metric = 0;
554 
555 	/* Only non-ospfd routes are considered for redistribution. */
556 	if (!(kr->flags & F_KERNEL))
557 		goto dont_redistribute;
558 
559 	/* Dynamic routes are not redistributable. */
560 	if (kr->flags & F_DYNAMIC)
561 		goto dont_redistribute;
562 
563 	/* interface is not up and running so don't announce */
564 	if (kr->flags & F_DOWN)
565 		goto dont_redistribute;
566 
567 	/*
568 	 * We consider the loopback net, multicast and experimental addresses
569 	 * as not redistributable.
570 	 */
571 	a = ntohl(kr->prefix.s_addr);
572 	if (IN_MULTICAST(a) || IN_BADCLASS(a) ||
573 	    (a >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET)
574 		goto dont_redistribute;
575 	/*
576 	 * Consider networks with nexthop loopback as not redistributable
577 	 * unless it is a reject or blackhole route.
578 	 */
579 	if (kr->nexthop.s_addr == htonl(INADDR_LOOPBACK) &&
580 	    !(kr->flags & (F_BLACKHOLE|F_REJECT)))
581 		goto dont_redistribute;
582 
583 	/* Should we redistribute this route? */
584 	if (!ospf_redistribute(kr, &metric))
585 		goto dont_redistribute;
586 
587 	/* prefix should be redistributed */
588 	kr->flags |= F_REDISTRIBUTED;
589 	/*
590 	 * only one of all multipath routes can be redistributed so
591 	 * redistribute the best one.
592 	 */
593 	if (new_kr->metric > metric) {
594 		*new_kr = *kr;
595 		new_kr->metric = metric;
596 	}
597 
598 	return (1);
599 
600 dont_redistribute:
601 	/* was the route redistributed? */
602 	if ((kr->flags & F_REDISTRIBUTED) == 0)
603 		return (0);
604 
605 	kr->flags &= ~F_REDISTRIBUTED;
606 	return (1);
607 }
608 
609 void
610 kr_redistribute(struct kroute_node *kh)
611 {
612 	struct kroute_node	*kn;
613 	struct kroute		 kr;
614 	int			 redistribute = 0;
615 
616 	/* only the highest prio route can be redistributed */
617 	if (kroute_find(kh->r.prefix.s_addr, kh->r.prefixlen, RTP_ANY) != kh)
618 		return;
619 
620 	bzero(&kr, sizeof(kr));
621 	kr.metric = UINT_MAX;
622 	for (kn = kh; kn; kn = kn->next)
623 		if (kr_redist_eval(&kn->r, &kr))
624 			redistribute = 1;
625 
626 	if (!redistribute)
627 		return;
628 
629 	if (kr.flags & F_REDISTRIBUTED) {
630 		main_imsg_compose_rde(IMSG_NETWORK_ADD, 0, &kr,
631 		    sizeof(struct kroute));
632 	} else {
633 		kr = kh->r;
634 		main_imsg_compose_rde(IMSG_NETWORK_DEL, 0, &kr,
635 		    sizeof(struct kroute));
636 	}
637 }
638 
639 void
640 kr_reload(int redis_label_or_prefix)
641 {
642 	struct kroute_node	*kr, *kn;
643 	u_int32_t		 dummy;
644 	int			 r;
645 	int			 filter_prio = kr_state.fib_prio;
646 
647 	/* update the priority filter */
648 	if (redis_label_or_prefix) {
649 		filter_prio = 0;
650 		log_info("%s: priority filter disabled", __func__);
651 	} else
652 		log_debug("%s: priority filter enabled", __func__);
653 
654 	if (setsockopt(kr_state.fd, AF_ROUTE, ROUTE_PRIOFILTER, &filter_prio,
655 	    sizeof(filter_prio)) == -1) {
656 		log_warn("%s: setsockopt AF_ROUTE ROUTE_PRIOFILTER", __func__);
657 		/* not fatal */
658 	}
659 
660 	/* update redistribute lists */
661 	RB_FOREACH(kr, kroute_tree, &krt) {
662 		for (kn = kr; kn; kn = kn->next) {
663 			r = ospf_redistribute(&kn->r, &dummy);
664 			/*
665 			 * if it is redistributed, redistribute again metric
666 			 * may have changed.
667 			 */
668 			if ((kn->r.flags & F_REDISTRIBUTED && !r) || r)
669 				break;
670 		}
671 		if (kn) {
672 			/*
673 			 * kr_redistribute copes with removes and RDE with
674 			 * duplicates
675 			 */
676 			kr_redistribute(kr);
677 		}
678 	}
679 }
680 
681 /* rb-tree compare */
682 int
683 kroute_compare(struct kroute_node *a, struct kroute_node *b)
684 {
685 	if (ntohl(a->r.prefix.s_addr) < ntohl(b->r.prefix.s_addr))
686 		return (-1);
687 	if (ntohl(a->r.prefix.s_addr) > ntohl(b->r.prefix.s_addr))
688 		return (1);
689 	if (a->r.prefixlen < b->r.prefixlen)
690 		return (-1);
691 	if (a->r.prefixlen > b->r.prefixlen)
692 		return (1);
693 
694 	/* if the priority is RTP_ANY finish on the first address hit */
695 	if (a->r.priority == RTP_ANY || b->r.priority == RTP_ANY)
696 		return (0);
697 	if (a->r.priority < b->r.priority)
698 		return (-1);
699 	if (a->r.priority > b->r.priority)
700 		return (1);
701 	return (0);
702 }
703 
704 int
705 kif_compare(struct kif_node *a, struct kif_node *b)
706 {
707 	return (b->k.ifindex - a->k.ifindex);
708 }
709 
710 /* tree management */
711 struct kroute_node *
712 kroute_find(in_addr_t prefix, u_int8_t prefixlen, u_int8_t prio)
713 {
714 	struct kroute_node	s;
715 	struct kroute_node	*kn, *tmp;
716 
717 	s.r.prefix.s_addr = prefix;
718 	s.r.prefixlen = prefixlen;
719 	s.r.priority = prio;
720 
721 	kn = RB_FIND(kroute_tree, &krt, &s);
722 	if (kn && prio == RTP_ANY) {
723 		tmp = RB_PREV(kroute_tree, &krt, kn);
724 		while (tmp) {
725 			if (kroute_compare(&s, tmp) == 0)
726 				kn = tmp;
727 			else
728 				break;
729 			tmp = RB_PREV(kroute_tree, &krt, kn);
730 		}
731 	}
732 	return (kn);
733 }
734 
735 struct kroute_node *
736 kroute_matchgw(struct kroute_node *kr, struct in_addr nh)
737 {
738 	in_addr_t	nexthop;
739 
740 	nexthop = nh.s_addr;
741 
742 	while (kr) {
743 		if (kr->r.nexthop.s_addr == nexthop)
744 			return (kr);
745 		kr = kr->next;
746 	}
747 
748 	return (NULL);
749 }
750 
751 int
752 kroute_insert(struct kroute_node *kr)
753 {
754 	struct kroute_node	*krm, *krh;
755 
756 	kr->serial = kr_state.fib_serial;
757 
758 	if ((krh = RB_INSERT(kroute_tree, &krt, kr)) != NULL) {
759 		/*
760 		 * Multipath route, add at end of list.
761 		 */
762 		krm = krh;
763 		while (krm->next != NULL)
764 			krm = krm->next;
765 		krm->next = kr;
766 		kr->next = NULL; /* to be sure */
767 	} else
768 		krh = kr;
769 
770 	if (!(kr->r.flags & F_KERNEL)) {
771 		/* don't validate or redistribute ospf route */
772 		kr->r.flags &= ~F_DOWN;
773 		return (0);
774 	}
775 
776 	if (kif_validate(kr->r.ifindex))
777 		kr->r.flags &= ~F_DOWN;
778 	else
779 		kr->r.flags |= F_DOWN;
780 
781 	kr_redistribute(krh);
782 	return (0);
783 }
784 
785 int
786 kroute_remove(struct kroute_node *kr)
787 {
788 	struct kroute_node	*krm;
789 
790 	if ((krm = RB_FIND(kroute_tree, &krt, kr)) == NULL) {
791 		log_warnx("kroute_remove failed to find %s/%u",
792 		    inet_ntoa(kr->r.prefix), kr->r.prefixlen);
793 		return (-1);
794 	}
795 
796 	if (krm == kr) {
797 		/* head element */
798 		if (RB_REMOVE(kroute_tree, &krt, kr) == NULL) {
799 			log_warnx("kroute_remove failed for %s/%u",
800 			    inet_ntoa(kr->r.prefix), kr->r.prefixlen);
801 			return (-1);
802 		}
803 		if (kr->next != NULL) {
804 			if (RB_INSERT(kroute_tree, &krt, kr->next) != NULL) {
805 				log_warnx("kroute_remove failed to add %s/%u",
806 				    inet_ntoa(kr->r.prefix), kr->r.prefixlen);
807 				return (-1);
808 			}
809 		}
810 	} else {
811 		/* somewhere in the list */
812 		while (krm->next != kr && krm->next != NULL)
813 			krm = krm->next;
814 		if (krm->next == NULL) {
815 			log_warnx("kroute_remove multipath list corrupted "
816 			    "for %s/%u", inet_ntoa(kr->r.prefix),
817 			    kr->r.prefixlen);
818 			return (-1);
819 		}
820 		krm->next = kr->next;
821 	}
822 
823 	kr_redist_remove(krm, kr);
824 	rtlabel_unref(kr->r.rtlabel);
825 
826 	free(kr);
827 	return (0);
828 }
829 
830 void
831 kroute_clear(void)
832 {
833 	struct kroute_node	*kr;
834 
835 	while ((kr = RB_MIN(kroute_tree, &krt)) != NULL)
836 		kroute_remove(kr);
837 }
838 
839 struct kif_node *
840 kif_find(u_short ifindex)
841 {
842 	struct kif_node	s;
843 
844 	bzero(&s, sizeof(s));
845 	s.k.ifindex = ifindex;
846 
847 	return (RB_FIND(kif_tree, &kit, &s));
848 }
849 
850 struct kif *
851 kif_findname(char *ifname, struct in_addr addr, struct kif_addr **kap)
852 {
853 	struct kif_node	*kif;
854 	struct kif_addr	*ka;
855 
856 	RB_FOREACH(kif, kif_tree, &kit)
857 		if (!strcmp(ifname, kif->k.ifname)) {
858 			ka = TAILQ_FIRST(&kif->addrs);
859 			if (addr.s_addr != 0) {
860 				TAILQ_FOREACH(ka, &kif->addrs, entry) {
861 					if (addr.s_addr == ka->addr.s_addr)
862 						break;
863 				}
864 			}
865 			if (kap != NULL)
866 				*kap = ka;
867 			return (&kif->k);
868 		}
869 
870 	return (NULL);
871 }
872 
873 struct kif_node *
874 kif_insert(u_short ifindex)
875 {
876 	struct kif_node	*kif;
877 
878 	if ((kif = calloc(1, sizeof(struct kif_node))) == NULL)
879 		return (NULL);
880 
881 	kif->k.ifindex = ifindex;
882 	TAILQ_INIT(&kif->addrs);
883 
884 	if (RB_INSERT(kif_tree, &kit, kif) != NULL)
885 		fatalx("kif_insert: RB_INSERT");
886 
887 	return (kif);
888 }
889 
890 int
891 kif_remove(struct kif_node *kif)
892 {
893 	struct kif_addr	*ka;
894 
895 	if (RB_REMOVE(kif_tree, &kit, kif) == NULL) {
896 		log_warnx("RB_REMOVE(kif_tree, &kit, kif)");
897 		return (-1);
898 	}
899 
900 	while ((ka = TAILQ_FIRST(&kif->addrs)) != NULL) {
901 		TAILQ_REMOVE(&kif->addrs, ka, entry);
902 		free(ka);
903 	}
904 	free(kif);
905 	return (0);
906 }
907 
908 void
909 kif_clear(void)
910 {
911 	struct kif_node	*kif;
912 
913 	while ((kif = RB_MIN(kif_tree, &kit)) != NULL)
914 		kif_remove(kif);
915 }
916 
917 struct kif *
918 kif_update(u_short ifindex, int flags, struct if_data *ifd,
919     struct sockaddr_dl *sdl)
920 {
921 	struct kif_node		*kif;
922 
923 	if ((kif = kif_find(ifindex)) == NULL) {
924 		if ((kif = kif_insert(ifindex)) == NULL)
925 			return (NULL);
926 		kif->k.nh_reachable = (flags & IFF_UP) &&
927 		    LINK_STATE_IS_UP(ifd->ifi_link_state);
928 	}
929 
930 	kif->k.flags = flags;
931 	kif->k.link_state = ifd->ifi_link_state;
932 	kif->k.if_type = ifd->ifi_type;
933 	kif->k.baudrate = ifd->ifi_baudrate;
934 	kif->k.mtu = ifd->ifi_mtu;
935 	kif->k.rdomain = ifd->ifi_rdomain;
936 
937 	if (sdl && sdl->sdl_family == AF_LINK) {
938 		if (sdl->sdl_nlen >= sizeof(kif->k.ifname))
939 			memcpy(kif->k.ifname, sdl->sdl_data,
940 			    sizeof(kif->k.ifname) - 1);
941 		else if (sdl->sdl_nlen > 0)
942 			memcpy(kif->k.ifname, sdl->sdl_data,
943 			    sdl->sdl_nlen);
944 		/* string already terminated via calloc() */
945 	}
946 
947 	return (&kif->k);
948 }
949 
950 int
951 kif_validate(u_short ifindex)
952 {
953 	struct kif_node		*kif;
954 
955 	if ((kif = kif_find(ifindex)) == NULL) {
956 		log_warnx("interface with index %u not found", ifindex);
957 		return (1);
958 	}
959 
960 	return (kif->k.nh_reachable);
961 }
962 
963 struct kroute_node *
964 kroute_match(in_addr_t key)
965 {
966 	int			 i;
967 	struct kroute_node	*kr;
968 
969 	/* we will never match the default route */
970 	for (i = 32; i > 0; i--)
971 		if ((kr = kroute_find(key & prefixlen2mask(i), i,
972 		    RTP_ANY)) != NULL)
973 			return (kr);
974 
975 	/* if we don't have a match yet, try to find a default route */
976 	if ((kr = kroute_find(0, 0, RTP_ANY)) != NULL)
977 			return (kr);
978 
979 	return (NULL);
980 }
981 
982 /* misc */
983 int
984 protect_lo(void)
985 {
986 	struct kroute_node	*kr;
987 
988 	/* special protection for 127/8 */
989 	if ((kr = calloc(1, sizeof(struct kroute_node))) == NULL) {
990 		log_warn("protect_lo");
991 		return (-1);
992 	}
993 	kr->r.prefix.s_addr = htonl(INADDR_LOOPBACK & IN_CLASSA_NET);
994 	kr->r.prefixlen = 8;
995 	kr->r.flags = F_KERNEL|F_CONNECTED;
996 
997 	if (RB_INSERT(kroute_tree, &krt, kr) != NULL)
998 		free(kr);	/* kernel route already there, no problem */
999 
1000 	return (0);
1001 }
1002 
1003 u_int8_t
1004 prefixlen_classful(in_addr_t ina)
1005 {
1006 	/* it hurt to write this. */
1007 
1008 	if (ina >= 0xf0000000U)		/* class E */
1009 		return (32);
1010 	else if (ina >= 0xe0000000U)	/* class D */
1011 		return (4);
1012 	else if (ina >= 0xc0000000U)	/* class C */
1013 		return (24);
1014 	else if (ina >= 0x80000000U)	/* class B */
1015 		return (16);
1016 	else				/* class A */
1017 		return (8);
1018 }
1019 
1020 u_int8_t
1021 mask2prefixlen(in_addr_t ina)
1022 {
1023 	if (ina == 0)
1024 		return (0);
1025 	else
1026 		return (33 - ffs(ntohl(ina)));
1027 }
1028 
1029 in_addr_t
1030 prefixlen2mask(u_int8_t prefixlen)
1031 {
1032 	if (prefixlen == 0)
1033 		return (0);
1034 
1035 	return (htonl(0xffffffff << (32 - prefixlen)));
1036 }
1037 
1038 #define ROUNDUP(a) \
1039 	((a) > 0 ? (1 + (((a) - 1) | (sizeof(long) - 1))) : sizeof(long))
1040 
1041 void
1042 get_rtaddrs(int addrs, struct sockaddr *sa, struct sockaddr **rti_info)
1043 {
1044 	int	i;
1045 
1046 	for (i = 0; i < RTAX_MAX; i++) {
1047 		if (addrs & (1 << i)) {
1048 			rti_info[i] = sa;
1049 			sa = (struct sockaddr *)((char *)(sa) +
1050 			    ROUNDUP(sa->sa_len));
1051 		} else
1052 			rti_info[i] = NULL;
1053 	}
1054 }
1055 
1056 void
1057 if_change(u_short ifindex, int flags, struct if_data *ifd,
1058     struct sockaddr_dl *sdl)
1059 {
1060 	struct kroute_node	*kr, *tkr;
1061 	struct kif		*kif;
1062 	u_int8_t		 reachable;
1063 
1064 	if ((kif = kif_update(ifindex, flags, ifd, sdl)) == NULL) {
1065 		log_warn("if_change:  kif_update(%u)", ifindex);
1066 		return;
1067 	}
1068 
1069 	/* notify ospfe about interface link state */
1070 	main_imsg_compose_ospfe(IMSG_IFINFO, 0, kif, sizeof(struct kif));
1071 
1072 	reachable = (kif->flags & IFF_UP) &&
1073 	    LINK_STATE_IS_UP(kif->link_state);
1074 
1075 	if (reachable == kif->nh_reachable)
1076 		return;		/* nothing changed wrt nexthop validity */
1077 
1078 	kif->nh_reachable = reachable;
1079 
1080 	/* update redistribute list */
1081 	RB_FOREACH(kr, kroute_tree, &krt) {
1082 		for (tkr = kr; tkr != NULL; tkr = tkr->next) {
1083 			if (tkr->r.ifindex == ifindex) {
1084 				if (reachable)
1085 					tkr->r.flags &= ~F_DOWN;
1086 				else
1087 					tkr->r.flags |= F_DOWN;
1088 
1089 			}
1090 		}
1091 		kr_redistribute(kr);
1092 	}
1093 }
1094 
1095 void
1096 if_newaddr(u_short ifindex, struct sockaddr_in *ifa, struct sockaddr_in *mask,
1097     struct sockaddr_in *brd)
1098 {
1099 	struct kif_node 	*kif;
1100 	struct kif_addr 	*ka;
1101 	struct ifaddrchange	 ifn;
1102 
1103 	if (ifa == NULL || ifa->sin_family != AF_INET)
1104 		return;
1105 	if ((kif = kif_find(ifindex)) == NULL) {
1106 		log_warnx("if_newaddr: corresponding if %d not found", ifindex);
1107 		return;
1108 	}
1109 	if ((ka = calloc(1, sizeof(struct kif_addr))) == NULL)
1110 		fatal("if_newaddr");
1111 	ka->addr = ifa->sin_addr;
1112 	if (mask)
1113 		ka->mask = mask->sin_addr;
1114 	else
1115 		ka->mask.s_addr = INADDR_NONE;
1116 	if (brd)
1117 		ka->dstbrd = brd->sin_addr;
1118 	else
1119 		ka->dstbrd.s_addr = INADDR_NONE;
1120 
1121 	TAILQ_INSERT_TAIL(&kif->addrs, ka, entry);
1122 
1123 	ifn.addr = ka->addr;
1124 	ifn.mask = ka->mask;
1125 	ifn.dst = ka->dstbrd;
1126 	ifn.ifindex = ifindex;
1127 	main_imsg_compose_ospfe(IMSG_IFADDRADD, 0, &ifn, sizeof(ifn));
1128 }
1129 
1130 void
1131 if_deladdr(u_short ifindex, struct sockaddr_in *ifa, struct sockaddr_in *mask,
1132     struct sockaddr_in *brd)
1133 {
1134 	struct kif_node 	*kif;
1135 	struct kif_addr		*ka, *nka;
1136 	struct ifaddrchange	 ifc;
1137 
1138 	if (ifa == NULL || ifa->sin_family != AF_INET)
1139 		return;
1140 	if ((kif = kif_find(ifindex)) == NULL) {
1141 		log_warnx("if_deladdr: corresponding if %d not found", ifindex);
1142 		return;
1143 	}
1144 
1145 	for (ka = TAILQ_FIRST(&kif->addrs); ka != NULL; ka = nka) {
1146 		nka = TAILQ_NEXT(ka, entry);
1147 
1148 		if (ka->addr.s_addr == ifa->sin_addr.s_addr) {
1149 			TAILQ_REMOVE(&kif->addrs, ka, entry);
1150 			ifc.addr = ifa->sin_addr;
1151 			ifc.ifindex = ifindex;
1152 			main_imsg_compose_ospfe(IMSG_IFADDRDEL, 0, &ifc,
1153 			    sizeof(ifc));
1154 			free(ka);
1155 			return;
1156 		}
1157 	}
1158 }
1159 
1160 void
1161 if_announce(void *msg)
1162 {
1163 	struct if_announcemsghdr	*ifan;
1164 	struct kif_node			*kif;
1165 
1166 	ifan = msg;
1167 
1168 	switch (ifan->ifan_what) {
1169 	case IFAN_ARRIVAL:
1170 		kif = kif_insert(ifan->ifan_index);
1171 		strlcpy(kif->k.ifname, ifan->ifan_name, sizeof(kif->k.ifname));
1172 		break;
1173 	case IFAN_DEPARTURE:
1174 		kif = kif_find(ifan->ifan_index);
1175 		kif_remove(kif);
1176 		break;
1177 	}
1178 }
1179 
1180 /* rtsock */
1181 int
1182 send_rtmsg(int fd, int action, struct kroute *kroute)
1183 {
1184 	struct iovec		iov[5];
1185 	struct rt_msghdr	hdr;
1186 	struct sockaddr_in	prefix;
1187 	struct sockaddr_in	nexthop;
1188 	struct sockaddr_in	mask;
1189 	struct sockaddr_rtlabel	sa_rl;
1190 	int			iovcnt = 0;
1191 	const char		*label;
1192 
1193 	if (kr_state.fib_sync == 0)
1194 		return (0);
1195 
1196 	/* initialize header */
1197 	bzero(&hdr, sizeof(hdr));
1198 	hdr.rtm_version = RTM_VERSION;
1199 	hdr.rtm_type = action;
1200 	hdr.rtm_priority = kr_state.fib_prio;
1201 	hdr.rtm_tableid = kr_state.rdomain;	/* rtableid */
1202 	if (action == RTM_CHANGE)
1203 		hdr.rtm_fmask = RTF_REJECT|RTF_BLACKHOLE;
1204 	else
1205 		hdr.rtm_flags = RTF_MPATH;
1206 	hdr.rtm_seq = kr_state.rtseq++;	/* overflow doesn't matter */
1207 	hdr.rtm_msglen = sizeof(hdr);
1208 	/* adjust iovec */
1209 	iov[iovcnt].iov_base = &hdr;
1210 	iov[iovcnt++].iov_len = sizeof(hdr);
1211 
1212 	bzero(&prefix, sizeof(prefix));
1213 	prefix.sin_len = sizeof(prefix);
1214 	prefix.sin_family = AF_INET;
1215 	prefix.sin_addr.s_addr = kroute->prefix.s_addr;
1216 	/* adjust header */
1217 	hdr.rtm_addrs |= RTA_DST;
1218 	hdr.rtm_msglen += sizeof(prefix);
1219 	/* adjust iovec */
1220 	iov[iovcnt].iov_base = &prefix;
1221 	iov[iovcnt++].iov_len = sizeof(prefix);
1222 
1223 	if (kroute->nexthop.s_addr != 0) {
1224 		bzero(&nexthop, sizeof(nexthop));
1225 		nexthop.sin_len = sizeof(nexthop);
1226 		nexthop.sin_family = AF_INET;
1227 		nexthop.sin_addr.s_addr = kroute->nexthop.s_addr;
1228 		/* adjust header */
1229 		hdr.rtm_flags |= RTF_GATEWAY;
1230 		hdr.rtm_addrs |= RTA_GATEWAY;
1231 		hdr.rtm_msglen += sizeof(nexthop);
1232 		/* adjust iovec */
1233 		iov[iovcnt].iov_base = &nexthop;
1234 		iov[iovcnt++].iov_len = sizeof(nexthop);
1235 	}
1236 
1237 	bzero(&mask, sizeof(mask));
1238 	mask.sin_len = sizeof(mask);
1239 	mask.sin_family = AF_INET;
1240 	mask.sin_addr.s_addr = prefixlen2mask(kroute->prefixlen);
1241 	/* adjust header */
1242 	hdr.rtm_addrs |= RTA_NETMASK;
1243 	hdr.rtm_msglen += sizeof(mask);
1244 	/* adjust iovec */
1245 	iov[iovcnt].iov_base = &mask;
1246 	iov[iovcnt++].iov_len = sizeof(mask);
1247 
1248 	if (kroute->rtlabel != 0) {
1249 		sa_rl.sr_len = sizeof(sa_rl);
1250 		sa_rl.sr_family = AF_UNSPEC;
1251 		label = rtlabel_id2name(kroute->rtlabel);
1252 		if (strlcpy(sa_rl.sr_label, label,
1253 		    sizeof(sa_rl.sr_label)) >= sizeof(sa_rl.sr_label)) {
1254 			log_warnx("send_rtmsg: invalid rtlabel");
1255 			return (-1);
1256 		}
1257 		/* adjust header */
1258 		hdr.rtm_addrs |= RTA_LABEL;
1259 		hdr.rtm_msglen += sizeof(sa_rl);
1260 		/* adjust iovec */
1261 		iov[iovcnt].iov_base = &sa_rl;
1262 		iov[iovcnt++].iov_len = sizeof(sa_rl);
1263 	}
1264 
1265 retry:
1266 	if (writev(fd, iov, iovcnt) == -1) {
1267 		if (errno == ESRCH) {
1268 			if (hdr.rtm_type == RTM_CHANGE) {
1269 				hdr.rtm_type = RTM_ADD;
1270 				goto retry;
1271 			} else if (hdr.rtm_type == RTM_DELETE) {
1272 				log_info("route %s/%u vanished before delete",
1273 				    inet_ntoa(kroute->prefix),
1274 				    kroute->prefixlen);
1275 				return (0);
1276 			}
1277 		}
1278 		log_warn("send_rtmsg: action %u, prefix %s/%u", hdr.rtm_type,
1279 		    inet_ntoa(kroute->prefix), kroute->prefixlen);
1280 		return (0);
1281 	}
1282 
1283 	return (0);
1284 }
1285 
1286 int
1287 fetchtable(void)
1288 {
1289 	size_t			 len;
1290 	int			 mib[7];
1291 	char			*buf;
1292 	int			 rv;
1293 
1294 	mib[0] = CTL_NET;
1295 	mib[1] = PF_ROUTE;
1296 	mib[2] = 0;
1297 	mib[3] = AF_INET;
1298 	mib[4] = NET_RT_DUMP;
1299 	mib[5] = 0;
1300 	mib[6] = kr_state.rdomain;	/* rtableid */
1301 
1302 	if (sysctl(mib, 7, NULL, &len, NULL, 0) == -1) {
1303 		log_warn("sysctl");
1304 		return (-1);
1305 	}
1306 	if ((buf = malloc(len)) == NULL) {
1307 		log_warn("fetchtable");
1308 		return (-1);
1309 	}
1310 	if (sysctl(mib, 7, buf, &len, NULL, 0) == -1) {
1311 		log_warn("sysctl");
1312 		free(buf);
1313 		return (-1);
1314 	}
1315 
1316 	rv = rtmsg_process(buf, len);
1317 	free(buf);
1318 
1319 	return (rv);
1320 }
1321 
1322 int
1323 fetchifs(u_short ifindex)
1324 {
1325 	size_t			 len;
1326 	int			 mib[6];
1327 	char			*buf;
1328 	int			 rv;
1329 
1330 	mib[0] = CTL_NET;
1331 	mib[1] = PF_ROUTE;
1332 	mib[2] = 0;
1333 	mib[3] = AF_INET;
1334 	mib[4] = NET_RT_IFLIST;
1335 	mib[5] = ifindex;
1336 
1337 	if (sysctl(mib, 6, NULL, &len, NULL, 0) == -1) {
1338 		log_warn("sysctl");
1339 		return (-1);
1340 	}
1341 	if ((buf = malloc(len)) == NULL) {
1342 		log_warn("fetchif");
1343 		return (-1);
1344 	}
1345 	if (sysctl(mib, 6, buf, &len, NULL, 0) == -1) {
1346 		log_warn("sysctl");
1347 		free(buf);
1348 		return (-1);
1349 	}
1350 
1351 	rv = rtmsg_process(buf, len);
1352 	free(buf);
1353 
1354 	return (rv);
1355 }
1356 
1357 int
1358 dispatch_rtmsg(void)
1359 {
1360 	char			 buf[RT_BUF_SIZE];
1361 	ssize_t			 n;
1362 
1363 	if ((n = read(kr_state.fd, &buf, sizeof(buf))) == -1) {
1364 		if (errno == EAGAIN || errno == EINTR)
1365 			return (0);
1366 		log_warn("dispatch_rtmsg: read error");
1367 		return (-1);
1368 	}
1369 
1370 	if (n == 0) {
1371 		log_warnx("routing socket closed");
1372 		return (-1);
1373 	}
1374 
1375 	return (rtmsg_process(buf, n));
1376 }
1377 
1378 int
1379 rtmsg_process(char *buf, size_t len)
1380 {
1381 	struct rt_msghdr	*rtm;
1382 	struct if_msghdr	 ifm;
1383 	struct ifa_msghdr	*ifam;
1384 	struct sockaddr		*sa, *rti_info[RTAX_MAX];
1385 	struct sockaddr_in	*sa_in;
1386 	struct sockaddr_rtlabel	*label;
1387 	struct kroute_node	*kr, *okr;
1388 	struct in_addr		 prefix, nexthop;
1389 	u_int8_t		 prefixlen, prio;
1390 	int			 flags, mpath;
1391 	u_short			 ifindex = 0;
1392 	int			 rv, delay;
1393 
1394 	size_t			 offset;
1395 	char			*next;
1396 
1397 	for (offset = 0; offset < len; offset += rtm->rtm_msglen) {
1398 		next = buf + offset;
1399 		rtm = (struct rt_msghdr *)next;
1400 		if (len < offset + sizeof(u_short) ||
1401 		    len < offset + rtm->rtm_msglen)
1402 			fatalx("%s: partial rtm in buffer", __func__);
1403 		if (rtm->rtm_version != RTM_VERSION)
1404 			continue;
1405 
1406 		prefix.s_addr = 0;
1407 		prefixlen = 0;
1408 		nexthop.s_addr = 0;
1409 		mpath = 0;
1410 		prio = 0;
1411 		flags = F_KERNEL;
1412 
1413 		sa = (struct sockaddr *)(next + rtm->rtm_hdrlen);
1414 		get_rtaddrs(rtm->rtm_addrs, sa, rti_info);
1415 
1416 		switch (rtm->rtm_type) {
1417 		case RTM_ADD:
1418 		case RTM_GET:
1419 		case RTM_CHANGE:
1420 		case RTM_DELETE:
1421 			if (rtm->rtm_errno)		/* failed attempts... */
1422 				continue;
1423 
1424 			if (rtm->rtm_tableid != kr_state.rdomain)
1425 				continue;
1426 
1427 			if (rtm->rtm_type == RTM_GET &&
1428 			    rtm->rtm_pid != kr_state.pid)
1429 				continue;
1430 
1431 			if ((sa = rti_info[RTAX_DST]) == NULL)
1432 				continue;
1433 
1434 			/* Skip ARP/ND cache and broadcast routes. */
1435 			if (rtm->rtm_flags & (RTF_LLINFO|RTF_BROADCAST))
1436 				continue;
1437 
1438 			if (rtm->rtm_flags & RTF_MPATH)
1439 				mpath = 1;
1440 			prio = rtm->rtm_priority;
1441 			flags = (prio == kr_state.fib_prio) ?
1442 			    F_OSPFD_INSERTED : F_KERNEL;
1443 
1444 			switch (sa->sa_family) {
1445 			case AF_INET:
1446 				prefix.s_addr =
1447 				    ((struct sockaddr_in *)sa)->sin_addr.s_addr;
1448 				sa_in = (struct sockaddr_in *)
1449 				    rti_info[RTAX_NETMASK];
1450 				if (sa_in != NULL) {
1451 					if (sa_in->sin_len != 0)
1452 						prefixlen = mask2prefixlen(
1453 						    sa_in->sin_addr.s_addr);
1454 				} else if (rtm->rtm_flags & RTF_HOST)
1455 					prefixlen = 32;
1456 				else
1457 					prefixlen =
1458 					    prefixlen_classful(prefix.s_addr);
1459 				if (rtm->rtm_flags & RTF_STATIC)
1460 					flags |= F_STATIC;
1461 				if (rtm->rtm_flags & RTF_BLACKHOLE)
1462 					flags |= F_BLACKHOLE;
1463 				if (rtm->rtm_flags & RTF_REJECT)
1464 					flags |= F_REJECT;
1465 				if (rtm->rtm_flags & RTF_DYNAMIC)
1466 					flags |= F_DYNAMIC;
1467 				break;
1468 			default:
1469 				continue;
1470 			}
1471 
1472 			ifindex = rtm->rtm_index;
1473 			if ((sa = rti_info[RTAX_GATEWAY]) != NULL) {
1474 				switch (sa->sa_family) {
1475 				case AF_INET:
1476 					if (rtm->rtm_flags & RTF_CONNECTED)
1477 						flags |= F_CONNECTED;
1478 
1479 					nexthop.s_addr = ((struct
1480 					    sockaddr_in *)sa)->sin_addr.s_addr;
1481 					break;
1482 				case AF_LINK:
1483 					/*
1484 					 * Traditional BSD connected routes have
1485 					 * a gateway of type AF_LINK.
1486 					 */
1487 					flags |= F_CONNECTED;
1488 					break;
1489 				}
1490 			}
1491 		}
1492 
1493 		switch (rtm->rtm_type) {
1494 		case RTM_ADD:
1495 		case RTM_GET:
1496 		case RTM_CHANGE:
1497 			if (nexthop.s_addr == 0 && !(flags & F_CONNECTED)) {
1498 				log_warnx("no nexthop for %s/%u",
1499 				    inet_ntoa(prefix), prefixlen);
1500 				continue;
1501 			}
1502 
1503 			if ((okr = kroute_find(prefix.s_addr, prefixlen, prio))
1504 			    != NULL) {
1505 				/* get the correct route */
1506 				kr = okr;
1507 				if ((mpath || prio == kr_state.fib_prio) &&
1508 				    (kr = kroute_matchgw(okr, nexthop)) ==
1509 				    NULL) {
1510 					log_warnx("%s: mpath route not found",
1511 					    __func__);
1512 					/* add routes we missed out earlier */
1513 					goto add;
1514 				}
1515 
1516 				if (kr->r.flags & F_REDISTRIBUTED)
1517 					flags |= F_REDISTRIBUTED;
1518 				kr->r.nexthop.s_addr = nexthop.s_addr;
1519 				kr->r.flags = flags;
1520 				kr->r.ifindex = ifindex;
1521 
1522 				rtlabel_unref(kr->r.rtlabel);
1523 				kr->r.rtlabel = 0;
1524 				kr->r.ext_tag = 0;
1525 				if ((label = (struct sockaddr_rtlabel *)
1526 				    rti_info[RTAX_LABEL]) != NULL) {
1527 					kr->r.rtlabel =
1528 					    rtlabel_name2id(label->sr_label);
1529 					kr->r.ext_tag =
1530 					    rtlabel_id2tag(kr->r.rtlabel);
1531 				}
1532 
1533 				if (kif_validate(kr->r.ifindex))
1534 					kr->r.flags &= ~F_DOWN;
1535 				else
1536 					kr->r.flags |= F_DOWN;
1537 
1538 				/* just readd, the RDE will care */
1539 				kr->serial = kr_state.fib_serial;
1540 				kr_redistribute(okr);
1541 			} else {
1542 add:
1543 				if ((kr = calloc(1,
1544 				    sizeof(struct kroute_node))) == NULL) {
1545 					log_warn("%s: calloc", __func__);
1546 					return (-1);
1547 				}
1548 
1549 				kr->r.prefix.s_addr = prefix.s_addr;
1550 				kr->r.prefixlen = prefixlen;
1551 				kr->r.nexthop.s_addr = nexthop.s_addr;
1552 				kr->r.flags = flags;
1553 				kr->r.ifindex = ifindex;
1554 				kr->r.priority = prio;
1555 
1556 				if (rtm->rtm_priority == kr_state.fib_prio) {
1557 					log_warnx("alien OSPF route %s/%d",
1558 					    inet_ntoa(prefix), prefixlen);
1559 					rv = send_rtmsg(kr_state.fd,
1560 					    RTM_DELETE, &kr->r);
1561 					free(kr);
1562 					if (rv == -1)
1563 						return (-1);
1564 				} else {
1565 					if ((label = (struct sockaddr_rtlabel *)
1566 					    rti_info[RTAX_LABEL]) != NULL) {
1567 						kr->r.rtlabel =
1568 						    rtlabel_name2id(
1569 						    label->sr_label);
1570 						kr->r.ext_tag =
1571 						    rtlabel_id2tag(
1572 						    kr->r.rtlabel);
1573 					}
1574 
1575 					kroute_insert(kr);
1576 				}
1577 			}
1578 			break;
1579 		case RTM_DELETE:
1580 			if ((kr = kroute_find(prefix.s_addr, prefixlen, prio))
1581 			    == NULL)
1582 				continue;
1583 			if (!(kr->r.flags & F_KERNEL))
1584 				continue;
1585 			/* get the correct route */
1586 			okr = kr;
1587 			if (mpath &&
1588 			    (kr = kroute_matchgw(kr, nexthop)) == NULL) {
1589 				log_warnx("%s: mpath route not found",
1590 				    __func__);
1591 				return (-1);
1592 			}
1593 			if (kroute_remove(kr) == -1)
1594 				return (-1);
1595 			break;
1596 		case RTM_IFINFO:
1597 			memcpy(&ifm, next, sizeof(ifm));
1598 			if_change(ifm.ifm_index, ifm.ifm_flags, &ifm.ifm_data,
1599 			    (struct sockaddr_dl *)rti_info[RTAX_IFP]);
1600 			break;
1601 		case RTM_NEWADDR:
1602 			ifam = (struct ifa_msghdr *)rtm;
1603 			if ((ifam->ifam_addrs & (RTA_NETMASK | RTA_IFA |
1604 			    RTA_BRD)) == 0)
1605 				break;
1606 
1607 			if_newaddr(ifam->ifam_index,
1608 			    (struct sockaddr_in *)rti_info[RTAX_IFA],
1609 			    (struct sockaddr_in *)rti_info[RTAX_NETMASK],
1610 			    (struct sockaddr_in *)rti_info[RTAX_BRD]);
1611 			break;
1612 		case RTM_DELADDR:
1613 			ifam = (struct ifa_msghdr *)rtm;
1614 			if ((ifam->ifam_addrs & (RTA_NETMASK | RTA_IFA |
1615 			    RTA_BRD)) == 0)
1616 				break;
1617 
1618 			if_deladdr(ifam->ifam_index,
1619 			    (struct sockaddr_in *)rti_info[RTAX_IFA],
1620 			    (struct sockaddr_in *)rti_info[RTAX_NETMASK],
1621 			    (struct sockaddr_in *)rti_info[RTAX_BRD]);
1622 			break;
1623 		case RTM_IFANNOUNCE:
1624 			if_announce(next);
1625 			break;
1626 		case RTM_DESYNC:
1627 			/*
1628 			 * We lost some routing packets. Schedule a reload
1629 			 * of the kernel route/interface information.
1630 			 */
1631 			if (kr_state.reload_state == KR_RELOAD_IDLE) {
1632 				delay = KR_RELOAD_TIMER;
1633 				log_info("desync; scheduling fib reload");
1634 			} else {
1635 				delay = KR_RELOAD_HOLD_TIMER;
1636 				log_debug("desync during KR_RELOAD_%s",
1637 				    kr_state.reload_state ==
1638 				    KR_RELOAD_FETCH ? "FETCH" : "HOLD");
1639 			}
1640 			kr_state.reload_state = KR_RELOAD_FETCH;
1641 			kr_fib_reload_arm_timer(delay);
1642 			break;
1643 		default:
1644 			/* ignore for now */
1645 			break;
1646 		}
1647 	}
1648 
1649 	return (offset);
1650 }
1651