xref: /openbsd-src/usr.sbin/ospfd/kroute.c (revision d13be5d47e4149db2549a9828e244d59dbc43f15)
1 /*	$OpenBSD: kroute.c,v 1.90 2011/07/04 04:34:14 claudio 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/param.h>
21 #include <sys/types.h>
22 #include <sys/socket.h>
23 #include <sys/sysctl.h>
24 #include <sys/tree.h>
25 #include <sys/uio.h>
26 #include <netinet/in.h>
27 #include <arpa/inet.h>
28 #include <net/if.h>
29 #include <net/if_dl.h>
30 #include <net/if_types.h>
31 #include <net/route.h>
32 #include <err.h>
33 #include <errno.h>
34 #include <fcntl.h>
35 #include <stdio.h>
36 #include <stdlib.h>
37 #include <string.h>
38 #include <unistd.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 	int			fd;
49 	struct event		ev;
50 	u_int			rdomain;
51 } kr_state;
52 
53 struct kroute_node {
54 	RB_ENTRY(kroute_node)	 entry;
55 	struct kroute_node	*next;
56 	struct kroute		 r;
57 	int			 serial;
58 };
59 
60 struct kif_node {
61 	RB_ENTRY(kif_node)	 entry;
62 	TAILQ_HEAD(, kif_addr)	 addrs;
63 	struct kif		 k;
64 };
65 
66 void	kr_redist_remove(struct kroute_node *, struct kroute_node *);
67 int	kr_redist_eval(struct kroute *, struct kroute *);
68 void	kr_redistribute(struct kroute_node *);
69 int	kroute_compare(struct kroute_node *, struct kroute_node *);
70 int	kif_compare(struct kif_node *, struct kif_node *);
71 int	kr_change_fib(struct kroute_node *, struct kroute *, int, int);
72 int	kr_delete_fib(struct kroute_node *);
73 
74 struct kroute_node	*kroute_find(in_addr_t, u_int8_t, u_int8_t);
75 struct kroute_node	*kroute_matchgw(struct kroute_node *, struct in_addr);
76 int			 kroute_insert(struct kroute_node *);
77 int			 kroute_remove(struct kroute_node *);
78 void			 kroute_clear(void);
79 
80 struct kif_node		*kif_find(u_short);
81 struct kif_node		*kif_insert(u_short);
82 int			 kif_remove(struct kif_node *);
83 void			 kif_clear(void);
84 struct kif		*kif_update(u_short, int, struct if_data *,
85 			    struct sockaddr_dl *);
86 int			 kif_validate(u_short);
87 
88 struct kroute_node	*kroute_match(in_addr_t);
89 
90 int		protect_lo(void);
91 u_int8_t	prefixlen_classful(in_addr_t);
92 void		get_rtaddrs(int, struct sockaddr *, struct sockaddr **);
93 void		if_change(u_short, int, struct if_data *, struct sockaddr_dl *);
94 void		if_newaddr(u_short, struct sockaddr_in *, struct sockaddr_in *,
95 		    struct sockaddr_in *);
96 void		if_deladdr(u_short, struct sockaddr_in *, struct sockaddr_in *,
97 		    struct sockaddr_in *);
98 void		if_announce(void *);
99 
100 int		send_rtmsg(int, int, struct kroute *);
101 int		dispatch_rtmsg(void);
102 int		fetchtable(void);
103 int		fetchifs(u_short);
104 int		rtmsg_process(char *, int);
105 
106 RB_HEAD(kroute_tree, kroute_node)	krt;
107 RB_PROTOTYPE(kroute_tree, kroute_node, entry, kroute_compare)
108 RB_GENERATE(kroute_tree, kroute_node, entry, kroute_compare)
109 
110 RB_HEAD(kif_tree, kif_node)		kit;
111 RB_PROTOTYPE(kif_tree, kif_node, entry, kif_compare)
112 RB_GENERATE(kif_tree, kif_node, entry, kif_compare)
113 
114 int
115 kif_init(void)
116 {
117 	RB_INIT(&kit);
118 	/* init also krt tree so that we can call kr_shutdown() */
119 	RB_INIT(&krt);
120 	kr_state.fib_sync = 0;	/* decoupled */
121 
122 	if (fetchifs(0) == -1)
123 		return (-1);
124 
125 	return (0);
126 }
127 
128 int
129 kr_init(int fs, u_int rdomain)
130 {
131 	int		opt = 0, rcvbuf, default_rcvbuf;
132 	socklen_t	optlen;
133 
134 	kr_state.fib_sync = fs;
135 	kr_state.rdomain = rdomain;
136 
137 	if ((kr_state.fd = socket(AF_ROUTE, SOCK_RAW, AF_INET)) == -1) {
138 		log_warn("kr_init: socket");
139 		return (-1);
140 	}
141 
142 	/* not interested in my own messages */
143 	if (setsockopt(kr_state.fd, SOL_SOCKET, SO_USELOOPBACK,
144 	    &opt, sizeof(opt)) == -1)
145 		log_warn("kr_init: setsockopt");	/* not fatal */
146 
147 	/* grow receive buffer, don't wanna miss messages */
148 	optlen = sizeof(default_rcvbuf);
149 	if (getsockopt(kr_state.fd, SOL_SOCKET, SO_RCVBUF,
150 	    &default_rcvbuf, &optlen) == -1)
151 		log_warn("kr_init getsockopt SOL_SOCKET SO_RCVBUF");
152 	else
153 		for (rcvbuf = MAX_RTSOCK_BUF;
154 		    rcvbuf > default_rcvbuf &&
155 		    setsockopt(kr_state.fd, SOL_SOCKET, SO_RCVBUF,
156 		    &rcvbuf, sizeof(rcvbuf)) == -1 && errno == ENOBUFS;
157 		    rcvbuf /= 2)
158 			;	/* nothing */
159 
160 	kr_state.pid = getpid();
161 	kr_state.rtseq = 1;
162 
163 	if (fetchtable() == -1)
164 		return (-1);
165 
166 	if (protect_lo() == -1)
167 		return (-1);
168 
169 	event_set(&kr_state.ev, kr_state.fd, EV_READ | EV_PERSIST,
170 	    kr_dispatch_msg, NULL);
171 	event_add(&kr_state.ev, NULL);
172 
173 	return (0);
174 }
175 
176 int
177 kr_change_fib(struct kroute_node *kr, struct kroute *kroute, int krcount,
178     int action)
179 {
180 	int			 i;
181 	struct kroute_node	*kn, *nkn;
182 
183 	if (action == RTM_ADD) {
184 		/*
185 		 * First remove all stale multipath routes.
186 		 * This step must be skipped when the action is RTM_CHANGE
187 		 * because it is already a single path route that will be
188 		 * changed.
189 		 */
190 		for (kn = kr; kn != NULL; kn = nkn) {
191 			for (i = 0; i < krcount; i++) {
192 				if (kn->r.nexthop.s_addr ==
193 				    kroute[i].nexthop.s_addr)
194 					break;
195 			}
196 			nkn = kn->next;
197 			if (i == krcount) {
198 				/* stale route */
199 				if (kr_delete_fib(kn) == -1)
200 					log_warnx("kr_delete_fib failed");
201 				/*
202 				 * if head element was removed we need to adjust
203 				 * the head
204 				 */
205 				if (kr == kn)
206 					kr = nkn;
207 			}
208 		}
209 	}
210 
211 	/*
212 	 * now add or change the route
213 	 */
214 	for (i = 0; i < krcount; i++) {
215 		/* nexthop within 127/8 -> ignore silently */
216 		if ((kroute[i].nexthop.s_addr & htonl(IN_CLASSA_NET)) ==
217 		    htonl(INADDR_LOOPBACK & IN_CLASSA_NET))
218 			continue;
219 
220 		if (action == RTM_ADD && kr) {
221 			for (kn = kr; kn != NULL; kn = kn->next) {
222 				if (kn->r.nexthop.s_addr ==
223 				    kroute[i].nexthop.s_addr)
224 					break;
225 			}
226 
227 			if (kn != NULL)
228 				/* nexthop already present, skip it */
229 				continue;
230 		} else
231 			/* modify first entry */
232 			kn = kr;
233 
234 		/* send update */
235 		if (send_rtmsg(kr_state.fd, action, &kroute[i]) == -1)
236 			return (-1);
237 
238 		/* create new entry unless we are changing the first entry */
239 		if (action == RTM_ADD)
240 			if ((kn = calloc(1, sizeof(*kn))) == NULL)
241 				fatal(NULL);
242 
243 		kn->r.prefix.s_addr = kroute[i].prefix.s_addr;
244 		kn->r.prefixlen = kroute[i].prefixlen;
245 		kn->r.nexthop.s_addr = kroute[i].nexthop.s_addr;
246 		kn->r.flags = kroute[i].flags | F_OSPFD_INSERTED;
247 		kn->r.priority = RTP_OSPF;
248 		kn->r.ext_tag = kroute[i].ext_tag;
249 		rtlabel_unref(kn->r.rtlabel);	/* for RTM_CHANGE */
250 		kn->r.rtlabel = kroute[i].rtlabel;
251 
252 		if (action == RTM_ADD)
253 			if (kroute_insert(kn) == -1) {
254 				log_debug("kr_update_fib: cannot insert %s",
255 				    inet_ntoa(kn->r.nexthop));
256 				free(kn);
257 			}
258 		action = RTM_ADD;
259 	}
260 	return  (0);
261 }
262 
263 int
264 kr_change(struct kroute *kroute, int krcount)
265 {
266 	struct kroute_node	*kr;
267 	int			 action = RTM_ADD;
268 
269 	kroute->rtlabel = rtlabel_tag2id(kroute->ext_tag);
270 
271 	kr = kroute_find(kroute->prefix.s_addr, kroute->prefixlen, RTP_OSPF);
272 	if (kr != NULL && kr->next == NULL && krcount == 1)
273 		/* single path OSPF route */
274 		action = RTM_CHANGE;
275 
276 	return (kr_change_fib(kr, kroute, krcount, action));
277 }
278 
279 int
280 kr_delete_fib(struct kroute_node *kr)
281 {
282 	if (kr->r.priority != RTP_OSPF)
283 		log_warn("kr_delete_fib: %s/%d has wrong priority %d",
284 		    inet_ntoa(kr->r.prefix), kr->r.prefixlen, kr->r.priority);
285 
286 	if (send_rtmsg(kr_state.fd, RTM_DELETE, &kr->r) == -1)
287 		return (-1);
288 
289 	if (kroute_remove(kr) == -1)
290 		return (-1);
291 
292 	return (0);
293 }
294 
295 int
296 kr_delete(struct kroute *kroute)
297 {
298 	struct kroute_node	*kr, *nkr;
299 
300 	if ((kr = kroute_find(kroute->prefix.s_addr, kroute->prefixlen,
301 	    RTP_OSPF)) == NULL)
302 		return (0);
303 
304 	while (kr != NULL) {
305 		nkr = kr->next;
306 		if (kr_delete_fib(kr) == -1)
307 			return (-1);
308 		kr = nkr;
309 	}
310 	return (0);
311 }
312 
313 void
314 kr_shutdown(void)
315 {
316 	kr_fib_decouple();
317 	kroute_clear();
318 	kif_clear();
319 }
320 
321 void
322 kr_fib_couple(void)
323 {
324 	struct kroute_node	*kr;
325 	struct kroute_node	*kn;
326 
327 	if (kr_state.fib_sync == 1)	/* already coupled */
328 		return;
329 
330 	kr_state.fib_sync = 1;
331 
332 	RB_FOREACH(kr, kroute_tree, &krt)
333 		if (kr->r.priority == RTP_OSPF)
334 			for (kn = kr; kn != NULL; kn = kn->next)
335 				send_rtmsg(kr_state.fd, RTM_ADD, &kn->r);
336 
337 	log_info("kernel routing table coupled");
338 }
339 
340 void
341 kr_fib_decouple(void)
342 {
343 	struct kroute_node	*kr;
344 	struct kroute_node	*kn;
345 
346 	if (kr_state.fib_sync == 0)	/* already decoupled */
347 		return;
348 
349 	RB_FOREACH(kr, kroute_tree, &krt)
350 		if (kr->r.priority == RTP_OSPF)
351 			for (kn = kr; kn != NULL; kn = kn->next)
352 				send_rtmsg(kr_state.fd, RTM_DELETE, &kn->r);
353 
354 	kr_state.fib_sync = 0;
355 
356 	log_info("kernel routing table decoupled");
357 }
358 
359 void
360 kr_fib_reload(void)
361 {
362 	struct kroute_node	*krn, *kr, *kn;
363 
364 	log_info("reloading interface list and routing table");
365 
366 	kr_state.fib_serial++;
367 
368 	if (fetchifs(0) == -1 || fetchtable() == -1)
369 		return;
370 
371 	for (kr = RB_MIN(kroute_tree, &krt); kr != NULL; kr = krn) {
372 		krn = RB_NEXT(kroute_tree, &krt, kr);
373 
374 		do {
375 			kn = kr->next;
376 
377 			if (kr->serial != kr_state.fib_serial) {
378 				if (kr->r.priority == RTP_OSPF) {
379 					kr->serial = kr_state.fib_serial;
380 					if (send_rtmsg(kr_state.fd,
381 					    RTM_ADD, &kr->r) != 0)
382 						break;
383 				} else
384 					kroute_remove(kr);
385 			}
386 
387 		} while ((kr = kn) != NULL);
388 	}
389 }
390 
391 /* ARGSUSED */
392 void
393 kr_dispatch_msg(int fd, short event, void *bula)
394 {
395 	/* XXX this is stupid */
396 	if (dispatch_rtmsg() == -1)
397 		event_loopexit(NULL);
398 }
399 
400 void
401 kr_show_route(struct imsg *imsg)
402 {
403 	struct kroute_node	*kr;
404 	struct kroute_node	*kn;
405 	int			 flags;
406 	struct in_addr		 addr;
407 
408 	switch (imsg->hdr.type) {
409 	case IMSG_CTL_KROUTE:
410 		if (imsg->hdr.len != IMSG_HEADER_SIZE + sizeof(flags)) {
411 			log_warnx("kr_show_route: wrong imsg len");
412 			return;
413 		}
414 		memcpy(&flags, imsg->data, sizeof(flags));
415 		RB_FOREACH(kr, kroute_tree, &krt)
416 			if (!flags || kr->r.flags & flags) {
417 				kn = kr;
418 				do {
419 					main_imsg_compose_ospfe(IMSG_CTL_KROUTE,
420 					    imsg->hdr.pid,
421 					    &kn->r, sizeof(kn->r));
422 				} while ((kn = kn->next) != NULL);
423 			}
424 		break;
425 	case IMSG_CTL_KROUTE_ADDR:
426 		if (imsg->hdr.len != IMSG_HEADER_SIZE +
427 		    sizeof(struct in_addr)) {
428 			log_warnx("kr_show_route: wrong imsg len");
429 			return;
430 		}
431 		memcpy(&addr, imsg->data, sizeof(addr));
432 		kr = NULL;
433 		kr = kroute_match(addr.s_addr);
434 		if (kr != NULL)
435 			main_imsg_compose_ospfe(IMSG_CTL_KROUTE, imsg->hdr.pid,
436 			    &kr->r, sizeof(kr->r));
437 		break;
438 	default:
439 		log_debug("kr_show_route: error handling imsg");
440 		break;
441 	}
442 
443 	main_imsg_compose_ospfe(IMSG_CTL_END, imsg->hdr.pid, NULL, 0);
444 }
445 
446 void
447 kr_ifinfo(char *ifname, pid_t pid)
448 {
449 	struct kif_node	*kif;
450 
451 	RB_FOREACH(kif, kif_tree, &kit)
452 		if (ifname == NULL || !strcmp(ifname, kif->k.ifname)) {
453 			main_imsg_compose_ospfe(IMSG_CTL_IFINFO,
454 			    pid, &kif->k, sizeof(kif->k));
455 		}
456 
457 	main_imsg_compose_ospfe(IMSG_CTL_END, pid, NULL, 0);
458 }
459 
460 void
461 kr_redist_remove(struct kroute_node *kh, struct kroute_node *kn)
462 {
463 	struct kroute	*rr;
464 
465 	/* was the route redistributed? */
466 	if ((kn->r.flags & F_REDISTRIBUTED) == 0)
467 		return;
468 
469 	/* remove redistributed flag */
470 	kn->r.flags &= ~F_REDISTRIBUTED;
471 	rr = &kn->r;
472 
473 	/* probably inform the RDE (check if no other path is redistributed) */
474 	for (kn = kh; kn; kn = kn->next)
475 		if (kn->r.flags & F_REDISTRIBUTED)
476 			break;
477 
478 	if (kn == NULL)
479 		main_imsg_compose_rde(IMSG_NETWORK_DEL, 0, rr,
480 		    sizeof(struct kroute));
481 }
482 
483 int
484 kr_redist_eval(struct kroute *kr, struct kroute *rr)
485 {
486 	u_int32_t	 a, metric = 0;
487 
488 	/* Only non-ospfd routes are considered for redistribution. */
489 	if (!(kr->flags & F_KERNEL))
490 		goto dont_redistribute;
491 
492 	/* Dynamic routes are not redistributable. */
493 	if (kr->flags & F_DYNAMIC)
494 		goto dont_redistribute;
495 
496 	/* interface is not up and running so don't announce */
497 	if (kr->flags & F_DOWN)
498 		goto dont_redistribute;
499 
500 	/*
501 	 * We consider the loopback net, multicast and experimental addresses
502 	 * as not redistributable.
503 	 */
504 	a = ntohl(kr->prefix.s_addr);
505 	if (IN_MULTICAST(a) || IN_BADCLASS(a) ||
506 	    (a >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET)
507 		goto dont_redistribute;
508 	/*
509 	 * Consider networks with nexthop loopback as not redistributable
510 	 * unless it is a reject or blackhole route.
511 	 */
512 	if (kr->nexthop.s_addr == htonl(INADDR_LOOPBACK) &&
513 	    !(kr->flags & (F_BLACKHOLE|F_REJECT)))
514 		goto dont_redistribute;
515 
516 	/* Should we redistribute this route? */
517 	if (!ospf_redistribute(kr, &metric))
518 		goto dont_redistribute;
519 
520 	/* prefix should be redistributed */
521 	kr->flags |= F_REDISTRIBUTED;
522 	/*
523 	 * only one of all multipath routes can be redistributed so
524 	 * redistribute the best one.
525 	 */
526 	if (rr->metric > metric) {
527 		*rr = *kr;
528 		rr->metric = metric;
529 	}
530 
531 	return (1);
532 
533 dont_redistribute:
534 	/* was the route redistributed? */
535 	if ((kr->flags & F_REDISTRIBUTED) == 0)
536 		return (0);
537 
538 	kr->flags &= ~F_REDISTRIBUTED;
539 	return (1);
540 }
541 
542 void
543 kr_redistribute(struct kroute_node *kh)
544 {
545 	struct kroute_node	*kn;
546 	struct kroute		 rr;
547 	int			 redistribute = 0;
548 
549 	/* only the highest prio route can be redistributed */
550 	if (kroute_find(kh->r.prefix.s_addr, kh->r.prefixlen, RTP_ANY) != kh)
551 		return;
552 
553 	bzero(&rr, sizeof(rr));
554 	rr.metric = UINT_MAX;
555 	for (kn = kh; kn; kn = kn->next)
556 		if (kr_redist_eval(&kn->r, &rr))
557 			redistribute = 1;
558 
559 	if (!redistribute)
560 		return;
561 
562 	if (rr.flags & F_REDISTRIBUTED) {
563 		main_imsg_compose_rde(IMSG_NETWORK_ADD, 0, &rr,
564 		    sizeof(struct kroute));
565 	} else {
566 		rr = kh->r;
567 		main_imsg_compose_rde(IMSG_NETWORK_DEL, 0, &rr,
568 		    sizeof(struct kroute));
569 	}
570 }
571 
572 void
573 kr_reload(void)
574 {
575 	struct kroute_node	*kr, *kn;
576 	u_int32_t		 dummy;
577 	int			 r;
578 
579 	RB_FOREACH(kr, kroute_tree, &krt) {
580 		for (kn = kr; kn; kn = kn->next) {
581 			r = ospf_redistribute(&kn->r, &dummy);
582 			/*
583 			 * if it is redistributed, redistribute again metric
584 			 * may have changed.
585 			 */
586 			if ((kn->r.flags & F_REDISTRIBUTED && !r) || r)
587 				break;
588 		}
589 		if (kn) {
590 			/*
591 			 * kr_redistribute copes with removes and RDE with
592 			 * duplicates
593 			 */
594 			kr_redistribute(kr);
595 		}
596 	}
597 }
598 
599 /* rb-tree compare */
600 int
601 kroute_compare(struct kroute_node *a, struct kroute_node *b)
602 {
603 	if (ntohl(a->r.prefix.s_addr) < ntohl(b->r.prefix.s_addr))
604 		return (-1);
605 	if (ntohl(a->r.prefix.s_addr) > ntohl(b->r.prefix.s_addr))
606 		return (1);
607 	if (a->r.prefixlen < b->r.prefixlen)
608 		return (-1);
609 	if (a->r.prefixlen > b->r.prefixlen)
610 		return (1);
611 
612 	/* if the priority is RTP_ANY finish on the first address hit */
613 	if (a->r.priority == RTP_ANY || b->r.priority == RTP_ANY)
614 		return (0);
615 	if (a->r.priority < b->r.priority)
616 		return (-1);
617 	if (a->r.priority > b->r.priority)
618 		return (1);
619 	return (0);
620 }
621 
622 int
623 kif_compare(struct kif_node *a, struct kif_node *b)
624 {
625 	return (b->k.ifindex - a->k.ifindex);
626 }
627 
628 /* tree management */
629 struct kroute_node *
630 kroute_find(in_addr_t prefix, u_int8_t prefixlen, u_int8_t prio)
631 {
632 	struct kroute_node	s;
633 	struct kroute_node	*kn, *tmp;
634 
635 	s.r.prefix.s_addr = prefix;
636 	s.r.prefixlen = prefixlen;
637 	s.r.priority = prio;
638 
639 	kn = RB_FIND(kroute_tree, &krt, &s);
640 	if (kn && prio == RTP_ANY) {
641 		tmp = RB_PREV(kroute_tree, &krt, kn);
642 		while (tmp) {
643 			if (kroute_compare(&s, tmp) == 0)
644 				kn = tmp;
645 			else
646 				break;
647 			tmp = RB_PREV(kroute_tree, &krt, kn);
648 		}
649 	}
650 	return (kn);
651 }
652 
653 struct kroute_node *
654 kroute_matchgw(struct kroute_node *kr, struct in_addr nh)
655 {
656 	in_addr_t	nexthop;
657 
658 	nexthop = nh.s_addr;
659 
660 	while (kr) {
661 		if (kr->r.nexthop.s_addr == nexthop)
662 			return (kr);
663 		kr = kr->next;
664 	}
665 
666 	return (NULL);
667 }
668 
669 int
670 kroute_insert(struct kroute_node *kr)
671 {
672 	struct kroute_node	*krm, *krh;
673 
674 	kr->serial = kr_state.fib_serial;
675 
676 	if ((krh = RB_INSERT(kroute_tree, &krt, kr)) != NULL) {
677 		/*
678 		 * Multipath route, add at end of list.
679 		 */
680 		krm = krh;
681 		while (krm->next != NULL)
682 			krm = krm->next;
683 		krm->next = kr;
684 		kr->next = NULL; /* to be sure */
685 	} else
686 		krh = kr;
687 
688 	if (!(kr->r.flags & F_KERNEL)) {
689 		/* don't validate or redistribute ospf route */
690 		kr->r.flags &= ~F_DOWN;
691 		return (0);
692 	}
693 
694 	if (kif_validate(kr->r.ifindex))
695 		kr->r.flags &= ~F_DOWN;
696 	else
697 		kr->r.flags |= F_DOWN;
698 
699 	kr_redistribute(krh);
700 	return (0);
701 }
702 
703 int
704 kroute_remove(struct kroute_node *kr)
705 {
706 	struct kroute_node	*krm;
707 
708 	if ((krm = RB_FIND(kroute_tree, &krt, kr)) == NULL) {
709 		log_warnx("kroute_remove failed to find %s/%u",
710 		    inet_ntoa(kr->r.prefix), kr->r.prefixlen);
711 		return (-1);
712 	}
713 
714 	if (krm == kr) {
715 		/* head element */
716 		if (RB_REMOVE(kroute_tree, &krt, kr) == NULL) {
717 			log_warnx("kroute_remove failed for %s/%u",
718 			    inet_ntoa(kr->r.prefix), kr->r.prefixlen);
719 			return (-1);
720 		}
721 		if (kr->next != NULL) {
722 			if (RB_INSERT(kroute_tree, &krt, kr->next) != NULL) {
723 				log_warnx("kroute_remove failed to add %s/%u",
724 				    inet_ntoa(kr->r.prefix), kr->r.prefixlen);
725 				return (-1);
726 			}
727 		}
728 	} else {
729 		/* somewhere in the list */
730 		while (krm->next != kr && krm->next != NULL)
731 			krm = krm->next;
732 		if (krm->next == NULL) {
733 			log_warnx("kroute_remove multipath list corrupted "
734 			    "for %s/%u", inet_ntoa(kr->r.prefix),
735 			    kr->r.prefixlen);
736 			return (-1);
737 		}
738 		krm->next = kr->next;
739 	}
740 
741 	kr_redist_remove(krm, kr);
742 	rtlabel_unref(kr->r.rtlabel);
743 
744 	free(kr);
745 	return (0);
746 }
747 
748 void
749 kroute_clear(void)
750 {
751 	struct kroute_node	*kr;
752 
753 	while ((kr = RB_MIN(kroute_tree, &krt)) != NULL)
754 		kroute_remove(kr);
755 }
756 
757 struct kif_node *
758 kif_find(u_short ifindex)
759 {
760 	struct kif_node	s;
761 
762 	bzero(&s, sizeof(s));
763 	s.k.ifindex = ifindex;
764 
765 	return (RB_FIND(kif_tree, &kit, &s));
766 }
767 
768 struct kif *
769 kif_findname(char *ifname, struct in_addr addr, struct kif_addr **kap)
770 {
771 	struct kif_node	*kif;
772 	struct kif_addr	*ka;
773 
774 	RB_FOREACH(kif, kif_tree, &kit)
775 		if (!strcmp(ifname, kif->k.ifname)) {
776 			ka = TAILQ_FIRST(&kif->addrs);
777 			if (addr.s_addr != 0) {
778 				TAILQ_FOREACH(ka, &kif->addrs, entry) {
779 					if (addr.s_addr == ka->addr.s_addr)
780 						break;
781 				}
782 			}
783 			if (kap != NULL)
784 				*kap = ka;
785 			return (&kif->k);
786 		}
787 
788 	return (NULL);
789 }
790 
791 struct kif_node *
792 kif_insert(u_short ifindex)
793 {
794 	struct kif_node	*kif;
795 
796 	if ((kif = calloc(1, sizeof(struct kif_node))) == NULL)
797 		return (NULL);
798 
799 	kif->k.ifindex = ifindex;
800 	TAILQ_INIT(&kif->addrs);
801 
802 	if (RB_INSERT(kif_tree, &kit, kif) != NULL)
803 		fatalx("kif_insert: RB_INSERT");
804 
805 	return (kif);
806 }
807 
808 int
809 kif_remove(struct kif_node *kif)
810 {
811 	struct kif_addr	*ka;
812 
813 	if (RB_REMOVE(kif_tree, &kit, kif) == NULL) {
814 		log_warnx("RB_REMOVE(kif_tree, &kit, kif)");
815 		return (-1);
816 	}
817 
818 	while ((ka = TAILQ_FIRST(&kif->addrs)) != NULL) {
819 		TAILQ_REMOVE(&kif->addrs, ka, entry);
820 		free(ka);
821 	}
822 	free(kif);
823 	return (0);
824 }
825 
826 void
827 kif_clear(void)
828 {
829 	struct kif_node	*kif;
830 
831 	while ((kif = RB_MIN(kif_tree, &kit)) != NULL)
832 		kif_remove(kif);
833 }
834 
835 struct kif *
836 kif_update(u_short ifindex, int flags, struct if_data *ifd,
837     struct sockaddr_dl *sdl)
838 {
839 	struct kif_node		*kif;
840 
841 	if ((kif = kif_find(ifindex)) == NULL) {
842 		if ((kif = kif_insert(ifindex)) == NULL)
843 			return (NULL);
844 		kif->k.nh_reachable = (flags & IFF_UP) &&
845 		    LINK_STATE_IS_UP(ifd->ifi_link_state);
846 	}
847 
848 	kif->k.flags = flags;
849 	kif->k.link_state = ifd->ifi_link_state;
850 	kif->k.media_type = ifd->ifi_type;
851 	kif->k.baudrate = ifd->ifi_baudrate;
852 	kif->k.mtu = ifd->ifi_mtu;
853 
854 	if (sdl && sdl->sdl_family == AF_LINK) {
855 		if (sdl->sdl_nlen >= sizeof(kif->k.ifname))
856 			memcpy(kif->k.ifname, sdl->sdl_data,
857 			    sizeof(kif->k.ifname) - 1);
858 		else if (sdl->sdl_nlen > 0)
859 			memcpy(kif->k.ifname, sdl->sdl_data,
860 			    sdl->sdl_nlen);
861 		/* string already terminated via calloc() */
862 	}
863 
864 	return (&kif->k);
865 }
866 
867 int
868 kif_validate(u_short ifindex)
869 {
870 	struct kif_node		*kif;
871 
872 	if ((kif = kif_find(ifindex)) == NULL) {
873 		log_warnx("interface with index %u not found", ifindex);
874 		return (1);
875 	}
876 
877 	return (kif->k.nh_reachable);
878 }
879 
880 struct kroute_node *
881 kroute_match(in_addr_t key)
882 {
883 	int			 i;
884 	struct kroute_node	*kr;
885 
886 	/* we will never match the default route */
887 	for (i = 32; i > 0; i--)
888 		if ((kr = kroute_find(key & prefixlen2mask(i), i,
889 		    RTP_ANY)) != NULL)
890 			return (kr);
891 
892 	/* if we don't have a match yet, try to find a default route */
893 	if ((kr = kroute_find(0, 0, RTP_ANY)) != NULL)
894 			return (kr);
895 
896 	return (NULL);
897 }
898 
899 /* misc */
900 int
901 protect_lo(void)
902 {
903 	struct kroute_node	*kr;
904 
905 	/* special protection for 127/8 */
906 	if ((kr = calloc(1, sizeof(struct kroute_node))) == NULL) {
907 		log_warn("protect_lo");
908 		return (-1);
909 	}
910 	kr->r.prefix.s_addr = htonl(INADDR_LOOPBACK & IN_CLASSA_NET);
911 	kr->r.prefixlen = 8;
912 	kr->r.flags = F_KERNEL|F_CONNECTED;
913 
914 	if (RB_INSERT(kroute_tree, &krt, kr) != NULL)
915 		free(kr);	/* kernel route already there, no problem */
916 
917 	return (0);
918 }
919 
920 u_int8_t
921 prefixlen_classful(in_addr_t ina)
922 {
923 	/* it hurt to write this. */
924 
925 	if (ina >= 0xf0000000U)		/* class E */
926 		return (32);
927 	else if (ina >= 0xe0000000U)	/* class D */
928 		return (4);
929 	else if (ina >= 0xc0000000U)	/* class C */
930 		return (24);
931 	else if (ina >= 0x80000000U)	/* class B */
932 		return (16);
933 	else				/* class A */
934 		return (8);
935 }
936 
937 u_int8_t
938 mask2prefixlen(in_addr_t ina)
939 {
940 	if (ina == 0)
941 		return (0);
942 	else
943 		return (33 - ffs(ntohl(ina)));
944 }
945 
946 in_addr_t
947 prefixlen2mask(u_int8_t prefixlen)
948 {
949 	if (prefixlen == 0)
950 		return (0);
951 
952 	return (htonl(0xffffffff << (32 - prefixlen)));
953 }
954 
955 #define	ROUNDUP(a)	\
956     (((a) & (sizeof(long) - 1)) ? (1 + ((a) | (sizeof(long) - 1))) : (a))
957 
958 void
959 get_rtaddrs(int addrs, struct sockaddr *sa, struct sockaddr **rti_info)
960 {
961 	int	i;
962 
963 	for (i = 0; i < RTAX_MAX; i++) {
964 		if (addrs & (1 << i)) {
965 			rti_info[i] = sa;
966 			sa = (struct sockaddr *)((char *)(sa) +
967 			    ROUNDUP(sa->sa_len));
968 		} else
969 			rti_info[i] = NULL;
970 	}
971 }
972 
973 void
974 if_change(u_short ifindex, int flags, struct if_data *ifd,
975     struct sockaddr_dl *sdl)
976 {
977 	struct kroute_node	*kr, *tkr;
978 	struct kif		*kif;
979 	u_int8_t		 reachable;
980 
981 	if ((kif = kif_update(ifindex, flags, ifd, sdl)) == NULL) {
982 		log_warn("if_change:  kif_update(%u)", ifindex);
983 		return;
984 	}
985 
986 	reachable = (kif->flags & IFF_UP) &&
987 	    LINK_STATE_IS_UP(kif->link_state);
988 
989 	if (reachable == kif->nh_reachable)
990 		return;		/* nothing changed wrt nexthop validity */
991 
992 	kif->nh_reachable = reachable;
993 
994 	/* notify ospfe about interface link state */
995 	main_imsg_compose_ospfe(IMSG_IFINFO, 0, kif, sizeof(struct kif));
996 
997 	/* update redistribute list */
998 	RB_FOREACH(kr, kroute_tree, &krt) {
999 		for (tkr = kr; tkr != NULL; tkr = tkr->next) {
1000 			if (tkr->r.ifindex == ifindex) {
1001 				if (reachable)
1002 					tkr->r.flags &= ~F_DOWN;
1003 				else
1004 					tkr->r.flags |= F_DOWN;
1005 
1006 			}
1007 		}
1008 		kr_redistribute(kr);
1009 	}
1010 }
1011 
1012 void
1013 if_newaddr(u_short ifindex, struct sockaddr_in *ifa, struct sockaddr_in *mask,
1014     struct sockaddr_in *brd)
1015 {
1016 	struct kif_node *kif;
1017 	struct kif_addr *ka;
1018 
1019 	if (ifa == NULL || ifa->sin_family != AF_INET)
1020 		return;
1021 	if ((kif = kif_find(ifindex)) == NULL) {
1022 		log_warnx("if_newaddr: corresponding if %i not found", ifindex);
1023 		return;
1024 	}
1025 	if ((ka = calloc(1, sizeof(struct kif_addr))) == NULL)
1026 		fatal("if_newaddr");
1027 	ka->addr = ifa->sin_addr;
1028 	if (mask)
1029 		ka->mask = mask->sin_addr;
1030 	else
1031 		ka->mask.s_addr = INADDR_NONE;
1032 	if (brd)
1033 		ka->dstbrd = brd->sin_addr;
1034 	else
1035 		ka->dstbrd.s_addr = INADDR_NONE;
1036 
1037 	TAILQ_INSERT_TAIL(&kif->addrs, ka, entry);
1038 }
1039 
1040 void
1041 if_deladdr(u_short ifindex, struct sockaddr_in *ifa, struct sockaddr_in *mask,
1042     struct sockaddr_in *brd)
1043 {
1044 	struct kif_node *kif;
1045 	struct kif_addr *ka, *nka;
1046 	struct ifaddrdel ifc;
1047 
1048 	if (ifa == NULL || ifa->sin_family != AF_INET)
1049 		return;
1050 	if ((kif = kif_find(ifindex)) == NULL) {
1051 		log_warnx("if_deladdr: corresponding if %i not found", ifindex);
1052 		return;
1053 	}
1054 
1055 	for (ka = TAILQ_FIRST(&kif->addrs); ka != NULL; ka = nka) {
1056 		nka = TAILQ_NEXT(ka, entry);
1057 
1058 		if (ka->addr.s_addr == ifa->sin_addr.s_addr) {
1059 			TAILQ_REMOVE(&kif->addrs, ka, entry);
1060 			ifc.addr = ifa->sin_addr;
1061 			ifc.ifindex = ifindex;
1062 			main_imsg_compose_ospfe(IMSG_IFADDRDEL, 0, &ifc,
1063 			    sizeof(ifc));
1064 			free(ka);
1065 			return;
1066 		}
1067 	}
1068 }
1069 
1070 void
1071 if_announce(void *msg)
1072 {
1073 	struct if_announcemsghdr	*ifan;
1074 	struct kif_node			*kif;
1075 
1076 	ifan = msg;
1077 
1078 	switch (ifan->ifan_what) {
1079 	case IFAN_ARRIVAL:
1080 		kif = kif_insert(ifan->ifan_index);
1081 		strlcpy(kif->k.ifname, ifan->ifan_name, sizeof(kif->k.ifname));
1082 		break;
1083 	case IFAN_DEPARTURE:
1084 		kif = kif_find(ifan->ifan_index);
1085 		kif_remove(kif);
1086 		break;
1087 	}
1088 }
1089 
1090 /* rtsock */
1091 int
1092 send_rtmsg(int fd, int action, struct kroute *kroute)
1093 {
1094 	struct iovec		iov[5];
1095 	struct rt_msghdr	hdr;
1096 	struct sockaddr_in	prefix;
1097 	struct sockaddr_in	nexthop;
1098 	struct sockaddr_in	mask;
1099 	struct sockaddr_rtlabel	sa_rl;
1100 	int			iovcnt = 0;
1101 	const char		*label;
1102 
1103 	if (kr_state.fib_sync == 0)
1104 		return (0);
1105 
1106 	/* initialize header */
1107 	bzero(&hdr, sizeof(hdr));
1108 	hdr.rtm_version = RTM_VERSION;
1109 	hdr.rtm_type = action;
1110 	hdr.rtm_priority = RTP_OSPF;
1111 	hdr.rtm_tableid = kr_state.rdomain;	/* rtableid */
1112 	if (action == RTM_CHANGE)
1113 		hdr.rtm_fmask = RTF_REJECT|RTF_BLACKHOLE;
1114 	else
1115 		hdr.rtm_flags = RTF_MPATH;
1116 	hdr.rtm_seq = kr_state.rtseq++;	/* overflow doesn't matter */
1117 	hdr.rtm_msglen = sizeof(hdr);
1118 	/* adjust iovec */
1119 	iov[iovcnt].iov_base = &hdr;
1120 	iov[iovcnt++].iov_len = sizeof(hdr);
1121 
1122 	bzero(&prefix, sizeof(prefix));
1123 	prefix.sin_len = sizeof(prefix);
1124 	prefix.sin_family = AF_INET;
1125 	prefix.sin_addr.s_addr = kroute->prefix.s_addr;
1126 	/* adjust header */
1127 	hdr.rtm_addrs |= RTA_DST;
1128 	hdr.rtm_msglen += sizeof(prefix);
1129 	/* adjust iovec */
1130 	iov[iovcnt].iov_base = &prefix;
1131 	iov[iovcnt++].iov_len = sizeof(prefix);
1132 
1133 	if (kroute->nexthop.s_addr != 0) {
1134 		bzero(&nexthop, sizeof(nexthop));
1135 		nexthop.sin_len = sizeof(nexthop);
1136 		nexthop.sin_family = AF_INET;
1137 		nexthop.sin_addr.s_addr = kroute->nexthop.s_addr;
1138 		/* adjust header */
1139 		hdr.rtm_flags |= RTF_GATEWAY;
1140 		hdr.rtm_addrs |= RTA_GATEWAY;
1141 		hdr.rtm_msglen += sizeof(nexthop);
1142 		/* adjust iovec */
1143 		iov[iovcnt].iov_base = &nexthop;
1144 		iov[iovcnt++].iov_len = sizeof(nexthop);
1145 	}
1146 
1147 	bzero(&mask, sizeof(mask));
1148 	mask.sin_len = sizeof(mask);
1149 	mask.sin_family = AF_INET;
1150 	mask.sin_addr.s_addr = prefixlen2mask(kroute->prefixlen);
1151 	/* adjust header */
1152 	hdr.rtm_addrs |= RTA_NETMASK;
1153 	hdr.rtm_msglen += sizeof(mask);
1154 	/* adjust iovec */
1155 	iov[iovcnt].iov_base = &mask;
1156 	iov[iovcnt++].iov_len = sizeof(mask);
1157 
1158 	if (kroute->rtlabel != 0) {
1159 		sa_rl.sr_len = sizeof(sa_rl);
1160 		sa_rl.sr_family = AF_UNSPEC;
1161 		label = rtlabel_id2name(kroute->rtlabel);
1162 		if (strlcpy(sa_rl.sr_label, label,
1163 		    sizeof(sa_rl.sr_label)) >= sizeof(sa_rl.sr_label)) {
1164 			log_warnx("send_rtmsg: invalid rtlabel");
1165 			return (-1);
1166 		}
1167 		/* adjust header */
1168 		hdr.rtm_addrs |= RTA_LABEL;
1169 		hdr.rtm_msglen += sizeof(sa_rl);
1170 		/* adjust iovec */
1171 		iov[iovcnt].iov_base = &sa_rl;
1172 		iov[iovcnt++].iov_len = sizeof(sa_rl);
1173 	}
1174 
1175 
1176 retry:
1177 	if (writev(fd, iov, iovcnt) == -1) {
1178 		if (errno == ESRCH) {
1179 			if (hdr.rtm_type == RTM_CHANGE) {
1180 				hdr.rtm_type = RTM_ADD;
1181 				goto retry;
1182 			} else if (hdr.rtm_type == RTM_DELETE) {
1183 				log_info("route %s/%u vanished before delete",
1184 				    inet_ntoa(kroute->prefix),
1185 				    kroute->prefixlen);
1186 				return (0);
1187 			}
1188 		}
1189 		log_warn("send_rtmsg: action %u, prefix %s/%u", hdr.rtm_type,
1190 		    inet_ntoa(kroute->prefix), kroute->prefixlen);
1191 		return (0);
1192 	}
1193 
1194 	return (0);
1195 }
1196 
1197 int
1198 fetchtable(void)
1199 {
1200 	size_t			 len;
1201 	int			 mib[7];
1202 	char			*buf;
1203 	int			 rv;
1204 
1205 	mib[0] = CTL_NET;
1206 	mib[1] = AF_ROUTE;
1207 	mib[2] = 0;
1208 	mib[3] = AF_INET;
1209 	mib[4] = NET_RT_DUMP;
1210 	mib[5] = 0;
1211 	mib[6] = kr_state.rdomain;	/* rtableid */
1212 
1213 	if (sysctl(mib, 7, NULL, &len, NULL, 0) == -1) {
1214 		log_warn("sysctl");
1215 		return (-1);
1216 	}
1217 	if ((buf = malloc(len)) == NULL) {
1218 		log_warn("fetchtable");
1219 		return (-1);
1220 	}
1221 	if (sysctl(mib, 7, buf, &len, NULL, 0) == -1) {
1222 		log_warn("sysctl");
1223 		free(buf);
1224 		return (-1);
1225 	}
1226 
1227 	rv = rtmsg_process(buf, len);
1228 	free(buf);
1229 
1230 	return (rv);
1231 }
1232 
1233 int
1234 fetchifs(u_short ifindex)
1235 {
1236 	size_t			 len;
1237 	int			 mib[6];
1238 	char			*buf;
1239 	int			 rv;
1240 
1241 	mib[0] = CTL_NET;
1242 	mib[1] = AF_ROUTE;
1243 	mib[2] = 0;
1244 	mib[3] = AF_INET;
1245 	mib[4] = NET_RT_IFLIST;
1246 	mib[5] = ifindex;
1247 
1248 	if (sysctl(mib, 6, NULL, &len, NULL, 0) == -1) {
1249 		log_warn("sysctl");
1250 		return (-1);
1251 	}
1252 	if ((buf = malloc(len)) == NULL) {
1253 		log_warn("fetchif");
1254 		return (-1);
1255 	}
1256 	if (sysctl(mib, 6, buf, &len, NULL, 0) == -1) {
1257 		log_warn("sysctl");
1258 		free(buf);
1259 		return (-1);
1260 	}
1261 
1262 	rv = rtmsg_process(buf, len);
1263 	free(buf);
1264 
1265 	return (rv);
1266 }
1267 
1268 int
1269 dispatch_rtmsg(void)
1270 {
1271 	char			 buf[RT_BUF_SIZE];
1272 	ssize_t			 n;
1273 
1274 	if ((n = read(kr_state.fd, &buf, sizeof(buf))) == -1) {
1275 		log_warn("dispatch_rtmsg: read error");
1276 		return (-1);
1277 	}
1278 
1279 	if (n == 0) {
1280 		log_warnx("routing socket closed");
1281 		return (-1);
1282 	}
1283 
1284 	return (rtmsg_process(buf, n));
1285 }
1286 
1287 int
1288 rtmsg_process(char *buf, int len)
1289 {
1290 	struct rt_msghdr	*rtm;
1291 	struct if_msghdr	 ifm;
1292 	struct ifa_msghdr	*ifam;
1293 	struct sockaddr		*sa, *rti_info[RTAX_MAX];
1294 	struct sockaddr_in	*sa_in;
1295 	struct sockaddr_rtlabel	*label;
1296 	struct kroute_node	*kr, *okr;
1297 	struct in_addr		 prefix, nexthop;
1298 	u_int8_t		 prefixlen, prio;
1299 	int			 flags, mpath;
1300 	u_short			 ifindex = 0;
1301 	int			 rv;
1302 
1303 	int			 offset;
1304 	char			*next;
1305 
1306 	for (offset = 0; offset < len; offset += rtm->rtm_msglen) {
1307 		next = buf + offset;
1308 		rtm = (struct rt_msghdr *)next;
1309 		if (rtm->rtm_version != RTM_VERSION)
1310 			continue;
1311 
1312 		prefix.s_addr = 0;
1313 		prefixlen = 0;
1314 		flags = F_KERNEL;
1315 		nexthop.s_addr = 0;
1316 		mpath = 0;
1317 		prio = 0;
1318 
1319 		sa = (struct sockaddr *)(next + rtm->rtm_hdrlen);
1320 		get_rtaddrs(rtm->rtm_addrs, sa, rti_info);
1321 
1322 		switch (rtm->rtm_type) {
1323 		case RTM_ADD:
1324 		case RTM_GET:
1325 		case RTM_CHANGE:
1326 		case RTM_DELETE:
1327 			prefix.s_addr = 0;
1328 			prefixlen = 0;
1329 			nexthop.s_addr = 0;
1330 			mpath = 0;
1331 			prio = 0;
1332 
1333 			if (rtm->rtm_errno)		/* failed attempts... */
1334 				continue;
1335 
1336 			if (rtm->rtm_tableid != kr_state.rdomain)
1337 				continue;
1338 
1339 			if ((sa = rti_info[RTAX_DST]) == NULL)
1340 				continue;
1341 
1342 			if (rtm->rtm_flags & RTF_LLINFO)	/* arp cache */
1343 				continue;
1344 
1345 			if (rtm->rtm_flags & RTF_MPATH)
1346 				mpath = 1;
1347 			prio = rtm->rtm_priority;
1348 			flags = (prio == RTP_OSPF) ?
1349 			    F_OSPFD_INSERTED : F_KERNEL;
1350 
1351 			switch (sa->sa_family) {
1352 			case AF_INET:
1353 				prefix.s_addr =
1354 				    ((struct sockaddr_in *)sa)->sin_addr.s_addr;
1355 				sa_in = (struct sockaddr_in *)
1356 				    rti_info[RTAX_NETMASK];
1357 				if (sa_in != NULL) {
1358 					if (sa_in->sin_len != 0)
1359 						prefixlen = mask2prefixlen(
1360 						    sa_in->sin_addr.s_addr);
1361 				} else if (rtm->rtm_flags & RTF_HOST)
1362 					prefixlen = 32;
1363 				else
1364 					prefixlen =
1365 					    prefixlen_classful(prefix.s_addr);
1366 				if (rtm->rtm_flags & RTF_STATIC)
1367 					flags |= F_STATIC;
1368 				if (rtm->rtm_flags & RTF_BLACKHOLE)
1369 					flags |= F_BLACKHOLE;
1370 				if (rtm->rtm_flags & RTF_REJECT)
1371 					flags |= F_REJECT;
1372 				if (rtm->rtm_flags & RTF_DYNAMIC)
1373 					flags |= F_DYNAMIC;
1374 				break;
1375 			default:
1376 				continue;
1377 			}
1378 
1379 			ifindex = rtm->rtm_index;
1380 			if ((sa = rti_info[RTAX_GATEWAY]) != NULL) {
1381 				switch (sa->sa_family) {
1382 				case AF_INET:
1383 					nexthop.s_addr = ((struct
1384 					    sockaddr_in *)sa)->sin_addr.s_addr;
1385 					break;
1386 				case AF_LINK:
1387 					flags |= F_CONNECTED;
1388 					break;
1389 				}
1390 			}
1391 		}
1392 
1393 		switch (rtm->rtm_type) {
1394 		case RTM_ADD:
1395 		case RTM_GET:
1396 		case RTM_CHANGE:
1397 			if (nexthop.s_addr == 0 && !(flags & F_CONNECTED)) {
1398 				log_warnx("no nexthop for %s/%u",
1399 				    inet_ntoa(prefix), prefixlen);
1400 				continue;
1401 			}
1402 
1403 			if ((okr = kroute_find(prefix.s_addr, prefixlen, prio))
1404 			    != NULL) {
1405 				/* get the correct route */
1406 				kr = okr;
1407 				if ((mpath || prio == RTP_OSPF) &&
1408 				    (kr = kroute_matchgw(okr, nexthop)) ==
1409 				    NULL) {
1410 					log_warnx("mpath route not found");
1411 					/* add routes we missed out earlier */
1412 					goto add;
1413 				}
1414 
1415 				if (kr->r.flags & F_REDISTRIBUTED)
1416 					flags |= F_REDISTRIBUTED;
1417 				kr->r.nexthop.s_addr = nexthop.s_addr;
1418 				kr->r.flags = flags;
1419 				kr->r.ifindex = ifindex;
1420 
1421 				rtlabel_unref(kr->r.rtlabel);
1422 				kr->r.rtlabel = 0;
1423 				kr->r.ext_tag = 0;
1424 				if ((label = (struct sockaddr_rtlabel *)
1425 				    rti_info[RTAX_LABEL]) != NULL) {
1426 					kr->r.rtlabel =
1427 					    rtlabel_name2id(label->sr_label);
1428 					kr->r.ext_tag =
1429 					    rtlabel_id2tag(kr->r.rtlabel);
1430 				}
1431 
1432 				if (kif_validate(kr->r.ifindex))
1433 					kr->r.flags &= ~F_DOWN;
1434 				else
1435 					kr->r.flags |= F_DOWN;
1436 
1437 				/* just readd, the RDE will care */
1438 				kr->serial = kr_state.fib_serial;
1439 				kr_redistribute(okr);
1440 			} else {
1441 add:
1442 				if ((kr = calloc(1,
1443 				    sizeof(struct kroute_node))) == NULL) {
1444 					log_warn("dispatch calloc");
1445 					return (-1);
1446 				}
1447 
1448 				kr->r.prefix.s_addr = prefix.s_addr;
1449 				kr->r.prefixlen = prefixlen;
1450 				kr->r.nexthop.s_addr = nexthop.s_addr;
1451 				kr->r.flags = flags;
1452 				kr->r.ifindex = ifindex;
1453 				kr->r.priority = prio;
1454 
1455 				if (rtm->rtm_priority == RTP_OSPF) {
1456 					log_warnx("alien OSPF route %s/%d",
1457 					    inet_ntoa(prefix), prefixlen);
1458 					rv = send_rtmsg(kr_state.fd,
1459 					    RTM_DELETE, &kr->r);
1460 					free(kr);
1461 					if (rv == -1)
1462 						return (-1);
1463 				} else {
1464 					if ((label = (struct sockaddr_rtlabel *)
1465 					    rti_info[RTAX_LABEL]) != NULL) {
1466 						kr->r.rtlabel =
1467 						    rtlabel_name2id(
1468 						    label->sr_label);
1469 						kr->r.ext_tag =
1470 						    rtlabel_id2tag(
1471 						    kr->r.rtlabel);
1472 					}
1473 
1474 					kroute_insert(kr);
1475 				}
1476 			}
1477 			break;
1478 		case RTM_DELETE:
1479 			if ((kr = kroute_find(prefix.s_addr, prefixlen, prio))
1480 			    == NULL)
1481 				continue;
1482 			if (!(kr->r.flags & F_KERNEL))
1483 				continue;
1484 			/* get the correct route */
1485 			okr = kr;
1486 			if (mpath &&
1487 			    (kr = kroute_matchgw(kr, nexthop)) == NULL) {
1488 				log_warnx("dispatch_rtmsg mpath route"
1489 				    " not found");
1490 				return (-1);
1491 			}
1492 			if (kroute_remove(kr) == -1)
1493 				return (-1);
1494 			break;
1495 		case RTM_IFINFO:
1496 			memcpy(&ifm, next, sizeof(ifm));
1497 			if_change(ifm.ifm_index, ifm.ifm_flags, &ifm.ifm_data,
1498 			    (struct sockaddr_dl *)rti_info[RTAX_IFP]);
1499 			break;
1500 		case RTM_NEWADDR:
1501 			ifam = (struct ifa_msghdr *)rtm;
1502 			if ((ifam->ifam_addrs & (RTA_NETMASK | RTA_IFA |
1503 			    RTA_BRD)) == 0)
1504 				break;
1505 
1506 			if_newaddr(ifam->ifam_index,
1507 			    (struct sockaddr_in *)rti_info[RTAX_IFA],
1508 			    (struct sockaddr_in *)rti_info[RTAX_NETMASK],
1509 			    (struct sockaddr_in *)rti_info[RTAX_BRD]);
1510 			break;
1511 		case RTM_DELADDR:
1512 			ifam = (struct ifa_msghdr *)rtm;
1513 			if ((ifam->ifam_addrs & (RTA_NETMASK | RTA_IFA |
1514 			    RTA_BRD)) == 0)
1515 				break;
1516 
1517 			if_deladdr(ifam->ifam_index,
1518 			    (struct sockaddr_in *)rti_info[RTAX_IFA],
1519 			    (struct sockaddr_in *)rti_info[RTAX_NETMASK],
1520 			    (struct sockaddr_in *)rti_info[RTAX_BRD]);
1521 			break;
1522 		case RTM_IFANNOUNCE:
1523 			if_announce(next);
1524 			break;
1525 		case RTM_DESYNC:
1526 			/*
1527 			 * We lost some routing packets. Force a reload of
1528 			 * the kernel route/interface information.
1529 			 */
1530 			kr_fib_reload();
1531 			break;
1532 		default:
1533 			/* ignore for now */
1534 			break;
1535 		}
1536 	}
1537 
1538 	return (offset);
1539 }
1540