xref: /freebsd/sys/net/route.c (revision 86c9d9918f1db7cdd968b60f8902466887bcd9e9)
1 /*-
2  * Copyright (c) 1980, 1986, 1991, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 4. Neither the name of the University nor the names of its contributors
14  *    may be used to endorse or promote products derived from this software
15  *    without specific prior written permission.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  *
29  *	@(#)route.c	8.3.1.1 (Berkeley) 2/23/95
30  * $FreeBSD$
31  */
32 /************************************************************************
33  * Note: In this file a 'fib' is a "forwarding information base"	*
34  * Which is the new name for an in kernel routing (next hop) table.	*
35  ***********************************************************************/
36 
37 #include "opt_inet.h"
38 #include "opt_inet6.h"
39 #include "opt_route.h"
40 #include "opt_sctp.h"
41 #include "opt_mrouting.h"
42 #include "opt_mpath.h"
43 
44 #include <sys/param.h>
45 #include <sys/systm.h>
46 #include <sys/malloc.h>
47 #include <sys/mbuf.h>
48 #include <sys/socket.h>
49 #include <sys/sysctl.h>
50 #include <sys/syslog.h>
51 #include <sys/sysproto.h>
52 #include <sys/proc.h>
53 #include <sys/domain.h>
54 #include <sys/kernel.h>
55 
56 #include <net/if.h>
57 #include <net/if_var.h>
58 #include <net/if_dl.h>
59 #include <net/route.h>
60 #include <net/route_var.h>
61 #include <net/vnet.h>
62 #include <net/flowtable.h>
63 
64 #ifdef RADIX_MPATH
65 #include <net/radix_mpath.h>
66 #endif
67 
68 #include <netinet/in.h>
69 #include <netinet/ip_mroute.h>
70 
71 #include <vm/uma.h>
72 
73 #define	RT_MAXFIBS	UINT16_MAX
74 
75 /* Kernel config default option. */
76 #ifdef ROUTETABLES
77 #if ROUTETABLES <= 0
78 #error "ROUTETABLES defined too low"
79 #endif
80 #if ROUTETABLES > RT_MAXFIBS
81 #error "ROUTETABLES defined too big"
82 #endif
83 #define	RT_NUMFIBS	ROUTETABLES
84 #endif /* ROUTETABLES */
85 /* Initialize to default if not otherwise set. */
86 #ifndef	RT_NUMFIBS
87 #define	RT_NUMFIBS	1
88 #endif
89 
90 #if defined(INET) || defined(INET6)
91 #ifdef SCTP
92 extern void sctp_addr_change(struct ifaddr *ifa, int cmd);
93 #endif /* SCTP */
94 #endif
95 
96 
97 /* This is read-only.. */
98 u_int rt_numfibs = RT_NUMFIBS;
99 SYSCTL_UINT(_net, OID_AUTO, fibs, CTLFLAG_RDTUN, &rt_numfibs, 0, "");
100 
101 /*
102  * By default add routes to all fibs for new interfaces.
103  * Once this is set to 0 then only allocate routes on interface
104  * changes for the FIB of the caller when adding a new set of addresses
105  * to an interface.  XXX this is a shotgun aproach to a problem that needs
106  * a more fine grained solution.. that will come.
107  * XXX also has the problems getting the FIB from curthread which will not
108  * always work given the fib can be overridden and prefixes can be added
109  * from the network stack context.
110  */
111 VNET_DEFINE(u_int, rt_add_addr_allfibs) = 1;
112 SYSCTL_UINT(_net, OID_AUTO, add_addr_allfibs, CTLFLAG_RWTUN | CTLFLAG_VNET,
113     &VNET_NAME(rt_add_addr_allfibs), 0, "");
114 
115 VNET_DEFINE(struct rtstat, rtstat);
116 #define	V_rtstat	VNET(rtstat)
117 
118 VNET_DEFINE(struct rib_head *, rt_tables);
119 #define	V_rt_tables	VNET(rt_tables)
120 
121 VNET_DEFINE(int, rttrash);		/* routes not in table but not freed */
122 #define	V_rttrash	VNET(rttrash)
123 
124 
125 /*
126  * Convert a 'struct radix_node *' to a 'struct rtentry *'.
127  * The operation can be done safely (in this code) because a
128  * 'struct rtentry' starts with two 'struct radix_node''s, the first
129  * one representing leaf nodes in the routing tree, which is
130  * what the code in radix.c passes us as a 'struct radix_node'.
131  *
132  * But because there are a lot of assumptions in this conversion,
133  * do not cast explicitly, but always use the macro below.
134  */
135 #define RNTORT(p)	((struct rtentry *)(p))
136 
137 static VNET_DEFINE(uma_zone_t, rtzone);		/* Routing table UMA zone. */
138 #define	V_rtzone	VNET(rtzone)
139 
140 static int rtrequest1_fib_change(struct rib_head *, struct rt_addrinfo *,
141     struct rtentry **, u_int);
142 static void rt_setmetrics(const struct rt_addrinfo *, struct rtentry *);
143 static int rt_ifdelroute(const struct rtentry *rt, void *arg);
144 static struct rtentry *rt_unlinkrte(struct rib_head *rnh,
145     struct rt_addrinfo *info, int *perror);
146 static void rt_notifydelete(struct rtentry *rt, struct rt_addrinfo *info);
147 #ifdef RADIX_MPATH
148 static struct radix_node *rt_mpath_unlink(struct rib_head *rnh,
149     struct rt_addrinfo *info, struct rtentry *rto, int *perror);
150 #endif
151 static int rt_exportinfo(struct rtentry *rt, struct rt_addrinfo *info,
152     int flags);
153 
154 struct if_mtuinfo
155 {
156 	struct ifnet	*ifp;
157 	int		mtu;
158 };
159 
160 static int	if_updatemtu_cb(struct radix_node *, void *);
161 
162 /*
163  * handler for net.my_fibnum
164  */
165 static int
166 sysctl_my_fibnum(SYSCTL_HANDLER_ARGS)
167 {
168         int fibnum;
169         int error;
170 
171         fibnum = curthread->td_proc->p_fibnum;
172         error = sysctl_handle_int(oidp, &fibnum, 0, req);
173         return (error);
174 }
175 
176 SYSCTL_PROC(_net, OID_AUTO, my_fibnum, CTLTYPE_INT|CTLFLAG_RD,
177             NULL, 0, &sysctl_my_fibnum, "I", "default FIB of caller");
178 
179 static __inline struct rib_head **
180 rt_tables_get_rnh_ptr(int table, int fam)
181 {
182 	struct rib_head **rnh;
183 
184 	KASSERT(table >= 0 && table < rt_numfibs, ("%s: table out of bounds.",
185 	    __func__));
186 	KASSERT(fam >= 0 && fam < (AF_MAX+1), ("%s: fam out of bounds.",
187 	    __func__));
188 
189 	/* rnh is [fib=0][af=0]. */
190 	rnh = (struct rib_head **)V_rt_tables;
191 	/* Get the offset to the requested table and fam. */
192 	rnh += table * (AF_MAX+1) + fam;
193 
194 	return (rnh);
195 }
196 
197 struct rib_head *
198 rt_tables_get_rnh(int table, int fam)
199 {
200 
201 	return (*rt_tables_get_rnh_ptr(table, fam));
202 }
203 
204 u_int
205 rt_tables_get_gen(int table, int fam)
206 {
207 	struct rib_head *rnh;
208 
209 	rnh = *rt_tables_get_rnh_ptr(table, fam);
210 	KASSERT(rnh != NULL, ("%s: NULL rib_head pointer table %d fam %d",
211 	    __func__, table, fam));
212 	return (rnh->rnh_gen);
213 }
214 
215 
216 /*
217  * route initialization must occur before ip6_init2(), which happenas at
218  * SI_ORDER_MIDDLE.
219  */
220 static void
221 route_init(void)
222 {
223 
224 	/* whack the tunable ints into  line. */
225 	if (rt_numfibs > RT_MAXFIBS)
226 		rt_numfibs = RT_MAXFIBS;
227 	if (rt_numfibs == 0)
228 		rt_numfibs = 1;
229 }
230 SYSINIT(route_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD, route_init, 0);
231 
232 static int
233 rtentry_zinit(void *mem, int size, int how)
234 {
235 	struct rtentry *rt = mem;
236 
237 	rt->rt_pksent = counter_u64_alloc(how);
238 	if (rt->rt_pksent == NULL)
239 		return (ENOMEM);
240 
241 	RT_LOCK_INIT(rt);
242 
243 	return (0);
244 }
245 
246 static void
247 rtentry_zfini(void *mem, int size)
248 {
249 	struct rtentry *rt = mem;
250 
251 	RT_LOCK_DESTROY(rt);
252 	counter_u64_free(rt->rt_pksent);
253 }
254 
255 static int
256 rtentry_ctor(void *mem, int size, void *arg, int how)
257 {
258 	struct rtentry *rt = mem;
259 
260 	bzero(rt, offsetof(struct rtentry, rt_endzero));
261 	counter_u64_zero(rt->rt_pksent);
262 	rt->rt_chain = NULL;
263 
264 	return (0);
265 }
266 
267 static void
268 rtentry_dtor(void *mem, int size, void *arg)
269 {
270 	struct rtentry *rt = mem;
271 
272 	RT_UNLOCK_COND(rt);
273 }
274 
275 static void
276 vnet_route_init(const void *unused __unused)
277 {
278 	struct domain *dom;
279 	struct rib_head **rnh;
280 	int table;
281 	int fam;
282 
283 	V_rt_tables = malloc(rt_numfibs * (AF_MAX+1) *
284 	    sizeof(struct rib_head *), M_RTABLE, M_WAITOK|M_ZERO);
285 
286 	V_rtzone = uma_zcreate("rtentry", sizeof(struct rtentry),
287 	    rtentry_ctor, rtentry_dtor,
288 	    rtentry_zinit, rtentry_zfini, UMA_ALIGN_PTR, 0);
289 	for (dom = domains; dom; dom = dom->dom_next) {
290 		if (dom->dom_rtattach == NULL)
291 			continue;
292 
293 		for  (table = 0; table < rt_numfibs; table++) {
294 			fam = dom->dom_family;
295 			if (table != 0 && fam != AF_INET6 && fam != AF_INET)
296 				break;
297 
298 			rnh = rt_tables_get_rnh_ptr(table, fam);
299 			if (rnh == NULL)
300 				panic("%s: rnh NULL", __func__);
301 			dom->dom_rtattach((void **)rnh, 0);
302 		}
303 	}
304 }
305 VNET_SYSINIT(vnet_route_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH,
306     vnet_route_init, 0);
307 
308 #ifdef VIMAGE
309 static void
310 vnet_route_uninit(const void *unused __unused)
311 {
312 	int table;
313 	int fam;
314 	struct domain *dom;
315 	struct rib_head **rnh;
316 
317 	for (dom = domains; dom; dom = dom->dom_next) {
318 		if (dom->dom_rtdetach == NULL)
319 			continue;
320 
321 		for (table = 0; table < rt_numfibs; table++) {
322 			fam = dom->dom_family;
323 
324 			if (table != 0 && fam != AF_INET6 && fam != AF_INET)
325 				break;
326 
327 			rnh = rt_tables_get_rnh_ptr(table, fam);
328 			if (rnh == NULL)
329 				panic("%s: rnh NULL", __func__);
330 			dom->dom_rtdetach((void **)rnh, 0);
331 		}
332 	}
333 
334 	free(V_rt_tables, M_RTABLE);
335 	uma_zdestroy(V_rtzone);
336 }
337 VNET_SYSUNINIT(vnet_route_uninit, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD,
338     vnet_route_uninit, 0);
339 #endif
340 
341 struct rib_head *
342 rt_table_init(int offset)
343 {
344 	struct rib_head *rh;
345 
346 	rh = malloc(sizeof(struct rib_head), M_RTABLE, M_WAITOK | M_ZERO);
347 
348 	/* TODO: These details should be hidded inside radix.c */
349 	/* Init masks tree */
350 	rn_inithead_internal(&rh->head, rh->rnh_nodes, offset);
351 	rn_inithead_internal(&rh->rmhead.head, rh->rmhead.mask_nodes, 0);
352 	rh->head.rnh_masks = &rh->rmhead;
353 
354 	/* Init locks */
355 	rw_init(&rh->rib_lock, "rib head lock");
356 
357 	/* Finally, set base callbacks */
358 	rh->rnh_addaddr = rn_addroute;
359 	rh->rnh_deladdr = rn_delete;
360 	rh->rnh_matchaddr = rn_match;
361 	rh->rnh_lookup = rn_lookup;
362 	rh->rnh_walktree = rn_walktree;
363 	rh->rnh_walktree_from = rn_walktree_from;
364 
365 	return (rh);
366 }
367 
368 static int
369 rt_freeentry(struct radix_node *rn, void *arg)
370 {
371 	struct radix_head * const rnh = arg;
372 	struct radix_node *x;
373 
374 	x = (struct radix_node *)rn_delete(rn + 2, NULL, rnh);
375 	if (x != NULL)
376 		R_Free(x);
377 	return (0);
378 }
379 
380 void
381 rt_table_destroy(struct rib_head *rh)
382 {
383 
384 	rn_walktree(&rh->rmhead.head, rt_freeentry, &rh->rmhead.head);
385 
386 	/* Assume table is already empty */
387 	rw_destroy(&rh->rib_lock);
388 	free(rh, M_RTABLE);
389 }
390 
391 
392 #ifndef _SYS_SYSPROTO_H_
393 struct setfib_args {
394 	int     fibnum;
395 };
396 #endif
397 int
398 sys_setfib(struct thread *td, struct setfib_args *uap)
399 {
400 	if (uap->fibnum < 0 || uap->fibnum >= rt_numfibs)
401 		return EINVAL;
402 	td->td_proc->p_fibnum = uap->fibnum;
403 	return (0);
404 }
405 
406 /*
407  * Packet routing routines.
408  */
409 void
410 rtalloc_ign_fib(struct route *ro, u_long ignore, u_int fibnum)
411 {
412 	struct rtentry *rt;
413 
414 	if ((rt = ro->ro_rt) != NULL) {
415 		if (rt->rt_ifp != NULL && rt->rt_flags & RTF_UP)
416 			return;
417 		RTFREE(rt);
418 		ro->ro_rt = NULL;
419 	}
420 	ro->ro_rt = rtalloc1_fib(&ro->ro_dst, 1, ignore, fibnum);
421 	if (ro->ro_rt)
422 		RT_UNLOCK(ro->ro_rt);
423 }
424 
425 /*
426  * Look up the route that matches the address given
427  * Or, at least try.. Create a cloned route if needed.
428  *
429  * The returned route, if any, is locked.
430  */
431 struct rtentry *
432 rtalloc1(struct sockaddr *dst, int report, u_long ignflags)
433 {
434 
435 	return (rtalloc1_fib(dst, report, ignflags, RT_DEFAULT_FIB));
436 }
437 
438 struct rtentry *
439 rtalloc1_fib(struct sockaddr *dst, int report, u_long ignflags,
440 		    u_int fibnum)
441 {
442 	struct rib_head *rh;
443 	struct radix_node *rn;
444 	struct rtentry *newrt;
445 	struct rt_addrinfo info;
446 	int err = 0, msgtype = RTM_MISS;
447 
448 	KASSERT((fibnum < rt_numfibs), ("rtalloc1_fib: bad fibnum"));
449 	rh = rt_tables_get_rnh(fibnum, dst->sa_family);
450 	newrt = NULL;
451 	if (rh == NULL)
452 		goto miss;
453 
454 	/*
455 	 * Look up the address in the table for that Address Family
456 	 */
457 	RIB_RLOCK(rh);
458 	rn = rh->rnh_matchaddr(dst, &rh->head);
459 	if (rn && ((rn->rn_flags & RNF_ROOT) == 0)) {
460 		newrt = RNTORT(rn);
461 		RT_LOCK(newrt);
462 		RT_ADDREF(newrt);
463 		RIB_RUNLOCK(rh);
464 		return (newrt);
465 
466 	} else
467 		RIB_RUNLOCK(rh);
468 
469 	/*
470 	 * Either we hit the root or couldn't find any match,
471 	 * Which basically means
472 	 * "caint get there frm here"
473 	 */
474 miss:
475 	V_rtstat.rts_unreach++;
476 
477 	if (report) {
478 		/*
479 		 * If required, report the failure to the supervising
480 		 * Authorities.
481 		 * For a delete, this is not an error. (report == 0)
482 		 */
483 		bzero(&info, sizeof(info));
484 		info.rti_info[RTAX_DST] = dst;
485 		rt_missmsg_fib(msgtype, &info, 0, err, fibnum);
486 	}
487 	return (newrt);
488 }
489 
490 /*
491  * Remove a reference count from an rtentry.
492  * If the count gets low enough, take it out of the routing table
493  */
494 void
495 rtfree(struct rtentry *rt)
496 {
497 	struct rib_head *rnh;
498 
499 	KASSERT(rt != NULL,("%s: NULL rt", __func__));
500 	rnh = rt_tables_get_rnh(rt->rt_fibnum, rt_key(rt)->sa_family);
501 	KASSERT(rnh != NULL,("%s: NULL rnh", __func__));
502 
503 	RT_LOCK_ASSERT(rt);
504 
505 	/*
506 	 * The callers should use RTFREE_LOCKED() or RTFREE(), so
507 	 * we should come here exactly with the last reference.
508 	 */
509 	RT_REMREF(rt);
510 	if (rt->rt_refcnt > 0) {
511 		log(LOG_DEBUG, "%s: %p has %d refs\n", __func__, rt, rt->rt_refcnt);
512 		goto done;
513 	}
514 
515 	/*
516 	 * On last reference give the "close method" a chance
517 	 * to cleanup private state.  This also permits (for
518 	 * IPv4 and IPv6) a chance to decide if the routing table
519 	 * entry should be purged immediately or at a later time.
520 	 * When an immediate purge is to happen the close routine
521 	 * typically calls rtexpunge which clears the RTF_UP flag
522 	 * on the entry so that the code below reclaims the storage.
523 	 */
524 	if (rt->rt_refcnt == 0 && rnh->rnh_close)
525 		rnh->rnh_close((struct radix_node *)rt, &rnh->head);
526 
527 	/*
528 	 * If we are no longer "up" (and ref == 0)
529 	 * then we can free the resources associated
530 	 * with the route.
531 	 */
532 	if ((rt->rt_flags & RTF_UP) == 0) {
533 		if (rt->rt_nodes->rn_flags & (RNF_ACTIVE | RNF_ROOT))
534 			panic("rtfree 2");
535 		/*
536 		 * the rtentry must have been removed from the routing table
537 		 * so it is represented in rttrash.. remove that now.
538 		 */
539 		V_rttrash--;
540 #ifdef	DIAGNOSTIC
541 		if (rt->rt_refcnt < 0) {
542 			printf("rtfree: %p not freed (neg refs)\n", rt);
543 			goto done;
544 		}
545 #endif
546 		/*
547 		 * release references on items we hold them on..
548 		 * e.g other routes and ifaddrs.
549 		 */
550 		if (rt->rt_ifa)
551 			ifa_free(rt->rt_ifa);
552 		/*
553 		 * The key is separatly alloc'd so free it (see rt_setgate()).
554 		 * This also frees the gateway, as they are always malloc'd
555 		 * together.
556 		 */
557 		R_Free(rt_key(rt));
558 
559 		/*
560 		 * and the rtentry itself of course
561 		 */
562 		uma_zfree(V_rtzone, rt);
563 		return;
564 	}
565 done:
566 	RT_UNLOCK(rt);
567 }
568 
569 
570 /*
571  * Force a routing table entry to the specified
572  * destination to go through the given gateway.
573  * Normally called as a result of a routing redirect
574  * message from the network layer.
575  */
576 void
577 rtredirect_fib(struct sockaddr *dst,
578 	struct sockaddr *gateway,
579 	struct sockaddr *netmask,
580 	int flags,
581 	struct sockaddr *src,
582 	u_int fibnum)
583 {
584 	struct rtentry *rt;
585 	int error = 0;
586 	short *stat = NULL;
587 	struct rt_addrinfo info;
588 	struct ifaddr *ifa;
589 	struct rib_head *rnh;
590 
591 	ifa = NULL;
592 	rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
593 	if (rnh == NULL) {
594 		error = EAFNOSUPPORT;
595 		goto out;
596 	}
597 
598 	/* verify the gateway is directly reachable */
599 	if ((ifa = ifa_ifwithnet(gateway, 0, fibnum)) == NULL) {
600 		error = ENETUNREACH;
601 		goto out;
602 	}
603 	rt = rtalloc1_fib(dst, 0, 0UL, fibnum);	/* NB: rt is locked */
604 	/*
605 	 * If the redirect isn't from our current router for this dst,
606 	 * it's either old or wrong.  If it redirects us to ourselves,
607 	 * we have a routing loop, perhaps as a result of an interface
608 	 * going down recently.
609 	 */
610 	if (!(flags & RTF_DONE) && rt) {
611 		if (!sa_equal(src, rt->rt_gateway)) {
612 			error = EINVAL;
613 			goto done;
614 		}
615 		if (rt->rt_ifa != ifa && ifa->ifa_addr->sa_family != AF_LINK) {
616 			error = EINVAL;
617 			goto done;
618 		}
619 	}
620 	if ((flags & RTF_GATEWAY) && ifa_ifwithaddr_check(gateway)) {
621 		error = EHOSTUNREACH;
622 		goto done;
623 	}
624 	/*
625 	 * Create a new entry if we just got back a wildcard entry
626 	 * or the lookup failed.  This is necessary for hosts
627 	 * which use routing redirects generated by smart gateways
628 	 * to dynamically build the routing tables.
629 	 */
630 	if (rt == NULL || (rt_mask(rt) && rt_mask(rt)->sa_len < 2))
631 		goto create;
632 	/*
633 	 * Don't listen to the redirect if it's
634 	 * for a route to an interface.
635 	 */
636 	if (rt->rt_flags & RTF_GATEWAY) {
637 		if (((rt->rt_flags & RTF_HOST) == 0) && (flags & RTF_HOST)) {
638 			/*
639 			 * Changing from route to net => route to host.
640 			 * Create new route, rather than smashing route to net.
641 			 */
642 		create:
643 			if (rt != NULL)
644 				RTFREE_LOCKED(rt);
645 
646 			flags |= RTF_DYNAMIC;
647 			bzero((caddr_t)&info, sizeof(info));
648 			info.rti_info[RTAX_DST] = dst;
649 			info.rti_info[RTAX_GATEWAY] = gateway;
650 			info.rti_info[RTAX_NETMASK] = netmask;
651 			info.rti_ifa = ifa;
652 			info.rti_flags = flags;
653 			error = rtrequest1_fib(RTM_ADD, &info, &rt, fibnum);
654 			if (rt != NULL) {
655 				RT_LOCK(rt);
656 				flags = rt->rt_flags;
657 			}
658 
659 			stat = &V_rtstat.rts_dynamic;
660 		} else {
661 
662 			/*
663 			 * Smash the current notion of the gateway to
664 			 * this destination.  Should check about netmask!!!
665 			 */
666 			if ((flags & RTF_GATEWAY) == 0)
667 				rt->rt_flags &= ~RTF_GATEWAY;
668 			rt->rt_flags |= RTF_MODIFIED;
669 			flags |= RTF_MODIFIED;
670 			stat = &V_rtstat.rts_newgateway;
671 			/*
672 			 * add the key and gateway (in one malloc'd chunk).
673 			 */
674 			RT_UNLOCK(rt);
675 			RIB_WLOCK(rnh);
676 			RT_LOCK(rt);
677 			rt_setgate(rt, rt_key(rt), gateway);
678 			RIB_WUNLOCK(rnh);
679 		}
680 	} else
681 		error = EHOSTUNREACH;
682 done:
683 	if (rt)
684 		RTFREE_LOCKED(rt);
685 out:
686 	if (error)
687 		V_rtstat.rts_badredirect++;
688 	else if (stat != NULL)
689 		(*stat)++;
690 	bzero((caddr_t)&info, sizeof(info));
691 	info.rti_info[RTAX_DST] = dst;
692 	info.rti_info[RTAX_GATEWAY] = gateway;
693 	info.rti_info[RTAX_NETMASK] = netmask;
694 	info.rti_info[RTAX_AUTHOR] = src;
695 	rt_missmsg_fib(RTM_REDIRECT, &info, flags, error, fibnum);
696 	if (ifa != NULL)
697 		ifa_free(ifa);
698 }
699 
700 /*
701  * Routing table ioctl interface.
702  */
703 int
704 rtioctl_fib(u_long req, caddr_t data, u_int fibnum)
705 {
706 
707 	/*
708 	 * If more ioctl commands are added here, make sure the proper
709 	 * super-user checks are being performed because it is possible for
710 	 * prison-root to make it this far if raw sockets have been enabled
711 	 * in jails.
712 	 */
713 #ifdef INET
714 	/* Multicast goop, grrr... */
715 	return mrt_ioctl ? mrt_ioctl(req, data, fibnum) : EOPNOTSUPP;
716 #else /* INET */
717 	return ENXIO;
718 #endif /* INET */
719 }
720 
721 struct ifaddr *
722 ifa_ifwithroute(int flags, const struct sockaddr *dst, struct sockaddr *gateway,
723 				u_int fibnum)
724 {
725 	struct ifaddr *ifa;
726 	int not_found = 0;
727 
728 	if ((flags & RTF_GATEWAY) == 0) {
729 		/*
730 		 * If we are adding a route to an interface,
731 		 * and the interface is a pt to pt link
732 		 * we should search for the destination
733 		 * as our clue to the interface.  Otherwise
734 		 * we can use the local address.
735 		 */
736 		ifa = NULL;
737 		if (flags & RTF_HOST)
738 			ifa = ifa_ifwithdstaddr(dst, fibnum);
739 		if (ifa == NULL)
740 			ifa = ifa_ifwithaddr(gateway);
741 	} else {
742 		/*
743 		 * If we are adding a route to a remote net
744 		 * or host, the gateway may still be on the
745 		 * other end of a pt to pt link.
746 		 */
747 		ifa = ifa_ifwithdstaddr(gateway, fibnum);
748 	}
749 	if (ifa == NULL)
750 		ifa = ifa_ifwithnet(gateway, 0, fibnum);
751 	if (ifa == NULL) {
752 		struct rtentry *rt = rtalloc1_fib(gateway, 0, 0, fibnum);
753 		if (rt == NULL)
754 			return (NULL);
755 		/*
756 		 * dismiss a gateway that is reachable only
757 		 * through the default router
758 		 */
759 		switch (gateway->sa_family) {
760 		case AF_INET:
761 			if (satosin(rt_key(rt))->sin_addr.s_addr == INADDR_ANY)
762 				not_found = 1;
763 			break;
764 		case AF_INET6:
765 			if (IN6_IS_ADDR_UNSPECIFIED(&satosin6(rt_key(rt))->sin6_addr))
766 				not_found = 1;
767 			break;
768 		default:
769 			break;
770 		}
771 		if (!not_found && rt->rt_ifa != NULL) {
772 			ifa = rt->rt_ifa;
773 			ifa_ref(ifa);
774 		}
775 		RT_REMREF(rt);
776 		RT_UNLOCK(rt);
777 		if (not_found || ifa == NULL)
778 			return (NULL);
779 	}
780 	if (ifa->ifa_addr->sa_family != dst->sa_family) {
781 		struct ifaddr *oifa = ifa;
782 		ifa = ifaof_ifpforaddr(dst, ifa->ifa_ifp);
783 		if (ifa == NULL)
784 			ifa = oifa;
785 		else
786 			ifa_free(oifa);
787 	}
788 	return (ifa);
789 }
790 
791 /*
792  * Do appropriate manipulations of a routing tree given
793  * all the bits of info needed
794  */
795 int
796 rtrequest_fib(int req,
797 	struct sockaddr *dst,
798 	struct sockaddr *gateway,
799 	struct sockaddr *netmask,
800 	int flags,
801 	struct rtentry **ret_nrt,
802 	u_int fibnum)
803 {
804 	struct rt_addrinfo info;
805 
806 	if (dst->sa_len == 0)
807 		return(EINVAL);
808 
809 	bzero((caddr_t)&info, sizeof(info));
810 	info.rti_flags = flags;
811 	info.rti_info[RTAX_DST] = dst;
812 	info.rti_info[RTAX_GATEWAY] = gateway;
813 	info.rti_info[RTAX_NETMASK] = netmask;
814 	return rtrequest1_fib(req, &info, ret_nrt, fibnum);
815 }
816 
817 
818 /*
819  * Copy most of @rt data into @info.
820  *
821  * If @flags contains NHR_COPY, copies dst,netmask and gw to the
822  * pointers specified by @info structure. Assume such pointers
823  * are zeroed sockaddr-like structures with sa_len field initialized
824  * to reflect size of the provided buffer. if no NHR_COPY is specified,
825  * point dst,netmask and gw @info fields to appropriate @rt values.
826  *
827  * if @flags contains NHR_REF, do refcouting on rt_ifp.
828  *
829  * Returns 0 on success.
830  */
831 int
832 rt_exportinfo(struct rtentry *rt, struct rt_addrinfo *info, int flags)
833 {
834 	struct rt_metrics *rmx;
835 	struct sockaddr *src, *dst;
836 	int sa_len;
837 
838 	if (flags & NHR_COPY) {
839 		/* Copy destination if dst is non-zero */
840 		src = rt_key(rt);
841 		dst = info->rti_info[RTAX_DST];
842 		sa_len = src->sa_len;
843 		if (dst != NULL) {
844 			if (src->sa_len > dst->sa_len)
845 				return (ENOMEM);
846 			memcpy(dst, src, src->sa_len);
847 			info->rti_addrs |= RTA_DST;
848 		}
849 
850 		/* Copy mask if set && dst is non-zero */
851 		src = rt_mask(rt);
852 		dst = info->rti_info[RTAX_NETMASK];
853 		if (src != NULL && dst != NULL) {
854 
855 			/*
856 			 * Radix stores different value in sa_len,
857 			 * assume rt_mask() to have the same length
858 			 * as rt_key()
859 			 */
860 			if (sa_len > dst->sa_len)
861 				return (ENOMEM);
862 			memcpy(dst, src, src->sa_len);
863 			info->rti_addrs |= RTA_NETMASK;
864 		}
865 
866 		/* Copy gateway is set && dst is non-zero */
867 		src = rt->rt_gateway;
868 		dst = info->rti_info[RTAX_GATEWAY];
869 		if ((rt->rt_flags & RTF_GATEWAY) && src != NULL && dst != NULL){
870 			if (src->sa_len > dst->sa_len)
871 				return (ENOMEM);
872 			memcpy(dst, src, src->sa_len);
873 			info->rti_addrs |= RTA_GATEWAY;
874 		}
875 	} else {
876 		info->rti_info[RTAX_DST] = rt_key(rt);
877 		info->rti_addrs |= RTA_DST;
878 		if (rt_mask(rt) != NULL) {
879 			info->rti_info[RTAX_NETMASK] = rt_mask(rt);
880 			info->rti_addrs |= RTA_NETMASK;
881 		}
882 		if (rt->rt_flags & RTF_GATEWAY) {
883 			info->rti_info[RTAX_GATEWAY] = rt->rt_gateway;
884 			info->rti_addrs |= RTA_GATEWAY;
885 		}
886 	}
887 
888 	rmx = info->rti_rmx;
889 	if (rmx != NULL) {
890 		info->rti_mflags |= RTV_MTU;
891 		rmx->rmx_mtu = rt->rt_mtu;
892 	}
893 
894 	info->rti_flags = rt->rt_flags;
895 	info->rti_ifp = rt->rt_ifp;
896 	info->rti_ifa = rt->rt_ifa;
897 
898 	if (flags & NHR_REF) {
899 		/* Do 'traditional' refcouting */
900 		if_ref(info->rti_ifp);
901 	}
902 
903 	return (0);
904 }
905 
906 /*
907  * Lookups up route entry for @dst in RIB database for fib @fibnum.
908  * Exports entry data to @info using rt_exportinfo().
909  *
910  * if @flags contains NHR_REF, refcouting is performed on rt_ifp.
911  *   All references can be released later by calling rib_free_info()
912  *
913  * Returns 0 on success.
914  * Returns ENOENT for lookup failure, ENOMEM for export failure.
915  */
916 int
917 rib_lookup_info(uint32_t fibnum, const struct sockaddr *dst, uint32_t flags,
918     uint32_t flowid, struct rt_addrinfo *info)
919 {
920 	struct rib_head *rh;
921 	struct radix_node *rn;
922 	struct rtentry *rt;
923 	int error;
924 
925 	KASSERT((fibnum < rt_numfibs), ("rib_lookup_rte: bad fibnum"));
926 	rh = rt_tables_get_rnh(fibnum, dst->sa_family);
927 	if (rh == NULL)
928 		return (ENOENT);
929 
930 	RIB_RLOCK(rh);
931 	rn = rh->rnh_matchaddr(__DECONST(void *, dst), &rh->head);
932 	if (rn != NULL && ((rn->rn_flags & RNF_ROOT) == 0)) {
933 		rt = RNTORT(rn);
934 		/* Ensure route & ifp is UP */
935 		if (RT_LINK_IS_UP(rt->rt_ifp)) {
936 			flags = (flags & NHR_REF) | NHR_COPY;
937 			error = rt_exportinfo(rt, info, flags);
938 			RIB_RUNLOCK(rh);
939 
940 			return (error);
941 		}
942 	}
943 	RIB_RUNLOCK(rh);
944 
945 	return (ENOENT);
946 }
947 
948 /*
949  * Releases all references acquired by rib_lookup_info() when
950  * called with NHR_REF flags.
951  */
952 void
953 rib_free_info(struct rt_addrinfo *info)
954 {
955 
956 	if_rele(info->rti_ifp);
957 }
958 
959 /*
960  * Iterates over all existing fibs in system calling
961  *  @setwa_f function prior to traversing each fib.
962  *  Calls @wa_f function for each element in current fib.
963  * If af is not AF_UNSPEC, iterates over fibs in particular
964  * address family.
965  */
966 void
967 rt_foreach_fib_walk(int af, rt_setwarg_t *setwa_f, rt_walktree_f_t *wa_f,
968     void *arg)
969 {
970 	struct rib_head *rnh;
971 	uint32_t fibnum;
972 	int i;
973 
974 	for (fibnum = 0; fibnum < rt_numfibs; fibnum++) {
975 		/* Do we want some specific family? */
976 		if (af != AF_UNSPEC) {
977 			rnh = rt_tables_get_rnh(fibnum, af);
978 			if (rnh == NULL)
979 				continue;
980 			if (setwa_f != NULL)
981 				setwa_f(rnh, fibnum, af, arg);
982 
983 			RIB_WLOCK(rnh);
984 			rnh->rnh_walktree(&rnh->head, (walktree_f_t *)wa_f,arg);
985 			RIB_WUNLOCK(rnh);
986 			continue;
987 		}
988 
989 		for (i = 1; i <= AF_MAX; i++) {
990 			rnh = rt_tables_get_rnh(fibnum, i);
991 			if (rnh == NULL)
992 				continue;
993 			if (setwa_f != NULL)
994 				setwa_f(rnh, fibnum, i, arg);
995 
996 			RIB_WLOCK(rnh);
997 			rnh->rnh_walktree(&rnh->head, (walktree_f_t *)wa_f,arg);
998 			RIB_WUNLOCK(rnh);
999 		}
1000 	}
1001 }
1002 
1003 struct rt_delinfo
1004 {
1005 	struct rt_addrinfo info;
1006 	struct rib_head *rnh;
1007 	struct rtentry *head;
1008 };
1009 
1010 /*
1011  * Conditionally unlinks @rn from radix tree based
1012  * on info data passed in @arg.
1013  */
1014 static int
1015 rt_checkdelroute(struct radix_node *rn, void *arg)
1016 {
1017 	struct rt_delinfo *di;
1018 	struct rt_addrinfo *info;
1019 	struct rtentry *rt;
1020 	int error;
1021 
1022 	di = (struct rt_delinfo *)arg;
1023 	rt = (struct rtentry *)rn;
1024 	info = &di->info;
1025 	error = 0;
1026 
1027 	info->rti_info[RTAX_DST] = rt_key(rt);
1028 	info->rti_info[RTAX_NETMASK] = rt_mask(rt);
1029 	info->rti_info[RTAX_GATEWAY] = rt->rt_gateway;
1030 
1031 	rt = rt_unlinkrte(di->rnh, info, &error);
1032 	if (rt == NULL) {
1033 		/* Either not allowed or not matched. Skip entry */
1034 		return (0);
1035 	}
1036 
1037 	/* Entry was unlinked. Add to the list and return */
1038 	rt->rt_chain = di->head;
1039 	di->head = rt;
1040 
1041 	return (0);
1042 }
1043 
1044 /*
1045  * Iterates over all existing fibs in system.
1046  * Deletes each element for which @filter_f function returned
1047  * non-zero value.
1048  * If @af is not AF_UNSPEC, iterates over fibs in particular
1049  * address family.
1050  */
1051 void
1052 rt_foreach_fib_walk_del(int af, rt_filter_f_t *filter_f, void *arg)
1053 {
1054 	struct rib_head *rnh;
1055 	struct rt_delinfo di;
1056 	struct rtentry *rt;
1057 	uint32_t fibnum;
1058 	int i, start, end;
1059 
1060 	bzero(&di, sizeof(di));
1061 	di.info.rti_filter = filter_f;
1062 	di.info.rti_filterdata = arg;
1063 
1064 	for (fibnum = 0; fibnum < rt_numfibs; fibnum++) {
1065 		/* Do we want some specific family? */
1066 		if (af != AF_UNSPEC) {
1067 			start = af;
1068 			end = af;
1069 		} else {
1070 			start = 1;
1071 			end = AF_MAX;
1072 		}
1073 
1074 		for (i = start; i <= end; i++) {
1075 			rnh = rt_tables_get_rnh(fibnum, i);
1076 			if (rnh == NULL)
1077 				continue;
1078 			di.rnh = rnh;
1079 
1080 			RIB_WLOCK(rnh);
1081 			rnh->rnh_walktree(&rnh->head, rt_checkdelroute, &di);
1082 			RIB_WUNLOCK(rnh);
1083 
1084 			if (di.head == NULL)
1085 				continue;
1086 
1087 			/* We might have something to reclaim */
1088 			while (di.head != NULL) {
1089 				rt = di.head;
1090 				di.head = rt->rt_chain;
1091 				rt->rt_chain = NULL;
1092 
1093 				/* TODO std rt -> rt_addrinfo export */
1094 				di.info.rti_info[RTAX_DST] = rt_key(rt);
1095 				di.info.rti_info[RTAX_NETMASK] = rt_mask(rt);
1096 
1097 				rt_notifydelete(rt, &di.info);
1098 				RTFREE_LOCKED(rt);
1099 			}
1100 
1101 		}
1102 	}
1103 }
1104 
1105 /*
1106  * Delete Routes for a Network Interface
1107  *
1108  * Called for each routing entry via the rnh->rnh_walktree() call above
1109  * to delete all route entries referencing a detaching network interface.
1110  *
1111  * Arguments:
1112  *	rt	pointer to rtentry
1113  *	arg	argument passed to rnh->rnh_walktree() - detaching interface
1114  *
1115  * Returns:
1116  *	0	successful
1117  *	errno	failed - reason indicated
1118  */
1119 static int
1120 rt_ifdelroute(const struct rtentry *rt, void *arg)
1121 {
1122 	struct ifnet	*ifp = arg;
1123 
1124 	if (rt->rt_ifp != ifp)
1125 		return (0);
1126 
1127 	/*
1128 	 * Protect (sorta) against walktree recursion problems
1129 	 * with cloned routes
1130 	 */
1131 	if ((rt->rt_flags & RTF_UP) == 0)
1132 		return (0);
1133 
1134 	return (1);
1135 }
1136 
1137 /*
1138  * Delete all remaining routes using this interface
1139  * Unfortuneatly the only way to do this is to slog through
1140  * the entire routing table looking for routes which point
1141  * to this interface...oh well...
1142  */
1143 void
1144 rt_flushifroutes_af(struct ifnet *ifp, int af)
1145 {
1146 	KASSERT((af >= 1 && af <= AF_MAX), ("%s: af %d not >= 1 and <= %d",
1147 	    __func__, af, AF_MAX));
1148 
1149 	rt_foreach_fib_walk_del(af, rt_ifdelroute, ifp);
1150 }
1151 
1152 void
1153 rt_flushifroutes(struct ifnet *ifp)
1154 {
1155 
1156 	rt_foreach_fib_walk_del(AF_UNSPEC, rt_ifdelroute, ifp);
1157 }
1158 
1159 /*
1160  * Conditionally unlinks rtentry matching data inside @info from @rnh.
1161  * Returns unlinked, locked and referenced @rtentry on success,
1162  * Returns NULL and sets @perror to:
1163  * ESRCH - if prefix was not found,
1164  * EADDRINUSE - if trying to delete PINNED route without appropriate flag.
1165  * ENOENT - if supplied filter function returned 0 (not matched).
1166  */
1167 static struct rtentry *
1168 rt_unlinkrte(struct rib_head *rnh, struct rt_addrinfo *info, int *perror)
1169 {
1170 	struct sockaddr *dst, *netmask;
1171 	struct rtentry *rt;
1172 	struct radix_node *rn;
1173 
1174 	dst = info->rti_info[RTAX_DST];
1175 	netmask = info->rti_info[RTAX_NETMASK];
1176 
1177 	rt = (struct rtentry *)rnh->rnh_lookup(dst, netmask, &rnh->head);
1178 	if (rt == NULL) {
1179 		*perror = ESRCH;
1180 		return (NULL);
1181 	}
1182 
1183 	if ((info->rti_flags & RTF_PINNED) == 0) {
1184 		/* Check if target route can be deleted */
1185 		if (rt->rt_flags & RTF_PINNED) {
1186 			*perror = EADDRINUSE;
1187 			return (NULL);
1188 		}
1189 	}
1190 
1191 	if (info->rti_filter != NULL) {
1192 		if (info->rti_filter(rt, info->rti_filterdata) == 0) {
1193 			/* Not matched */
1194 			*perror = ENOENT;
1195 			return (NULL);
1196 		}
1197 
1198 		/*
1199 		 * Filter function requested rte deletion.
1200 		 * Ease the caller work by filling in remaining info
1201 		 * from that particular entry.
1202 		 */
1203 		info->rti_info[RTAX_GATEWAY] = rt->rt_gateway;
1204 	}
1205 
1206 	/*
1207 	 * Remove the item from the tree and return it.
1208 	 * Complain if it is not there and do no more processing.
1209 	 */
1210 	*perror = ESRCH;
1211 #ifdef RADIX_MPATH
1212 	if (rt_mpath_capable(rnh))
1213 		rn = rt_mpath_unlink(rnh, info, rt, perror);
1214 	else
1215 #endif
1216 	rn = rnh->rnh_deladdr(dst, netmask, &rnh->head);
1217 	if (rn == NULL)
1218 		return (NULL);
1219 
1220 	if (rn->rn_flags & (RNF_ACTIVE | RNF_ROOT))
1221 		panic ("rtrequest delete");
1222 
1223 	rt = RNTORT(rn);
1224 	RT_LOCK(rt);
1225 	RT_ADDREF(rt);
1226 	rt->rt_flags &= ~RTF_UP;
1227 
1228 	*perror = 0;
1229 
1230 	return (rt);
1231 }
1232 
1233 static void
1234 rt_notifydelete(struct rtentry *rt, struct rt_addrinfo *info)
1235 {
1236 	struct ifaddr *ifa;
1237 
1238 	/*
1239 	 * give the protocol a chance to keep things in sync.
1240 	 */
1241 	ifa = rt->rt_ifa;
1242 	if (ifa != NULL && ifa->ifa_rtrequest != NULL)
1243 		ifa->ifa_rtrequest(RTM_DELETE, rt, info);
1244 
1245 	/*
1246 	 * One more rtentry floating around that is not
1247 	 * linked to the routing table. rttrash will be decremented
1248 	 * when RTFREE(rt) is eventually called.
1249 	 */
1250 	V_rttrash++;
1251 }
1252 
1253 
1254 /*
1255  * These (questionable) definitions of apparent local variables apply
1256  * to the next two functions.  XXXXXX!!!
1257  */
1258 #define	dst	info->rti_info[RTAX_DST]
1259 #define	gateway	info->rti_info[RTAX_GATEWAY]
1260 #define	netmask	info->rti_info[RTAX_NETMASK]
1261 #define	ifaaddr	info->rti_info[RTAX_IFA]
1262 #define	ifpaddr	info->rti_info[RTAX_IFP]
1263 #define	flags	info->rti_flags
1264 
1265 /*
1266  * Look up rt_addrinfo for a specific fib.  Note that if rti_ifa is defined,
1267  * it will be referenced so the caller must free it.
1268  */
1269 int
1270 rt_getifa_fib(struct rt_addrinfo *info, u_int fibnum)
1271 {
1272 	struct ifaddr *ifa;
1273 	int error = 0;
1274 
1275 	/*
1276 	 * ifp may be specified by sockaddr_dl
1277 	 * when protocol address is ambiguous.
1278 	 */
1279 	if (info->rti_ifp == NULL && ifpaddr != NULL &&
1280 	    ifpaddr->sa_family == AF_LINK &&
1281 	    (ifa = ifa_ifwithnet(ifpaddr, 0, fibnum)) != NULL) {
1282 		info->rti_ifp = ifa->ifa_ifp;
1283 		ifa_free(ifa);
1284 	}
1285 	if (info->rti_ifa == NULL && ifaaddr != NULL)
1286 		info->rti_ifa = ifa_ifwithaddr(ifaaddr);
1287 	if (info->rti_ifa == NULL) {
1288 		struct sockaddr *sa;
1289 
1290 		sa = ifaaddr != NULL ? ifaaddr :
1291 		    (gateway != NULL ? gateway : dst);
1292 		if (sa != NULL && info->rti_ifp != NULL)
1293 			info->rti_ifa = ifaof_ifpforaddr(sa, info->rti_ifp);
1294 		else if (dst != NULL && gateway != NULL)
1295 			info->rti_ifa = ifa_ifwithroute(flags, dst, gateway,
1296 							fibnum);
1297 		else if (sa != NULL)
1298 			info->rti_ifa = ifa_ifwithroute(flags, sa, sa,
1299 							fibnum);
1300 	}
1301 	if ((ifa = info->rti_ifa) != NULL) {
1302 		if (info->rti_ifp == NULL)
1303 			info->rti_ifp = ifa->ifa_ifp;
1304 	} else
1305 		error = ENETUNREACH;
1306 	return (error);
1307 }
1308 
1309 static int
1310 if_updatemtu_cb(struct radix_node *rn, void *arg)
1311 {
1312 	struct rtentry *rt;
1313 	struct if_mtuinfo *ifmtu;
1314 
1315 	rt = (struct rtentry *)rn;
1316 	ifmtu = (struct if_mtuinfo *)arg;
1317 
1318 	if (rt->rt_ifp != ifmtu->ifp)
1319 		return (0);
1320 
1321 	if (rt->rt_mtu >= ifmtu->mtu) {
1322 		/* We have to decrease mtu regardless of flags */
1323 		rt->rt_mtu = ifmtu->mtu;
1324 		return (0);
1325 	}
1326 
1327 	/*
1328 	 * New MTU is bigger. Check if are allowed to alter it
1329 	 */
1330 	if ((rt->rt_flags & (RTF_FIXEDMTU | RTF_GATEWAY | RTF_HOST)) != 0) {
1331 
1332 		/*
1333 		 * Skip routes with user-supplied MTU and
1334 		 * non-interface routes
1335 		 */
1336 		return (0);
1337 	}
1338 
1339 	/* We are safe to update route MTU */
1340 	rt->rt_mtu = ifmtu->mtu;
1341 
1342 	return (0);
1343 }
1344 
1345 void
1346 rt_updatemtu(struct ifnet *ifp)
1347 {
1348 	struct if_mtuinfo ifmtu;
1349 	struct rib_head *rnh;
1350 	int i, j;
1351 
1352 	ifmtu.ifp = ifp;
1353 
1354 	/*
1355 	 * Try to update rt_mtu for all routes using this interface
1356 	 * Unfortunately the only way to do this is to traverse all
1357 	 * routing tables in all fibs/domains.
1358 	 */
1359 	for (i = 1; i <= AF_MAX; i++) {
1360 		ifmtu.mtu = if_getmtu_family(ifp, i);
1361 		for (j = 0; j < rt_numfibs; j++) {
1362 			rnh = rt_tables_get_rnh(j, i);
1363 			if (rnh == NULL)
1364 				continue;
1365 			RIB_WLOCK(rnh);
1366 			rnh->rnh_walktree(&rnh->head, if_updatemtu_cb, &ifmtu);
1367 			RIB_WUNLOCK(rnh);
1368 		}
1369 	}
1370 }
1371 
1372 
1373 #if 0
1374 int p_sockaddr(char *buf, int buflen, struct sockaddr *s);
1375 int rt_print(char *buf, int buflen, struct rtentry *rt);
1376 
1377 int
1378 p_sockaddr(char *buf, int buflen, struct sockaddr *s)
1379 {
1380 	void *paddr = NULL;
1381 
1382 	switch (s->sa_family) {
1383 	case AF_INET:
1384 		paddr = &((struct sockaddr_in *)s)->sin_addr;
1385 		break;
1386 	case AF_INET6:
1387 		paddr = &((struct sockaddr_in6 *)s)->sin6_addr;
1388 		break;
1389 	}
1390 
1391 	if (paddr == NULL)
1392 		return (0);
1393 
1394 	if (inet_ntop(s->sa_family, paddr, buf, buflen) == NULL)
1395 		return (0);
1396 
1397 	return (strlen(buf));
1398 }
1399 
1400 int
1401 rt_print(char *buf, int buflen, struct rtentry *rt)
1402 {
1403 	struct sockaddr *addr, *mask;
1404 	int i = 0;
1405 
1406 	addr = rt_key(rt);
1407 	mask = rt_mask(rt);
1408 
1409 	i = p_sockaddr(buf, buflen, addr);
1410 	if (!(rt->rt_flags & RTF_HOST)) {
1411 		buf[i++] = '/';
1412 		i += p_sockaddr(buf + i, buflen - i, mask);
1413 	}
1414 
1415 	if (rt->rt_flags & RTF_GATEWAY) {
1416 		buf[i++] = '>';
1417 		i += p_sockaddr(buf + i, buflen - i, rt->rt_gateway);
1418 	}
1419 
1420 	return (i);
1421 }
1422 #endif
1423 
1424 #ifdef RADIX_MPATH
1425 /*
1426  * Deletes key for single-path routes, unlinks rtentry with
1427  * gateway specified in @info from multi-path routes.
1428  *
1429  * Returnes unlinked entry. In case of failure, returns NULL
1430  * and sets @perror to ESRCH.
1431  */
1432 static struct radix_node *
1433 rt_mpath_unlink(struct rib_head *rnh, struct rt_addrinfo *info,
1434     struct rtentry *rto, int *perror)
1435 {
1436 	/*
1437 	 * if we got multipath routes, we require users to specify
1438 	 * a matching RTAX_GATEWAY.
1439 	 */
1440 	struct rtentry *rt; // *rto = NULL;
1441 	struct radix_node *rn;
1442 	struct sockaddr *gw;
1443 
1444 	gw = info->rti_info[RTAX_GATEWAY];
1445 	rt = rt_mpath_matchgate(rto, gw);
1446 	if (rt == NULL) {
1447 		*perror = ESRCH;
1448 		return (NULL);
1449 	}
1450 
1451 	/*
1452 	 * this is the first entry in the chain
1453 	 */
1454 	if (rto == rt) {
1455 		rn = rn_mpath_next((struct radix_node *)rt);
1456 		/*
1457 		 * there is another entry, now it's active
1458 		 */
1459 		if (rn) {
1460 			rto = RNTORT(rn);
1461 			RT_LOCK(rto);
1462 			rto->rt_flags |= RTF_UP;
1463 			RT_UNLOCK(rto);
1464 		} else if (rt->rt_flags & RTF_GATEWAY) {
1465 			/*
1466 			 * For gateway routes, we need to
1467 			 * make sure that we we are deleting
1468 			 * the correct gateway.
1469 			 * rt_mpath_matchgate() does not
1470 			 * check the case when there is only
1471 			 * one route in the chain.
1472 			 */
1473 			if (gw &&
1474 			    (rt->rt_gateway->sa_len != gw->sa_len ||
1475 				memcmp(rt->rt_gateway, gw, gw->sa_len))) {
1476 				*perror = ESRCH;
1477 				return (NULL);
1478 			}
1479 		}
1480 
1481 		/*
1482 		 * use the normal delete code to remove
1483 		 * the first entry
1484 		 */
1485 		rn = rnh->rnh_deladdr(dst, netmask, &rnh->head);
1486 		*perror = 0;
1487 		return (rn);
1488 	}
1489 
1490 	/*
1491 	 * if the entry is 2nd and on up
1492 	 */
1493 	if (rt_mpath_deldup(rto, rt) == 0)
1494 		panic ("rtrequest1: rt_mpath_deldup");
1495 	*perror = 0;
1496 	rn = (struct radix_node *)rt;
1497 	return (rn);
1498 }
1499 #endif
1500 
1501 #ifdef FLOWTABLE
1502 static struct rtentry *
1503 rt_flowtable_check_route(struct rib_head *rnh, struct rt_addrinfo *info)
1504 {
1505 #if defined(INET6) || defined(INET)
1506 	struct radix_node *rn;
1507 #endif
1508 	struct rtentry *rt0;
1509 
1510 	rt0 = NULL;
1511 	/* "flow-table" only supports IPv6 and IPv4 at the moment. */
1512 	switch (dst->sa_family) {
1513 #ifdef INET6
1514 	case AF_INET6:
1515 #endif
1516 #ifdef INET
1517 	case AF_INET:
1518 #endif
1519 #if defined(INET6) || defined(INET)
1520 		rn = rnh->rnh_matchaddr(dst, &rnh->head);
1521 		if (rn && ((rn->rn_flags & RNF_ROOT) == 0)) {
1522 			struct sockaddr *mask;
1523 			u_char *m, *n;
1524 			int len;
1525 
1526 			/*
1527 			 * compare mask to see if the new route is
1528 			 * more specific than the existing one
1529 			 */
1530 			rt0 = RNTORT(rn);
1531 			RT_LOCK(rt0);
1532 			RT_ADDREF(rt0);
1533 			RT_UNLOCK(rt0);
1534 			/*
1535 			 * A host route is already present, so
1536 			 * leave the flow-table entries as is.
1537 			 */
1538 			if (rt0->rt_flags & RTF_HOST) {
1539 				RTFREE(rt0);
1540 				rt0 = NULL;
1541 			} else if (!(flags & RTF_HOST) && netmask) {
1542 				mask = rt_mask(rt0);
1543 				len = mask->sa_len;
1544 				m = (u_char *)mask;
1545 				n = (u_char *)netmask;
1546 				while (len-- > 0) {
1547 					if (*n != *m)
1548 						break;
1549 					n++;
1550 					m++;
1551 				}
1552 				if (len == 0 || (*n < *m)) {
1553 					RTFREE(rt0);
1554 					rt0 = NULL;
1555 				}
1556 			}
1557 		}
1558 #endif/* INET6 || INET */
1559 	}
1560 
1561 	return (rt0);
1562 }
1563 #endif
1564 
1565 int
1566 rtrequest1_fib(int req, struct rt_addrinfo *info, struct rtentry **ret_nrt,
1567 				u_int fibnum)
1568 {
1569 	int error = 0;
1570 	struct rtentry *rt, *rt_old;
1571 #ifdef FLOWTABLE
1572 	struct rtentry *rt0;
1573 #endif
1574 	struct radix_node *rn;
1575 	struct rib_head *rnh;
1576 	struct ifaddr *ifa;
1577 	struct sockaddr *ndst;
1578 	struct sockaddr_storage mdst;
1579 
1580 	KASSERT((fibnum < rt_numfibs), ("rtrequest1_fib: bad fibnum"));
1581 	KASSERT((flags & RTF_RNH_LOCKED) == 0, ("rtrequest1_fib: locked"));
1582 	switch (dst->sa_family) {
1583 	case AF_INET6:
1584 	case AF_INET:
1585 		/* We support multiple FIBs. */
1586 		break;
1587 	default:
1588 		fibnum = RT_DEFAULT_FIB;
1589 		break;
1590 	}
1591 
1592 	/*
1593 	 * Find the correct routing tree to use for this Address Family
1594 	 */
1595 	rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
1596 	if (rnh == NULL)
1597 		return (EAFNOSUPPORT);
1598 
1599 	/*
1600 	 * If we are adding a host route then we don't want to put
1601 	 * a netmask in the tree, nor do we want to clone it.
1602 	 */
1603 	if (flags & RTF_HOST)
1604 		netmask = NULL;
1605 
1606 	switch (req) {
1607 	case RTM_DELETE:
1608 		if (netmask) {
1609 			rt_maskedcopy(dst, (struct sockaddr *)&mdst, netmask);
1610 			dst = (struct sockaddr *)&mdst;
1611 		}
1612 
1613 		RIB_WLOCK(rnh);
1614 		rt = rt_unlinkrte(rnh, info, &error);
1615 		RIB_WUNLOCK(rnh);
1616 		if (error != 0)
1617 			return (error);
1618 
1619 		rt_notifydelete(rt, info);
1620 
1621 		/*
1622 		 * If the caller wants it, then it can have it,
1623 		 * but it's up to it to free the rtentry as we won't be
1624 		 * doing it.
1625 		 */
1626 		if (ret_nrt) {
1627 			*ret_nrt = rt;
1628 			RT_UNLOCK(rt);
1629 		} else
1630 			RTFREE_LOCKED(rt);
1631 		break;
1632 	case RTM_RESOLVE:
1633 		/*
1634 		 * resolve was only used for route cloning
1635 		 * here for compat
1636 		 */
1637 		break;
1638 	case RTM_ADD:
1639 		if ((flags & RTF_GATEWAY) && !gateway)
1640 			return (EINVAL);
1641 		if (dst && gateway && (dst->sa_family != gateway->sa_family) &&
1642 		    (gateway->sa_family != AF_UNSPEC) && (gateway->sa_family != AF_LINK))
1643 			return (EINVAL);
1644 
1645 		if (info->rti_ifa == NULL) {
1646 			error = rt_getifa_fib(info, fibnum);
1647 			if (error)
1648 				return (error);
1649 		} else
1650 			ifa_ref(info->rti_ifa);
1651 		ifa = info->rti_ifa;
1652 		rt = uma_zalloc(V_rtzone, M_NOWAIT);
1653 		if (rt == NULL) {
1654 			ifa_free(ifa);
1655 			return (ENOBUFS);
1656 		}
1657 		rt->rt_flags = RTF_UP | flags;
1658 		rt->rt_fibnum = fibnum;
1659 		/*
1660 		 * Add the gateway. Possibly re-malloc-ing the storage for it.
1661 		 */
1662 		if ((error = rt_setgate(rt, dst, gateway)) != 0) {
1663 			ifa_free(ifa);
1664 			uma_zfree(V_rtzone, rt);
1665 			return (error);
1666 		}
1667 
1668 		/*
1669 		 * point to the (possibly newly malloc'd) dest address.
1670 		 */
1671 		ndst = (struct sockaddr *)rt_key(rt);
1672 
1673 		/*
1674 		 * make sure it contains the value we want (masked if needed).
1675 		 */
1676 		if (netmask) {
1677 			rt_maskedcopy(dst, ndst, netmask);
1678 		} else
1679 			bcopy(dst, ndst, dst->sa_len);
1680 
1681 		/*
1682 		 * We use the ifa reference returned by rt_getifa_fib().
1683 		 * This moved from below so that rnh->rnh_addaddr() can
1684 		 * examine the ifa and  ifa->ifa_ifp if it so desires.
1685 		 */
1686 		rt->rt_ifa = ifa;
1687 		rt->rt_ifp = ifa->ifa_ifp;
1688 		rt->rt_weight = 1;
1689 
1690 		rt_setmetrics(info, rt);
1691 
1692 		RIB_WLOCK(rnh);
1693 		RT_LOCK(rt);
1694 #ifdef RADIX_MPATH
1695 		/* do not permit exactly the same dst/mask/gw pair */
1696 		if (rt_mpath_capable(rnh) &&
1697 			rt_mpath_conflict(rnh, rt, netmask)) {
1698 			RIB_WUNLOCK(rnh);
1699 
1700 			ifa_free(rt->rt_ifa);
1701 			R_Free(rt_key(rt));
1702 			uma_zfree(V_rtzone, rt);
1703 			return (EEXIST);
1704 		}
1705 #endif
1706 
1707 #ifdef FLOWTABLE
1708 		rt0 = rt_flowtable_check_route(rnh, info);
1709 #endif /* FLOWTABLE */
1710 
1711 		/* XXX mtu manipulation will be done in rnh_addaddr -- itojun */
1712 		rn = rnh->rnh_addaddr(ndst, netmask, &rnh->head, rt->rt_nodes);
1713 
1714 		rt_old = NULL;
1715 		if (rn == NULL && (info->rti_flags & RTF_PINNED) != 0) {
1716 
1717 			/*
1718 			 * Force removal and re-try addition
1719 			 * TODO: better multipath&pinned support
1720 			 */
1721 			struct sockaddr *info_dst = info->rti_info[RTAX_DST];
1722 			info->rti_info[RTAX_DST] = ndst;
1723 			/* Do not delete existing PINNED(interface) routes */
1724 			info->rti_flags &= ~RTF_PINNED;
1725 			rt_old = rt_unlinkrte(rnh, info, &error);
1726 			info->rti_flags |= RTF_PINNED;
1727 			info->rti_info[RTAX_DST] = info_dst;
1728 			if (rt_old != NULL)
1729 				rn = rnh->rnh_addaddr(ndst, netmask, &rnh->head,
1730 				    rt->rt_nodes);
1731 		}
1732 		RIB_WUNLOCK(rnh);
1733 
1734 		if (rt_old != NULL)
1735 			RT_UNLOCK(rt_old);
1736 
1737 		/*
1738 		 * If it still failed to go into the tree,
1739 		 * then un-make it (this should be a function)
1740 		 */
1741 		if (rn == NULL) {
1742 			ifa_free(rt->rt_ifa);
1743 			R_Free(rt_key(rt));
1744 			uma_zfree(V_rtzone, rt);
1745 #ifdef FLOWTABLE
1746 			if (rt0 != NULL)
1747 				RTFREE(rt0);
1748 #endif
1749 			return (EEXIST);
1750 		}
1751 #ifdef FLOWTABLE
1752 		else if (rt0 != NULL) {
1753 			flowtable_route_flush(dst->sa_family, rt0);
1754 			RTFREE(rt0);
1755 		}
1756 #endif
1757 
1758 		if (rt_old != NULL) {
1759 			rt_notifydelete(rt_old, info);
1760 			RTFREE(rt_old);
1761 		}
1762 
1763 		/*
1764 		 * If this protocol has something to add to this then
1765 		 * allow it to do that as well.
1766 		 */
1767 		if (ifa->ifa_rtrequest)
1768 			ifa->ifa_rtrequest(req, rt, info);
1769 
1770 		/*
1771 		 * actually return a resultant rtentry and
1772 		 * give the caller a single reference.
1773 		 */
1774 		if (ret_nrt) {
1775 			*ret_nrt = rt;
1776 			RT_ADDREF(rt);
1777 		}
1778 		rnh->rnh_gen++;		/* Routing table updated */
1779 		RT_UNLOCK(rt);
1780 		break;
1781 	case RTM_CHANGE:
1782 		RIB_WLOCK(rnh);
1783 		error = rtrequest1_fib_change(rnh, info, ret_nrt, fibnum);
1784 		RIB_WUNLOCK(rnh);
1785 		break;
1786 	default:
1787 		error = EOPNOTSUPP;
1788 	}
1789 
1790 	return (error);
1791 }
1792 
1793 #undef dst
1794 #undef gateway
1795 #undef netmask
1796 #undef ifaaddr
1797 #undef ifpaddr
1798 #undef flags
1799 
1800 static int
1801 rtrequest1_fib_change(struct rib_head *rnh, struct rt_addrinfo *info,
1802     struct rtentry **ret_nrt, u_int fibnum)
1803 {
1804 	struct rtentry *rt = NULL;
1805 	int error = 0;
1806 	int free_ifa = 0;
1807 	int family, mtu;
1808 	struct if_mtuinfo ifmtu;
1809 
1810 	rt = (struct rtentry *)rnh->rnh_lookup(info->rti_info[RTAX_DST],
1811 	    info->rti_info[RTAX_NETMASK], &rnh->head);
1812 
1813 	if (rt == NULL)
1814 		return (ESRCH);
1815 
1816 #ifdef RADIX_MPATH
1817 	/*
1818 	 * If we got multipath routes,
1819 	 * we require users to specify a matching RTAX_GATEWAY.
1820 	 */
1821 	if (rt_mpath_capable(rnh)) {
1822 		rt = rt_mpath_matchgate(rt, info->rti_info[RTAX_GATEWAY]);
1823 		if (rt == NULL)
1824 			return (ESRCH);
1825 	}
1826 #endif
1827 
1828 	RT_LOCK(rt);
1829 
1830 	rt_setmetrics(info, rt);
1831 
1832 	/*
1833 	 * New gateway could require new ifaddr, ifp;
1834 	 * flags may also be different; ifp may be specified
1835 	 * by ll sockaddr when protocol address is ambiguous
1836 	 */
1837 	if (((rt->rt_flags & RTF_GATEWAY) &&
1838 	    info->rti_info[RTAX_GATEWAY] != NULL) ||
1839 	    info->rti_info[RTAX_IFP] != NULL ||
1840 	    (info->rti_info[RTAX_IFA] != NULL &&
1841 	     !sa_equal(info->rti_info[RTAX_IFA], rt->rt_ifa->ifa_addr))) {
1842 
1843 		error = rt_getifa_fib(info, fibnum);
1844 		if (info->rti_ifa != NULL)
1845 			free_ifa = 1;
1846 
1847 		if (error != 0)
1848 			goto bad;
1849 	}
1850 
1851 	/* Check if outgoing interface has changed */
1852 	if (info->rti_ifa != NULL && info->rti_ifa != rt->rt_ifa &&
1853 	    rt->rt_ifa != NULL && rt->rt_ifa->ifa_rtrequest != NULL) {
1854 		rt->rt_ifa->ifa_rtrequest(RTM_DELETE, rt, info);
1855 		ifa_free(rt->rt_ifa);
1856 	}
1857 	/* Update gateway address */
1858 	if (info->rti_info[RTAX_GATEWAY] != NULL) {
1859 		error = rt_setgate(rt, rt_key(rt), info->rti_info[RTAX_GATEWAY]);
1860 		if (error != 0)
1861 			goto bad;
1862 
1863 		rt->rt_flags &= ~RTF_GATEWAY;
1864 		rt->rt_flags |= (RTF_GATEWAY & info->rti_flags);
1865 	}
1866 
1867 	if (info->rti_ifa != NULL && info->rti_ifa != rt->rt_ifa) {
1868 		ifa_ref(info->rti_ifa);
1869 		rt->rt_ifa = info->rti_ifa;
1870 		rt->rt_ifp = info->rti_ifp;
1871 	}
1872 	/* Allow some flags to be toggled on change. */
1873 	rt->rt_flags &= ~RTF_FMASK;
1874 	rt->rt_flags |= info->rti_flags & RTF_FMASK;
1875 
1876 	if (rt->rt_ifa && rt->rt_ifa->ifa_rtrequest != NULL)
1877 	       rt->rt_ifa->ifa_rtrequest(RTM_ADD, rt, info);
1878 
1879 	/* Alter route MTU if necessary */
1880 	if (rt->rt_ifp != NULL) {
1881 		family = info->rti_info[RTAX_DST]->sa_family;
1882 		mtu = if_getmtu_family(rt->rt_ifp, family);
1883 		/* Set default MTU */
1884 		if (rt->rt_mtu == 0)
1885 			rt->rt_mtu = mtu;
1886 		if (rt->rt_mtu != mtu) {
1887 			/* Check if we really need to update */
1888 			ifmtu.ifp = rt->rt_ifp;
1889 			ifmtu.mtu = mtu;
1890 			if_updatemtu_cb(rt->rt_nodes, &ifmtu);
1891 		}
1892 	}
1893 
1894 	if (ret_nrt) {
1895 		*ret_nrt = rt;
1896 		RT_ADDREF(rt);
1897 	}
1898 bad:
1899 	RT_UNLOCK(rt);
1900 	if (free_ifa != 0)
1901 		ifa_free(info->rti_ifa);
1902 	return (error);
1903 }
1904 
1905 static void
1906 rt_setmetrics(const struct rt_addrinfo *info, struct rtentry *rt)
1907 {
1908 
1909 	if (info->rti_mflags & RTV_MTU) {
1910 		if (info->rti_rmx->rmx_mtu != 0) {
1911 
1912 			/*
1913 			 * MTU was explicitly provided by user.
1914 			 * Keep it.
1915 			 */
1916 			rt->rt_flags |= RTF_FIXEDMTU;
1917 		} else {
1918 
1919 			/*
1920 			 * User explicitly sets MTU to 0.
1921 			 * Assume rollback to default.
1922 			 */
1923 			rt->rt_flags &= ~RTF_FIXEDMTU;
1924 		}
1925 		rt->rt_mtu = info->rti_rmx->rmx_mtu;
1926 	}
1927 	if (info->rti_mflags & RTV_WEIGHT)
1928 		rt->rt_weight = info->rti_rmx->rmx_weight;
1929 	/* Kernel -> userland timebase conversion. */
1930 	if (info->rti_mflags & RTV_EXPIRE)
1931 		rt->rt_expire = info->rti_rmx->rmx_expire ?
1932 		    info->rti_rmx->rmx_expire - time_second + time_uptime : 0;
1933 }
1934 
1935 int
1936 rt_setgate(struct rtentry *rt, struct sockaddr *dst, struct sockaddr *gate)
1937 {
1938 	/* XXX dst may be overwritten, can we move this to below */
1939 	int dlen = SA_SIZE(dst), glen = SA_SIZE(gate);
1940 
1941 	/*
1942 	 * Prepare to store the gateway in rt->rt_gateway.
1943 	 * Both dst and gateway are stored one after the other in the same
1944 	 * malloc'd chunk. If we have room, we can reuse the old buffer,
1945 	 * rt_gateway already points to the right place.
1946 	 * Otherwise, malloc a new block and update the 'dst' address.
1947 	 */
1948 	if (rt->rt_gateway == NULL || glen > SA_SIZE(rt->rt_gateway)) {
1949 		caddr_t new;
1950 
1951 		R_Malloc(new, caddr_t, dlen + glen);
1952 		if (new == NULL)
1953 			return ENOBUFS;
1954 		/*
1955 		 * XXX note, we copy from *dst and not *rt_key(rt) because
1956 		 * rt_setgate() can be called to initialize a newly
1957 		 * allocated route entry, in which case rt_key(rt) == NULL
1958 		 * (and also rt->rt_gateway == NULL).
1959 		 * Free()/free() handle a NULL argument just fine.
1960 		 */
1961 		bcopy(dst, new, dlen);
1962 		R_Free(rt_key(rt));	/* free old block, if any */
1963 		rt_key(rt) = (struct sockaddr *)new;
1964 		rt->rt_gateway = (struct sockaddr *)(new + dlen);
1965 	}
1966 
1967 	/*
1968 	 * Copy the new gateway value into the memory chunk.
1969 	 */
1970 	bcopy(gate, rt->rt_gateway, glen);
1971 
1972 	return (0);
1973 }
1974 
1975 void
1976 rt_maskedcopy(struct sockaddr *src, struct sockaddr *dst, struct sockaddr *netmask)
1977 {
1978 	u_char *cp1 = (u_char *)src;
1979 	u_char *cp2 = (u_char *)dst;
1980 	u_char *cp3 = (u_char *)netmask;
1981 	u_char *cplim = cp2 + *cp3;
1982 	u_char *cplim2 = cp2 + *cp1;
1983 
1984 	*cp2++ = *cp1++; *cp2++ = *cp1++; /* copies sa_len & sa_family */
1985 	cp3 += 2;
1986 	if (cplim > cplim2)
1987 		cplim = cplim2;
1988 	while (cp2 < cplim)
1989 		*cp2++ = *cp1++ & *cp3++;
1990 	if (cp2 < cplim2)
1991 		bzero((caddr_t)cp2, (unsigned)(cplim2 - cp2));
1992 }
1993 
1994 /*
1995  * Set up a routing table entry, normally
1996  * for an interface.
1997  */
1998 #define _SOCKADDR_TMPSIZE 128 /* Not too big.. kernel stack size is limited */
1999 static inline  int
2000 rtinit1(struct ifaddr *ifa, int cmd, int flags, int fibnum)
2001 {
2002 	struct sockaddr *dst;
2003 	struct sockaddr *netmask;
2004 	struct rtentry *rt = NULL;
2005 	struct rt_addrinfo info;
2006 	int error = 0;
2007 	int startfib, endfib;
2008 	char tempbuf[_SOCKADDR_TMPSIZE];
2009 	int didwork = 0;
2010 	int a_failure = 0;
2011 	static struct sockaddr_dl null_sdl = {sizeof(null_sdl), AF_LINK};
2012 	struct rib_head *rnh;
2013 
2014 	if (flags & RTF_HOST) {
2015 		dst = ifa->ifa_dstaddr;
2016 		netmask = NULL;
2017 	} else {
2018 		dst = ifa->ifa_addr;
2019 		netmask = ifa->ifa_netmask;
2020 	}
2021 	if (dst->sa_len == 0)
2022 		return(EINVAL);
2023 	switch (dst->sa_family) {
2024 	case AF_INET6:
2025 	case AF_INET:
2026 		/* We support multiple FIBs. */
2027 		break;
2028 	default:
2029 		fibnum = RT_DEFAULT_FIB;
2030 		break;
2031 	}
2032 	if (fibnum == RT_ALL_FIBS) {
2033 		if (V_rt_add_addr_allfibs == 0 && cmd == (int)RTM_ADD)
2034 			startfib = endfib = ifa->ifa_ifp->if_fib;
2035 		else {
2036 			startfib = 0;
2037 			endfib = rt_numfibs - 1;
2038 		}
2039 	} else {
2040 		KASSERT((fibnum < rt_numfibs), ("rtinit1: bad fibnum"));
2041 		startfib = fibnum;
2042 		endfib = fibnum;
2043 	}
2044 
2045 	/*
2046 	 * If it's a delete, check that if it exists,
2047 	 * it's on the correct interface or we might scrub
2048 	 * a route to another ifa which would
2049 	 * be confusing at best and possibly worse.
2050 	 */
2051 	if (cmd == RTM_DELETE) {
2052 		/*
2053 		 * It's a delete, so it should already exist..
2054 		 * If it's a net, mask off the host bits
2055 		 * (Assuming we have a mask)
2056 		 * XXX this is kinda inet specific..
2057 		 */
2058 		if (netmask != NULL) {
2059 			rt_maskedcopy(dst, (struct sockaddr *)tempbuf, netmask);
2060 			dst = (struct sockaddr *)tempbuf;
2061 		}
2062 	}
2063 	/*
2064 	 * Now go through all the requested tables (fibs) and do the
2065 	 * requested action. Realistically, this will either be fib 0
2066 	 * for protocols that don't do multiple tables or all the
2067 	 * tables for those that do.
2068 	 */
2069 	for ( fibnum = startfib; fibnum <= endfib; fibnum++) {
2070 		if (cmd == RTM_DELETE) {
2071 			struct radix_node *rn;
2072 			/*
2073 			 * Look up an rtentry that is in the routing tree and
2074 			 * contains the correct info.
2075 			 */
2076 			rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
2077 			if (rnh == NULL)
2078 				/* this table doesn't exist but others might */
2079 				continue;
2080 			RIB_RLOCK(rnh);
2081 			rn = rnh->rnh_lookup(dst, netmask, &rnh->head);
2082 #ifdef RADIX_MPATH
2083 			if (rt_mpath_capable(rnh)) {
2084 
2085 				if (rn == NULL)
2086 					error = ESRCH;
2087 				else {
2088 					rt = RNTORT(rn);
2089 					/*
2090 					 * for interface route the
2091 					 * rt->rt_gateway is sockaddr_intf
2092 					 * for cloning ARP entries, so
2093 					 * rt_mpath_matchgate must use the
2094 					 * interface address
2095 					 */
2096 					rt = rt_mpath_matchgate(rt,
2097 					    ifa->ifa_addr);
2098 					if (rt == NULL)
2099 						error = ESRCH;
2100 				}
2101 			}
2102 #endif
2103 			error = (rn == NULL ||
2104 			    (rn->rn_flags & RNF_ROOT) ||
2105 			    RNTORT(rn)->rt_ifa != ifa);
2106 			RIB_RUNLOCK(rnh);
2107 			if (error) {
2108 				/* this is only an error if bad on ALL tables */
2109 				continue;
2110 			}
2111 		}
2112 		/*
2113 		 * Do the actual request
2114 		 */
2115 		bzero((caddr_t)&info, sizeof(info));
2116 		info.rti_ifa = ifa;
2117 		info.rti_flags = flags |
2118 		    (ifa->ifa_flags & ~IFA_RTSELF) | RTF_PINNED;
2119 		info.rti_info[RTAX_DST] = dst;
2120 		/*
2121 		 * doing this for compatibility reasons
2122 		 */
2123 		if (cmd == RTM_ADD)
2124 			info.rti_info[RTAX_GATEWAY] =
2125 			    (struct sockaddr *)&null_sdl;
2126 		else
2127 			info.rti_info[RTAX_GATEWAY] = ifa->ifa_addr;
2128 		info.rti_info[RTAX_NETMASK] = netmask;
2129 		error = rtrequest1_fib(cmd, &info, &rt, fibnum);
2130 
2131 		if (error == 0 && rt != NULL) {
2132 			/*
2133 			 * notify any listening routing agents of the change
2134 			 */
2135 			RT_LOCK(rt);
2136 #ifdef RADIX_MPATH
2137 			/*
2138 			 * in case address alias finds the first address
2139 			 * e.g. ifconfig bge0 192.0.2.246/24
2140 			 * e.g. ifconfig bge0 192.0.2.247/24
2141 			 * the address set in the route is 192.0.2.246
2142 			 * so we need to replace it with 192.0.2.247
2143 			 */
2144 			if (memcmp(rt->rt_ifa->ifa_addr,
2145 			    ifa->ifa_addr, ifa->ifa_addr->sa_len)) {
2146 				ifa_free(rt->rt_ifa);
2147 				ifa_ref(ifa);
2148 				rt->rt_ifp = ifa->ifa_ifp;
2149 				rt->rt_ifa = ifa;
2150 			}
2151 #endif
2152 			/*
2153 			 * doing this for compatibility reasons
2154 			 */
2155 			if (cmd == RTM_ADD) {
2156 			    ((struct sockaddr_dl *)rt->rt_gateway)->sdl_type  =
2157 				rt->rt_ifp->if_type;
2158 			    ((struct sockaddr_dl *)rt->rt_gateway)->sdl_index =
2159 				rt->rt_ifp->if_index;
2160 			}
2161 			RT_ADDREF(rt);
2162 			RT_UNLOCK(rt);
2163 			rt_newaddrmsg_fib(cmd, ifa, error, rt, fibnum);
2164 			RT_LOCK(rt);
2165 			RT_REMREF(rt);
2166 			if (cmd == RTM_DELETE) {
2167 				/*
2168 				 * If we are deleting, and we found an entry,
2169 				 * then it's been removed from the tree..
2170 				 * now throw it away.
2171 				 */
2172 				RTFREE_LOCKED(rt);
2173 			} else {
2174 				if (cmd == RTM_ADD) {
2175 					/*
2176 					 * We just wanted to add it..
2177 					 * we don't actually need a reference.
2178 					 */
2179 					RT_REMREF(rt);
2180 				}
2181 				RT_UNLOCK(rt);
2182 			}
2183 			didwork = 1;
2184 		}
2185 		if (error)
2186 			a_failure = error;
2187 	}
2188 	if (cmd == RTM_DELETE) {
2189 		if (didwork) {
2190 			error = 0;
2191 		} else {
2192 			/* we only give an error if it wasn't in any table */
2193 			error = ((flags & RTF_HOST) ?
2194 			    EHOSTUNREACH : ENETUNREACH);
2195 		}
2196 	} else {
2197 		if (a_failure) {
2198 			/* return an error if any of them failed */
2199 			error = a_failure;
2200 		}
2201 	}
2202 	return (error);
2203 }
2204 
2205 /*
2206  * Set up a routing table entry, normally
2207  * for an interface.
2208  */
2209 int
2210 rtinit(struct ifaddr *ifa, int cmd, int flags)
2211 {
2212 	struct sockaddr *dst;
2213 	int fib = RT_DEFAULT_FIB;
2214 
2215 	if (flags & RTF_HOST) {
2216 		dst = ifa->ifa_dstaddr;
2217 	} else {
2218 		dst = ifa->ifa_addr;
2219 	}
2220 
2221 	switch (dst->sa_family) {
2222 	case AF_INET6:
2223 	case AF_INET:
2224 		/* We do support multiple FIBs. */
2225 		fib = RT_ALL_FIBS;
2226 		break;
2227 	}
2228 	return (rtinit1(ifa, cmd, flags, fib));
2229 }
2230 
2231 /*
2232  * Announce interface address arrival/withdraw
2233  * Returns 0 on success.
2234  */
2235 int
2236 rt_addrmsg(int cmd, struct ifaddr *ifa, int fibnum)
2237 {
2238 
2239 	KASSERT(cmd == RTM_ADD || cmd == RTM_DELETE,
2240 	    ("unexpected cmd %d", cmd));
2241 
2242 	KASSERT(fibnum == RT_ALL_FIBS || (fibnum >= 0 && fibnum < rt_numfibs),
2243 	    ("%s: fib out of range 0 <=%d<%d", __func__, fibnum, rt_numfibs));
2244 
2245 #if defined(INET) || defined(INET6)
2246 #ifdef SCTP
2247 	/*
2248 	 * notify the SCTP stack
2249 	 * this will only get called when an address is added/deleted
2250 	 * XXX pass the ifaddr struct instead if ifa->ifa_addr...
2251 	 */
2252 	sctp_addr_change(ifa, cmd);
2253 #endif /* SCTP */
2254 #endif
2255 	return (rtsock_addrmsg(cmd, ifa, fibnum));
2256 }
2257 
2258 /*
2259  * Announce route addition/removal.
2260  * Users of this function MUST validate input data BEFORE calling.
2261  * However we have to be able to handle invalid data:
2262  * if some userland app sends us "invalid" route message (invalid mask,
2263  * no dst, wrong address families, etc...) we need to pass it back
2264  * to app (and any other rtsock consumers) with rtm_errno field set to
2265  * non-zero value.
2266  * Returns 0 on success.
2267  */
2268 int
2269 rt_routemsg(int cmd, struct ifnet *ifp, int error, struct rtentry *rt,
2270     int fibnum)
2271 {
2272 
2273 	KASSERT(cmd == RTM_ADD || cmd == RTM_DELETE,
2274 	    ("unexpected cmd %d", cmd));
2275 
2276 	KASSERT(fibnum == RT_ALL_FIBS || (fibnum >= 0 && fibnum < rt_numfibs),
2277 	    ("%s: fib out of range 0 <=%d<%d", __func__, fibnum, rt_numfibs));
2278 
2279 	KASSERT(rt_key(rt) != NULL, (":%s: rt_key must be supplied", __func__));
2280 
2281 	return (rtsock_routemsg(cmd, ifp, error, rt, fibnum));
2282 }
2283 
2284 void
2285 rt_newaddrmsg(int cmd, struct ifaddr *ifa, int error, struct rtentry *rt)
2286 {
2287 
2288 	rt_newaddrmsg_fib(cmd, ifa, error, rt, RT_ALL_FIBS);
2289 }
2290 
2291 /*
2292  * This is called to generate messages from the routing socket
2293  * indicating a network interface has had addresses associated with it.
2294  */
2295 void
2296 rt_newaddrmsg_fib(int cmd, struct ifaddr *ifa, int error, struct rtentry *rt,
2297     int fibnum)
2298 {
2299 
2300 	KASSERT(cmd == RTM_ADD || cmd == RTM_DELETE,
2301 		("unexpected cmd %u", cmd));
2302 	KASSERT(fibnum == RT_ALL_FIBS || (fibnum >= 0 && fibnum < rt_numfibs),
2303 	    ("%s: fib out of range 0 <=%d<%d", __func__, fibnum, rt_numfibs));
2304 
2305 	if (cmd == RTM_ADD) {
2306 		rt_addrmsg(cmd, ifa, fibnum);
2307 		if (rt != NULL)
2308 			rt_routemsg(cmd, ifa->ifa_ifp, error, rt, fibnum);
2309 	} else {
2310 		if (rt != NULL)
2311 			rt_routemsg(cmd, ifa->ifa_ifp, error, rt, fibnum);
2312 		rt_addrmsg(cmd, ifa, fibnum);
2313 	}
2314 }
2315 
2316