xref: /freebsd/sys/net/route.c (revision 683b025adebfc2bb0d488c1d8d89e14379045ccc)
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  * 3. 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_FIRST,
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 	RIB_LOCK_INIT(rh);
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 	RIB_LOCK_DESTROY(rh);
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 	if ((ignflags & RTF_RNH_LOCKED) == 0)
458 		RIB_RLOCK(rh);
459 #ifdef INVARIANTS
460 	else
461 		RIB_LOCK_ASSERT(rh);
462 #endif
463 	rn = rh->rnh_matchaddr(dst, &rh->head);
464 	if (rn && ((rn->rn_flags & RNF_ROOT) == 0)) {
465 		newrt = RNTORT(rn);
466 		RT_LOCK(newrt);
467 		RT_ADDREF(newrt);
468 		if ((ignflags & RTF_RNH_LOCKED) == 0)
469 			RIB_RUNLOCK(rh);
470 		return (newrt);
471 
472 	} else if ((ignflags & RTF_RNH_LOCKED) == 0)
473 		RIB_RUNLOCK(rh);
474 	/*
475 	 * Either we hit the root or could not find any match,
476 	 * which basically means: "cannot get there from here".
477 	 */
478 miss:
479 	V_rtstat.rts_unreach++;
480 
481 	if (report) {
482 		/*
483 		 * If required, report the failure to the supervising
484 		 * Authorities.
485 		 * For a delete, this is not an error. (report == 0)
486 		 */
487 		bzero(&info, sizeof(info));
488 		info.rti_info[RTAX_DST] = dst;
489 		rt_missmsg_fib(msgtype, &info, 0, err, fibnum);
490 	}
491 	return (newrt);
492 }
493 
494 /*
495  * Remove a reference count from an rtentry.
496  * If the count gets low enough, take it out of the routing table
497  */
498 void
499 rtfree(struct rtentry *rt)
500 {
501 	struct rib_head *rnh;
502 
503 	KASSERT(rt != NULL,("%s: NULL rt", __func__));
504 	rnh = rt_tables_get_rnh(rt->rt_fibnum, rt_key(rt)->sa_family);
505 	KASSERT(rnh != NULL,("%s: NULL rnh", __func__));
506 
507 	RT_LOCK_ASSERT(rt);
508 
509 	/*
510 	 * The callers should use RTFREE_LOCKED() or RTFREE(), so
511 	 * we should come here exactly with the last reference.
512 	 */
513 	RT_REMREF(rt);
514 	if (rt->rt_refcnt > 0) {
515 		log(LOG_DEBUG, "%s: %p has %d refs\n", __func__, rt, rt->rt_refcnt);
516 		goto done;
517 	}
518 
519 	/*
520 	 * On last reference give the "close method" a chance
521 	 * to cleanup private state.  This also permits (for
522 	 * IPv4 and IPv6) a chance to decide if the routing table
523 	 * entry should be purged immediately or at a later time.
524 	 * When an immediate purge is to happen the close routine
525 	 * typically calls rtexpunge which clears the RTF_UP flag
526 	 * on the entry so that the code below reclaims the storage.
527 	 */
528 	if (rt->rt_refcnt == 0 && rnh->rnh_close)
529 		rnh->rnh_close((struct radix_node *)rt, &rnh->head);
530 
531 	/*
532 	 * If we are no longer "up" (and ref == 0)
533 	 * then we can free the resources associated
534 	 * with the route.
535 	 */
536 	if ((rt->rt_flags & RTF_UP) == 0) {
537 		if (rt->rt_nodes->rn_flags & (RNF_ACTIVE | RNF_ROOT))
538 			panic("rtfree 2");
539 		/*
540 		 * the rtentry must have been removed from the routing table
541 		 * so it is represented in rttrash.. remove that now.
542 		 */
543 		V_rttrash--;
544 #ifdef	DIAGNOSTIC
545 		if (rt->rt_refcnt < 0) {
546 			printf("rtfree: %p not freed (neg refs)\n", rt);
547 			goto done;
548 		}
549 #endif
550 		/*
551 		 * release references on items we hold them on..
552 		 * e.g other routes and ifaddrs.
553 		 */
554 		if (rt->rt_ifa)
555 			ifa_free(rt->rt_ifa);
556 		/*
557 		 * The key is separatly alloc'd so free it (see rt_setgate()).
558 		 * This also frees the gateway, as they are always malloc'd
559 		 * together.
560 		 */
561 		R_Free(rt_key(rt));
562 
563 		/*
564 		 * and the rtentry itself of course
565 		 */
566 		uma_zfree(V_rtzone, rt);
567 		return;
568 	}
569 done:
570 	RT_UNLOCK(rt);
571 }
572 
573 
574 /*
575  * Force a routing table entry to the specified
576  * destination to go through the given gateway.
577  * Normally called as a result of a routing redirect
578  * message from the network layer.
579  */
580 void
581 rtredirect_fib(struct sockaddr *dst,
582 	struct sockaddr *gateway,
583 	struct sockaddr *netmask,
584 	int flags,
585 	struct sockaddr *src,
586 	u_int fibnum)
587 {
588 	struct rtentry *rt;
589 	int error = 0;
590 	short *stat = NULL;
591 	struct rt_addrinfo info;
592 	struct ifaddr *ifa;
593 	struct rib_head *rnh;
594 
595 	ifa = NULL;
596 	rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
597 	if (rnh == NULL) {
598 		error = EAFNOSUPPORT;
599 		goto out;
600 	}
601 
602 	/* verify the gateway is directly reachable */
603 	if ((ifa = ifa_ifwithnet(gateway, 0, fibnum)) == NULL) {
604 		error = ENETUNREACH;
605 		goto out;
606 	}
607 	rt = rtalloc1_fib(dst, 0, 0UL, fibnum);	/* NB: rt is locked */
608 	/*
609 	 * If the redirect isn't from our current router for this dst,
610 	 * it's either old or wrong.  If it redirects us to ourselves,
611 	 * we have a routing loop, perhaps as a result of an interface
612 	 * going down recently.
613 	 */
614 	if (!(flags & RTF_DONE) && rt) {
615 		if (!sa_equal(src, rt->rt_gateway)) {
616 			error = EINVAL;
617 			goto done;
618 		}
619 		if (rt->rt_ifa != ifa && ifa->ifa_addr->sa_family != AF_LINK) {
620 			error = EINVAL;
621 			goto done;
622 		}
623 	}
624 	if ((flags & RTF_GATEWAY) && ifa_ifwithaddr_check(gateway)) {
625 		error = EHOSTUNREACH;
626 		goto done;
627 	}
628 	/*
629 	 * Create a new entry if we just got back a wildcard entry
630 	 * or the lookup failed.  This is necessary for hosts
631 	 * which use routing redirects generated by smart gateways
632 	 * to dynamically build the routing tables.
633 	 */
634 	if (rt == NULL || (rt_mask(rt) && rt_mask(rt)->sa_len < 2))
635 		goto create;
636 	/*
637 	 * Don't listen to the redirect if it's
638 	 * for a route to an interface.
639 	 */
640 	if (rt->rt_flags & RTF_GATEWAY) {
641 		if (((rt->rt_flags & RTF_HOST) == 0) && (flags & RTF_HOST)) {
642 			/*
643 			 * Changing from route to net => route to host.
644 			 * Create new route, rather than smashing route to net.
645 			 */
646 		create:
647 			if (rt != NULL)
648 				RTFREE_LOCKED(rt);
649 
650 			flags |= RTF_DYNAMIC;
651 			bzero((caddr_t)&info, sizeof(info));
652 			info.rti_info[RTAX_DST] = dst;
653 			info.rti_info[RTAX_GATEWAY] = gateway;
654 			info.rti_info[RTAX_NETMASK] = netmask;
655 			info.rti_ifa = ifa;
656 			info.rti_flags = flags;
657 			error = rtrequest1_fib(RTM_ADD, &info, &rt, fibnum);
658 			if (rt != NULL) {
659 				RT_LOCK(rt);
660 				flags = rt->rt_flags;
661 			}
662 
663 			stat = &V_rtstat.rts_dynamic;
664 		} else {
665 
666 			/*
667 			 * Smash the current notion of the gateway to
668 			 * this destination.  Should check about netmask!!!
669 			 */
670 			if ((flags & RTF_GATEWAY) == 0)
671 				rt->rt_flags &= ~RTF_GATEWAY;
672 			rt->rt_flags |= RTF_MODIFIED;
673 			flags |= RTF_MODIFIED;
674 			stat = &V_rtstat.rts_newgateway;
675 			/*
676 			 * add the key and gateway (in one malloc'd chunk).
677 			 */
678 			RT_UNLOCK(rt);
679 			RIB_WLOCK(rnh);
680 			RT_LOCK(rt);
681 			rt_setgate(rt, rt_key(rt), gateway);
682 			RIB_WUNLOCK(rnh);
683 		}
684 	} else
685 		error = EHOSTUNREACH;
686 done:
687 	if (rt)
688 		RTFREE_LOCKED(rt);
689 out:
690 	if (error)
691 		V_rtstat.rts_badredirect++;
692 	else if (stat != NULL)
693 		(*stat)++;
694 	bzero((caddr_t)&info, sizeof(info));
695 	info.rti_info[RTAX_DST] = dst;
696 	info.rti_info[RTAX_GATEWAY] = gateway;
697 	info.rti_info[RTAX_NETMASK] = netmask;
698 	info.rti_info[RTAX_AUTHOR] = src;
699 	rt_missmsg_fib(RTM_REDIRECT, &info, flags, error, fibnum);
700 	if (ifa != NULL)
701 		ifa_free(ifa);
702 }
703 
704 /*
705  * Routing table ioctl interface.
706  */
707 int
708 rtioctl_fib(u_long req, caddr_t data, u_int fibnum)
709 {
710 
711 	/*
712 	 * If more ioctl commands are added here, make sure the proper
713 	 * super-user checks are being performed because it is possible for
714 	 * prison-root to make it this far if raw sockets have been enabled
715 	 * in jails.
716 	 */
717 #ifdef INET
718 	/* Multicast goop, grrr... */
719 	return mrt_ioctl ? mrt_ioctl(req, data, fibnum) : EOPNOTSUPP;
720 #else /* INET */
721 	return ENXIO;
722 #endif /* INET */
723 }
724 
725 struct ifaddr *
726 ifa_ifwithroute(int flags, const struct sockaddr *dst, struct sockaddr *gateway,
727 				u_int fibnum)
728 {
729 	struct ifaddr *ifa;
730 	int not_found = 0;
731 
732 	if ((flags & RTF_GATEWAY) == 0) {
733 		/*
734 		 * If we are adding a route to an interface,
735 		 * and the interface is a pt to pt link
736 		 * we should search for the destination
737 		 * as our clue to the interface.  Otherwise
738 		 * we can use the local address.
739 		 */
740 		ifa = NULL;
741 		if (flags & RTF_HOST)
742 			ifa = ifa_ifwithdstaddr(dst, fibnum);
743 		if (ifa == NULL)
744 			ifa = ifa_ifwithaddr(gateway);
745 	} else {
746 		/*
747 		 * If we are adding a route to a remote net
748 		 * or host, the gateway may still be on the
749 		 * other end of a pt to pt link.
750 		 */
751 		ifa = ifa_ifwithdstaddr(gateway, fibnum);
752 	}
753 	if (ifa == NULL)
754 		ifa = ifa_ifwithnet(gateway, 0, fibnum);
755 	if (ifa == NULL) {
756 		struct rtentry *rt;
757 
758 		rt = rtalloc1_fib(gateway, 0, flags, fibnum);
759 		if (rt == NULL)
760 			return (NULL);
761 		/*
762 		 * dismiss a gateway that is reachable only
763 		 * through the default router
764 		 */
765 		switch (gateway->sa_family) {
766 		case AF_INET:
767 			if (satosin(rt_key(rt))->sin_addr.s_addr == INADDR_ANY)
768 				not_found = 1;
769 			break;
770 		case AF_INET6:
771 			if (IN6_IS_ADDR_UNSPECIFIED(&satosin6(rt_key(rt))->sin6_addr))
772 				not_found = 1;
773 			break;
774 		default:
775 			break;
776 		}
777 		if (!not_found && rt->rt_ifa != NULL) {
778 			ifa = rt->rt_ifa;
779 			ifa_ref(ifa);
780 		}
781 		RT_REMREF(rt);
782 		RT_UNLOCK(rt);
783 		if (not_found || ifa == NULL)
784 			return (NULL);
785 	}
786 	if (ifa->ifa_addr->sa_family != dst->sa_family) {
787 		struct ifaddr *oifa = ifa;
788 		ifa = ifaof_ifpforaddr(dst, ifa->ifa_ifp);
789 		if (ifa == NULL)
790 			ifa = oifa;
791 		else
792 			ifa_free(oifa);
793 	}
794 	return (ifa);
795 }
796 
797 /*
798  * Do appropriate manipulations of a routing tree given
799  * all the bits of info needed
800  */
801 int
802 rtrequest_fib(int req,
803 	struct sockaddr *dst,
804 	struct sockaddr *gateway,
805 	struct sockaddr *netmask,
806 	int flags,
807 	struct rtentry **ret_nrt,
808 	u_int fibnum)
809 {
810 	struct rt_addrinfo info;
811 
812 	if (dst->sa_len == 0)
813 		return(EINVAL);
814 
815 	bzero((caddr_t)&info, sizeof(info));
816 	info.rti_flags = flags;
817 	info.rti_info[RTAX_DST] = dst;
818 	info.rti_info[RTAX_GATEWAY] = gateway;
819 	info.rti_info[RTAX_NETMASK] = netmask;
820 	return rtrequest1_fib(req, &info, ret_nrt, fibnum);
821 }
822 
823 
824 /*
825  * Copy most of @rt data into @info.
826  *
827  * If @flags contains NHR_COPY, copies dst,netmask and gw to the
828  * pointers specified by @info structure. Assume such pointers
829  * are zeroed sockaddr-like structures with sa_len field initialized
830  * to reflect size of the provided buffer. if no NHR_COPY is specified,
831  * point dst,netmask and gw @info fields to appropriate @rt values.
832  *
833  * if @flags contains NHR_REF, do refcouting on rt_ifp.
834  *
835  * Returns 0 on success.
836  */
837 int
838 rt_exportinfo(struct rtentry *rt, struct rt_addrinfo *info, int flags)
839 {
840 	struct rt_metrics *rmx;
841 	struct sockaddr *src, *dst;
842 	int sa_len;
843 
844 	if (flags & NHR_COPY) {
845 		/* Copy destination if dst is non-zero */
846 		src = rt_key(rt);
847 		dst = info->rti_info[RTAX_DST];
848 		sa_len = src->sa_len;
849 		if (dst != NULL) {
850 			if (src->sa_len > dst->sa_len)
851 				return (ENOMEM);
852 			memcpy(dst, src, src->sa_len);
853 			info->rti_addrs |= RTA_DST;
854 		}
855 
856 		/* Copy mask if set && dst is non-zero */
857 		src = rt_mask(rt);
858 		dst = info->rti_info[RTAX_NETMASK];
859 		if (src != NULL && dst != NULL) {
860 
861 			/*
862 			 * Radix stores different value in sa_len,
863 			 * assume rt_mask() to have the same length
864 			 * as rt_key()
865 			 */
866 			if (sa_len > dst->sa_len)
867 				return (ENOMEM);
868 			memcpy(dst, src, src->sa_len);
869 			info->rti_addrs |= RTA_NETMASK;
870 		}
871 
872 		/* Copy gateway is set && dst is non-zero */
873 		src = rt->rt_gateway;
874 		dst = info->rti_info[RTAX_GATEWAY];
875 		if ((rt->rt_flags & RTF_GATEWAY) && src != NULL && dst != NULL){
876 			if (src->sa_len > dst->sa_len)
877 				return (ENOMEM);
878 			memcpy(dst, src, src->sa_len);
879 			info->rti_addrs |= RTA_GATEWAY;
880 		}
881 	} else {
882 		info->rti_info[RTAX_DST] = rt_key(rt);
883 		info->rti_addrs |= RTA_DST;
884 		if (rt_mask(rt) != NULL) {
885 			info->rti_info[RTAX_NETMASK] = rt_mask(rt);
886 			info->rti_addrs |= RTA_NETMASK;
887 		}
888 		if (rt->rt_flags & RTF_GATEWAY) {
889 			info->rti_info[RTAX_GATEWAY] = rt->rt_gateway;
890 			info->rti_addrs |= RTA_GATEWAY;
891 		}
892 	}
893 
894 	rmx = info->rti_rmx;
895 	if (rmx != NULL) {
896 		info->rti_mflags |= RTV_MTU;
897 		rmx->rmx_mtu = rt->rt_mtu;
898 	}
899 
900 	info->rti_flags = rt->rt_flags;
901 	info->rti_ifp = rt->rt_ifp;
902 	info->rti_ifa = rt->rt_ifa;
903 
904 	if (flags & NHR_REF) {
905 		/* Do 'traditional' refcouting */
906 		if_ref(info->rti_ifp);
907 	}
908 
909 	return (0);
910 }
911 
912 /*
913  * Lookups up route entry for @dst in RIB database for fib @fibnum.
914  * Exports entry data to @info using rt_exportinfo().
915  *
916  * if @flags contains NHR_REF, refcouting is performed on rt_ifp.
917  *   All references can be released later by calling rib_free_info()
918  *
919  * Returns 0 on success.
920  * Returns ENOENT for lookup failure, ENOMEM for export failure.
921  */
922 int
923 rib_lookup_info(uint32_t fibnum, const struct sockaddr *dst, uint32_t flags,
924     uint32_t flowid, struct rt_addrinfo *info)
925 {
926 	struct rib_head *rh;
927 	struct radix_node *rn;
928 	struct rtentry *rt;
929 	int error;
930 
931 	KASSERT((fibnum < rt_numfibs), ("rib_lookup_rte: bad fibnum"));
932 	rh = rt_tables_get_rnh(fibnum, dst->sa_family);
933 	if (rh == NULL)
934 		return (ENOENT);
935 
936 	RIB_RLOCK(rh);
937 	rn = rh->rnh_matchaddr(__DECONST(void *, dst), &rh->head);
938 	if (rn != NULL && ((rn->rn_flags & RNF_ROOT) == 0)) {
939 		rt = RNTORT(rn);
940 		/* Ensure route & ifp is UP */
941 		if (RT_LINK_IS_UP(rt->rt_ifp)) {
942 			flags = (flags & NHR_REF) | NHR_COPY;
943 			error = rt_exportinfo(rt, info, flags);
944 			RIB_RUNLOCK(rh);
945 
946 			return (error);
947 		}
948 	}
949 	RIB_RUNLOCK(rh);
950 
951 	return (ENOENT);
952 }
953 
954 /*
955  * Releases all references acquired by rib_lookup_info() when
956  * called with NHR_REF flags.
957  */
958 void
959 rib_free_info(struct rt_addrinfo *info)
960 {
961 
962 	if_rele(info->rti_ifp);
963 }
964 
965 /*
966  * Iterates over all existing fibs in system calling
967  *  @setwa_f function prior to traversing each fib.
968  *  Calls @wa_f function for each element in current fib.
969  * If af is not AF_UNSPEC, iterates over fibs in particular
970  * address family.
971  */
972 void
973 rt_foreach_fib_walk(int af, rt_setwarg_t *setwa_f, rt_walktree_f_t *wa_f,
974     void *arg)
975 {
976 	struct rib_head *rnh;
977 	uint32_t fibnum;
978 	int i;
979 
980 	for (fibnum = 0; fibnum < rt_numfibs; fibnum++) {
981 		/* Do we want some specific family? */
982 		if (af != AF_UNSPEC) {
983 			rnh = rt_tables_get_rnh(fibnum, af);
984 			if (rnh == NULL)
985 				continue;
986 			if (setwa_f != NULL)
987 				setwa_f(rnh, fibnum, af, arg);
988 
989 			RIB_WLOCK(rnh);
990 			rnh->rnh_walktree(&rnh->head, (walktree_f_t *)wa_f,arg);
991 			RIB_WUNLOCK(rnh);
992 			continue;
993 		}
994 
995 		for (i = 1; i <= AF_MAX; i++) {
996 			rnh = rt_tables_get_rnh(fibnum, i);
997 			if (rnh == NULL)
998 				continue;
999 			if (setwa_f != NULL)
1000 				setwa_f(rnh, fibnum, i, arg);
1001 
1002 			RIB_WLOCK(rnh);
1003 			rnh->rnh_walktree(&rnh->head, (walktree_f_t *)wa_f,arg);
1004 			RIB_WUNLOCK(rnh);
1005 		}
1006 	}
1007 }
1008 
1009 struct rt_delinfo
1010 {
1011 	struct rt_addrinfo info;
1012 	struct rib_head *rnh;
1013 	struct rtentry *head;
1014 };
1015 
1016 /*
1017  * Conditionally unlinks @rn from radix tree based
1018  * on info data passed in @arg.
1019  */
1020 static int
1021 rt_checkdelroute(struct radix_node *rn, void *arg)
1022 {
1023 	struct rt_delinfo *di;
1024 	struct rt_addrinfo *info;
1025 	struct rtentry *rt;
1026 	int error;
1027 
1028 	di = (struct rt_delinfo *)arg;
1029 	rt = (struct rtentry *)rn;
1030 	info = &di->info;
1031 	error = 0;
1032 
1033 	info->rti_info[RTAX_DST] = rt_key(rt);
1034 	info->rti_info[RTAX_NETMASK] = rt_mask(rt);
1035 	info->rti_info[RTAX_GATEWAY] = rt->rt_gateway;
1036 
1037 	rt = rt_unlinkrte(di->rnh, info, &error);
1038 	if (rt == NULL) {
1039 		/* Either not allowed or not matched. Skip entry */
1040 		return (0);
1041 	}
1042 
1043 	/* Entry was unlinked. Add to the list and return */
1044 	rt->rt_chain = di->head;
1045 	di->head = rt;
1046 
1047 	return (0);
1048 }
1049 
1050 /*
1051  * Iterates over all existing fibs in system.
1052  * Deletes each element for which @filter_f function returned
1053  * non-zero value.
1054  * If @af is not AF_UNSPEC, iterates over fibs in particular
1055  * address family.
1056  */
1057 void
1058 rt_foreach_fib_walk_del(int af, rt_filter_f_t *filter_f, void *arg)
1059 {
1060 	struct rib_head *rnh;
1061 	struct rt_delinfo di;
1062 	struct rtentry *rt;
1063 	uint32_t fibnum;
1064 	int i, start, end;
1065 
1066 	bzero(&di, sizeof(di));
1067 	di.info.rti_filter = filter_f;
1068 	di.info.rti_filterdata = arg;
1069 
1070 	for (fibnum = 0; fibnum < rt_numfibs; fibnum++) {
1071 		/* Do we want some specific family? */
1072 		if (af != AF_UNSPEC) {
1073 			start = af;
1074 			end = af;
1075 		} else {
1076 			start = 1;
1077 			end = AF_MAX;
1078 		}
1079 
1080 		for (i = start; i <= end; i++) {
1081 			rnh = rt_tables_get_rnh(fibnum, i);
1082 			if (rnh == NULL)
1083 				continue;
1084 			di.rnh = rnh;
1085 
1086 			RIB_WLOCK(rnh);
1087 			rnh->rnh_walktree(&rnh->head, rt_checkdelroute, &di);
1088 			RIB_WUNLOCK(rnh);
1089 
1090 			if (di.head == NULL)
1091 				continue;
1092 
1093 			/* We might have something to reclaim */
1094 			while (di.head != NULL) {
1095 				rt = di.head;
1096 				di.head = rt->rt_chain;
1097 				rt->rt_chain = NULL;
1098 
1099 				/* TODO std rt -> rt_addrinfo export */
1100 				di.info.rti_info[RTAX_DST] = rt_key(rt);
1101 				di.info.rti_info[RTAX_NETMASK] = rt_mask(rt);
1102 
1103 				rt_notifydelete(rt, &di.info);
1104 				RTFREE_LOCKED(rt);
1105 			}
1106 
1107 		}
1108 	}
1109 }
1110 
1111 /*
1112  * Delete Routes for a Network Interface
1113  *
1114  * Called for each routing entry via the rnh->rnh_walktree() call above
1115  * to delete all route entries referencing a detaching network interface.
1116  *
1117  * Arguments:
1118  *	rt	pointer to rtentry
1119  *	arg	argument passed to rnh->rnh_walktree() - detaching interface
1120  *
1121  * Returns:
1122  *	0	successful
1123  *	errno	failed - reason indicated
1124  */
1125 static int
1126 rt_ifdelroute(const struct rtentry *rt, void *arg)
1127 {
1128 	struct ifnet	*ifp = arg;
1129 
1130 	if (rt->rt_ifp != ifp)
1131 		return (0);
1132 
1133 	/*
1134 	 * Protect (sorta) against walktree recursion problems
1135 	 * with cloned routes
1136 	 */
1137 	if ((rt->rt_flags & RTF_UP) == 0)
1138 		return (0);
1139 
1140 	return (1);
1141 }
1142 
1143 /*
1144  * Delete all remaining routes using this interface
1145  * Unfortuneatly the only way to do this is to slog through
1146  * the entire routing table looking for routes which point
1147  * to this interface...oh well...
1148  */
1149 void
1150 rt_flushifroutes_af(struct ifnet *ifp, int af)
1151 {
1152 	KASSERT((af >= 1 && af <= AF_MAX), ("%s: af %d not >= 1 and <= %d",
1153 	    __func__, af, AF_MAX));
1154 
1155 	rt_foreach_fib_walk_del(af, rt_ifdelroute, ifp);
1156 }
1157 
1158 void
1159 rt_flushifroutes(struct ifnet *ifp)
1160 {
1161 
1162 	rt_foreach_fib_walk_del(AF_UNSPEC, rt_ifdelroute, ifp);
1163 }
1164 
1165 /*
1166  * Conditionally unlinks rtentry matching data inside @info from @rnh.
1167  * Returns unlinked, locked and referenced @rtentry on success,
1168  * Returns NULL and sets @perror to:
1169  * ESRCH - if prefix was not found,
1170  * EADDRINUSE - if trying to delete PINNED route without appropriate flag.
1171  * ENOENT - if supplied filter function returned 0 (not matched).
1172  */
1173 static struct rtentry *
1174 rt_unlinkrte(struct rib_head *rnh, struct rt_addrinfo *info, int *perror)
1175 {
1176 	struct sockaddr *dst, *netmask;
1177 	struct rtentry *rt;
1178 	struct radix_node *rn;
1179 
1180 	dst = info->rti_info[RTAX_DST];
1181 	netmask = info->rti_info[RTAX_NETMASK];
1182 
1183 	rt = (struct rtentry *)rnh->rnh_lookup(dst, netmask, &rnh->head);
1184 	if (rt == NULL) {
1185 		*perror = ESRCH;
1186 		return (NULL);
1187 	}
1188 
1189 	if ((info->rti_flags & RTF_PINNED) == 0) {
1190 		/* Check if target route can be deleted */
1191 		if (rt->rt_flags & RTF_PINNED) {
1192 			*perror = EADDRINUSE;
1193 			return (NULL);
1194 		}
1195 	}
1196 
1197 	if (info->rti_filter != NULL) {
1198 		if (info->rti_filter(rt, info->rti_filterdata) == 0) {
1199 			/* Not matched */
1200 			*perror = ENOENT;
1201 			return (NULL);
1202 		}
1203 
1204 		/*
1205 		 * Filter function requested rte deletion.
1206 		 * Ease the caller work by filling in remaining info
1207 		 * from that particular entry.
1208 		 */
1209 		info->rti_info[RTAX_GATEWAY] = rt->rt_gateway;
1210 	}
1211 
1212 	/*
1213 	 * Remove the item from the tree and return it.
1214 	 * Complain if it is not there and do no more processing.
1215 	 */
1216 	*perror = ESRCH;
1217 #ifdef RADIX_MPATH
1218 	if (rt_mpath_capable(rnh))
1219 		rn = rt_mpath_unlink(rnh, info, rt, perror);
1220 	else
1221 #endif
1222 	rn = rnh->rnh_deladdr(dst, netmask, &rnh->head);
1223 	if (rn == NULL)
1224 		return (NULL);
1225 
1226 	if (rn->rn_flags & (RNF_ACTIVE | RNF_ROOT))
1227 		panic ("rtrequest delete");
1228 
1229 	rt = RNTORT(rn);
1230 	RT_LOCK(rt);
1231 	RT_ADDREF(rt);
1232 	rt->rt_flags &= ~RTF_UP;
1233 
1234 	*perror = 0;
1235 
1236 	return (rt);
1237 }
1238 
1239 static void
1240 rt_notifydelete(struct rtentry *rt, struct rt_addrinfo *info)
1241 {
1242 	struct ifaddr *ifa;
1243 
1244 	/*
1245 	 * give the protocol a chance to keep things in sync.
1246 	 */
1247 	ifa = rt->rt_ifa;
1248 	if (ifa != NULL && ifa->ifa_rtrequest != NULL)
1249 		ifa->ifa_rtrequest(RTM_DELETE, rt, info);
1250 
1251 	/*
1252 	 * One more rtentry floating around that is not
1253 	 * linked to the routing table. rttrash will be decremented
1254 	 * when RTFREE(rt) is eventually called.
1255 	 */
1256 	V_rttrash++;
1257 }
1258 
1259 
1260 /*
1261  * These (questionable) definitions of apparent local variables apply
1262  * to the next two functions.  XXXXXX!!!
1263  */
1264 #define	dst	info->rti_info[RTAX_DST]
1265 #define	gateway	info->rti_info[RTAX_GATEWAY]
1266 #define	netmask	info->rti_info[RTAX_NETMASK]
1267 #define	ifaaddr	info->rti_info[RTAX_IFA]
1268 #define	ifpaddr	info->rti_info[RTAX_IFP]
1269 #define	flags	info->rti_flags
1270 
1271 /*
1272  * Look up rt_addrinfo for a specific fib.  Note that if rti_ifa is defined,
1273  * it will be referenced so the caller must free it.
1274  */
1275 int
1276 rt_getifa_fib(struct rt_addrinfo *info, u_int fibnum)
1277 {
1278 	struct ifaddr *ifa;
1279 	int error = 0;
1280 
1281 	/*
1282 	 * ifp may be specified by sockaddr_dl
1283 	 * when protocol address is ambiguous.
1284 	 */
1285 	if (info->rti_ifp == NULL && ifpaddr != NULL &&
1286 	    ifpaddr->sa_family == AF_LINK &&
1287 	    (ifa = ifa_ifwithnet(ifpaddr, 0, fibnum)) != NULL) {
1288 		info->rti_ifp = ifa->ifa_ifp;
1289 		ifa_free(ifa);
1290 	}
1291 	if (info->rti_ifa == NULL && ifaaddr != NULL)
1292 		info->rti_ifa = ifa_ifwithaddr(ifaaddr);
1293 	if (info->rti_ifa == NULL) {
1294 		struct sockaddr *sa;
1295 
1296 		sa = ifaaddr != NULL ? ifaaddr :
1297 		    (gateway != NULL ? gateway : dst);
1298 		if (sa != NULL && info->rti_ifp != NULL)
1299 			info->rti_ifa = ifaof_ifpforaddr(sa, info->rti_ifp);
1300 		else if (dst != NULL && gateway != NULL)
1301 			info->rti_ifa = ifa_ifwithroute(flags, dst, gateway,
1302 							fibnum);
1303 		else if (sa != NULL)
1304 			info->rti_ifa = ifa_ifwithroute(flags, sa, sa,
1305 							fibnum);
1306 	}
1307 	if ((ifa = info->rti_ifa) != NULL) {
1308 		if (info->rti_ifp == NULL)
1309 			info->rti_ifp = ifa->ifa_ifp;
1310 	} else
1311 		error = ENETUNREACH;
1312 	return (error);
1313 }
1314 
1315 static int
1316 if_updatemtu_cb(struct radix_node *rn, void *arg)
1317 {
1318 	struct rtentry *rt;
1319 	struct if_mtuinfo *ifmtu;
1320 
1321 	rt = (struct rtentry *)rn;
1322 	ifmtu = (struct if_mtuinfo *)arg;
1323 
1324 	if (rt->rt_ifp != ifmtu->ifp)
1325 		return (0);
1326 
1327 	if (rt->rt_mtu >= ifmtu->mtu) {
1328 		/* We have to decrease mtu regardless of flags */
1329 		rt->rt_mtu = ifmtu->mtu;
1330 		return (0);
1331 	}
1332 
1333 	/*
1334 	 * New MTU is bigger. Check if are allowed to alter it
1335 	 */
1336 	if ((rt->rt_flags & (RTF_FIXEDMTU | RTF_GATEWAY | RTF_HOST)) != 0) {
1337 
1338 		/*
1339 		 * Skip routes with user-supplied MTU and
1340 		 * non-interface routes
1341 		 */
1342 		return (0);
1343 	}
1344 
1345 	/* We are safe to update route MTU */
1346 	rt->rt_mtu = ifmtu->mtu;
1347 
1348 	return (0);
1349 }
1350 
1351 void
1352 rt_updatemtu(struct ifnet *ifp)
1353 {
1354 	struct if_mtuinfo ifmtu;
1355 	struct rib_head *rnh;
1356 	int i, j;
1357 
1358 	ifmtu.ifp = ifp;
1359 
1360 	/*
1361 	 * Try to update rt_mtu for all routes using this interface
1362 	 * Unfortunately the only way to do this is to traverse all
1363 	 * routing tables in all fibs/domains.
1364 	 */
1365 	for (i = 1; i <= AF_MAX; i++) {
1366 		ifmtu.mtu = if_getmtu_family(ifp, i);
1367 		for (j = 0; j < rt_numfibs; j++) {
1368 			rnh = rt_tables_get_rnh(j, i);
1369 			if (rnh == NULL)
1370 				continue;
1371 			RIB_WLOCK(rnh);
1372 			rnh->rnh_walktree(&rnh->head, if_updatemtu_cb, &ifmtu);
1373 			RIB_WUNLOCK(rnh);
1374 		}
1375 	}
1376 }
1377 
1378 
1379 #if 0
1380 int p_sockaddr(char *buf, int buflen, struct sockaddr *s);
1381 int rt_print(char *buf, int buflen, struct rtentry *rt);
1382 
1383 int
1384 p_sockaddr(char *buf, int buflen, struct sockaddr *s)
1385 {
1386 	void *paddr = NULL;
1387 
1388 	switch (s->sa_family) {
1389 	case AF_INET:
1390 		paddr = &((struct sockaddr_in *)s)->sin_addr;
1391 		break;
1392 	case AF_INET6:
1393 		paddr = &((struct sockaddr_in6 *)s)->sin6_addr;
1394 		break;
1395 	}
1396 
1397 	if (paddr == NULL)
1398 		return (0);
1399 
1400 	if (inet_ntop(s->sa_family, paddr, buf, buflen) == NULL)
1401 		return (0);
1402 
1403 	return (strlen(buf));
1404 }
1405 
1406 int
1407 rt_print(char *buf, int buflen, struct rtentry *rt)
1408 {
1409 	struct sockaddr *addr, *mask;
1410 	int i = 0;
1411 
1412 	addr = rt_key(rt);
1413 	mask = rt_mask(rt);
1414 
1415 	i = p_sockaddr(buf, buflen, addr);
1416 	if (!(rt->rt_flags & RTF_HOST)) {
1417 		buf[i++] = '/';
1418 		i += p_sockaddr(buf + i, buflen - i, mask);
1419 	}
1420 
1421 	if (rt->rt_flags & RTF_GATEWAY) {
1422 		buf[i++] = '>';
1423 		i += p_sockaddr(buf + i, buflen - i, rt->rt_gateway);
1424 	}
1425 
1426 	return (i);
1427 }
1428 #endif
1429 
1430 #ifdef RADIX_MPATH
1431 /*
1432  * Deletes key for single-path routes, unlinks rtentry with
1433  * gateway specified in @info from multi-path routes.
1434  *
1435  * Returnes unlinked entry. In case of failure, returns NULL
1436  * and sets @perror to ESRCH.
1437  */
1438 static struct radix_node *
1439 rt_mpath_unlink(struct rib_head *rnh, struct rt_addrinfo *info,
1440     struct rtentry *rto, int *perror)
1441 {
1442 	/*
1443 	 * if we got multipath routes, we require users to specify
1444 	 * a matching RTAX_GATEWAY.
1445 	 */
1446 	struct rtentry *rt; // *rto = NULL;
1447 	struct radix_node *rn;
1448 	struct sockaddr *gw;
1449 
1450 	gw = info->rti_info[RTAX_GATEWAY];
1451 	rt = rt_mpath_matchgate(rto, gw);
1452 	if (rt == NULL) {
1453 		*perror = ESRCH;
1454 		return (NULL);
1455 	}
1456 
1457 	/*
1458 	 * this is the first entry in the chain
1459 	 */
1460 	if (rto == rt) {
1461 		rn = rn_mpath_next((struct radix_node *)rt);
1462 		/*
1463 		 * there is another entry, now it's active
1464 		 */
1465 		if (rn) {
1466 			rto = RNTORT(rn);
1467 			RT_LOCK(rto);
1468 			rto->rt_flags |= RTF_UP;
1469 			RT_UNLOCK(rto);
1470 		} else if (rt->rt_flags & RTF_GATEWAY) {
1471 			/*
1472 			 * For gateway routes, we need to
1473 			 * make sure that we we are deleting
1474 			 * the correct gateway.
1475 			 * rt_mpath_matchgate() does not
1476 			 * check the case when there is only
1477 			 * one route in the chain.
1478 			 */
1479 			if (gw &&
1480 			    (rt->rt_gateway->sa_len != gw->sa_len ||
1481 				memcmp(rt->rt_gateway, gw, gw->sa_len))) {
1482 				*perror = ESRCH;
1483 				return (NULL);
1484 			}
1485 		}
1486 
1487 		/*
1488 		 * use the normal delete code to remove
1489 		 * the first entry
1490 		 */
1491 		rn = rnh->rnh_deladdr(dst, netmask, &rnh->head);
1492 		*perror = 0;
1493 		return (rn);
1494 	}
1495 
1496 	/*
1497 	 * if the entry is 2nd and on up
1498 	 */
1499 	if (rt_mpath_deldup(rto, rt) == 0)
1500 		panic ("rtrequest1: rt_mpath_deldup");
1501 	*perror = 0;
1502 	rn = (struct radix_node *)rt;
1503 	return (rn);
1504 }
1505 #endif
1506 
1507 #ifdef FLOWTABLE
1508 static struct rtentry *
1509 rt_flowtable_check_route(struct rib_head *rnh, struct rt_addrinfo *info)
1510 {
1511 #if defined(INET6) || defined(INET)
1512 	struct radix_node *rn;
1513 #endif
1514 	struct rtentry *rt0;
1515 
1516 	rt0 = NULL;
1517 	/* "flow-table" only supports IPv6 and IPv4 at the moment. */
1518 	switch (dst->sa_family) {
1519 #ifdef INET6
1520 	case AF_INET6:
1521 #endif
1522 #ifdef INET
1523 	case AF_INET:
1524 #endif
1525 #if defined(INET6) || defined(INET)
1526 		rn = rnh->rnh_matchaddr(dst, &rnh->head);
1527 		if (rn && ((rn->rn_flags & RNF_ROOT) == 0)) {
1528 			struct sockaddr *mask;
1529 			u_char *m, *n;
1530 			int len;
1531 
1532 			/*
1533 			 * compare mask to see if the new route is
1534 			 * more specific than the existing one
1535 			 */
1536 			rt0 = RNTORT(rn);
1537 			RT_LOCK(rt0);
1538 			RT_ADDREF(rt0);
1539 			RT_UNLOCK(rt0);
1540 			/*
1541 			 * A host route is already present, so
1542 			 * leave the flow-table entries as is.
1543 			 */
1544 			if (rt0->rt_flags & RTF_HOST) {
1545 				RTFREE(rt0);
1546 				rt0 = NULL;
1547 			} else if (!(flags & RTF_HOST) && netmask) {
1548 				mask = rt_mask(rt0);
1549 				len = mask->sa_len;
1550 				m = (u_char *)mask;
1551 				n = (u_char *)netmask;
1552 				while (len-- > 0) {
1553 					if (*n != *m)
1554 						break;
1555 					n++;
1556 					m++;
1557 				}
1558 				if (len == 0 || (*n < *m)) {
1559 					RTFREE(rt0);
1560 					rt0 = NULL;
1561 				}
1562 			}
1563 		}
1564 #endif/* INET6 || INET */
1565 	}
1566 
1567 	return (rt0);
1568 }
1569 #endif
1570 
1571 int
1572 rtrequest1_fib(int req, struct rt_addrinfo *info, struct rtentry **ret_nrt,
1573 				u_int fibnum)
1574 {
1575 	int error = 0;
1576 	struct rtentry *rt, *rt_old;
1577 #ifdef FLOWTABLE
1578 	struct rtentry *rt0;
1579 #endif
1580 	struct radix_node *rn;
1581 	struct rib_head *rnh;
1582 	struct ifaddr *ifa;
1583 	struct sockaddr *ndst;
1584 	struct sockaddr_storage mdst;
1585 
1586 	KASSERT((fibnum < rt_numfibs), ("rtrequest1_fib: bad fibnum"));
1587 	KASSERT((flags & RTF_RNH_LOCKED) == 0, ("rtrequest1_fib: locked"));
1588 	switch (dst->sa_family) {
1589 	case AF_INET6:
1590 	case AF_INET:
1591 		/* We support multiple FIBs. */
1592 		break;
1593 	default:
1594 		fibnum = RT_DEFAULT_FIB;
1595 		break;
1596 	}
1597 
1598 	/*
1599 	 * Find the correct routing tree to use for this Address Family
1600 	 */
1601 	rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
1602 	if (rnh == NULL)
1603 		return (EAFNOSUPPORT);
1604 
1605 	/*
1606 	 * If we are adding a host route then we don't want to put
1607 	 * a netmask in the tree, nor do we want to clone it.
1608 	 */
1609 	if (flags & RTF_HOST)
1610 		netmask = NULL;
1611 
1612 	switch (req) {
1613 	case RTM_DELETE:
1614 		if (netmask) {
1615 			rt_maskedcopy(dst, (struct sockaddr *)&mdst, netmask);
1616 			dst = (struct sockaddr *)&mdst;
1617 		}
1618 
1619 		RIB_WLOCK(rnh);
1620 		rt = rt_unlinkrte(rnh, info, &error);
1621 		RIB_WUNLOCK(rnh);
1622 		if (error != 0)
1623 			return (error);
1624 
1625 		rt_notifydelete(rt, info);
1626 
1627 		/*
1628 		 * If the caller wants it, then it can have it,
1629 		 * but it's up to it to free the rtentry as we won't be
1630 		 * doing it.
1631 		 */
1632 		if (ret_nrt) {
1633 			*ret_nrt = rt;
1634 			RT_UNLOCK(rt);
1635 		} else
1636 			RTFREE_LOCKED(rt);
1637 		break;
1638 	case RTM_RESOLVE:
1639 		/*
1640 		 * resolve was only used for route cloning
1641 		 * here for compat
1642 		 */
1643 		break;
1644 	case RTM_ADD:
1645 		if ((flags & RTF_GATEWAY) && !gateway)
1646 			return (EINVAL);
1647 		if (dst && gateway && (dst->sa_family != gateway->sa_family) &&
1648 		    (gateway->sa_family != AF_UNSPEC) && (gateway->sa_family != AF_LINK))
1649 			return (EINVAL);
1650 
1651 		if (info->rti_ifa == NULL) {
1652 			error = rt_getifa_fib(info, fibnum);
1653 			if (error)
1654 				return (error);
1655 		} else
1656 			ifa_ref(info->rti_ifa);
1657 		ifa = info->rti_ifa;
1658 		rt = uma_zalloc(V_rtzone, M_NOWAIT);
1659 		if (rt == NULL) {
1660 			ifa_free(ifa);
1661 			return (ENOBUFS);
1662 		}
1663 		rt->rt_flags = RTF_UP | flags;
1664 		rt->rt_fibnum = fibnum;
1665 		/*
1666 		 * Add the gateway. Possibly re-malloc-ing the storage for it.
1667 		 */
1668 		if ((error = rt_setgate(rt, dst, gateway)) != 0) {
1669 			ifa_free(ifa);
1670 			uma_zfree(V_rtzone, rt);
1671 			return (error);
1672 		}
1673 
1674 		/*
1675 		 * point to the (possibly newly malloc'd) dest address.
1676 		 */
1677 		ndst = (struct sockaddr *)rt_key(rt);
1678 
1679 		/*
1680 		 * make sure it contains the value we want (masked if needed).
1681 		 */
1682 		if (netmask) {
1683 			rt_maskedcopy(dst, ndst, netmask);
1684 		} else
1685 			bcopy(dst, ndst, dst->sa_len);
1686 
1687 		/*
1688 		 * We use the ifa reference returned by rt_getifa_fib().
1689 		 * This moved from below so that rnh->rnh_addaddr() can
1690 		 * examine the ifa and  ifa->ifa_ifp if it so desires.
1691 		 */
1692 		rt->rt_ifa = ifa;
1693 		rt->rt_ifp = ifa->ifa_ifp;
1694 		rt->rt_weight = 1;
1695 
1696 		rt_setmetrics(info, rt);
1697 
1698 		RIB_WLOCK(rnh);
1699 		RT_LOCK(rt);
1700 #ifdef RADIX_MPATH
1701 		/* do not permit exactly the same dst/mask/gw pair */
1702 		if (rt_mpath_capable(rnh) &&
1703 			rt_mpath_conflict(rnh, rt, netmask)) {
1704 			RIB_WUNLOCK(rnh);
1705 
1706 			ifa_free(rt->rt_ifa);
1707 			R_Free(rt_key(rt));
1708 			uma_zfree(V_rtzone, rt);
1709 			return (EEXIST);
1710 		}
1711 #endif
1712 
1713 #ifdef FLOWTABLE
1714 		rt0 = rt_flowtable_check_route(rnh, info);
1715 #endif /* FLOWTABLE */
1716 
1717 		/* XXX mtu manipulation will be done in rnh_addaddr -- itojun */
1718 		rn = rnh->rnh_addaddr(ndst, netmask, &rnh->head, rt->rt_nodes);
1719 
1720 		rt_old = NULL;
1721 		if (rn == NULL && (info->rti_flags & RTF_PINNED) != 0) {
1722 
1723 			/*
1724 			 * Force removal and re-try addition
1725 			 * TODO: better multipath&pinned support
1726 			 */
1727 			struct sockaddr *info_dst = info->rti_info[RTAX_DST];
1728 			info->rti_info[RTAX_DST] = ndst;
1729 			/* Do not delete existing PINNED(interface) routes */
1730 			info->rti_flags &= ~RTF_PINNED;
1731 			rt_old = rt_unlinkrte(rnh, info, &error);
1732 			info->rti_flags |= RTF_PINNED;
1733 			info->rti_info[RTAX_DST] = info_dst;
1734 			if (rt_old != NULL)
1735 				rn = rnh->rnh_addaddr(ndst, netmask, &rnh->head,
1736 				    rt->rt_nodes);
1737 		}
1738 		RIB_WUNLOCK(rnh);
1739 
1740 		if (rt_old != NULL)
1741 			RT_UNLOCK(rt_old);
1742 
1743 		/*
1744 		 * If it still failed to go into the tree,
1745 		 * then un-make it (this should be a function)
1746 		 */
1747 		if (rn == NULL) {
1748 			ifa_free(rt->rt_ifa);
1749 			R_Free(rt_key(rt));
1750 			uma_zfree(V_rtzone, rt);
1751 #ifdef FLOWTABLE
1752 			if (rt0 != NULL)
1753 				RTFREE(rt0);
1754 #endif
1755 			return (EEXIST);
1756 		}
1757 #ifdef FLOWTABLE
1758 		else if (rt0 != NULL) {
1759 			flowtable_route_flush(dst->sa_family, rt0);
1760 			RTFREE(rt0);
1761 		}
1762 #endif
1763 
1764 		if (rt_old != NULL) {
1765 			rt_notifydelete(rt_old, info);
1766 			RTFREE(rt_old);
1767 		}
1768 
1769 		/*
1770 		 * If this protocol has something to add to this then
1771 		 * allow it to do that as well.
1772 		 */
1773 		if (ifa->ifa_rtrequest)
1774 			ifa->ifa_rtrequest(req, rt, info);
1775 
1776 		/*
1777 		 * actually return a resultant rtentry and
1778 		 * give the caller a single reference.
1779 		 */
1780 		if (ret_nrt) {
1781 			*ret_nrt = rt;
1782 			RT_ADDREF(rt);
1783 		}
1784 		rnh->rnh_gen++;		/* Routing table updated */
1785 		RT_UNLOCK(rt);
1786 		break;
1787 	case RTM_CHANGE:
1788 		RIB_WLOCK(rnh);
1789 		error = rtrequest1_fib_change(rnh, info, ret_nrt, fibnum);
1790 		RIB_WUNLOCK(rnh);
1791 		break;
1792 	default:
1793 		error = EOPNOTSUPP;
1794 	}
1795 
1796 	return (error);
1797 }
1798 
1799 #undef dst
1800 #undef gateway
1801 #undef netmask
1802 #undef ifaaddr
1803 #undef ifpaddr
1804 #undef flags
1805 
1806 static int
1807 rtrequest1_fib_change(struct rib_head *rnh, struct rt_addrinfo *info,
1808     struct rtentry **ret_nrt, u_int fibnum)
1809 {
1810 	struct rtentry *rt = NULL;
1811 	int error = 0;
1812 	int free_ifa = 0;
1813 	int family, mtu;
1814 	struct if_mtuinfo ifmtu;
1815 
1816 	rt = (struct rtentry *)rnh->rnh_lookup(info->rti_info[RTAX_DST],
1817 	    info->rti_info[RTAX_NETMASK], &rnh->head);
1818 
1819 	if (rt == NULL)
1820 		return (ESRCH);
1821 
1822 #ifdef RADIX_MPATH
1823 	/*
1824 	 * If we got multipath routes,
1825 	 * we require users to specify a matching RTAX_GATEWAY.
1826 	 */
1827 	if (rt_mpath_capable(rnh)) {
1828 		rt = rt_mpath_matchgate(rt, info->rti_info[RTAX_GATEWAY]);
1829 		if (rt == NULL)
1830 			return (ESRCH);
1831 	}
1832 #endif
1833 
1834 	RT_LOCK(rt);
1835 
1836 	rt_setmetrics(info, rt);
1837 
1838 	/*
1839 	 * New gateway could require new ifaddr, ifp;
1840 	 * flags may also be different; ifp may be specified
1841 	 * by ll sockaddr when protocol address is ambiguous
1842 	 */
1843 	if (((rt->rt_flags & RTF_GATEWAY) &&
1844 	    info->rti_info[RTAX_GATEWAY] != NULL) ||
1845 	    info->rti_info[RTAX_IFP] != NULL ||
1846 	    (info->rti_info[RTAX_IFA] != NULL &&
1847 	     !sa_equal(info->rti_info[RTAX_IFA], rt->rt_ifa->ifa_addr))) {
1848 		/*
1849 		 * XXX: Temporarily set RTF_RNH_LOCKED flag in the rti_flags
1850 		 *	to avoid rlock in the ifa_ifwithroute().
1851 		 */
1852 		info->rti_flags |= RTF_RNH_LOCKED;
1853 		error = rt_getifa_fib(info, fibnum);
1854 		info->rti_flags &= ~RTF_RNH_LOCKED;
1855 		if (info->rti_ifa != NULL)
1856 			free_ifa = 1;
1857 
1858 		if (error != 0)
1859 			goto bad;
1860 	}
1861 
1862 	/* Check if outgoing interface has changed */
1863 	if (info->rti_ifa != NULL && info->rti_ifa != rt->rt_ifa &&
1864 	    rt->rt_ifa != NULL && rt->rt_ifa->ifa_rtrequest != NULL) {
1865 		rt->rt_ifa->ifa_rtrequest(RTM_DELETE, rt, info);
1866 		ifa_free(rt->rt_ifa);
1867 	}
1868 	/* Update gateway address */
1869 	if (info->rti_info[RTAX_GATEWAY] != NULL) {
1870 		error = rt_setgate(rt, rt_key(rt), info->rti_info[RTAX_GATEWAY]);
1871 		if (error != 0)
1872 			goto bad;
1873 
1874 		rt->rt_flags &= ~RTF_GATEWAY;
1875 		rt->rt_flags |= (RTF_GATEWAY & info->rti_flags);
1876 	}
1877 
1878 	if (info->rti_ifa != NULL && info->rti_ifa != rt->rt_ifa) {
1879 		ifa_ref(info->rti_ifa);
1880 		rt->rt_ifa = info->rti_ifa;
1881 		rt->rt_ifp = info->rti_ifp;
1882 	}
1883 	/* Allow some flags to be toggled on change. */
1884 	rt->rt_flags &= ~RTF_FMASK;
1885 	rt->rt_flags |= info->rti_flags & RTF_FMASK;
1886 
1887 	if (rt->rt_ifa && rt->rt_ifa->ifa_rtrequest != NULL)
1888 	       rt->rt_ifa->ifa_rtrequest(RTM_ADD, rt, info);
1889 
1890 	/* Alter route MTU if necessary */
1891 	if (rt->rt_ifp != NULL) {
1892 		family = info->rti_info[RTAX_DST]->sa_family;
1893 		mtu = if_getmtu_family(rt->rt_ifp, family);
1894 		/* Set default MTU */
1895 		if (rt->rt_mtu == 0)
1896 			rt->rt_mtu = mtu;
1897 		if (rt->rt_mtu != mtu) {
1898 			/* Check if we really need to update */
1899 			ifmtu.ifp = rt->rt_ifp;
1900 			ifmtu.mtu = mtu;
1901 			if_updatemtu_cb(rt->rt_nodes, &ifmtu);
1902 		}
1903 	}
1904 
1905 	if (ret_nrt) {
1906 		*ret_nrt = rt;
1907 		RT_ADDREF(rt);
1908 	}
1909 bad:
1910 	RT_UNLOCK(rt);
1911 	if (free_ifa != 0)
1912 		ifa_free(info->rti_ifa);
1913 	return (error);
1914 }
1915 
1916 static void
1917 rt_setmetrics(const struct rt_addrinfo *info, struct rtentry *rt)
1918 {
1919 
1920 	if (info->rti_mflags & RTV_MTU) {
1921 		if (info->rti_rmx->rmx_mtu != 0) {
1922 
1923 			/*
1924 			 * MTU was explicitly provided by user.
1925 			 * Keep it.
1926 			 */
1927 			rt->rt_flags |= RTF_FIXEDMTU;
1928 		} else {
1929 
1930 			/*
1931 			 * User explicitly sets MTU to 0.
1932 			 * Assume rollback to default.
1933 			 */
1934 			rt->rt_flags &= ~RTF_FIXEDMTU;
1935 		}
1936 		rt->rt_mtu = info->rti_rmx->rmx_mtu;
1937 	}
1938 	if (info->rti_mflags & RTV_WEIGHT)
1939 		rt->rt_weight = info->rti_rmx->rmx_weight;
1940 	/* Kernel -> userland timebase conversion. */
1941 	if (info->rti_mflags & RTV_EXPIRE)
1942 		rt->rt_expire = info->rti_rmx->rmx_expire ?
1943 		    info->rti_rmx->rmx_expire - time_second + time_uptime : 0;
1944 }
1945 
1946 int
1947 rt_setgate(struct rtentry *rt, struct sockaddr *dst, struct sockaddr *gate)
1948 {
1949 	/* XXX dst may be overwritten, can we move this to below */
1950 	int dlen = SA_SIZE(dst), glen = SA_SIZE(gate);
1951 
1952 	/*
1953 	 * Prepare to store the gateway in rt->rt_gateway.
1954 	 * Both dst and gateway are stored one after the other in the same
1955 	 * malloc'd chunk. If we have room, we can reuse the old buffer,
1956 	 * rt_gateway already points to the right place.
1957 	 * Otherwise, malloc a new block and update the 'dst' address.
1958 	 */
1959 	if (rt->rt_gateway == NULL || glen > SA_SIZE(rt->rt_gateway)) {
1960 		caddr_t new;
1961 
1962 		R_Malloc(new, caddr_t, dlen + glen);
1963 		if (new == NULL)
1964 			return ENOBUFS;
1965 		/*
1966 		 * XXX note, we copy from *dst and not *rt_key(rt) because
1967 		 * rt_setgate() can be called to initialize a newly
1968 		 * allocated route entry, in which case rt_key(rt) == NULL
1969 		 * (and also rt->rt_gateway == NULL).
1970 		 * Free()/free() handle a NULL argument just fine.
1971 		 */
1972 		bcopy(dst, new, dlen);
1973 		R_Free(rt_key(rt));	/* free old block, if any */
1974 		rt_key(rt) = (struct sockaddr *)new;
1975 		rt->rt_gateway = (struct sockaddr *)(new + dlen);
1976 	}
1977 
1978 	/*
1979 	 * Copy the new gateway value into the memory chunk.
1980 	 */
1981 	bcopy(gate, rt->rt_gateway, glen);
1982 
1983 	return (0);
1984 }
1985 
1986 void
1987 rt_maskedcopy(struct sockaddr *src, struct sockaddr *dst, struct sockaddr *netmask)
1988 {
1989 	u_char *cp1 = (u_char *)src;
1990 	u_char *cp2 = (u_char *)dst;
1991 	u_char *cp3 = (u_char *)netmask;
1992 	u_char *cplim = cp2 + *cp3;
1993 	u_char *cplim2 = cp2 + *cp1;
1994 
1995 	*cp2++ = *cp1++; *cp2++ = *cp1++; /* copies sa_len & sa_family */
1996 	cp3 += 2;
1997 	if (cplim > cplim2)
1998 		cplim = cplim2;
1999 	while (cp2 < cplim)
2000 		*cp2++ = *cp1++ & *cp3++;
2001 	if (cp2 < cplim2)
2002 		bzero((caddr_t)cp2, (unsigned)(cplim2 - cp2));
2003 }
2004 
2005 /*
2006  * Set up a routing table entry, normally
2007  * for an interface.
2008  */
2009 #define _SOCKADDR_TMPSIZE 128 /* Not too big.. kernel stack size is limited */
2010 static inline  int
2011 rtinit1(struct ifaddr *ifa, int cmd, int flags, int fibnum)
2012 {
2013 	struct sockaddr *dst;
2014 	struct sockaddr *netmask;
2015 	struct rtentry *rt = NULL;
2016 	struct rt_addrinfo info;
2017 	int error = 0;
2018 	int startfib, endfib;
2019 	char tempbuf[_SOCKADDR_TMPSIZE];
2020 	int didwork = 0;
2021 	int a_failure = 0;
2022 	static struct sockaddr_dl null_sdl = {sizeof(null_sdl), AF_LINK};
2023 	struct rib_head *rnh;
2024 
2025 	if (flags & RTF_HOST) {
2026 		dst = ifa->ifa_dstaddr;
2027 		netmask = NULL;
2028 	} else {
2029 		dst = ifa->ifa_addr;
2030 		netmask = ifa->ifa_netmask;
2031 	}
2032 	if (dst->sa_len == 0)
2033 		return(EINVAL);
2034 	switch (dst->sa_family) {
2035 	case AF_INET6:
2036 	case AF_INET:
2037 		/* We support multiple FIBs. */
2038 		break;
2039 	default:
2040 		fibnum = RT_DEFAULT_FIB;
2041 		break;
2042 	}
2043 	if (fibnum == RT_ALL_FIBS) {
2044 		if (V_rt_add_addr_allfibs == 0 && cmd == (int)RTM_ADD)
2045 			startfib = endfib = ifa->ifa_ifp->if_fib;
2046 		else {
2047 			startfib = 0;
2048 			endfib = rt_numfibs - 1;
2049 		}
2050 	} else {
2051 		KASSERT((fibnum < rt_numfibs), ("rtinit1: bad fibnum"));
2052 		startfib = fibnum;
2053 		endfib = fibnum;
2054 	}
2055 
2056 	/*
2057 	 * If it's a delete, check that if it exists,
2058 	 * it's on the correct interface or we might scrub
2059 	 * a route to another ifa which would
2060 	 * be confusing at best and possibly worse.
2061 	 */
2062 	if (cmd == RTM_DELETE) {
2063 		/*
2064 		 * It's a delete, so it should already exist..
2065 		 * If it's a net, mask off the host bits
2066 		 * (Assuming we have a mask)
2067 		 * XXX this is kinda inet specific..
2068 		 */
2069 		if (netmask != NULL) {
2070 			rt_maskedcopy(dst, (struct sockaddr *)tempbuf, netmask);
2071 			dst = (struct sockaddr *)tempbuf;
2072 		}
2073 	}
2074 	/*
2075 	 * Now go through all the requested tables (fibs) and do the
2076 	 * requested action. Realistically, this will either be fib 0
2077 	 * for protocols that don't do multiple tables or all the
2078 	 * tables for those that do.
2079 	 */
2080 	for ( fibnum = startfib; fibnum <= endfib; fibnum++) {
2081 		if (cmd == RTM_DELETE) {
2082 			struct radix_node *rn;
2083 			/*
2084 			 * Look up an rtentry that is in the routing tree and
2085 			 * contains the correct info.
2086 			 */
2087 			rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
2088 			if (rnh == NULL)
2089 				/* this table doesn't exist but others might */
2090 				continue;
2091 			RIB_RLOCK(rnh);
2092 			rn = rnh->rnh_lookup(dst, netmask, &rnh->head);
2093 #ifdef RADIX_MPATH
2094 			if (rt_mpath_capable(rnh)) {
2095 
2096 				if (rn == NULL)
2097 					error = ESRCH;
2098 				else {
2099 					rt = RNTORT(rn);
2100 					/*
2101 					 * for interface route the
2102 					 * rt->rt_gateway is sockaddr_intf
2103 					 * for cloning ARP entries, so
2104 					 * rt_mpath_matchgate must use the
2105 					 * interface address
2106 					 */
2107 					rt = rt_mpath_matchgate(rt,
2108 					    ifa->ifa_addr);
2109 					if (rt == NULL)
2110 						error = ESRCH;
2111 				}
2112 			}
2113 #endif
2114 			error = (rn == NULL ||
2115 			    (rn->rn_flags & RNF_ROOT) ||
2116 			    RNTORT(rn)->rt_ifa != ifa);
2117 			RIB_RUNLOCK(rnh);
2118 			if (error) {
2119 				/* this is only an error if bad on ALL tables */
2120 				continue;
2121 			}
2122 		}
2123 		/*
2124 		 * Do the actual request
2125 		 */
2126 		bzero((caddr_t)&info, sizeof(info));
2127 		info.rti_ifa = ifa;
2128 		info.rti_flags = flags |
2129 		    (ifa->ifa_flags & ~IFA_RTSELF) | RTF_PINNED;
2130 		info.rti_info[RTAX_DST] = dst;
2131 		/*
2132 		 * doing this for compatibility reasons
2133 		 */
2134 		if (cmd == RTM_ADD)
2135 			info.rti_info[RTAX_GATEWAY] =
2136 			    (struct sockaddr *)&null_sdl;
2137 		else
2138 			info.rti_info[RTAX_GATEWAY] = ifa->ifa_addr;
2139 		info.rti_info[RTAX_NETMASK] = netmask;
2140 		error = rtrequest1_fib(cmd, &info, &rt, fibnum);
2141 
2142 		if (error == 0 && rt != NULL) {
2143 			/*
2144 			 * notify any listening routing agents of the change
2145 			 */
2146 			RT_LOCK(rt);
2147 #ifdef RADIX_MPATH
2148 			/*
2149 			 * in case address alias finds the first address
2150 			 * e.g. ifconfig bge0 192.0.2.246/24
2151 			 * e.g. ifconfig bge0 192.0.2.247/24
2152 			 * the address set in the route is 192.0.2.246
2153 			 * so we need to replace it with 192.0.2.247
2154 			 */
2155 			if (memcmp(rt->rt_ifa->ifa_addr,
2156 			    ifa->ifa_addr, ifa->ifa_addr->sa_len)) {
2157 				ifa_free(rt->rt_ifa);
2158 				ifa_ref(ifa);
2159 				rt->rt_ifp = ifa->ifa_ifp;
2160 				rt->rt_ifa = ifa;
2161 			}
2162 #endif
2163 			/*
2164 			 * doing this for compatibility reasons
2165 			 */
2166 			if (cmd == RTM_ADD) {
2167 			    ((struct sockaddr_dl *)rt->rt_gateway)->sdl_type  =
2168 				rt->rt_ifp->if_type;
2169 			    ((struct sockaddr_dl *)rt->rt_gateway)->sdl_index =
2170 				rt->rt_ifp->if_index;
2171 			}
2172 			RT_ADDREF(rt);
2173 			RT_UNLOCK(rt);
2174 			rt_newaddrmsg_fib(cmd, ifa, error, rt, fibnum);
2175 			RT_LOCK(rt);
2176 			RT_REMREF(rt);
2177 			if (cmd == RTM_DELETE) {
2178 				/*
2179 				 * If we are deleting, and we found an entry,
2180 				 * then it's been removed from the tree..
2181 				 * now throw it away.
2182 				 */
2183 				RTFREE_LOCKED(rt);
2184 			} else {
2185 				if (cmd == RTM_ADD) {
2186 					/*
2187 					 * We just wanted to add it..
2188 					 * we don't actually need a reference.
2189 					 */
2190 					RT_REMREF(rt);
2191 				}
2192 				RT_UNLOCK(rt);
2193 			}
2194 			didwork = 1;
2195 		}
2196 		if (error)
2197 			a_failure = error;
2198 	}
2199 	if (cmd == RTM_DELETE) {
2200 		if (didwork) {
2201 			error = 0;
2202 		} else {
2203 			/* we only give an error if it wasn't in any table */
2204 			error = ((flags & RTF_HOST) ?
2205 			    EHOSTUNREACH : ENETUNREACH);
2206 		}
2207 	} else {
2208 		if (a_failure) {
2209 			/* return an error if any of them failed */
2210 			error = a_failure;
2211 		}
2212 	}
2213 	return (error);
2214 }
2215 
2216 /*
2217  * Set up a routing table entry, normally
2218  * for an interface.
2219  */
2220 int
2221 rtinit(struct ifaddr *ifa, int cmd, int flags)
2222 {
2223 	struct sockaddr *dst;
2224 	int fib = RT_DEFAULT_FIB;
2225 
2226 	if (flags & RTF_HOST) {
2227 		dst = ifa->ifa_dstaddr;
2228 	} else {
2229 		dst = ifa->ifa_addr;
2230 	}
2231 
2232 	switch (dst->sa_family) {
2233 	case AF_INET6:
2234 	case AF_INET:
2235 		/* We do support multiple FIBs. */
2236 		fib = RT_ALL_FIBS;
2237 		break;
2238 	}
2239 	return (rtinit1(ifa, cmd, flags, fib));
2240 }
2241 
2242 /*
2243  * Announce interface address arrival/withdraw
2244  * Returns 0 on success.
2245  */
2246 int
2247 rt_addrmsg(int cmd, struct ifaddr *ifa, int fibnum)
2248 {
2249 
2250 	KASSERT(cmd == RTM_ADD || cmd == RTM_DELETE,
2251 	    ("unexpected cmd %d", cmd));
2252 
2253 	KASSERT(fibnum == RT_ALL_FIBS || (fibnum >= 0 && fibnum < rt_numfibs),
2254 	    ("%s: fib out of range 0 <=%d<%d", __func__, fibnum, rt_numfibs));
2255 
2256 #if defined(INET) || defined(INET6)
2257 #ifdef SCTP
2258 	/*
2259 	 * notify the SCTP stack
2260 	 * this will only get called when an address is added/deleted
2261 	 * XXX pass the ifaddr struct instead if ifa->ifa_addr...
2262 	 */
2263 	sctp_addr_change(ifa, cmd);
2264 #endif /* SCTP */
2265 #endif
2266 	return (rtsock_addrmsg(cmd, ifa, fibnum));
2267 }
2268 
2269 /*
2270  * Announce route addition/removal.
2271  * Users of this function MUST validate input data BEFORE calling.
2272  * However we have to be able to handle invalid data:
2273  * if some userland app sends us "invalid" route message (invalid mask,
2274  * no dst, wrong address families, etc...) we need to pass it back
2275  * to app (and any other rtsock consumers) with rtm_errno field set to
2276  * non-zero value.
2277  * Returns 0 on success.
2278  */
2279 int
2280 rt_routemsg(int cmd, struct ifnet *ifp, int error, struct rtentry *rt,
2281     int fibnum)
2282 {
2283 
2284 	KASSERT(cmd == RTM_ADD || cmd == RTM_DELETE,
2285 	    ("unexpected cmd %d", cmd));
2286 
2287 	KASSERT(fibnum == RT_ALL_FIBS || (fibnum >= 0 && fibnum < rt_numfibs),
2288 	    ("%s: fib out of range 0 <=%d<%d", __func__, fibnum, rt_numfibs));
2289 
2290 	KASSERT(rt_key(rt) != NULL, (":%s: rt_key must be supplied", __func__));
2291 
2292 	return (rtsock_routemsg(cmd, ifp, error, rt, fibnum));
2293 }
2294 
2295 void
2296 rt_newaddrmsg(int cmd, struct ifaddr *ifa, int error, struct rtentry *rt)
2297 {
2298 
2299 	rt_newaddrmsg_fib(cmd, ifa, error, rt, RT_ALL_FIBS);
2300 }
2301 
2302 /*
2303  * This is called to generate messages from the routing socket
2304  * indicating a network interface has had addresses associated with it.
2305  */
2306 void
2307 rt_newaddrmsg_fib(int cmd, struct ifaddr *ifa, int error, struct rtentry *rt,
2308     int fibnum)
2309 {
2310 
2311 	KASSERT(cmd == RTM_ADD || cmd == RTM_DELETE,
2312 		("unexpected cmd %u", cmd));
2313 	KASSERT(fibnum == RT_ALL_FIBS || (fibnum >= 0 && fibnum < rt_numfibs),
2314 	    ("%s: fib out of range 0 <=%d<%d", __func__, fibnum, rt_numfibs));
2315 
2316 	if (cmd == RTM_ADD) {
2317 		rt_addrmsg(cmd, ifa, fibnum);
2318 		if (rt != NULL)
2319 			rt_routemsg(cmd, ifa->ifa_ifp, error, rt, fibnum);
2320 	} else {
2321 		if (rt != NULL)
2322 			rt_routemsg(cmd, ifa->ifa_ifp, error, rt, fibnum);
2323 		rt_addrmsg(cmd, ifa, fibnum);
2324 	}
2325 }
2326 
2327