xref: /freebsd/sys/net/route.c (revision f84460dded7e032dbe829f79876163b974c9bed1)
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/vnet.h>
61 #include <net/flowtable.h>
62 
63 #ifdef RADIX_MPATH
64 #include <net/radix_mpath.h>
65 #endif
66 
67 #include <netinet/in.h>
68 #include <netinet/ip_mroute.h>
69 
70 #include <vm/uma.h>
71 
72 #define	RT_MAXFIBS	UINT16_MAX
73 
74 /* Kernel config default option. */
75 #ifdef ROUTETABLES
76 #if ROUTETABLES <= 0
77 #error "ROUTETABLES defined too low"
78 #endif
79 #if ROUTETABLES > RT_MAXFIBS
80 #error "ROUTETABLES defined too big"
81 #endif
82 #define	RT_NUMFIBS	ROUTETABLES
83 #endif /* ROUTETABLES */
84 /* Initialize to default if not otherwise set. */
85 #ifndef	RT_NUMFIBS
86 #define	RT_NUMFIBS	1
87 #endif
88 
89 #if defined(INET) || defined(INET6)
90 #ifdef SCTP
91 extern void sctp_addr_change(struct ifaddr *ifa, int cmd);
92 #endif /* SCTP */
93 #endif
94 
95 
96 /* This is read-only.. */
97 u_int rt_numfibs = RT_NUMFIBS;
98 SYSCTL_UINT(_net, OID_AUTO, fibs, CTLFLAG_RDTUN, &rt_numfibs, 0, "");
99 
100 /*
101  * By default add routes to all fibs for new interfaces.
102  * Once this is set to 0 then only allocate routes on interface
103  * changes for the FIB of the caller when adding a new set of addresses
104  * to an interface.  XXX this is a shotgun aproach to a problem that needs
105  * a more fine grained solution.. that will come.
106  * XXX also has the problems getting the FIB from curthread which will not
107  * always work given the fib can be overridden and prefixes can be added
108  * from the network stack context.
109  */
110 VNET_DEFINE(u_int, rt_add_addr_allfibs) = 1;
111 SYSCTL_UINT(_net, OID_AUTO, add_addr_allfibs, CTLFLAG_RWTUN | CTLFLAG_VNET,
112     &VNET_NAME(rt_add_addr_allfibs), 0, "");
113 
114 VNET_DEFINE(struct rtstat, rtstat);
115 #define	V_rtstat	VNET(rtstat)
116 
117 VNET_DEFINE(struct radix_node_head *, rt_tables);
118 #define	V_rt_tables	VNET(rt_tables)
119 
120 VNET_DEFINE(int, rttrash);		/* routes not in table but not freed */
121 #define	V_rttrash	VNET(rttrash)
122 
123 
124 /*
125  * Convert a 'struct radix_node *' to a 'struct rtentry *'.
126  * The operation can be done safely (in this code) because a
127  * 'struct rtentry' starts with two 'struct radix_node''s, the first
128  * one representing leaf nodes in the routing tree, which is
129  * what the code in radix.c passes us as a 'struct radix_node'.
130  *
131  * But because there are a lot of assumptions in this conversion,
132  * do not cast explicitly, but always use the macro below.
133  */
134 #define RNTORT(p)	((struct rtentry *)(p))
135 
136 static VNET_DEFINE(uma_zone_t, rtzone);		/* Routing table UMA zone. */
137 #define	V_rtzone	VNET(rtzone)
138 
139 static int rtrequest1_fib_change(struct radix_node_head *, struct rt_addrinfo *,
140     struct rtentry **, u_int);
141 static void rt_setmetrics(const struct rt_addrinfo *, struct rtentry *);
142 static int rt_ifdelroute(struct rtentry *rt, void *arg);
143 
144 struct if_mtuinfo
145 {
146 	struct ifnet	*ifp;
147 	int		mtu;
148 };
149 
150 static int	if_updatemtu_cb(struct radix_node *, void *);
151 
152 /*
153  * handler for net.my_fibnum
154  */
155 static int
156 sysctl_my_fibnum(SYSCTL_HANDLER_ARGS)
157 {
158         int fibnum;
159         int error;
160 
161         fibnum = curthread->td_proc->p_fibnum;
162         error = sysctl_handle_int(oidp, &fibnum, 0, req);
163         return (error);
164 }
165 
166 SYSCTL_PROC(_net, OID_AUTO, my_fibnum, CTLTYPE_INT|CTLFLAG_RD,
167             NULL, 0, &sysctl_my_fibnum, "I", "default FIB of caller");
168 
169 static __inline struct radix_node_head **
170 rt_tables_get_rnh_ptr(int table, int fam)
171 {
172 	struct radix_node_head **rnh;
173 
174 	KASSERT(table >= 0 && table < rt_numfibs, ("%s: table out of bounds.",
175 	    __func__));
176 	KASSERT(fam >= 0 && fam < (AF_MAX+1), ("%s: fam out of bounds.",
177 	    __func__));
178 
179 	/* rnh is [fib=0][af=0]. */
180 	rnh = (struct radix_node_head **)V_rt_tables;
181 	/* Get the offset to the requested table and fam. */
182 	rnh += table * (AF_MAX+1) + fam;
183 
184 	return (rnh);
185 }
186 
187 struct radix_node_head *
188 rt_tables_get_rnh(int table, int fam)
189 {
190 
191 	return (*rt_tables_get_rnh_ptr(table, fam));
192 }
193 
194 /*
195  * route initialization must occur before ip6_init2(), which happenas at
196  * SI_ORDER_MIDDLE.
197  */
198 static void
199 route_init(void)
200 {
201 
202 	/* whack the tunable ints into  line. */
203 	if (rt_numfibs > RT_MAXFIBS)
204 		rt_numfibs = RT_MAXFIBS;
205 	if (rt_numfibs == 0)
206 		rt_numfibs = 1;
207 }
208 SYSINIT(route_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD, route_init, 0);
209 
210 static int
211 rtentry_zinit(void *mem, int size, int how)
212 {
213 	struct rtentry *rt = mem;
214 
215 	rt->rt_pksent = counter_u64_alloc(how);
216 	if (rt->rt_pksent == NULL)
217 		return (ENOMEM);
218 
219 	RT_LOCK_INIT(rt);
220 
221 	return (0);
222 }
223 
224 static void
225 rtentry_zfini(void *mem, int size)
226 {
227 	struct rtentry *rt = mem;
228 
229 	RT_LOCK_DESTROY(rt);
230 	counter_u64_free(rt->rt_pksent);
231 }
232 
233 static int
234 rtentry_ctor(void *mem, int size, void *arg, int how)
235 {
236 	struct rtentry *rt = mem;
237 
238 	bzero(rt, offsetof(struct rtentry, rt_endzero));
239 	counter_u64_zero(rt->rt_pksent);
240 
241 	return (0);
242 }
243 
244 static void
245 rtentry_dtor(void *mem, int size, void *arg)
246 {
247 	struct rtentry *rt = mem;
248 
249 	RT_UNLOCK_COND(rt);
250 }
251 
252 static void
253 vnet_route_init(const void *unused __unused)
254 {
255 	struct domain *dom;
256 	struct radix_node_head **rnh;
257 	int table;
258 	int fam;
259 
260 	V_rt_tables = malloc(rt_numfibs * (AF_MAX+1) *
261 	    sizeof(struct radix_node_head *), M_RTABLE, M_WAITOK|M_ZERO);
262 
263 	V_rtzone = uma_zcreate("rtentry", sizeof(struct rtentry),
264 	    rtentry_ctor, rtentry_dtor,
265 	    rtentry_zinit, rtentry_zfini, UMA_ALIGN_PTR, 0);
266 	for (dom = domains; dom; dom = dom->dom_next) {
267 		if (dom->dom_rtattach == NULL)
268 			continue;
269 
270 		for  (table = 0; table < rt_numfibs; table++) {
271 			fam = dom->dom_family;
272 			if (table != 0 && fam != AF_INET6 && fam != AF_INET)
273 				break;
274 
275 			rnh = rt_tables_get_rnh_ptr(table, fam);
276 			if (rnh == NULL)
277 				panic("%s: rnh NULL", __func__);
278 			dom->dom_rtattach((void **)rnh, 0);
279 		}
280 	}
281 }
282 VNET_SYSINIT(vnet_route_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH,
283     vnet_route_init, 0);
284 
285 #ifdef VIMAGE
286 static void
287 vnet_route_uninit(const void *unused __unused)
288 {
289 	int table;
290 	int fam;
291 	struct domain *dom;
292 	struct radix_node_head **rnh;
293 
294 	for (dom = domains; dom; dom = dom->dom_next) {
295 		if (dom->dom_rtdetach == NULL)
296 			continue;
297 
298 		for (table = 0; table < rt_numfibs; table++) {
299 			fam = dom->dom_family;
300 
301 			if (table != 0 && fam != AF_INET6 && fam != AF_INET)
302 				break;
303 
304 			rnh = rt_tables_get_rnh_ptr(table, fam);
305 			if (rnh == NULL)
306 				panic("%s: rnh NULL", __func__);
307 			dom->dom_rtdetach((void **)rnh, 0);
308 		}
309 	}
310 
311 	free(V_rt_tables, M_RTABLE);
312 	uma_zdestroy(V_rtzone);
313 }
314 VNET_SYSUNINIT(vnet_route_uninit, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD,
315     vnet_route_uninit, 0);
316 #endif
317 
318 #ifndef _SYS_SYSPROTO_H_
319 struct setfib_args {
320 	int     fibnum;
321 };
322 #endif
323 int
324 sys_setfib(struct thread *td, struct setfib_args *uap)
325 {
326 	if (uap->fibnum < 0 || uap->fibnum >= rt_numfibs)
327 		return EINVAL;
328 	td->td_proc->p_fibnum = uap->fibnum;
329 	return (0);
330 }
331 
332 /*
333  * Packet routing routines.
334  */
335 void
336 rtalloc(struct route *ro)
337 {
338 
339 	rtalloc_ign_fib(ro, 0UL, RT_DEFAULT_FIB);
340 }
341 
342 void
343 rtalloc_fib(struct route *ro, u_int fibnum)
344 {
345 	rtalloc_ign_fib(ro, 0UL, fibnum);
346 }
347 
348 void
349 rtalloc_ign(struct route *ro, u_long ignore)
350 {
351 	struct rtentry *rt;
352 
353 	if ((rt = ro->ro_rt) != NULL) {
354 		if (rt->rt_ifp != NULL && rt->rt_flags & RTF_UP)
355 			return;
356 		RTFREE(rt);
357 		ro->ro_rt = NULL;
358 	}
359 	ro->ro_rt = rtalloc1_fib(&ro->ro_dst, 1, ignore, RT_DEFAULT_FIB);
360 	if (ro->ro_rt)
361 		RT_UNLOCK(ro->ro_rt);
362 }
363 
364 void
365 rtalloc_ign_fib(struct route *ro, u_long ignore, u_int fibnum)
366 {
367 	struct rtentry *rt;
368 
369 	if ((rt = ro->ro_rt) != NULL) {
370 		if (rt->rt_ifp != NULL && rt->rt_flags & RTF_UP)
371 			return;
372 		RTFREE(rt);
373 		ro->ro_rt = NULL;
374 	}
375 	ro->ro_rt = rtalloc1_fib(&ro->ro_dst, 1, ignore, fibnum);
376 	if (ro->ro_rt)
377 		RT_UNLOCK(ro->ro_rt);
378 }
379 
380 /*
381  * Look up the route that matches the address given
382  * Or, at least try.. Create a cloned route if needed.
383  *
384  * The returned route, if any, is locked.
385  */
386 struct rtentry *
387 rtalloc1(struct sockaddr *dst, int report, u_long ignflags)
388 {
389 
390 	return (rtalloc1_fib(dst, report, ignflags, RT_DEFAULT_FIB));
391 }
392 
393 struct rtentry *
394 rtalloc1_fib(struct sockaddr *dst, int report, u_long ignflags,
395 		    u_int fibnum)
396 {
397 	struct radix_node_head *rnh;
398 	struct radix_node *rn;
399 	struct rtentry *newrt;
400 	struct rt_addrinfo info;
401 	int err = 0, msgtype = RTM_MISS;
402 	int needlock;
403 
404 	KASSERT((fibnum < rt_numfibs), ("rtalloc1_fib: bad fibnum"));
405 	rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
406 	newrt = NULL;
407 	if (rnh == NULL)
408 		goto miss;
409 
410 	/*
411 	 * Look up the address in the table for that Address Family
412 	 */
413 	needlock = !(ignflags & RTF_RNH_LOCKED);
414 	if (needlock)
415 		RADIX_NODE_HEAD_RLOCK(rnh);
416 #ifdef INVARIANTS
417 	else
418 		RADIX_NODE_HEAD_LOCK_ASSERT(rnh);
419 #endif
420 	rn = rnh->rnh_matchaddr(dst, rnh);
421 	if (rn && ((rn->rn_flags & RNF_ROOT) == 0)) {
422 		newrt = RNTORT(rn);
423 		RT_LOCK(newrt);
424 		RT_ADDREF(newrt);
425 		if (needlock)
426 			RADIX_NODE_HEAD_RUNLOCK(rnh);
427 		goto done;
428 
429 	} else if (needlock)
430 		RADIX_NODE_HEAD_RUNLOCK(rnh);
431 
432 	/*
433 	 * Either we hit the root or couldn't find any match,
434 	 * Which basically means
435 	 * "caint get there frm here"
436 	 */
437 miss:
438 	V_rtstat.rts_unreach++;
439 
440 	if (report) {
441 		/*
442 		 * If required, report the failure to the supervising
443 		 * Authorities.
444 		 * For a delete, this is not an error. (report == 0)
445 		 */
446 		bzero(&info, sizeof(info));
447 		info.rti_info[RTAX_DST] = dst;
448 		rt_missmsg_fib(msgtype, &info, 0, err, fibnum);
449 	}
450 done:
451 	if (newrt)
452 		RT_LOCK_ASSERT(newrt);
453 	return (newrt);
454 }
455 
456 /*
457  * Remove a reference count from an rtentry.
458  * If the count gets low enough, take it out of the routing table
459  */
460 void
461 rtfree(struct rtentry *rt)
462 {
463 	struct radix_node_head *rnh;
464 
465 	KASSERT(rt != NULL,("%s: NULL rt", __func__));
466 	rnh = rt_tables_get_rnh(rt->rt_fibnum, rt_key(rt)->sa_family);
467 	KASSERT(rnh != NULL,("%s: NULL rnh", __func__));
468 
469 	RT_LOCK_ASSERT(rt);
470 
471 	/*
472 	 * The callers should use RTFREE_LOCKED() or RTFREE(), so
473 	 * we should come here exactly with the last reference.
474 	 */
475 	RT_REMREF(rt);
476 	if (rt->rt_refcnt > 0) {
477 		log(LOG_DEBUG, "%s: %p has %d refs\n", __func__, rt, rt->rt_refcnt);
478 		goto done;
479 	}
480 
481 	/*
482 	 * On last reference give the "close method" a chance
483 	 * to cleanup private state.  This also permits (for
484 	 * IPv4 and IPv6) a chance to decide if the routing table
485 	 * entry should be purged immediately or at a later time.
486 	 * When an immediate purge is to happen the close routine
487 	 * typically calls rtexpunge which clears the RTF_UP flag
488 	 * on the entry so that the code below reclaims the storage.
489 	 */
490 	if (rt->rt_refcnt == 0 && rnh->rnh_close)
491 		rnh->rnh_close((struct radix_node *)rt, rnh);
492 
493 	/*
494 	 * If we are no longer "up" (and ref == 0)
495 	 * then we can free the resources associated
496 	 * with the route.
497 	 */
498 	if ((rt->rt_flags & RTF_UP) == 0) {
499 		if (rt->rt_nodes->rn_flags & (RNF_ACTIVE | RNF_ROOT))
500 			panic("rtfree 2");
501 		/*
502 		 * the rtentry must have been removed from the routing table
503 		 * so it is represented in rttrash.. remove that now.
504 		 */
505 		V_rttrash--;
506 #ifdef	DIAGNOSTIC
507 		if (rt->rt_refcnt < 0) {
508 			printf("rtfree: %p not freed (neg refs)\n", rt);
509 			goto done;
510 		}
511 #endif
512 		/*
513 		 * release references on items we hold them on..
514 		 * e.g other routes and ifaddrs.
515 		 */
516 		if (rt->rt_ifa)
517 			ifa_free(rt->rt_ifa);
518 		/*
519 		 * The key is separatly alloc'd so free it (see rt_setgate()).
520 		 * This also frees the gateway, as they are always malloc'd
521 		 * together.
522 		 */
523 		R_Free(rt_key(rt));
524 
525 		/*
526 		 * and the rtentry itself of course
527 		 */
528 		uma_zfree(V_rtzone, rt);
529 		return;
530 	}
531 done:
532 	RT_UNLOCK(rt);
533 }
534 
535 
536 /*
537  * Force a routing table entry to the specified
538  * destination to go through the given gateway.
539  * Normally called as a result of a routing redirect
540  * message from the network layer.
541  */
542 void
543 rtredirect(struct sockaddr *dst,
544 	struct sockaddr *gateway,
545 	struct sockaddr *netmask,
546 	int flags,
547 	struct sockaddr *src)
548 {
549 
550 	rtredirect_fib(dst, gateway, netmask, flags, src, RT_DEFAULT_FIB);
551 }
552 
553 void
554 rtredirect_fib(struct sockaddr *dst,
555 	struct sockaddr *gateway,
556 	struct sockaddr *netmask,
557 	int flags,
558 	struct sockaddr *src,
559 	u_int fibnum)
560 {
561 	struct rtentry *rt, *rt0 = NULL;
562 	int error = 0;
563 	short *stat = NULL;
564 	struct rt_addrinfo info;
565 	struct ifaddr *ifa;
566 	struct radix_node_head *rnh;
567 
568 	ifa = NULL;
569 	rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
570 	if (rnh == NULL) {
571 		error = EAFNOSUPPORT;
572 		goto out;
573 	}
574 
575 	/* verify the gateway is directly reachable */
576 	if ((ifa = ifa_ifwithnet(gateway, 0, fibnum)) == NULL) {
577 		error = ENETUNREACH;
578 		goto out;
579 	}
580 	rt = rtalloc1_fib(dst, 0, 0UL, fibnum);	/* NB: rt is locked */
581 	/*
582 	 * If the redirect isn't from our current router for this dst,
583 	 * it's either old or wrong.  If it redirects us to ourselves,
584 	 * we have a routing loop, perhaps as a result of an interface
585 	 * going down recently.
586 	 */
587 	if (!(flags & RTF_DONE) && rt &&
588 	     (!sa_equal(src, rt->rt_gateway) || rt->rt_ifa != ifa))
589 		error = EINVAL;
590 	else if (ifa_ifwithaddr_check(gateway))
591 		error = EHOSTUNREACH;
592 	if (error)
593 		goto done;
594 	/*
595 	 * Create a new entry if we just got back a wildcard entry
596 	 * or the lookup failed.  This is necessary for hosts
597 	 * which use routing redirects generated by smart gateways
598 	 * to dynamically build the routing tables.
599 	 */
600 	if (rt == NULL || (rt_mask(rt) && rt_mask(rt)->sa_len < 2))
601 		goto create;
602 	/*
603 	 * Don't listen to the redirect if it's
604 	 * for a route to an interface.
605 	 */
606 	if (rt->rt_flags & RTF_GATEWAY) {
607 		if (((rt->rt_flags & RTF_HOST) == 0) && (flags & RTF_HOST)) {
608 			/*
609 			 * Changing from route to net => route to host.
610 			 * Create new route, rather than smashing route to net.
611 			 */
612 		create:
613 			rt0 = rt;
614 			rt = NULL;
615 
616 			flags |=  RTF_GATEWAY | RTF_DYNAMIC;
617 			bzero((caddr_t)&info, sizeof(info));
618 			info.rti_info[RTAX_DST] = dst;
619 			info.rti_info[RTAX_GATEWAY] = gateway;
620 			info.rti_info[RTAX_NETMASK] = netmask;
621 			info.rti_ifa = ifa;
622 			info.rti_flags = flags;
623 			if (rt0 != NULL)
624 				RT_UNLOCK(rt0);	/* drop lock to avoid LOR with RNH */
625 			error = rtrequest1_fib(RTM_ADD, &info, &rt, fibnum);
626 			if (rt != NULL) {
627 				RT_LOCK(rt);
628 				if (rt0 != NULL)
629 					EVENTHANDLER_INVOKE(route_redirect_event, rt0, rt, dst);
630 				flags = rt->rt_flags;
631 			}
632 			if (rt0 != NULL)
633 				RTFREE(rt0);
634 
635 			stat = &V_rtstat.rts_dynamic;
636 		} else {
637 			struct rtentry *gwrt;
638 
639 			/*
640 			 * Smash the current notion of the gateway to
641 			 * this destination.  Should check about netmask!!!
642 			 */
643 			rt->rt_flags |= RTF_MODIFIED;
644 			flags |= RTF_MODIFIED;
645 			stat = &V_rtstat.rts_newgateway;
646 			/*
647 			 * add the key and gateway (in one malloc'd chunk).
648 			 */
649 			RT_UNLOCK(rt);
650 			RADIX_NODE_HEAD_LOCK(rnh);
651 			RT_LOCK(rt);
652 			rt_setgate(rt, rt_key(rt), gateway);
653 			gwrt = rtalloc1(gateway, 1, RTF_RNH_LOCKED);
654 			RADIX_NODE_HEAD_UNLOCK(rnh);
655 			EVENTHANDLER_INVOKE(route_redirect_event, rt, gwrt, dst);
656 			RTFREE_LOCKED(gwrt);
657 		}
658 	} else
659 		error = EHOSTUNREACH;
660 done:
661 	if (rt)
662 		RTFREE_LOCKED(rt);
663 out:
664 	if (error)
665 		V_rtstat.rts_badredirect++;
666 	else if (stat != NULL)
667 		(*stat)++;
668 	bzero((caddr_t)&info, sizeof(info));
669 	info.rti_info[RTAX_DST] = dst;
670 	info.rti_info[RTAX_GATEWAY] = gateway;
671 	info.rti_info[RTAX_NETMASK] = netmask;
672 	info.rti_info[RTAX_AUTHOR] = src;
673 	rt_missmsg_fib(RTM_REDIRECT, &info, flags, error, fibnum);
674 	if (ifa != NULL)
675 		ifa_free(ifa);
676 }
677 
678 int
679 rtioctl(u_long req, caddr_t data)
680 {
681 
682 	return (rtioctl_fib(req, data, RT_DEFAULT_FIB));
683 }
684 
685 /*
686  * Routing table ioctl interface.
687  */
688 int
689 rtioctl_fib(u_long req, caddr_t data, u_int fibnum)
690 {
691 
692 	/*
693 	 * If more ioctl commands are added here, make sure the proper
694 	 * super-user checks are being performed because it is possible for
695 	 * prison-root to make it this far if raw sockets have been enabled
696 	 * in jails.
697 	 */
698 #ifdef INET
699 	/* Multicast goop, grrr... */
700 	return mrt_ioctl ? mrt_ioctl(req, data, fibnum) : EOPNOTSUPP;
701 #else /* INET */
702 	return ENXIO;
703 #endif /* INET */
704 }
705 
706 struct ifaddr *
707 ifa_ifwithroute(int flags, struct sockaddr *dst, struct sockaddr *gateway,
708 				u_int fibnum)
709 {
710 	struct ifaddr *ifa;
711 	int not_found = 0;
712 
713 	if ((flags & RTF_GATEWAY) == 0) {
714 		/*
715 		 * If we are adding a route to an interface,
716 		 * and the interface is a pt to pt link
717 		 * we should search for the destination
718 		 * as our clue to the interface.  Otherwise
719 		 * we can use the local address.
720 		 */
721 		ifa = NULL;
722 		if (flags & RTF_HOST)
723 			ifa = ifa_ifwithdstaddr(dst, fibnum);
724 		if (ifa == NULL)
725 			ifa = ifa_ifwithaddr(gateway);
726 	} else {
727 		/*
728 		 * If we are adding a route to a remote net
729 		 * or host, the gateway may still be on the
730 		 * other end of a pt to pt link.
731 		 */
732 		ifa = ifa_ifwithdstaddr(gateway, fibnum);
733 	}
734 	if (ifa == NULL)
735 		ifa = ifa_ifwithnet(gateway, 0, fibnum);
736 	if (ifa == NULL) {
737 		struct rtentry *rt = rtalloc1_fib(gateway, 0, RTF_RNH_LOCKED, fibnum);
738 		if (rt == NULL)
739 			return (NULL);
740 		/*
741 		 * dismiss a gateway that is reachable only
742 		 * through the default router
743 		 */
744 		switch (gateway->sa_family) {
745 		case AF_INET:
746 			if (satosin(rt_key(rt))->sin_addr.s_addr == INADDR_ANY)
747 				not_found = 1;
748 			break;
749 		case AF_INET6:
750 			if (IN6_IS_ADDR_UNSPECIFIED(&satosin6(rt_key(rt))->sin6_addr))
751 				not_found = 1;
752 			break;
753 		default:
754 			break;
755 		}
756 		if (!not_found && rt->rt_ifa != NULL) {
757 			ifa = rt->rt_ifa;
758 			ifa_ref(ifa);
759 		}
760 		RT_REMREF(rt);
761 		RT_UNLOCK(rt);
762 		if (not_found || ifa == NULL)
763 			return (NULL);
764 	}
765 	if (ifa->ifa_addr->sa_family != dst->sa_family) {
766 		struct ifaddr *oifa = ifa;
767 		ifa = ifaof_ifpforaddr(dst, ifa->ifa_ifp);
768 		if (ifa == NULL)
769 			ifa = oifa;
770 		else
771 			ifa_free(oifa);
772 	}
773 	return (ifa);
774 }
775 
776 /*
777  * Do appropriate manipulations of a routing tree given
778  * all the bits of info needed
779  */
780 int
781 rtrequest(int req,
782 	struct sockaddr *dst,
783 	struct sockaddr *gateway,
784 	struct sockaddr *netmask,
785 	int flags,
786 	struct rtentry **ret_nrt)
787 {
788 
789 	return (rtrequest_fib(req, dst, gateway, netmask, flags, ret_nrt,
790 	    RT_DEFAULT_FIB));
791 }
792 
793 int
794 rtrequest_fib(int req,
795 	struct sockaddr *dst,
796 	struct sockaddr *gateway,
797 	struct sockaddr *netmask,
798 	int flags,
799 	struct rtentry **ret_nrt,
800 	u_int fibnum)
801 {
802 	struct rt_addrinfo info;
803 
804 	if (dst->sa_len == 0)
805 		return(EINVAL);
806 
807 	bzero((caddr_t)&info, sizeof(info));
808 	info.rti_flags = flags;
809 	info.rti_info[RTAX_DST] = dst;
810 	info.rti_info[RTAX_GATEWAY] = gateway;
811 	info.rti_info[RTAX_NETMASK] = netmask;
812 	return rtrequest1_fib(req, &info, ret_nrt, fibnum);
813 }
814 
815 
816 /*
817  * Iterates over all existing fibs in system calling
818  *  @setwa_f function prior to traversing each fib.
819  *  Calls @wa_f function for each element in current fib.
820  * If af is not AF_UNSPEC, iterates over fibs in particular
821  * address family.
822  */
823 void
824 rt_foreach_fib_walk(int af, rt_setwarg_t *setwa_f, rt_walktree_f_t *wa_f,
825     void *arg)
826 {
827 	struct radix_node_head *rnh;
828 	uint32_t fibnum;
829 	int i;
830 
831 	for (fibnum = 0; fibnum < rt_numfibs; fibnum++) {
832 		/* Do we want some specific family? */
833 		if (af != AF_UNSPEC) {
834 			rnh = rt_tables_get_rnh(fibnum, af);
835 			if (rnh == NULL)
836 				continue;
837 			if (setwa_f != NULL)
838 				setwa_f(rnh, fibnum, i, arg);
839 
840 			RADIX_NODE_HEAD_LOCK(rnh);
841 			rnh->rnh_walktree(rnh, (walktree_f_t *)wa_f, arg);
842 			RADIX_NODE_HEAD_UNLOCK(rnh);
843 			continue;
844 		}
845 
846 		for (i = 1; i <= AF_MAX; i++) {
847 			rnh = rt_tables_get_rnh(fibnum, i);
848 			if (rnh == NULL)
849 				continue;
850 			if (setwa_f != NULL)
851 				setwa_f(rnh, fibnum, i, arg);
852 
853 			RADIX_NODE_HEAD_LOCK(rnh);
854 			rnh->rnh_walktree(rnh, (walktree_f_t *)wa_f, arg);
855 			RADIX_NODE_HEAD_UNLOCK(rnh);
856 		}
857 	}
858 }
859 
860 /*
861  * Delete Routes for a Network Interface
862  *
863  * Called for each routing entry via the rnh->rnh_walktree() call above
864  * to delete all route entries referencing a detaching network interface.
865  *
866  * Arguments:
867  *	rt	pointer to rtentry
868  *	arg	argument passed to rnh->rnh_walktree() - detaching interface
869  *
870  * Returns:
871  *	0	successful
872  *	errno	failed - reason indicated
873  */
874 static int
875 rt_ifdelroute(struct rtentry *rt, void *arg)
876 {
877 	struct ifnet	*ifp = arg;
878 	int		err;
879 
880 	if (rt->rt_ifp != ifp)
881 		return (0);
882 
883 	/*
884 	 * Protect (sorta) against walktree recursion problems
885 	 * with cloned routes
886 	 */
887 	if ((rt->rt_flags & RTF_UP) == 0)
888 		return (0);
889 
890 	err = rtrequest_fib(RTM_DELETE, rt_key(rt), rt->rt_gateway,
891 			rt_mask(rt),
892 			rt->rt_flags | RTF_RNH_LOCKED | RTF_PINNED,
893 			(struct rtentry **) NULL, rt->rt_fibnum);
894 	if (err != 0)
895 		log(LOG_WARNING, "rt_ifdelroute: error %d\n", err);
896 
897 	return (0);
898 }
899 
900 /*
901  * Delete all remaining routes using this interface
902  * Unfortuneatly the only way to do this is to slog through
903  * the entire routing table looking for routes which point
904  * to this interface...oh well...
905  */
906 void
907 rt_flushifroutes(struct ifnet *ifp)
908 {
909 
910 	rt_foreach_fib_walk(AF_UNSPEC, NULL, rt_ifdelroute, ifp);
911 }
912 
913 /*
914  * These (questionable) definitions of apparent local variables apply
915  * to the next two functions.  XXXXXX!!!
916  */
917 #define	dst	info->rti_info[RTAX_DST]
918 #define	gateway	info->rti_info[RTAX_GATEWAY]
919 #define	netmask	info->rti_info[RTAX_NETMASK]
920 #define	ifaaddr	info->rti_info[RTAX_IFA]
921 #define	ifpaddr	info->rti_info[RTAX_IFP]
922 #define	flags	info->rti_flags
923 
924 int
925 rt_getifa(struct rt_addrinfo *info)
926 {
927 
928 	return (rt_getifa_fib(info, RT_DEFAULT_FIB));
929 }
930 
931 /*
932  * Look up rt_addrinfo for a specific fib.  Note that if rti_ifa is defined,
933  * it will be referenced so the caller must free it.
934  */
935 int
936 rt_getifa_fib(struct rt_addrinfo *info, u_int fibnum)
937 {
938 	struct ifaddr *ifa;
939 	int error = 0;
940 
941 	/*
942 	 * ifp may be specified by sockaddr_dl
943 	 * when protocol address is ambiguous.
944 	 */
945 	if (info->rti_ifp == NULL && ifpaddr != NULL &&
946 	    ifpaddr->sa_family == AF_LINK &&
947 	    (ifa = ifa_ifwithnet(ifpaddr, 0, fibnum)) != NULL) {
948 		info->rti_ifp = ifa->ifa_ifp;
949 		ifa_free(ifa);
950 	}
951 	if (info->rti_ifa == NULL && ifaaddr != NULL)
952 		info->rti_ifa = ifa_ifwithaddr(ifaaddr);
953 	if (info->rti_ifa == NULL) {
954 		struct sockaddr *sa;
955 
956 		sa = ifaaddr != NULL ? ifaaddr :
957 		    (gateway != NULL ? gateway : dst);
958 		if (sa != NULL && info->rti_ifp != NULL)
959 			info->rti_ifa = ifaof_ifpforaddr(sa, info->rti_ifp);
960 		else if (dst != NULL && gateway != NULL)
961 			info->rti_ifa = ifa_ifwithroute(flags, dst, gateway,
962 							fibnum);
963 		else if (sa != NULL)
964 			info->rti_ifa = ifa_ifwithroute(flags, sa, sa,
965 							fibnum);
966 	}
967 	if ((ifa = info->rti_ifa) != NULL) {
968 		if (info->rti_ifp == NULL)
969 			info->rti_ifp = ifa->ifa_ifp;
970 	} else
971 		error = ENETUNREACH;
972 	return (error);
973 }
974 
975 /*
976  * Expunges references to a route that's about to be reclaimed.
977  * The route must be locked.
978  */
979 int
980 rt_expunge(struct radix_node_head *rnh, struct rtentry *rt)
981 {
982 #if !defined(RADIX_MPATH)
983 	struct radix_node *rn;
984 #else
985 	struct rt_addrinfo info;
986 	int fib;
987 	struct rtentry *rt0;
988 #endif
989 	struct ifaddr *ifa;
990 	int error = 0;
991 
992 	RT_LOCK_ASSERT(rt);
993 	RADIX_NODE_HEAD_LOCK_ASSERT(rnh);
994 
995 #ifdef RADIX_MPATH
996 	fib = rt->rt_fibnum;
997 	bzero(&info, sizeof(info));
998 	info.rti_ifp = rt->rt_ifp;
999 	info.rti_flags = RTF_RNH_LOCKED;
1000 	info.rti_info[RTAX_DST] = rt_key(rt);
1001 	info.rti_info[RTAX_GATEWAY] = rt->rt_ifa->ifa_addr;
1002 
1003 	RT_UNLOCK(rt);
1004 	error = rtrequest1_fib(RTM_DELETE, &info, &rt0, fib);
1005 
1006 	if (error == 0 && rt0 != NULL) {
1007 		rt = rt0;
1008 		RT_LOCK(rt);
1009 	} else if (error != 0) {
1010 		RT_LOCK(rt);
1011 		return (error);
1012 	}
1013 #else
1014 	/*
1015 	 * Remove the item from the tree; it should be there,
1016 	 * but when callers invoke us blindly it may not (sigh).
1017 	 */
1018 	rn = rnh->rnh_deladdr(rt_key(rt), rt_mask(rt), rnh);
1019 	if (rn == NULL) {
1020 		error = ESRCH;
1021 		goto bad;
1022 	}
1023 	KASSERT((rn->rn_flags & (RNF_ACTIVE | RNF_ROOT)) == 0,
1024 		("unexpected flags 0x%x", rn->rn_flags));
1025 	KASSERT(rt == RNTORT(rn),
1026 		("lookup mismatch, rt %p rn %p", rt, rn));
1027 #endif /* RADIX_MPATH */
1028 
1029 	rt->rt_flags &= ~RTF_UP;
1030 
1031 	/*
1032 	 * Give the protocol a chance to keep things in sync.
1033 	 */
1034 	if ((ifa = rt->rt_ifa) && ifa->ifa_rtrequest) {
1035 		struct rt_addrinfo info;
1036 
1037 		bzero((caddr_t)&info, sizeof(info));
1038 		info.rti_flags = rt->rt_flags;
1039 		info.rti_info[RTAX_DST] = rt_key(rt);
1040 		info.rti_info[RTAX_GATEWAY] = rt->rt_gateway;
1041 		info.rti_info[RTAX_NETMASK] = rt_mask(rt);
1042 		ifa->ifa_rtrequest(RTM_DELETE, rt, &info);
1043 	}
1044 
1045 	/*
1046 	 * one more rtentry floating around that is not
1047 	 * linked to the routing table.
1048 	 */
1049 	V_rttrash++;
1050 #if !defined(RADIX_MPATH)
1051 bad:
1052 #endif
1053 	return (error);
1054 }
1055 
1056 static int
1057 if_updatemtu_cb(struct radix_node *rn, void *arg)
1058 {
1059 	struct rtentry *rt;
1060 	struct if_mtuinfo *ifmtu;
1061 
1062 	rt = (struct rtentry *)rn;
1063 	ifmtu = (struct if_mtuinfo *)arg;
1064 
1065 	if (rt->rt_ifp != ifmtu->ifp)
1066 		return (0);
1067 
1068 	if (rt->rt_mtu >= ifmtu->mtu) {
1069 		/* We have to decrease mtu regardless of flags */
1070 		rt->rt_mtu = ifmtu->mtu;
1071 		return (0);
1072 	}
1073 
1074 	/*
1075 	 * New MTU is bigger. Check if are allowed to alter it
1076 	 */
1077 	if ((rt->rt_flags & (RTF_FIXEDMTU | RTF_GATEWAY | RTF_HOST)) != 0) {
1078 
1079 		/*
1080 		 * Skip routes with user-supplied MTU and
1081 		 * non-interface routes
1082 		 */
1083 		return (0);
1084 	}
1085 
1086 	/* We are safe to update route MTU */
1087 	rt->rt_mtu = ifmtu->mtu;
1088 
1089 	return (0);
1090 }
1091 
1092 void
1093 rt_updatemtu(struct ifnet *ifp)
1094 {
1095 	struct if_mtuinfo ifmtu;
1096 	struct radix_node_head *rnh;
1097 	int i, j;
1098 
1099 	ifmtu.ifp = ifp;
1100 
1101 	/*
1102 	 * Try to update rt_mtu for all routes using this interface
1103 	 * Unfortunately the only way to do this is to traverse all
1104 	 * routing tables in all fibs/domains.
1105 	 */
1106 	for (i = 1; i <= AF_MAX; i++) {
1107 		ifmtu.mtu = if_getmtu_family(ifp, i);
1108 		for (j = 0; j < rt_numfibs; j++) {
1109 			rnh = rt_tables_get_rnh(j, i);
1110 			if (rnh == NULL)
1111 				continue;
1112 			RADIX_NODE_HEAD_LOCK(rnh);
1113 			rnh->rnh_walktree(rnh, if_updatemtu_cb, &ifmtu);
1114 			RADIX_NODE_HEAD_UNLOCK(rnh);
1115 		}
1116 	}
1117 }
1118 
1119 
1120 #if 0
1121 int p_sockaddr(char *buf, int buflen, struct sockaddr *s);
1122 int rt_print(char *buf, int buflen, struct rtentry *rt);
1123 
1124 int
1125 p_sockaddr(char *buf, int buflen, struct sockaddr *s)
1126 {
1127 	void *paddr = NULL;
1128 
1129 	switch (s->sa_family) {
1130 	case AF_INET:
1131 		paddr = &((struct sockaddr_in *)s)->sin_addr;
1132 		break;
1133 	case AF_INET6:
1134 		paddr = &((struct sockaddr_in6 *)s)->sin6_addr;
1135 		break;
1136 	}
1137 
1138 	if (paddr == NULL)
1139 		return (0);
1140 
1141 	if (inet_ntop(s->sa_family, paddr, buf, buflen) == NULL)
1142 		return (0);
1143 
1144 	return (strlen(buf));
1145 }
1146 
1147 int
1148 rt_print(char *buf, int buflen, struct rtentry *rt)
1149 {
1150 	struct sockaddr *addr, *mask;
1151 	int i = 0;
1152 
1153 	addr = rt_key(rt);
1154 	mask = rt_mask(rt);
1155 
1156 	i = p_sockaddr(buf, buflen, addr);
1157 	if (!(rt->rt_flags & RTF_HOST)) {
1158 		buf[i++] = '/';
1159 		i += p_sockaddr(buf + i, buflen - i, mask);
1160 	}
1161 
1162 	if (rt->rt_flags & RTF_GATEWAY) {
1163 		buf[i++] = '>';
1164 		i += p_sockaddr(buf + i, buflen - i, rt->rt_gateway);
1165 	}
1166 
1167 	return (i);
1168 }
1169 #endif
1170 
1171 #ifdef RADIX_MPATH
1172 static int
1173 rn_mpath_update(int req, struct rt_addrinfo *info,
1174     struct radix_node_head *rnh, struct rtentry **ret_nrt)
1175 {
1176 	/*
1177 	 * if we got multipath routes, we require users to specify
1178 	 * a matching RTAX_GATEWAY.
1179 	 */
1180 	struct rtentry *rt, *rto = NULL;
1181 	struct radix_node *rn;
1182 	int error = 0;
1183 
1184 	rn = rnh->rnh_lookup(dst, netmask, rnh);
1185 	if (rn == NULL)
1186 		return (ESRCH);
1187 	rto = rt = RNTORT(rn);
1188 
1189 	rt = rt_mpath_matchgate(rt, gateway);
1190 	if (rt == NULL)
1191 		return (ESRCH);
1192 	/*
1193 	 * this is the first entry in the chain
1194 	 */
1195 	if (rto == rt) {
1196 		rn = rn_mpath_next((struct radix_node *)rt);
1197 		/*
1198 		 * there is another entry, now it's active
1199 		 */
1200 		if (rn) {
1201 			rto = RNTORT(rn);
1202 			RT_LOCK(rto);
1203 			rto->rt_flags |= RTF_UP;
1204 			RT_UNLOCK(rto);
1205 		} else if (rt->rt_flags & RTF_GATEWAY) {
1206 			/*
1207 			 * For gateway routes, we need to
1208 			 * make sure that we we are deleting
1209 			 * the correct gateway.
1210 			 * rt_mpath_matchgate() does not
1211 			 * check the case when there is only
1212 			 * one route in the chain.
1213 			 */
1214 			if (gateway &&
1215 			    (rt->rt_gateway->sa_len != gateway->sa_len ||
1216 				memcmp(rt->rt_gateway, gateway, gateway->sa_len)))
1217 				error = ESRCH;
1218 			else {
1219 				/*
1220 				 * remove from tree before returning it
1221 				 * to the caller
1222 				 */
1223 				rn = rnh->rnh_deladdr(dst, netmask, rnh);
1224 				KASSERT(rt == RNTORT(rn), ("radix node disappeared"));
1225 				goto gwdelete;
1226 			}
1227 
1228 		}
1229 		/*
1230 		 * use the normal delete code to remove
1231 		 * the first entry
1232 		 */
1233 		if (req != RTM_DELETE)
1234 			goto nondelete;
1235 
1236 		error = ENOENT;
1237 		goto done;
1238 	}
1239 
1240 	/*
1241 	 * if the entry is 2nd and on up
1242 	 */
1243 	if ((req == RTM_DELETE) && !rt_mpath_deldup(rto, rt))
1244 		panic ("rtrequest1: rt_mpath_deldup");
1245 gwdelete:
1246 	RT_LOCK(rt);
1247 	RT_ADDREF(rt);
1248 	if (req == RTM_DELETE) {
1249 		rt->rt_flags &= ~RTF_UP;
1250 		/*
1251 		 * One more rtentry floating around that is not
1252 		 * linked to the routing table. rttrash will be decremented
1253 		 * when RTFREE(rt) is eventually called.
1254 		 */
1255 		V_rttrash++;
1256 	}
1257 
1258 nondelete:
1259 	if (req != RTM_DELETE)
1260 		panic("unrecognized request %d", req);
1261 
1262 
1263 	/*
1264 	 * If the caller wants it, then it can have it,
1265 	 * but it's up to it to free the rtentry as we won't be
1266 	 * doing it.
1267 	 */
1268 	if (ret_nrt) {
1269 		*ret_nrt = rt;
1270 		RT_UNLOCK(rt);
1271 	} else
1272 		RTFREE_LOCKED(rt);
1273 done:
1274 	return (error);
1275 }
1276 #endif
1277 
1278 int
1279 rtrequest1_fib(int req, struct rt_addrinfo *info, struct rtentry **ret_nrt,
1280 				u_int fibnum)
1281 {
1282 	int error = 0, needlock = 0;
1283 	struct rtentry *rt;
1284 #ifdef FLOWTABLE
1285 	struct rtentry *rt0;
1286 #endif
1287 	struct radix_node *rn;
1288 	struct radix_node_head *rnh;
1289 	struct ifaddr *ifa;
1290 	struct sockaddr *ndst;
1291 	struct sockaddr_storage mdst;
1292 #define senderr(x) { error = x ; goto bad; }
1293 
1294 	KASSERT((fibnum < rt_numfibs), ("rtrequest1_fib: bad fibnum"));
1295 	switch (dst->sa_family) {
1296 	case AF_INET6:
1297 	case AF_INET:
1298 		/* We support multiple FIBs. */
1299 		break;
1300 	default:
1301 		fibnum = RT_DEFAULT_FIB;
1302 		break;
1303 	}
1304 
1305 	/*
1306 	 * Find the correct routing tree to use for this Address Family
1307 	 */
1308 	rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
1309 	if (rnh == NULL)
1310 		return (EAFNOSUPPORT);
1311 	needlock = ((flags & RTF_RNH_LOCKED) == 0);
1312 	flags &= ~RTF_RNH_LOCKED;
1313 	if (needlock)
1314 		RADIX_NODE_HEAD_LOCK(rnh);
1315 	else
1316 		RADIX_NODE_HEAD_LOCK_ASSERT(rnh);
1317 	/*
1318 	 * If we are adding a host route then we don't want to put
1319 	 * a netmask in the tree, nor do we want to clone it.
1320 	 */
1321 	if (flags & RTF_HOST)
1322 		netmask = NULL;
1323 
1324 	switch (req) {
1325 	case RTM_DELETE:
1326 		if (netmask) {
1327 			rt_maskedcopy(dst, (struct sockaddr *)&mdst, netmask);
1328 			dst = (struct sockaddr *)&mdst;
1329 		}
1330 #ifdef RADIX_MPATH
1331 		if (rn_mpath_capable(rnh)) {
1332 			error = rn_mpath_update(req, info, rnh, ret_nrt);
1333 			/*
1334 			 * "bad" holds true for the success case
1335 			 * as well
1336 			 */
1337 			if (error != ENOENT)
1338 				goto bad;
1339 			error = 0;
1340 		}
1341 #endif
1342 		if ((flags & RTF_PINNED) == 0) {
1343 			/* Check if target route can be deleted */
1344 			rt = (struct rtentry *)rnh->rnh_lookup(dst,
1345 			    netmask, rnh);
1346 			if ((rt != NULL) && (rt->rt_flags & RTF_PINNED))
1347 				senderr(EADDRINUSE);
1348 		}
1349 
1350 		/*
1351 		 * Remove the item from the tree and return it.
1352 		 * Complain if it is not there and do no more processing.
1353 		 */
1354 		rn = rnh->rnh_deladdr(dst, netmask, rnh);
1355 		if (rn == NULL)
1356 			senderr(ESRCH);
1357 		if (rn->rn_flags & (RNF_ACTIVE | RNF_ROOT))
1358 			panic ("rtrequest delete");
1359 		rt = RNTORT(rn);
1360 		RT_LOCK(rt);
1361 		RT_ADDREF(rt);
1362 		rt->rt_flags &= ~RTF_UP;
1363 
1364 		/*
1365 		 * give the protocol a chance to keep things in sync.
1366 		 */
1367 		if ((ifa = rt->rt_ifa) && ifa->ifa_rtrequest)
1368 			ifa->ifa_rtrequest(RTM_DELETE, rt, info);
1369 
1370 		/*
1371 		 * One more rtentry floating around that is not
1372 		 * linked to the routing table. rttrash will be decremented
1373 		 * when RTFREE(rt) is eventually called.
1374 		 */
1375 		V_rttrash++;
1376 
1377 		/*
1378 		 * If the caller wants it, then it can have it,
1379 		 * but it's up to it to free the rtentry as we won't be
1380 		 * doing it.
1381 		 */
1382 		if (ret_nrt) {
1383 			*ret_nrt = rt;
1384 			RT_UNLOCK(rt);
1385 		} else
1386 			RTFREE_LOCKED(rt);
1387 		break;
1388 	case RTM_RESOLVE:
1389 		/*
1390 		 * resolve was only used for route cloning
1391 		 * here for compat
1392 		 */
1393 		break;
1394 	case RTM_ADD:
1395 		if ((flags & RTF_GATEWAY) && !gateway)
1396 			senderr(EINVAL);
1397 		if (dst && gateway && (dst->sa_family != gateway->sa_family) &&
1398 		    (gateway->sa_family != AF_UNSPEC) && (gateway->sa_family != AF_LINK))
1399 			senderr(EINVAL);
1400 
1401 		if (info->rti_ifa == NULL) {
1402 			error = rt_getifa_fib(info, fibnum);
1403 			if (error)
1404 				senderr(error);
1405 		} else
1406 			ifa_ref(info->rti_ifa);
1407 		ifa = info->rti_ifa;
1408 		rt = uma_zalloc(V_rtzone, M_NOWAIT);
1409 		if (rt == NULL) {
1410 			ifa_free(ifa);
1411 			senderr(ENOBUFS);
1412 		}
1413 		rt->rt_flags = RTF_UP | flags;
1414 		rt->rt_fibnum = fibnum;
1415 		/*
1416 		 * Add the gateway. Possibly re-malloc-ing the storage for it.
1417 		 */
1418 		RT_LOCK(rt);
1419 		if ((error = rt_setgate(rt, dst, gateway)) != 0) {
1420 			ifa_free(ifa);
1421 			uma_zfree(V_rtzone, rt);
1422 			senderr(error);
1423 		}
1424 
1425 		/*
1426 		 * point to the (possibly newly malloc'd) dest address.
1427 		 */
1428 		ndst = (struct sockaddr *)rt_key(rt);
1429 
1430 		/*
1431 		 * make sure it contains the value we want (masked if needed).
1432 		 */
1433 		if (netmask) {
1434 			rt_maskedcopy(dst, ndst, netmask);
1435 		} else
1436 			bcopy(dst, ndst, dst->sa_len);
1437 
1438 		/*
1439 		 * We use the ifa reference returned by rt_getifa_fib().
1440 		 * This moved from below so that rnh->rnh_addaddr() can
1441 		 * examine the ifa and  ifa->ifa_ifp if it so desires.
1442 		 */
1443 		rt->rt_ifa = ifa;
1444 		rt->rt_ifp = ifa->ifa_ifp;
1445 		rt->rt_weight = 1;
1446 
1447 		rt_setmetrics(info, rt);
1448 
1449 #ifdef RADIX_MPATH
1450 		/* do not permit exactly the same dst/mask/gw pair */
1451 		if (rn_mpath_capable(rnh) &&
1452 			rt_mpath_conflict(rnh, rt, netmask)) {
1453 			ifa_free(rt->rt_ifa);
1454 			R_Free(rt_key(rt));
1455 			uma_zfree(V_rtzone, rt);
1456 			senderr(EEXIST);
1457 		}
1458 #endif
1459 
1460 #ifdef FLOWTABLE
1461 		rt0 = NULL;
1462 		/* "flow-table" only supports IPv6 and IPv4 at the moment. */
1463 		switch (dst->sa_family) {
1464 #ifdef INET6
1465 		case AF_INET6:
1466 #endif
1467 #ifdef INET
1468 		case AF_INET:
1469 #endif
1470 #if defined(INET6) || defined(INET)
1471 			rn = rnh->rnh_matchaddr(dst, rnh);
1472 			if (rn && ((rn->rn_flags & RNF_ROOT) == 0)) {
1473 				struct sockaddr *mask;
1474 				u_char *m, *n;
1475 				int len;
1476 
1477 				/*
1478 				 * compare mask to see if the new route is
1479 				 * more specific than the existing one
1480 				 */
1481 				rt0 = RNTORT(rn);
1482 				RT_LOCK(rt0);
1483 				RT_ADDREF(rt0);
1484 				RT_UNLOCK(rt0);
1485 				/*
1486 				 * A host route is already present, so
1487 				 * leave the flow-table entries as is.
1488 				 */
1489 				if (rt0->rt_flags & RTF_HOST) {
1490 					RTFREE(rt0);
1491 					rt0 = NULL;
1492 				} else if (!(flags & RTF_HOST) && netmask) {
1493 					mask = rt_mask(rt0);
1494 					len = mask->sa_len;
1495 					m = (u_char *)mask;
1496 					n = (u_char *)netmask;
1497 					while (len-- > 0) {
1498 						if (*n != *m)
1499 							break;
1500 						n++;
1501 						m++;
1502 					}
1503 					if (len == 0 || (*n < *m)) {
1504 						RTFREE(rt0);
1505 						rt0 = NULL;
1506 					}
1507 				}
1508 			}
1509 #endif/* INET6 || INET */
1510 		}
1511 #endif /* FLOWTABLE */
1512 
1513 		/* XXX mtu manipulation will be done in rnh_addaddr -- itojun */
1514 		rn = rnh->rnh_addaddr(ndst, netmask, rnh, rt->rt_nodes);
1515 		/*
1516 		 * If it still failed to go into the tree,
1517 		 * then un-make it (this should be a function)
1518 		 */
1519 		if (rn == NULL) {
1520 			ifa_free(rt->rt_ifa);
1521 			R_Free(rt_key(rt));
1522 			uma_zfree(V_rtzone, rt);
1523 #ifdef FLOWTABLE
1524 			if (rt0 != NULL)
1525 				RTFREE(rt0);
1526 #endif
1527 			senderr(EEXIST);
1528 		}
1529 #ifdef FLOWTABLE
1530 		else if (rt0 != NULL) {
1531 			flowtable_route_flush(dst->sa_family, rt0);
1532 			RTFREE(rt0);
1533 		}
1534 #endif
1535 
1536 		/*
1537 		 * If this protocol has something to add to this then
1538 		 * allow it to do that as well.
1539 		 */
1540 		if (ifa->ifa_rtrequest)
1541 			ifa->ifa_rtrequest(req, rt, info);
1542 
1543 		/*
1544 		 * actually return a resultant rtentry and
1545 		 * give the caller a single reference.
1546 		 */
1547 		if (ret_nrt) {
1548 			*ret_nrt = rt;
1549 			RT_ADDREF(rt);
1550 		}
1551 		RT_UNLOCK(rt);
1552 		break;
1553 	case RTM_CHANGE:
1554 		error = rtrequest1_fib_change(rnh, info, ret_nrt, fibnum);
1555 		break;
1556 	default:
1557 		error = EOPNOTSUPP;
1558 	}
1559 bad:
1560 	if (needlock)
1561 		RADIX_NODE_HEAD_UNLOCK(rnh);
1562 	return (error);
1563 #undef senderr
1564 }
1565 
1566 #undef dst
1567 #undef gateway
1568 #undef netmask
1569 #undef ifaaddr
1570 #undef ifpaddr
1571 #undef flags
1572 
1573 static int
1574 rtrequest1_fib_change(struct radix_node_head *rnh, struct rt_addrinfo *info,
1575     struct rtentry **ret_nrt, u_int fibnum)
1576 {
1577 	struct rtentry *rt = NULL;
1578 	int error = 0;
1579 	int free_ifa = 0;
1580 	int family, mtu;
1581 	struct if_mtuinfo ifmtu;
1582 
1583 	rt = (struct rtentry *)rnh->rnh_lookup(info->rti_info[RTAX_DST],
1584 	    info->rti_info[RTAX_NETMASK], rnh);
1585 
1586 	if (rt == NULL)
1587 		return (ESRCH);
1588 
1589 #ifdef RADIX_MPATH
1590 	/*
1591 	 * If we got multipath routes,
1592 	 * we require users to specify a matching RTAX_GATEWAY.
1593 	 */
1594 	if (rn_mpath_capable(rnh)) {
1595 		rt = rt_mpath_matchgate(rt, info->rti_info[RTAX_GATEWAY]);
1596 		if (rt == NULL)
1597 			return (ESRCH);
1598 	}
1599 #endif
1600 
1601 	RT_LOCK(rt);
1602 
1603 	rt_setmetrics(info, rt);
1604 
1605 	/*
1606 	 * New gateway could require new ifaddr, ifp;
1607 	 * flags may also be different; ifp may be specified
1608 	 * by ll sockaddr when protocol address is ambiguous
1609 	 */
1610 	if (((rt->rt_flags & RTF_GATEWAY) &&
1611 	    info->rti_info[RTAX_GATEWAY] != NULL) ||
1612 	    info->rti_info[RTAX_IFP] != NULL ||
1613 	    (info->rti_info[RTAX_IFA] != NULL &&
1614 	     !sa_equal(info->rti_info[RTAX_IFA], rt->rt_ifa->ifa_addr))) {
1615 
1616 		error = rt_getifa_fib(info, fibnum);
1617 		if (info->rti_ifa != NULL)
1618 			free_ifa = 1;
1619 
1620 		if (error != 0)
1621 			goto bad;
1622 	}
1623 
1624 	/* Check if outgoing interface has changed */
1625 	if (info->rti_ifa != NULL && info->rti_ifa != rt->rt_ifa &&
1626 	    rt->rt_ifa != NULL && rt->rt_ifa->ifa_rtrequest != NULL) {
1627 		rt->rt_ifa->ifa_rtrequest(RTM_DELETE, rt, info);
1628 		ifa_free(rt->rt_ifa);
1629 	}
1630 	/* Update gateway address */
1631 	if (info->rti_info[RTAX_GATEWAY] != NULL) {
1632 		error = rt_setgate(rt, rt_key(rt), info->rti_info[RTAX_GATEWAY]);
1633 		if (error != 0)
1634 			goto bad;
1635 
1636 		rt->rt_flags &= ~RTF_GATEWAY;
1637 		rt->rt_flags |= (RTF_GATEWAY & info->rti_flags);
1638 	}
1639 
1640 	if (info->rti_ifa != NULL && info->rti_ifa != rt->rt_ifa) {
1641 		ifa_ref(info->rti_ifa);
1642 		rt->rt_ifa = info->rti_ifa;
1643 		rt->rt_ifp = info->rti_ifp;
1644 	}
1645 	/* Allow some flags to be toggled on change. */
1646 	rt->rt_flags &= ~RTF_FMASK;
1647 	rt->rt_flags |= info->rti_flags & RTF_FMASK;
1648 
1649 	if (rt->rt_ifa && rt->rt_ifa->ifa_rtrequest != NULL)
1650 	       rt->rt_ifa->ifa_rtrequest(RTM_ADD, rt, info);
1651 
1652 	/* Alter route MTU if necessary */
1653 	if (rt->rt_ifp != NULL) {
1654 		family = info->rti_info[RTAX_DST]->sa_family;
1655 		mtu = if_getmtu_family(rt->rt_ifp, family);
1656 		/* Set default MTU */
1657 		if (rt->rt_mtu == 0)
1658 			rt->rt_mtu = mtu;
1659 		if (rt->rt_mtu != mtu) {
1660 			/* Check if we really need to update */
1661 			ifmtu.ifp = rt->rt_ifp;
1662 			ifmtu.mtu = mtu;
1663 			if_updatemtu_cb(rt->rt_nodes, &ifmtu);
1664 		}
1665 	}
1666 
1667 	if (ret_nrt) {
1668 		*ret_nrt = rt;
1669 		RT_ADDREF(rt);
1670 	}
1671 bad:
1672 	RT_UNLOCK(rt);
1673 	if (free_ifa != 0)
1674 		ifa_free(info->rti_ifa);
1675 	return (error);
1676 }
1677 
1678 static void
1679 rt_setmetrics(const struct rt_addrinfo *info, struct rtentry *rt)
1680 {
1681 
1682 	if (info->rti_mflags & RTV_MTU) {
1683 		if (info->rti_rmx->rmx_mtu != 0) {
1684 
1685 			/*
1686 			 * MTU was explicitly provided by user.
1687 			 * Keep it.
1688 			 */
1689 			rt->rt_flags |= RTF_FIXEDMTU;
1690 		} else {
1691 
1692 			/*
1693 			 * User explicitly sets MTU to 0.
1694 			 * Assume rollback to default.
1695 			 */
1696 			rt->rt_flags &= ~RTF_FIXEDMTU;
1697 		}
1698 		rt->rt_mtu = info->rti_rmx->rmx_mtu;
1699 	}
1700 	if (info->rti_mflags & RTV_WEIGHT)
1701 		rt->rt_weight = info->rti_rmx->rmx_weight;
1702 	/* Kernel -> userland timebase conversion. */
1703 	if (info->rti_mflags & RTV_EXPIRE)
1704 		rt->rt_expire = info->rti_rmx->rmx_expire ?
1705 		    info->rti_rmx->rmx_expire - time_second + time_uptime : 0;
1706 }
1707 
1708 int
1709 rt_setgate(struct rtentry *rt, struct sockaddr *dst, struct sockaddr *gate)
1710 {
1711 	/* XXX dst may be overwritten, can we move this to below */
1712 	int dlen = SA_SIZE(dst), glen = SA_SIZE(gate);
1713 #ifdef INVARIANTS
1714 	struct radix_node_head *rnh;
1715 
1716 	rnh = rt_tables_get_rnh(rt->rt_fibnum, dst->sa_family);
1717 #endif
1718 
1719 	RT_LOCK_ASSERT(rt);
1720 	RADIX_NODE_HEAD_LOCK_ASSERT(rnh);
1721 
1722 	/*
1723 	 * Prepare to store the gateway in rt->rt_gateway.
1724 	 * Both dst and gateway are stored one after the other in the same
1725 	 * malloc'd chunk. If we have room, we can reuse the old buffer,
1726 	 * rt_gateway already points to the right place.
1727 	 * Otherwise, malloc a new block and update the 'dst' address.
1728 	 */
1729 	if (rt->rt_gateway == NULL || glen > SA_SIZE(rt->rt_gateway)) {
1730 		caddr_t new;
1731 
1732 		R_Malloc(new, caddr_t, dlen + glen);
1733 		if (new == NULL)
1734 			return ENOBUFS;
1735 		/*
1736 		 * XXX note, we copy from *dst and not *rt_key(rt) because
1737 		 * rt_setgate() can be called to initialize a newly
1738 		 * allocated route entry, in which case rt_key(rt) == NULL
1739 		 * (and also rt->rt_gateway == NULL).
1740 		 * Free()/free() handle a NULL argument just fine.
1741 		 */
1742 		bcopy(dst, new, dlen);
1743 		R_Free(rt_key(rt));	/* free old block, if any */
1744 		rt_key(rt) = (struct sockaddr *)new;
1745 		rt->rt_gateway = (struct sockaddr *)(new + dlen);
1746 	}
1747 
1748 	/*
1749 	 * Copy the new gateway value into the memory chunk.
1750 	 */
1751 	bcopy(gate, rt->rt_gateway, glen);
1752 
1753 	return (0);
1754 }
1755 
1756 void
1757 rt_maskedcopy(struct sockaddr *src, struct sockaddr *dst, struct sockaddr *netmask)
1758 {
1759 	u_char *cp1 = (u_char *)src;
1760 	u_char *cp2 = (u_char *)dst;
1761 	u_char *cp3 = (u_char *)netmask;
1762 	u_char *cplim = cp2 + *cp3;
1763 	u_char *cplim2 = cp2 + *cp1;
1764 
1765 	*cp2++ = *cp1++; *cp2++ = *cp1++; /* copies sa_len & sa_family */
1766 	cp3 += 2;
1767 	if (cplim > cplim2)
1768 		cplim = cplim2;
1769 	while (cp2 < cplim)
1770 		*cp2++ = *cp1++ & *cp3++;
1771 	if (cp2 < cplim2)
1772 		bzero((caddr_t)cp2, (unsigned)(cplim2 - cp2));
1773 }
1774 
1775 /*
1776  * Set up a routing table entry, normally
1777  * for an interface.
1778  */
1779 #define _SOCKADDR_TMPSIZE 128 /* Not too big.. kernel stack size is limited */
1780 static inline  int
1781 rtinit1(struct ifaddr *ifa, int cmd, int flags, int fibnum)
1782 {
1783 	struct sockaddr *dst;
1784 	struct sockaddr *netmask;
1785 	struct rtentry *rt = NULL;
1786 	struct rt_addrinfo info;
1787 	int error = 0;
1788 	int startfib, endfib;
1789 	char tempbuf[_SOCKADDR_TMPSIZE];
1790 	int didwork = 0;
1791 	int a_failure = 0;
1792 	static struct sockaddr_dl null_sdl = {sizeof(null_sdl), AF_LINK};
1793 	struct radix_node_head *rnh;
1794 
1795 	if (flags & RTF_HOST) {
1796 		dst = ifa->ifa_dstaddr;
1797 		netmask = NULL;
1798 	} else {
1799 		dst = ifa->ifa_addr;
1800 		netmask = ifa->ifa_netmask;
1801 	}
1802 	if (dst->sa_len == 0)
1803 		return(EINVAL);
1804 	switch (dst->sa_family) {
1805 	case AF_INET6:
1806 	case AF_INET:
1807 		/* We support multiple FIBs. */
1808 		break;
1809 	default:
1810 		fibnum = RT_DEFAULT_FIB;
1811 		break;
1812 	}
1813 	if (fibnum == RT_ALL_FIBS) {
1814 		if (V_rt_add_addr_allfibs == 0 && cmd == (int)RTM_ADD)
1815 			startfib = endfib = ifa->ifa_ifp->if_fib;
1816 		else {
1817 			startfib = 0;
1818 			endfib = rt_numfibs - 1;
1819 		}
1820 	} else {
1821 		KASSERT((fibnum < rt_numfibs), ("rtinit1: bad fibnum"));
1822 		startfib = fibnum;
1823 		endfib = fibnum;
1824 	}
1825 
1826 	/*
1827 	 * If it's a delete, check that if it exists,
1828 	 * it's on the correct interface or we might scrub
1829 	 * a route to another ifa which would
1830 	 * be confusing at best and possibly worse.
1831 	 */
1832 	if (cmd == RTM_DELETE) {
1833 		/*
1834 		 * It's a delete, so it should already exist..
1835 		 * If it's a net, mask off the host bits
1836 		 * (Assuming we have a mask)
1837 		 * XXX this is kinda inet specific..
1838 		 */
1839 		if (netmask != NULL) {
1840 			rt_maskedcopy(dst, (struct sockaddr *)tempbuf, netmask);
1841 			dst = (struct sockaddr *)tempbuf;
1842 		}
1843 	}
1844 	/*
1845 	 * Now go through all the requested tables (fibs) and do the
1846 	 * requested action. Realistically, this will either be fib 0
1847 	 * for protocols that don't do multiple tables or all the
1848 	 * tables for those that do.
1849 	 */
1850 	for ( fibnum = startfib; fibnum <= endfib; fibnum++) {
1851 		if (cmd == RTM_DELETE) {
1852 			struct radix_node *rn;
1853 			/*
1854 			 * Look up an rtentry that is in the routing tree and
1855 			 * contains the correct info.
1856 			 */
1857 			rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
1858 			if (rnh == NULL)
1859 				/* this table doesn't exist but others might */
1860 				continue;
1861 			RADIX_NODE_HEAD_RLOCK(rnh);
1862 			rn = rnh->rnh_lookup(dst, netmask, rnh);
1863 #ifdef RADIX_MPATH
1864 			if (rn_mpath_capable(rnh)) {
1865 
1866 				if (rn == NULL)
1867 					error = ESRCH;
1868 				else {
1869 					rt = RNTORT(rn);
1870 					/*
1871 					 * for interface route the
1872 					 * rt->rt_gateway is sockaddr_intf
1873 					 * for cloning ARP entries, so
1874 					 * rt_mpath_matchgate must use the
1875 					 * interface address
1876 					 */
1877 					rt = rt_mpath_matchgate(rt,
1878 					    ifa->ifa_addr);
1879 					if (rt == NULL)
1880 						error = ESRCH;
1881 				}
1882 			}
1883 #endif
1884 			error = (rn == NULL ||
1885 			    (rn->rn_flags & RNF_ROOT) ||
1886 			    RNTORT(rn)->rt_ifa != ifa);
1887 			RADIX_NODE_HEAD_RUNLOCK(rnh);
1888 			if (error) {
1889 				/* this is only an error if bad on ALL tables */
1890 				continue;
1891 			}
1892 		}
1893 		/*
1894 		 * Do the actual request
1895 		 */
1896 		bzero((caddr_t)&info, sizeof(info));
1897 		info.rti_ifa = ifa;
1898 		info.rti_flags = flags |
1899 		    (ifa->ifa_flags & ~IFA_RTSELF) | RTF_PINNED;
1900 		info.rti_info[RTAX_DST] = dst;
1901 		/*
1902 		 * doing this for compatibility reasons
1903 		 */
1904 		if (cmd == RTM_ADD)
1905 			info.rti_info[RTAX_GATEWAY] =
1906 			    (struct sockaddr *)&null_sdl;
1907 		else
1908 			info.rti_info[RTAX_GATEWAY] = ifa->ifa_addr;
1909 		info.rti_info[RTAX_NETMASK] = netmask;
1910 		error = rtrequest1_fib(cmd, &info, &rt, fibnum);
1911 
1912 		if ((error == EEXIST) && (cmd == RTM_ADD)) {
1913 			/*
1914 			 * Interface route addition failed.
1915 			 * Atomically delete current prefix generating
1916 			 * RTM_DELETE message, and retry adding
1917 			 * interface prefix.
1918 			 */
1919 			rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
1920 			RADIX_NODE_HEAD_LOCK(rnh);
1921 
1922 			/* Delete old prefix */
1923 			info.rti_ifa = NULL;
1924 			info.rti_flags = RTF_RNH_LOCKED;
1925 
1926 			error = rtrequest1_fib(RTM_DELETE, &info, NULL, fibnum);
1927 			if (error == 0) {
1928 				info.rti_ifa = ifa;
1929 				info.rti_flags = flags | RTF_RNH_LOCKED |
1930 				    (ifa->ifa_flags & ~IFA_RTSELF) | RTF_PINNED;
1931 				error = rtrequest1_fib(cmd, &info, &rt, fibnum);
1932 			}
1933 
1934 			RADIX_NODE_HEAD_UNLOCK(rnh);
1935 		}
1936 
1937 
1938 		if (error == 0 && rt != NULL) {
1939 			/*
1940 			 * notify any listening routing agents of the change
1941 			 */
1942 			RT_LOCK(rt);
1943 #ifdef RADIX_MPATH
1944 			/*
1945 			 * in case address alias finds the first address
1946 			 * e.g. ifconfig bge0 192.0.2.246/24
1947 			 * e.g. ifconfig bge0 192.0.2.247/24
1948 			 * the address set in the route is 192.0.2.246
1949 			 * so we need to replace it with 192.0.2.247
1950 			 */
1951 			if (memcmp(rt->rt_ifa->ifa_addr,
1952 			    ifa->ifa_addr, ifa->ifa_addr->sa_len)) {
1953 				ifa_free(rt->rt_ifa);
1954 				ifa_ref(ifa);
1955 				rt->rt_ifp = ifa->ifa_ifp;
1956 				rt->rt_ifa = ifa;
1957 			}
1958 #endif
1959 			/*
1960 			 * doing this for compatibility reasons
1961 			 */
1962 			if (cmd == RTM_ADD) {
1963 			    ((struct sockaddr_dl *)rt->rt_gateway)->sdl_type  =
1964 				rt->rt_ifp->if_type;
1965 			    ((struct sockaddr_dl *)rt->rt_gateway)->sdl_index =
1966 				rt->rt_ifp->if_index;
1967 			}
1968 			RT_ADDREF(rt);
1969 			RT_UNLOCK(rt);
1970 			rt_newaddrmsg_fib(cmd, ifa, error, rt, fibnum);
1971 			RT_LOCK(rt);
1972 			RT_REMREF(rt);
1973 			if (cmd == RTM_DELETE) {
1974 				/*
1975 				 * If we are deleting, and we found an entry,
1976 				 * then it's been removed from the tree..
1977 				 * now throw it away.
1978 				 */
1979 				RTFREE_LOCKED(rt);
1980 			} else {
1981 				if (cmd == RTM_ADD) {
1982 					/*
1983 					 * We just wanted to add it..
1984 					 * we don't actually need a reference.
1985 					 */
1986 					RT_REMREF(rt);
1987 				}
1988 				RT_UNLOCK(rt);
1989 			}
1990 			didwork = 1;
1991 		}
1992 		if (error)
1993 			a_failure = error;
1994 	}
1995 	if (cmd == RTM_DELETE) {
1996 		if (didwork) {
1997 			error = 0;
1998 		} else {
1999 			/* we only give an error if it wasn't in any table */
2000 			error = ((flags & RTF_HOST) ?
2001 			    EHOSTUNREACH : ENETUNREACH);
2002 		}
2003 	} else {
2004 		if (a_failure) {
2005 			/* return an error if any of them failed */
2006 			error = a_failure;
2007 		}
2008 	}
2009 	return (error);
2010 }
2011 
2012 /*
2013  * Set up a routing table entry, normally
2014  * for an interface.
2015  */
2016 int
2017 rtinit(struct ifaddr *ifa, int cmd, int flags)
2018 {
2019 	struct sockaddr *dst;
2020 	int fib = RT_DEFAULT_FIB;
2021 
2022 	if (flags & RTF_HOST) {
2023 		dst = ifa->ifa_dstaddr;
2024 	} else {
2025 		dst = ifa->ifa_addr;
2026 	}
2027 
2028 	switch (dst->sa_family) {
2029 	case AF_INET6:
2030 	case AF_INET:
2031 		/* We do support multiple FIBs. */
2032 		fib = RT_ALL_FIBS;
2033 		break;
2034 	}
2035 	return (rtinit1(ifa, cmd, flags, fib));
2036 }
2037 
2038 /*
2039  * Announce interface address arrival/withdraw
2040  * Returns 0 on success.
2041  */
2042 int
2043 rt_addrmsg(int cmd, struct ifaddr *ifa, int fibnum)
2044 {
2045 
2046 	KASSERT(cmd == RTM_ADD || cmd == RTM_DELETE,
2047 	    ("unexpected cmd %d", cmd));
2048 
2049 	KASSERT(fibnum == RT_ALL_FIBS || (fibnum >= 0 && fibnum < rt_numfibs),
2050 	    ("%s: fib out of range 0 <=%d<%d", __func__, fibnum, rt_numfibs));
2051 
2052 #if defined(INET) || defined(INET6)
2053 #ifdef SCTP
2054 	/*
2055 	 * notify the SCTP stack
2056 	 * this will only get called when an address is added/deleted
2057 	 * XXX pass the ifaddr struct instead if ifa->ifa_addr...
2058 	 */
2059 	sctp_addr_change(ifa, cmd);
2060 #endif /* SCTP */
2061 #endif
2062 	return (rtsock_addrmsg(cmd, ifa, fibnum));
2063 }
2064 
2065 /*
2066  * Announce route addition/removal.
2067  * Users of this function MUST validate input data BEFORE calling.
2068  * However we have to be able to handle invalid data:
2069  * if some userland app sends us "invalid" route message (invalid mask,
2070  * no dst, wrong address families, etc...) we need to pass it back
2071  * to app (and any other rtsock consumers) with rtm_errno field set to
2072  * non-zero value.
2073  * Returns 0 on success.
2074  */
2075 int
2076 rt_routemsg(int cmd, struct ifnet *ifp, int error, struct rtentry *rt,
2077     int fibnum)
2078 {
2079 
2080 	KASSERT(cmd == RTM_ADD || cmd == RTM_DELETE,
2081 	    ("unexpected cmd %d", cmd));
2082 
2083 	KASSERT(fibnum == RT_ALL_FIBS || (fibnum >= 0 && fibnum < rt_numfibs),
2084 	    ("%s: fib out of range 0 <=%d<%d", __func__, fibnum, rt_numfibs));
2085 
2086 	KASSERT(rt_key(rt) != NULL, (":%s: rt_key must be supplied", __func__));
2087 
2088 	return (rtsock_routemsg(cmd, ifp, error, rt, fibnum));
2089 }
2090 
2091 void
2092 rt_newaddrmsg(int cmd, struct ifaddr *ifa, int error, struct rtentry *rt)
2093 {
2094 
2095 	rt_newaddrmsg_fib(cmd, ifa, error, rt, RT_ALL_FIBS);
2096 }
2097 
2098 /*
2099  * This is called to generate messages from the routing socket
2100  * indicating a network interface has had addresses associated with it.
2101  */
2102 void
2103 rt_newaddrmsg_fib(int cmd, struct ifaddr *ifa, int error, struct rtentry *rt,
2104     int fibnum)
2105 {
2106 
2107 	KASSERT(cmd == RTM_ADD || cmd == RTM_DELETE,
2108 		("unexpected cmd %u", cmd));
2109 	KASSERT(fibnum == RT_ALL_FIBS || (fibnum >= 0 && fibnum < rt_numfibs),
2110 	    ("%s: fib out of range 0 <=%d<%d", __func__, fibnum, rt_numfibs));
2111 
2112 	if (cmd == RTM_ADD) {
2113 		rt_addrmsg(cmd, ifa, fibnum);
2114 		if (rt != NULL)
2115 			rt_routemsg(cmd, ifa->ifa_ifp, error, rt, fibnum);
2116 	} else {
2117 		if (rt != NULL)
2118 			rt_routemsg(cmd, ifa->ifa_ifp, error, rt, fibnum);
2119 		rt_addrmsg(cmd, ifa, fibnum);
2120 	}
2121 }
2122 
2123