xref: /freebsd/sys/net/route.c (revision 868a59a7e296bed53dd7b986e79964eed8a59f71)
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(const struct rtentry *rt, void *arg);
143 static struct rtentry *rt_unlinkrte(struct radix_node_head *rnh,
144     struct rt_addrinfo *info, int *perror);
145 static void rt_notifydelete(struct rtentry *rt, struct rt_addrinfo *info);
146 #ifdef RADIX_MPATH
147 static struct radix_node *rt_mpath_unlink(struct radix_node_head *rnh,
148     struct rt_addrinfo *info, struct rtentry *rto, int *perror);
149 #endif
150 static int rt_exportinfo(struct rtentry *rt, struct rt_addrinfo *info,
151     int flags);
152 
153 struct if_mtuinfo
154 {
155 	struct ifnet	*ifp;
156 	int		mtu;
157 };
158 
159 static int	if_updatemtu_cb(struct radix_node *, void *);
160 
161 /*
162  * handler for net.my_fibnum
163  */
164 static int
165 sysctl_my_fibnum(SYSCTL_HANDLER_ARGS)
166 {
167         int fibnum;
168         int error;
169 
170         fibnum = curthread->td_proc->p_fibnum;
171         error = sysctl_handle_int(oidp, &fibnum, 0, req);
172         return (error);
173 }
174 
175 SYSCTL_PROC(_net, OID_AUTO, my_fibnum, CTLTYPE_INT|CTLFLAG_RD,
176             NULL, 0, &sysctl_my_fibnum, "I", "default FIB of caller");
177 
178 static __inline struct radix_node_head **
179 rt_tables_get_rnh_ptr(int table, int fam)
180 {
181 	struct radix_node_head **rnh;
182 
183 	KASSERT(table >= 0 && table < rt_numfibs, ("%s: table out of bounds.",
184 	    __func__));
185 	KASSERT(fam >= 0 && fam < (AF_MAX+1), ("%s: fam out of bounds.",
186 	    __func__));
187 
188 	/* rnh is [fib=0][af=0]. */
189 	rnh = (struct radix_node_head **)V_rt_tables;
190 	/* Get the offset to the requested table and fam. */
191 	rnh += table * (AF_MAX+1) + fam;
192 
193 	return (rnh);
194 }
195 
196 struct radix_node_head *
197 rt_tables_get_rnh(int table, int fam)
198 {
199 
200 	return (*rt_tables_get_rnh_ptr(table, fam));
201 }
202 
203 /*
204  * route initialization must occur before ip6_init2(), which happenas at
205  * SI_ORDER_MIDDLE.
206  */
207 static void
208 route_init(void)
209 {
210 
211 	/* whack the tunable ints into  line. */
212 	if (rt_numfibs > RT_MAXFIBS)
213 		rt_numfibs = RT_MAXFIBS;
214 	if (rt_numfibs == 0)
215 		rt_numfibs = 1;
216 }
217 SYSINIT(route_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD, route_init, 0);
218 
219 static int
220 rtentry_zinit(void *mem, int size, int how)
221 {
222 	struct rtentry *rt = mem;
223 
224 	rt->rt_pksent = counter_u64_alloc(how);
225 	if (rt->rt_pksent == NULL)
226 		return (ENOMEM);
227 
228 	RT_LOCK_INIT(rt);
229 
230 	return (0);
231 }
232 
233 static void
234 rtentry_zfini(void *mem, int size)
235 {
236 	struct rtentry *rt = mem;
237 
238 	RT_LOCK_DESTROY(rt);
239 	counter_u64_free(rt->rt_pksent);
240 }
241 
242 static int
243 rtentry_ctor(void *mem, int size, void *arg, int how)
244 {
245 	struct rtentry *rt = mem;
246 
247 	bzero(rt, offsetof(struct rtentry, rt_endzero));
248 	counter_u64_zero(rt->rt_pksent);
249 	rt->rt_chain = NULL;
250 
251 	return (0);
252 }
253 
254 static void
255 rtentry_dtor(void *mem, int size, void *arg)
256 {
257 	struct rtentry *rt = mem;
258 
259 	RT_UNLOCK_COND(rt);
260 }
261 
262 static void
263 vnet_route_init(const void *unused __unused)
264 {
265 	struct domain *dom;
266 	struct radix_node_head **rnh;
267 	int table;
268 	int fam;
269 
270 	V_rt_tables = malloc(rt_numfibs * (AF_MAX+1) *
271 	    sizeof(struct radix_node_head *), M_RTABLE, M_WAITOK|M_ZERO);
272 
273 	V_rtzone = uma_zcreate("rtentry", sizeof(struct rtentry),
274 	    rtentry_ctor, rtentry_dtor,
275 	    rtentry_zinit, rtentry_zfini, UMA_ALIGN_PTR, 0);
276 	for (dom = domains; dom; dom = dom->dom_next) {
277 		if (dom->dom_rtattach == NULL)
278 			continue;
279 
280 		for  (table = 0; table < rt_numfibs; table++) {
281 			fam = dom->dom_family;
282 			if (table != 0 && fam != AF_INET6 && fam != AF_INET)
283 				break;
284 
285 			rnh = rt_tables_get_rnh_ptr(table, fam);
286 			if (rnh == NULL)
287 				panic("%s: rnh NULL", __func__);
288 			dom->dom_rtattach((void **)rnh, 0);
289 		}
290 	}
291 }
292 VNET_SYSINIT(vnet_route_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH,
293     vnet_route_init, 0);
294 
295 #ifdef VIMAGE
296 static void
297 vnet_route_uninit(const void *unused __unused)
298 {
299 	int table;
300 	int fam;
301 	struct domain *dom;
302 	struct radix_node_head **rnh;
303 
304 	for (dom = domains; dom; dom = dom->dom_next) {
305 		if (dom->dom_rtdetach == NULL)
306 			continue;
307 
308 		for (table = 0; table < rt_numfibs; table++) {
309 			fam = dom->dom_family;
310 
311 			if (table != 0 && fam != AF_INET6 && fam != AF_INET)
312 				break;
313 
314 			rnh = rt_tables_get_rnh_ptr(table, fam);
315 			if (rnh == NULL)
316 				panic("%s: rnh NULL", __func__);
317 			dom->dom_rtdetach((void **)rnh, 0);
318 		}
319 	}
320 
321 	free(V_rt_tables, M_RTABLE);
322 	uma_zdestroy(V_rtzone);
323 }
324 VNET_SYSUNINIT(vnet_route_uninit, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD,
325     vnet_route_uninit, 0);
326 #endif
327 
328 #ifndef _SYS_SYSPROTO_H_
329 struct setfib_args {
330 	int     fibnum;
331 };
332 #endif
333 int
334 sys_setfib(struct thread *td, struct setfib_args *uap)
335 {
336 	if (uap->fibnum < 0 || uap->fibnum >= rt_numfibs)
337 		return EINVAL;
338 	td->td_proc->p_fibnum = uap->fibnum;
339 	return (0);
340 }
341 
342 /*
343  * Packet routing routines.
344  */
345 void
346 rtalloc(struct route *ro)
347 {
348 
349 	rtalloc_ign_fib(ro, 0UL, RT_DEFAULT_FIB);
350 }
351 
352 void
353 rtalloc_fib(struct route *ro, u_int fibnum)
354 {
355 	rtalloc_ign_fib(ro, 0UL, fibnum);
356 }
357 
358 void
359 rtalloc_ign(struct route *ro, u_long ignore)
360 {
361 	struct rtentry *rt;
362 
363 	if ((rt = ro->ro_rt) != NULL) {
364 		if (rt->rt_ifp != NULL && rt->rt_flags & RTF_UP)
365 			return;
366 		RTFREE(rt);
367 		ro->ro_rt = NULL;
368 	}
369 	ro->ro_rt = rtalloc1_fib(&ro->ro_dst, 1, ignore, RT_DEFAULT_FIB);
370 	if (ro->ro_rt)
371 		RT_UNLOCK(ro->ro_rt);
372 }
373 
374 void
375 rtalloc_ign_fib(struct route *ro, u_long ignore, u_int fibnum)
376 {
377 	struct rtentry *rt;
378 
379 	if ((rt = ro->ro_rt) != NULL) {
380 		if (rt->rt_ifp != NULL && rt->rt_flags & RTF_UP)
381 			return;
382 		RTFREE(rt);
383 		ro->ro_rt = NULL;
384 	}
385 	ro->ro_rt = rtalloc1_fib(&ro->ro_dst, 1, ignore, fibnum);
386 	if (ro->ro_rt)
387 		RT_UNLOCK(ro->ro_rt);
388 }
389 
390 /*
391  * Look up the route that matches the address given
392  * Or, at least try.. Create a cloned route if needed.
393  *
394  * The returned route, if any, is locked.
395  */
396 struct rtentry *
397 rtalloc1(struct sockaddr *dst, int report, u_long ignflags)
398 {
399 
400 	return (rtalloc1_fib(dst, report, ignflags, RT_DEFAULT_FIB));
401 }
402 
403 struct rtentry *
404 rtalloc1_fib(struct sockaddr *dst, int report, u_long ignflags,
405 		    u_int fibnum)
406 {
407 	struct radix_node_head *rnh;
408 	struct radix_node *rn;
409 	struct rtentry *newrt;
410 	struct rt_addrinfo info;
411 	int err = 0, msgtype = RTM_MISS;
412 	int needlock;
413 
414 	KASSERT((fibnum < rt_numfibs), ("rtalloc1_fib: bad fibnum"));
415 	rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
416 	newrt = NULL;
417 	if (rnh == NULL)
418 		goto miss;
419 
420 	/*
421 	 * Look up the address in the table for that Address Family
422 	 */
423 	needlock = !(ignflags & RTF_RNH_LOCKED);
424 	if (needlock)
425 		RADIX_NODE_HEAD_RLOCK(rnh);
426 #ifdef INVARIANTS
427 	else
428 		RADIX_NODE_HEAD_LOCK_ASSERT(rnh);
429 #endif
430 	rn = rnh->rnh_matchaddr(dst, rnh);
431 	if (rn && ((rn->rn_flags & RNF_ROOT) == 0)) {
432 		newrt = RNTORT(rn);
433 		RT_LOCK(newrt);
434 		RT_ADDREF(newrt);
435 		if (needlock)
436 			RADIX_NODE_HEAD_RUNLOCK(rnh);
437 		goto done;
438 
439 	} else if (needlock)
440 		RADIX_NODE_HEAD_RUNLOCK(rnh);
441 
442 	/*
443 	 * Either we hit the root or couldn't find any match,
444 	 * Which basically means
445 	 * "caint get there frm here"
446 	 */
447 miss:
448 	V_rtstat.rts_unreach++;
449 
450 	if (report) {
451 		/*
452 		 * If required, report the failure to the supervising
453 		 * Authorities.
454 		 * For a delete, this is not an error. (report == 0)
455 		 */
456 		bzero(&info, sizeof(info));
457 		info.rti_info[RTAX_DST] = dst;
458 		rt_missmsg_fib(msgtype, &info, 0, err, fibnum);
459 	}
460 done:
461 	if (newrt)
462 		RT_LOCK_ASSERT(newrt);
463 	return (newrt);
464 }
465 
466 /*
467  * Remove a reference count from an rtentry.
468  * If the count gets low enough, take it out of the routing table
469  */
470 void
471 rtfree(struct rtentry *rt)
472 {
473 	struct radix_node_head *rnh;
474 
475 	KASSERT(rt != NULL,("%s: NULL rt", __func__));
476 	rnh = rt_tables_get_rnh(rt->rt_fibnum, rt_key(rt)->sa_family);
477 	KASSERT(rnh != NULL,("%s: NULL rnh", __func__));
478 
479 	RT_LOCK_ASSERT(rt);
480 
481 	/*
482 	 * The callers should use RTFREE_LOCKED() or RTFREE(), so
483 	 * we should come here exactly with the last reference.
484 	 */
485 	RT_REMREF(rt);
486 	if (rt->rt_refcnt > 0) {
487 		log(LOG_DEBUG, "%s: %p has %d refs\n", __func__, rt, rt->rt_refcnt);
488 		goto done;
489 	}
490 
491 	/*
492 	 * On last reference give the "close method" a chance
493 	 * to cleanup private state.  This also permits (for
494 	 * IPv4 and IPv6) a chance to decide if the routing table
495 	 * entry should be purged immediately or at a later time.
496 	 * When an immediate purge is to happen the close routine
497 	 * typically calls rtexpunge which clears the RTF_UP flag
498 	 * on the entry so that the code below reclaims the storage.
499 	 */
500 	if (rt->rt_refcnt == 0 && rnh->rnh_close)
501 		rnh->rnh_close((struct radix_node *)rt, rnh);
502 
503 	/*
504 	 * If we are no longer "up" (and ref == 0)
505 	 * then we can free the resources associated
506 	 * with the route.
507 	 */
508 	if ((rt->rt_flags & RTF_UP) == 0) {
509 		if (rt->rt_nodes->rn_flags & (RNF_ACTIVE | RNF_ROOT))
510 			panic("rtfree 2");
511 		/*
512 		 * the rtentry must have been removed from the routing table
513 		 * so it is represented in rttrash.. remove that now.
514 		 */
515 		V_rttrash--;
516 #ifdef	DIAGNOSTIC
517 		if (rt->rt_refcnt < 0) {
518 			printf("rtfree: %p not freed (neg refs)\n", rt);
519 			goto done;
520 		}
521 #endif
522 		/*
523 		 * release references on items we hold them on..
524 		 * e.g other routes and ifaddrs.
525 		 */
526 		if (rt->rt_ifa)
527 			ifa_free(rt->rt_ifa);
528 		/*
529 		 * The key is separatly alloc'd so free it (see rt_setgate()).
530 		 * This also frees the gateway, as they are always malloc'd
531 		 * together.
532 		 */
533 		R_Free(rt_key(rt));
534 
535 		/*
536 		 * and the rtentry itself of course
537 		 */
538 		uma_zfree(V_rtzone, rt);
539 		return;
540 	}
541 done:
542 	RT_UNLOCK(rt);
543 }
544 
545 
546 /*
547  * Force a routing table entry to the specified
548  * destination to go through the given gateway.
549  * Normally called as a result of a routing redirect
550  * message from the network layer.
551  */
552 void
553 rtredirect(struct sockaddr *dst,
554 	struct sockaddr *gateway,
555 	struct sockaddr *netmask,
556 	int flags,
557 	struct sockaddr *src)
558 {
559 
560 	rtredirect_fib(dst, gateway, netmask, flags, src, RT_DEFAULT_FIB);
561 }
562 
563 void
564 rtredirect_fib(struct sockaddr *dst,
565 	struct sockaddr *gateway,
566 	struct sockaddr *netmask,
567 	int flags,
568 	struct sockaddr *src,
569 	u_int fibnum)
570 {
571 	struct rtentry *rt;
572 	int error = 0;
573 	short *stat = NULL;
574 	struct rt_addrinfo info;
575 	struct ifaddr *ifa;
576 	struct radix_node_head *rnh;
577 
578 	ifa = NULL;
579 	rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
580 	if (rnh == NULL) {
581 		error = EAFNOSUPPORT;
582 		goto out;
583 	}
584 
585 	/* verify the gateway is directly reachable */
586 	if ((ifa = ifa_ifwithnet(gateway, 0, fibnum)) == NULL) {
587 		error = ENETUNREACH;
588 		goto out;
589 	}
590 	rt = rtalloc1_fib(dst, 0, 0UL, fibnum);	/* NB: rt is locked */
591 	/*
592 	 * If the redirect isn't from our current router for this dst,
593 	 * it's either old or wrong.  If it redirects us to ourselves,
594 	 * we have a routing loop, perhaps as a result of an interface
595 	 * going down recently.
596 	 */
597 	if (!(flags & RTF_DONE) && rt) {
598 		if (!sa_equal(src, rt->rt_gateway)) {
599 			error = EINVAL;
600 			goto done;
601 		}
602 		if (rt->rt_ifa != ifa && ifa->ifa_addr->sa_family != AF_LINK) {
603 			error = EINVAL;
604 			goto done;
605 		}
606 	}
607 	if ((flags & RTF_GATEWAY) && ifa_ifwithaddr_check(gateway)) {
608 		error = EHOSTUNREACH;
609 		goto done;
610 	}
611 	/*
612 	 * Create a new entry if we just got back a wildcard entry
613 	 * or the lookup failed.  This is necessary for hosts
614 	 * which use routing redirects generated by smart gateways
615 	 * to dynamically build the routing tables.
616 	 */
617 	if (rt == NULL || (rt_mask(rt) && rt_mask(rt)->sa_len < 2))
618 		goto create;
619 	/*
620 	 * Don't listen to the redirect if it's
621 	 * for a route to an interface.
622 	 */
623 	if (rt->rt_flags & RTF_GATEWAY) {
624 		if (((rt->rt_flags & RTF_HOST) == 0) && (flags & RTF_HOST)) {
625 			/*
626 			 * Changing from route to net => route to host.
627 			 * Create new route, rather than smashing route to net.
628 			 */
629 		create:
630 			RTFREE(rt);
631 			rt = NULL;
632 
633 			flags |= RTF_DYNAMIC;
634 			bzero((caddr_t)&info, sizeof(info));
635 			info.rti_info[RTAX_DST] = dst;
636 			info.rti_info[RTAX_GATEWAY] = gateway;
637 			info.rti_info[RTAX_NETMASK] = netmask;
638 			info.rti_ifa = ifa;
639 			info.rti_flags = flags;
640 			error = rtrequest1_fib(RTM_ADD, &info, &rt, fibnum);
641 			if (rt != NULL) {
642 				RT_LOCK(rt);
643 				flags = rt->rt_flags;
644 			}
645 
646 			stat = &V_rtstat.rts_dynamic;
647 		} else {
648 
649 			/*
650 			 * Smash the current notion of the gateway to
651 			 * this destination.  Should check about netmask!!!
652 			 */
653 			if ((flags & RTF_GATEWAY) == 0)
654 				rt->rt_flags &= ~RTF_GATEWAY;
655 			rt->rt_flags |= RTF_MODIFIED;
656 			flags |= RTF_MODIFIED;
657 			stat = &V_rtstat.rts_newgateway;
658 			/*
659 			 * add the key and gateway (in one malloc'd chunk).
660 			 */
661 			RT_UNLOCK(rt);
662 			RADIX_NODE_HEAD_LOCK(rnh);
663 			RT_LOCK(rt);
664 			rt_setgate(rt, rt_key(rt), gateway);
665 			RADIX_NODE_HEAD_UNLOCK(rnh);
666 		}
667 	} else
668 		error = EHOSTUNREACH;
669 done:
670 	if (rt)
671 		RTFREE_LOCKED(rt);
672 out:
673 	if (error)
674 		V_rtstat.rts_badredirect++;
675 	else if (stat != NULL)
676 		(*stat)++;
677 	bzero((caddr_t)&info, sizeof(info));
678 	info.rti_info[RTAX_DST] = dst;
679 	info.rti_info[RTAX_GATEWAY] = gateway;
680 	info.rti_info[RTAX_NETMASK] = netmask;
681 	info.rti_info[RTAX_AUTHOR] = src;
682 	rt_missmsg_fib(RTM_REDIRECT, &info, flags, error, fibnum);
683 	if (ifa != NULL)
684 		ifa_free(ifa);
685 }
686 
687 int
688 rtioctl(u_long req, caddr_t data)
689 {
690 
691 	return (rtioctl_fib(req, data, RT_DEFAULT_FIB));
692 }
693 
694 /*
695  * Routing table ioctl interface.
696  */
697 int
698 rtioctl_fib(u_long req, caddr_t data, u_int fibnum)
699 {
700 
701 	/*
702 	 * If more ioctl commands are added here, make sure the proper
703 	 * super-user checks are being performed because it is possible for
704 	 * prison-root to make it this far if raw sockets have been enabled
705 	 * in jails.
706 	 */
707 #ifdef INET
708 	/* Multicast goop, grrr... */
709 	return mrt_ioctl ? mrt_ioctl(req, data, fibnum) : EOPNOTSUPP;
710 #else /* INET */
711 	return ENXIO;
712 #endif /* INET */
713 }
714 
715 struct ifaddr *
716 ifa_ifwithroute(int flags, const struct sockaddr *dst, struct sockaddr *gateway,
717 				u_int fibnum)
718 {
719 	struct ifaddr *ifa;
720 	int not_found = 0;
721 
722 	if ((flags & RTF_GATEWAY) == 0) {
723 		/*
724 		 * If we are adding a route to an interface,
725 		 * and the interface is a pt to pt link
726 		 * we should search for the destination
727 		 * as our clue to the interface.  Otherwise
728 		 * we can use the local address.
729 		 */
730 		ifa = NULL;
731 		if (flags & RTF_HOST)
732 			ifa = ifa_ifwithdstaddr(dst, fibnum);
733 		if (ifa == NULL)
734 			ifa = ifa_ifwithaddr(gateway);
735 	} else {
736 		/*
737 		 * If we are adding a route to a remote net
738 		 * or host, the gateway may still be on the
739 		 * other end of a pt to pt link.
740 		 */
741 		ifa = ifa_ifwithdstaddr(gateway, fibnum);
742 	}
743 	if (ifa == NULL)
744 		ifa = ifa_ifwithnet(gateway, 0, fibnum);
745 	if (ifa == NULL) {
746 		struct rtentry *rt = rtalloc1_fib(gateway, 0, 0, fibnum);
747 		if (rt == NULL)
748 			return (NULL);
749 		/*
750 		 * dismiss a gateway that is reachable only
751 		 * through the default router
752 		 */
753 		switch (gateway->sa_family) {
754 		case AF_INET:
755 			if (satosin(rt_key(rt))->sin_addr.s_addr == INADDR_ANY)
756 				not_found = 1;
757 			break;
758 		case AF_INET6:
759 			if (IN6_IS_ADDR_UNSPECIFIED(&satosin6(rt_key(rt))->sin6_addr))
760 				not_found = 1;
761 			break;
762 		default:
763 			break;
764 		}
765 		if (!not_found && rt->rt_ifa != NULL) {
766 			ifa = rt->rt_ifa;
767 			ifa_ref(ifa);
768 		}
769 		RT_REMREF(rt);
770 		RT_UNLOCK(rt);
771 		if (not_found || ifa == NULL)
772 			return (NULL);
773 	}
774 	if (ifa->ifa_addr->sa_family != dst->sa_family) {
775 		struct ifaddr *oifa = ifa;
776 		ifa = ifaof_ifpforaddr(dst, ifa->ifa_ifp);
777 		if (ifa == NULL)
778 			ifa = oifa;
779 		else
780 			ifa_free(oifa);
781 	}
782 	return (ifa);
783 }
784 
785 /*
786  * Do appropriate manipulations of a routing tree given
787  * all the bits of info needed
788  */
789 int
790 rtrequest(int req,
791 	struct sockaddr *dst,
792 	struct sockaddr *gateway,
793 	struct sockaddr *netmask,
794 	int flags,
795 	struct rtentry **ret_nrt)
796 {
797 
798 	return (rtrequest_fib(req, dst, gateway, netmask, flags, ret_nrt,
799 	    RT_DEFAULT_FIB));
800 }
801 
802 int
803 rtrequest_fib(int req,
804 	struct sockaddr *dst,
805 	struct sockaddr *gateway,
806 	struct sockaddr *netmask,
807 	int flags,
808 	struct rtentry **ret_nrt,
809 	u_int fibnum)
810 {
811 	struct rt_addrinfo info;
812 
813 	if (dst->sa_len == 0)
814 		return(EINVAL);
815 
816 	bzero((caddr_t)&info, sizeof(info));
817 	info.rti_flags = flags;
818 	info.rti_info[RTAX_DST] = dst;
819 	info.rti_info[RTAX_GATEWAY] = gateway;
820 	info.rti_info[RTAX_NETMASK] = netmask;
821 	return rtrequest1_fib(req, &info, ret_nrt, fibnum);
822 }
823 
824 
825 /*
826  * Copy most of @rt data into @info.
827  *
828  * If @flags contains NHR_COPY, copies dst,netmask and gw to the
829  * pointers specified by @info structure. Assume such pointers
830  * are zeroed sockaddr-like structures with sa_len field initialized
831  * to reflect size of the provided buffer. if no NHR_COPY is specified,
832  * point dst,netmask and gw @info fields to appropriate @rt values.
833  *
834  * if @flags contains NHR_REF, do refcouting on rt_ifp.
835  *
836  * Returns 0 on success.
837  */
838 int
839 rt_exportinfo(struct rtentry *rt, struct rt_addrinfo *info, int flags)
840 {
841 	struct rt_metrics *rmx;
842 	struct sockaddr *src, *dst;
843 	int sa_len;
844 
845 	if (flags & NHR_COPY) {
846 		/* Copy destination if dst is non-zero */
847 		src = rt_key(rt);
848 		dst = info->rti_info[RTAX_DST];
849 		sa_len = src->sa_len;
850 		if (dst != NULL) {
851 			if (src->sa_len > dst->sa_len)
852 				return (ENOMEM);
853 			memcpy(dst, src, src->sa_len);
854 			info->rti_addrs |= RTA_DST;
855 		}
856 
857 		/* Copy mask if set && dst is non-zero */
858 		src = rt_mask(rt);
859 		dst = info->rti_info[RTAX_NETMASK];
860 		if (src != NULL && dst != NULL) {
861 
862 			/*
863 			 * Radix stores different value in sa_len,
864 			 * assume rt_mask() to have the same length
865 			 * as rt_key()
866 			 */
867 			if (sa_len > dst->sa_len)
868 				return (ENOMEM);
869 			memcpy(dst, src, src->sa_len);
870 			info->rti_addrs |= RTA_NETMASK;
871 		}
872 
873 		/* Copy gateway is set && dst is non-zero */
874 		src = rt->rt_gateway;
875 		dst = info->rti_info[RTAX_GATEWAY];
876 		if ((rt->rt_flags & RTF_GATEWAY) && src != NULL && dst != NULL){
877 			if (src->sa_len > dst->sa_len)
878 				return (ENOMEM);
879 			memcpy(dst, src, src->sa_len);
880 			info->rti_addrs |= RTA_GATEWAY;
881 		}
882 	} else {
883 		info->rti_info[RTAX_DST] = rt_key(rt);
884 		info->rti_addrs |= RTA_DST;
885 		if (rt_mask(rt) != NULL) {
886 			info->rti_info[RTAX_NETMASK] = rt_mask(rt);
887 			info->rti_addrs |= RTA_NETMASK;
888 		}
889 		if (rt->rt_flags & RTF_GATEWAY) {
890 			info->rti_info[RTAX_GATEWAY] = rt->rt_gateway;
891 			info->rti_addrs |= RTA_GATEWAY;
892 		}
893 	}
894 
895 	rmx = info->rti_rmx;
896 	if (rmx != NULL) {
897 		info->rti_mflags |= RTV_MTU;
898 		rmx->rmx_mtu = rt->rt_mtu;
899 	}
900 
901 	info->rti_flags = rt->rt_flags;
902 	info->rti_ifp = rt->rt_ifp;
903 	info->rti_ifa = rt->rt_ifa;
904 
905 	if (flags & NHR_REF) {
906 		/* Do 'traditional' refcouting */
907 		if_ref(info->rti_ifp);
908 	}
909 
910 	return (0);
911 }
912 
913 /*
914  * Lookups up route entry for @dst in RIB database for fib @fibnum.
915  * Exports entry data to @info using rt_exportinfo().
916  *
917  * if @flags contains NHR_REF, refcouting is performed on rt_ifp.
918  *   All references can be released later by calling rib_free_info()
919  *
920  * Returns 0 on success.
921  * Returns ENOENT for lookup failure, ENOMEM for export failure.
922  */
923 int
924 rib_lookup_info(uint32_t fibnum, const struct sockaddr *dst, uint32_t flags,
925     uint32_t flowid, struct rt_addrinfo *info)
926 {
927 	struct radix_node_head *rh;
928 	struct radix_node *rn;
929 	struct rtentry *rt;
930 	int error;
931 
932 	KASSERT((fibnum < rt_numfibs), ("rib_lookup_rte: bad fibnum"));
933 	rh = rt_tables_get_rnh(fibnum, dst->sa_family);
934 	if (rh == NULL)
935 		return (ENOENT);
936 
937 	RADIX_NODE_HEAD_RLOCK(rh);
938 	rn = rh->rnh_matchaddr(__DECONST(void *, dst), rh);
939 	if (rn != NULL && ((rn->rn_flags & RNF_ROOT) == 0)) {
940 		rt = RNTORT(rn);
941 		/* Ensure route & ifp is UP */
942 		if (RT_LINK_IS_UP(rt->rt_ifp)) {
943 			flags = (flags & NHR_REF) | NHR_COPY;
944 			error = rt_exportinfo(rt, info, flags);
945 			RADIX_NODE_HEAD_RUNLOCK(rh);
946 
947 			return (error);
948 		}
949 	}
950 	RADIX_NODE_HEAD_RUNLOCK(rh);
951 
952 	return (ENOENT);
953 }
954 
955 /*
956  * Releases all references acquired by rib_lookup_info() when
957  * called with NHR_REF flags.
958  */
959 void
960 rib_free_info(struct rt_addrinfo *info)
961 {
962 
963 	if_rele(info->rti_ifp);
964 }
965 
966 /*
967  * Iterates over all existing fibs in system calling
968  *  @setwa_f function prior to traversing each fib.
969  *  Calls @wa_f function for each element in current fib.
970  * If af is not AF_UNSPEC, iterates over fibs in particular
971  * address family.
972  */
973 void
974 rt_foreach_fib_walk(int af, rt_setwarg_t *setwa_f, rt_walktree_f_t *wa_f,
975     void *arg)
976 {
977 	struct radix_node_head *rnh;
978 	uint32_t fibnum;
979 	int i;
980 
981 	for (fibnum = 0; fibnum < rt_numfibs; fibnum++) {
982 		/* Do we want some specific family? */
983 		if (af != AF_UNSPEC) {
984 			rnh = rt_tables_get_rnh(fibnum, af);
985 			if (rnh == NULL)
986 				continue;
987 			if (setwa_f != NULL)
988 				setwa_f(rnh, fibnum, af, arg);
989 
990 			RADIX_NODE_HEAD_LOCK(rnh);
991 			rnh->rnh_walktree(rnh, (walktree_f_t *)wa_f, arg);
992 			RADIX_NODE_HEAD_UNLOCK(rnh);
993 			continue;
994 		}
995 
996 		for (i = 1; i <= AF_MAX; i++) {
997 			rnh = rt_tables_get_rnh(fibnum, i);
998 			if (rnh == NULL)
999 				continue;
1000 			if (setwa_f != NULL)
1001 				setwa_f(rnh, fibnum, i, arg);
1002 
1003 			RADIX_NODE_HEAD_LOCK(rnh);
1004 			rnh->rnh_walktree(rnh, (walktree_f_t *)wa_f, arg);
1005 			RADIX_NODE_HEAD_UNLOCK(rnh);
1006 		}
1007 	}
1008 }
1009 
1010 struct rt_delinfo
1011 {
1012 	struct rt_addrinfo info;
1013 	struct radix_node_head *rnh;
1014 	struct rtentry *head;
1015 };
1016 
1017 /*
1018  * Conditionally unlinks @rn from radix tree based
1019  * on info data passed in @arg.
1020  */
1021 static int
1022 rt_checkdelroute(struct radix_node *rn, void *arg)
1023 {
1024 	struct rt_delinfo *di;
1025 	struct rt_addrinfo *info;
1026 	struct rtentry *rt;
1027 	int error;
1028 
1029 	di = (struct rt_delinfo *)arg;
1030 	rt = (struct rtentry *)rn;
1031 	info = &di->info;
1032 	error = 0;
1033 
1034 	info->rti_info[RTAX_DST] = rt_key(rt);
1035 	info->rti_info[RTAX_NETMASK] = rt_mask(rt);
1036 	info->rti_info[RTAX_GATEWAY] = rt->rt_gateway;
1037 
1038 	rt = rt_unlinkrte(di->rnh, info, &error);
1039 	if (rt == NULL) {
1040 		/* Either not allowed or not matched. Skip entry */
1041 		return (0);
1042 	}
1043 
1044 	/* Entry was unlinked. Add to the list and return */
1045 	rt->rt_chain = di->head;
1046 	di->head = rt;
1047 
1048 	return (0);
1049 }
1050 
1051 /*
1052  * Iterates over all existing fibs in system.
1053  * Deletes each element for which @filter_f function returned
1054  * non-zero value.
1055  * If @af is not AF_UNSPEC, iterates over fibs in particular
1056  * address family.
1057  */
1058 void
1059 rt_foreach_fib_walk_del(int af, rt_filter_f_t *filter_f, void *arg)
1060 {
1061 	struct radix_node_head *rnh;
1062 	struct rt_delinfo di;
1063 	struct rtentry *rt;
1064 	uint32_t fibnum;
1065 	int i, start, end;
1066 
1067 	bzero(&di, sizeof(di));
1068 	di.info.rti_filter = filter_f;
1069 	di.info.rti_filterdata = arg;
1070 
1071 	for (fibnum = 0; fibnum < rt_numfibs; fibnum++) {
1072 		/* Do we want some specific family? */
1073 		if (af != AF_UNSPEC) {
1074 			start = af;
1075 			end = af;
1076 		} else {
1077 			start = 1;
1078 			end = AF_MAX;
1079 		}
1080 
1081 		for (i = start; i <= end; i++) {
1082 			rnh = rt_tables_get_rnh(fibnum, i);
1083 			if (rnh == NULL)
1084 				continue;
1085 			di.rnh = rnh;
1086 
1087 			RADIX_NODE_HEAD_LOCK(rnh);
1088 			rnh->rnh_walktree(rnh, rt_checkdelroute, &di);
1089 			RADIX_NODE_HEAD_UNLOCK(rnh);
1090 
1091 			if (di.head == NULL)
1092 				continue;
1093 
1094 			/* We might have something to reclaim */
1095 			while (di.head != NULL) {
1096 				rt = di.head;
1097 				di.head = rt->rt_chain;
1098 				rt->rt_chain = NULL;
1099 
1100 				/* TODO std rt -> rt_addrinfo export */
1101 				di.info.rti_info[RTAX_DST] = rt_key(rt);
1102 				di.info.rti_info[RTAX_NETMASK] = rt_mask(rt);
1103 
1104 				rt_notifydelete(rt, &di.info);
1105 				RTFREE_LOCKED(rt);
1106 			}
1107 
1108 		}
1109 	}
1110 }
1111 
1112 /*
1113  * Delete Routes for a Network Interface
1114  *
1115  * Called for each routing entry via the rnh->rnh_walktree() call above
1116  * to delete all route entries referencing a detaching network interface.
1117  *
1118  * Arguments:
1119  *	rt	pointer to rtentry
1120  *	arg	argument passed to rnh->rnh_walktree() - detaching interface
1121  *
1122  * Returns:
1123  *	0	successful
1124  *	errno	failed - reason indicated
1125  */
1126 static int
1127 rt_ifdelroute(const struct rtentry *rt, void *arg)
1128 {
1129 	struct ifnet	*ifp = arg;
1130 
1131 	if (rt->rt_ifp != ifp)
1132 		return (0);
1133 
1134 	/*
1135 	 * Protect (sorta) against walktree recursion problems
1136 	 * with cloned routes
1137 	 */
1138 	if ((rt->rt_flags & RTF_UP) == 0)
1139 		return (0);
1140 
1141 	return (1);
1142 }
1143 
1144 /*
1145  * Delete all remaining routes using this interface
1146  * Unfortuneatly the only way to do this is to slog through
1147  * the entire routing table looking for routes which point
1148  * to this interface...oh well...
1149  */
1150 void
1151 rt_flushifroutes(struct ifnet *ifp)
1152 {
1153 
1154 	rt_foreach_fib_walk_del(AF_UNSPEC, rt_ifdelroute, ifp);
1155 }
1156 
1157 /*
1158  * Conditionally unlinks rtentry matching data inside @info from @rnh.
1159  * Returns unlinked, locked and referenced @rtentry on success,
1160  * Returns NULL and sets @perror to:
1161  * ESRCH - if prefix was not found,
1162  * EADDRINUSE - if trying to delete PINNED route without appropriate flag.
1163  * ENOENT - if supplied filter function returned 0 (not matched).
1164  */
1165 static struct rtentry *
1166 rt_unlinkrte(struct radix_node_head *rnh, struct rt_addrinfo *info, int *perror)
1167 {
1168 	struct sockaddr *dst, *netmask;
1169 	struct rtentry *rt;
1170 	struct radix_node *rn;
1171 
1172 	dst = info->rti_info[RTAX_DST];
1173 	netmask = info->rti_info[RTAX_NETMASK];
1174 
1175 	rt = (struct rtentry *)rnh->rnh_lookup(dst, netmask, rnh);
1176 	if (rt == NULL) {
1177 		*perror = ESRCH;
1178 		return (NULL);
1179 	}
1180 
1181 	if ((info->rti_flags & RTF_PINNED) == 0) {
1182 		/* Check if target route can be deleted */
1183 		if (rt->rt_flags & RTF_PINNED) {
1184 			*perror = EADDRINUSE;
1185 			return (NULL);
1186 		}
1187 	}
1188 
1189 	if (info->rti_filter != NULL) {
1190 		if (info->rti_filter(rt, info->rti_filterdata) == 0) {
1191 			/* Not matched */
1192 			*perror = ENOENT;
1193 			return (NULL);
1194 		}
1195 
1196 		/*
1197 		 * Filter function requested rte deletion.
1198 		 * Ease the caller work by filling in remaining info
1199 		 * from that particular entry.
1200 		 */
1201 		info->rti_info[RTAX_GATEWAY] = rt->rt_gateway;
1202 	}
1203 
1204 	/*
1205 	 * Remove the item from the tree and return it.
1206 	 * Complain if it is not there and do no more processing.
1207 	 */
1208 	*perror = ESRCH;
1209 #ifdef RADIX_MPATH
1210 	if (rn_mpath_capable(rnh))
1211 		rn = rt_mpath_unlink(rnh, info, rt, perror);
1212 	else
1213 #endif
1214 	rn = rnh->rnh_deladdr(dst, netmask, rnh);
1215 	if (rn == NULL)
1216 		return (NULL);
1217 
1218 	if (rn->rn_flags & (RNF_ACTIVE | RNF_ROOT))
1219 		panic ("rtrequest delete");
1220 
1221 	rt = RNTORT(rn);
1222 	RT_LOCK(rt);
1223 	RT_ADDREF(rt);
1224 	rt->rt_flags &= ~RTF_UP;
1225 
1226 	*perror = 0;
1227 
1228 	return (rt);
1229 }
1230 
1231 static void
1232 rt_notifydelete(struct rtentry *rt, struct rt_addrinfo *info)
1233 {
1234 	struct ifaddr *ifa;
1235 
1236 	/*
1237 	 * give the protocol a chance to keep things in sync.
1238 	 */
1239 	ifa = rt->rt_ifa;
1240 	if (ifa != NULL && ifa->ifa_rtrequest != NULL)
1241 		ifa->ifa_rtrequest(RTM_DELETE, rt, info);
1242 
1243 	/*
1244 	 * One more rtentry floating around that is not
1245 	 * linked to the routing table. rttrash will be decremented
1246 	 * when RTFREE(rt) is eventually called.
1247 	 */
1248 	V_rttrash++;
1249 }
1250 
1251 
1252 /*
1253  * These (questionable) definitions of apparent local variables apply
1254  * to the next two functions.  XXXXXX!!!
1255  */
1256 #define	dst	info->rti_info[RTAX_DST]
1257 #define	gateway	info->rti_info[RTAX_GATEWAY]
1258 #define	netmask	info->rti_info[RTAX_NETMASK]
1259 #define	ifaaddr	info->rti_info[RTAX_IFA]
1260 #define	ifpaddr	info->rti_info[RTAX_IFP]
1261 #define	flags	info->rti_flags
1262 
1263 /*
1264  * Look up rt_addrinfo for a specific fib.  Note that if rti_ifa is defined,
1265  * it will be referenced so the caller must free it.
1266  */
1267 int
1268 rt_getifa_fib(struct rt_addrinfo *info, u_int fibnum)
1269 {
1270 	struct ifaddr *ifa;
1271 	int error = 0;
1272 
1273 	/*
1274 	 * ifp may be specified by sockaddr_dl
1275 	 * when protocol address is ambiguous.
1276 	 */
1277 	if (info->rti_ifp == NULL && ifpaddr != NULL &&
1278 	    ifpaddr->sa_family == AF_LINK &&
1279 	    (ifa = ifa_ifwithnet(ifpaddr, 0, fibnum)) != NULL) {
1280 		info->rti_ifp = ifa->ifa_ifp;
1281 		ifa_free(ifa);
1282 	}
1283 	if (info->rti_ifa == NULL && ifaaddr != NULL)
1284 		info->rti_ifa = ifa_ifwithaddr(ifaaddr);
1285 	if (info->rti_ifa == NULL) {
1286 		struct sockaddr *sa;
1287 
1288 		sa = ifaaddr != NULL ? ifaaddr :
1289 		    (gateway != NULL ? gateway : dst);
1290 		if (sa != NULL && info->rti_ifp != NULL)
1291 			info->rti_ifa = ifaof_ifpforaddr(sa, info->rti_ifp);
1292 		else if (dst != NULL && gateway != NULL)
1293 			info->rti_ifa = ifa_ifwithroute(flags, dst, gateway,
1294 							fibnum);
1295 		else if (sa != NULL)
1296 			info->rti_ifa = ifa_ifwithroute(flags, sa, sa,
1297 							fibnum);
1298 	}
1299 	if ((ifa = info->rti_ifa) != NULL) {
1300 		if (info->rti_ifp == NULL)
1301 			info->rti_ifp = ifa->ifa_ifp;
1302 	} else
1303 		error = ENETUNREACH;
1304 	return (error);
1305 }
1306 
1307 static int
1308 if_updatemtu_cb(struct radix_node *rn, void *arg)
1309 {
1310 	struct rtentry *rt;
1311 	struct if_mtuinfo *ifmtu;
1312 
1313 	rt = (struct rtentry *)rn;
1314 	ifmtu = (struct if_mtuinfo *)arg;
1315 
1316 	if (rt->rt_ifp != ifmtu->ifp)
1317 		return (0);
1318 
1319 	if (rt->rt_mtu >= ifmtu->mtu) {
1320 		/* We have to decrease mtu regardless of flags */
1321 		rt->rt_mtu = ifmtu->mtu;
1322 		return (0);
1323 	}
1324 
1325 	/*
1326 	 * New MTU is bigger. Check if are allowed to alter it
1327 	 */
1328 	if ((rt->rt_flags & (RTF_FIXEDMTU | RTF_GATEWAY | RTF_HOST)) != 0) {
1329 
1330 		/*
1331 		 * Skip routes with user-supplied MTU and
1332 		 * non-interface routes
1333 		 */
1334 		return (0);
1335 	}
1336 
1337 	/* We are safe to update route MTU */
1338 	rt->rt_mtu = ifmtu->mtu;
1339 
1340 	return (0);
1341 }
1342 
1343 void
1344 rt_updatemtu(struct ifnet *ifp)
1345 {
1346 	struct if_mtuinfo ifmtu;
1347 	struct radix_node_head *rnh;
1348 	int i, j;
1349 
1350 	ifmtu.ifp = ifp;
1351 
1352 	/*
1353 	 * Try to update rt_mtu for all routes using this interface
1354 	 * Unfortunately the only way to do this is to traverse all
1355 	 * routing tables in all fibs/domains.
1356 	 */
1357 	for (i = 1; i <= AF_MAX; i++) {
1358 		ifmtu.mtu = if_getmtu_family(ifp, i);
1359 		for (j = 0; j < rt_numfibs; j++) {
1360 			rnh = rt_tables_get_rnh(j, i);
1361 			if (rnh == NULL)
1362 				continue;
1363 			RADIX_NODE_HEAD_LOCK(rnh);
1364 			rnh->rnh_walktree(rnh, if_updatemtu_cb, &ifmtu);
1365 			RADIX_NODE_HEAD_UNLOCK(rnh);
1366 		}
1367 	}
1368 }
1369 
1370 
1371 #if 0
1372 int p_sockaddr(char *buf, int buflen, struct sockaddr *s);
1373 int rt_print(char *buf, int buflen, struct rtentry *rt);
1374 
1375 int
1376 p_sockaddr(char *buf, int buflen, struct sockaddr *s)
1377 {
1378 	void *paddr = NULL;
1379 
1380 	switch (s->sa_family) {
1381 	case AF_INET:
1382 		paddr = &((struct sockaddr_in *)s)->sin_addr;
1383 		break;
1384 	case AF_INET6:
1385 		paddr = &((struct sockaddr_in6 *)s)->sin6_addr;
1386 		break;
1387 	}
1388 
1389 	if (paddr == NULL)
1390 		return (0);
1391 
1392 	if (inet_ntop(s->sa_family, paddr, buf, buflen) == NULL)
1393 		return (0);
1394 
1395 	return (strlen(buf));
1396 }
1397 
1398 int
1399 rt_print(char *buf, int buflen, struct rtentry *rt)
1400 {
1401 	struct sockaddr *addr, *mask;
1402 	int i = 0;
1403 
1404 	addr = rt_key(rt);
1405 	mask = rt_mask(rt);
1406 
1407 	i = p_sockaddr(buf, buflen, addr);
1408 	if (!(rt->rt_flags & RTF_HOST)) {
1409 		buf[i++] = '/';
1410 		i += p_sockaddr(buf + i, buflen - i, mask);
1411 	}
1412 
1413 	if (rt->rt_flags & RTF_GATEWAY) {
1414 		buf[i++] = '>';
1415 		i += p_sockaddr(buf + i, buflen - i, rt->rt_gateway);
1416 	}
1417 
1418 	return (i);
1419 }
1420 #endif
1421 
1422 #ifdef RADIX_MPATH
1423 /*
1424  * Deletes key for single-path routes, unlinks rtentry with
1425  * gateway specified in @info from multi-path routes.
1426  *
1427  * Returnes unlinked entry. In case of failure, returns NULL
1428  * and sets @perror to ESRCH.
1429  */
1430 static struct radix_node *
1431 rt_mpath_unlink(struct radix_node_head *rnh, struct rt_addrinfo *info,
1432     struct rtentry *rto, int *perror)
1433 {
1434 	/*
1435 	 * if we got multipath routes, we require users to specify
1436 	 * a matching RTAX_GATEWAY.
1437 	 */
1438 	struct rtentry *rt; // *rto = NULL;
1439 	struct radix_node *rn;
1440 	struct sockaddr *gw;
1441 
1442 	gw = info->rti_info[RTAX_GATEWAY];
1443 	rt = rt_mpath_matchgate(rto, gw);
1444 	if (rt == NULL) {
1445 		*perror = ESRCH;
1446 		return (NULL);
1447 	}
1448 
1449 	/*
1450 	 * this is the first entry in the chain
1451 	 */
1452 	if (rto == rt) {
1453 		rn = rn_mpath_next((struct radix_node *)rt);
1454 		/*
1455 		 * there is another entry, now it's active
1456 		 */
1457 		if (rn) {
1458 			rto = RNTORT(rn);
1459 			RT_LOCK(rto);
1460 			rto->rt_flags |= RTF_UP;
1461 			RT_UNLOCK(rto);
1462 		} else if (rt->rt_flags & RTF_GATEWAY) {
1463 			/*
1464 			 * For gateway routes, we need to
1465 			 * make sure that we we are deleting
1466 			 * the correct gateway.
1467 			 * rt_mpath_matchgate() does not
1468 			 * check the case when there is only
1469 			 * one route in the chain.
1470 			 */
1471 			if (gw &&
1472 			    (rt->rt_gateway->sa_len != gw->sa_len ||
1473 				memcmp(rt->rt_gateway, gw, gw->sa_len))) {
1474 				*perror = ESRCH;
1475 				return (NULL);
1476 			}
1477 		}
1478 
1479 		/*
1480 		 * use the normal delete code to remove
1481 		 * the first entry
1482 		 */
1483 		rn = rnh->rnh_deladdr(dst, netmask, rnh);
1484 		*perror = 0;
1485 		return (rn);
1486 	}
1487 
1488 	/*
1489 	 * if the entry is 2nd and on up
1490 	 */
1491 	if (rt_mpath_deldup(rto, rt) == 0)
1492 		panic ("rtrequest1: rt_mpath_deldup");
1493 	*perror = 0;
1494 	rn = (struct radix_node *)rt;
1495 	return (rn);
1496 }
1497 #endif
1498 
1499 #ifdef FLOWTABLE
1500 static struct rtentry *
1501 rt_flowtable_check_route(struct radix_node_head *rnh, struct rt_addrinfo *info)
1502 {
1503 #if defined(INET6) || defined(INET)
1504 	struct radix_node *rn;
1505 #endif
1506 	struct rtentry *rt0;
1507 
1508 	rt0 = NULL;
1509 	/* "flow-table" only supports IPv6 and IPv4 at the moment. */
1510 	switch (dst->sa_family) {
1511 #ifdef INET6
1512 	case AF_INET6:
1513 #endif
1514 #ifdef INET
1515 	case AF_INET:
1516 #endif
1517 #if defined(INET6) || defined(INET)
1518 		rn = rnh->rnh_matchaddr(dst, rnh);
1519 		if (rn && ((rn->rn_flags & RNF_ROOT) == 0)) {
1520 			struct sockaddr *mask;
1521 			u_char *m, *n;
1522 			int len;
1523 
1524 			/*
1525 			 * compare mask to see if the new route is
1526 			 * more specific than the existing one
1527 			 */
1528 			rt0 = RNTORT(rn);
1529 			RT_LOCK(rt0);
1530 			RT_ADDREF(rt0);
1531 			RT_UNLOCK(rt0);
1532 			/*
1533 			 * A host route is already present, so
1534 			 * leave the flow-table entries as is.
1535 			 */
1536 			if (rt0->rt_flags & RTF_HOST) {
1537 				RTFREE(rt0);
1538 				rt0 = NULL;
1539 			} else if (!(flags & RTF_HOST) && netmask) {
1540 				mask = rt_mask(rt0);
1541 				len = mask->sa_len;
1542 				m = (u_char *)mask;
1543 				n = (u_char *)netmask;
1544 				while (len-- > 0) {
1545 					if (*n != *m)
1546 						break;
1547 					n++;
1548 					m++;
1549 				}
1550 				if (len == 0 || (*n < *m)) {
1551 					RTFREE(rt0);
1552 					rt0 = NULL;
1553 				}
1554 			}
1555 		}
1556 #endif/* INET6 || INET */
1557 	}
1558 
1559 	return (rt0);
1560 }
1561 #endif
1562 
1563 int
1564 rtrequest1_fib(int req, struct rt_addrinfo *info, struct rtentry **ret_nrt,
1565 				u_int fibnum)
1566 {
1567 	int error = 0;
1568 	struct rtentry *rt, *rt_old;
1569 #ifdef FLOWTABLE
1570 	struct rtentry *rt0;
1571 #endif
1572 	struct radix_node *rn;
1573 	struct radix_node_head *rnh;
1574 	struct ifaddr *ifa;
1575 	struct sockaddr *ndst;
1576 	struct sockaddr_storage mdst;
1577 
1578 	KASSERT((fibnum < rt_numfibs), ("rtrequest1_fib: bad fibnum"));
1579 	KASSERT((flags & RTF_RNH_LOCKED) == 0, ("rtrequest1_fib: locked"));
1580 	switch (dst->sa_family) {
1581 	case AF_INET6:
1582 	case AF_INET:
1583 		/* We support multiple FIBs. */
1584 		break;
1585 	default:
1586 		fibnum = RT_DEFAULT_FIB;
1587 		break;
1588 	}
1589 
1590 	/*
1591 	 * Find the correct routing tree to use for this Address Family
1592 	 */
1593 	rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
1594 	if (rnh == NULL)
1595 		return (EAFNOSUPPORT);
1596 
1597 	/*
1598 	 * If we are adding a host route then we don't want to put
1599 	 * a netmask in the tree, nor do we want to clone it.
1600 	 */
1601 	if (flags & RTF_HOST)
1602 		netmask = NULL;
1603 
1604 	switch (req) {
1605 	case RTM_DELETE:
1606 		if (netmask) {
1607 			rt_maskedcopy(dst, (struct sockaddr *)&mdst, netmask);
1608 			dst = (struct sockaddr *)&mdst;
1609 		}
1610 
1611 		RADIX_NODE_HEAD_LOCK(rnh);
1612 		rt = rt_unlinkrte(rnh, info, &error);
1613 		RADIX_NODE_HEAD_UNLOCK(rnh);
1614 		if (error != 0)
1615 			return (error);
1616 
1617 		rt_notifydelete(rt, info);
1618 
1619 		/*
1620 		 * If the caller wants it, then it can have it,
1621 		 * but it's up to it to free the rtentry as we won't be
1622 		 * doing it.
1623 		 */
1624 		if (ret_nrt) {
1625 			*ret_nrt = rt;
1626 			RT_UNLOCK(rt);
1627 		} else
1628 			RTFREE_LOCKED(rt);
1629 		break;
1630 	case RTM_RESOLVE:
1631 		/*
1632 		 * resolve was only used for route cloning
1633 		 * here for compat
1634 		 */
1635 		break;
1636 	case RTM_ADD:
1637 		if ((flags & RTF_GATEWAY) && !gateway)
1638 			return (EINVAL);
1639 		if (dst && gateway && (dst->sa_family != gateway->sa_family) &&
1640 		    (gateway->sa_family != AF_UNSPEC) && (gateway->sa_family != AF_LINK))
1641 			return (EINVAL);
1642 
1643 		if (info->rti_ifa == NULL) {
1644 			error = rt_getifa_fib(info, fibnum);
1645 			if (error)
1646 				return (error);
1647 		} else
1648 			ifa_ref(info->rti_ifa);
1649 		ifa = info->rti_ifa;
1650 		rt = uma_zalloc(V_rtzone, M_NOWAIT);
1651 		if (rt == NULL) {
1652 			ifa_free(ifa);
1653 			return (ENOBUFS);
1654 		}
1655 		rt->rt_flags = RTF_UP | flags;
1656 		rt->rt_fibnum = fibnum;
1657 		/*
1658 		 * Add the gateway. Possibly re-malloc-ing the storage for it.
1659 		 */
1660 		if ((error = rt_setgate(rt, dst, gateway)) != 0) {
1661 			ifa_free(ifa);
1662 			uma_zfree(V_rtzone, rt);
1663 			return (error);
1664 		}
1665 
1666 		/*
1667 		 * point to the (possibly newly malloc'd) dest address.
1668 		 */
1669 		ndst = (struct sockaddr *)rt_key(rt);
1670 
1671 		/*
1672 		 * make sure it contains the value we want (masked if needed).
1673 		 */
1674 		if (netmask) {
1675 			rt_maskedcopy(dst, ndst, netmask);
1676 		} else
1677 			bcopy(dst, ndst, dst->sa_len);
1678 
1679 		/*
1680 		 * We use the ifa reference returned by rt_getifa_fib().
1681 		 * This moved from below so that rnh->rnh_addaddr() can
1682 		 * examine the ifa and  ifa->ifa_ifp if it so desires.
1683 		 */
1684 		rt->rt_ifa = ifa;
1685 		rt->rt_ifp = ifa->ifa_ifp;
1686 		rt->rt_weight = 1;
1687 
1688 		rt_setmetrics(info, rt);
1689 
1690 		RADIX_NODE_HEAD_LOCK(rnh);
1691 		RT_LOCK(rt);
1692 #ifdef RADIX_MPATH
1693 		/* do not permit exactly the same dst/mask/gw pair */
1694 		if (rn_mpath_capable(rnh) &&
1695 			rt_mpath_conflict(rnh, rt, netmask)) {
1696 			RADIX_NODE_HEAD_UNLOCK(rnh);
1697 
1698 			ifa_free(rt->rt_ifa);
1699 			R_Free(rt_key(rt));
1700 			uma_zfree(V_rtzone, rt);
1701 			return (EEXIST);
1702 		}
1703 #endif
1704 
1705 #ifdef FLOWTABLE
1706 		rt0 = rt_flowtable_check_route(rnh, info);
1707 #endif /* FLOWTABLE */
1708 
1709 		/* XXX mtu manipulation will be done in rnh_addaddr -- itojun */
1710 		rn = rnh->rnh_addaddr(ndst, netmask, rnh, rt->rt_nodes);
1711 
1712 		rt_old = NULL;
1713 		if (rn == NULL && (info->rti_flags & RTF_PINNED) != 0) {
1714 
1715 			/*
1716 			 * Force removal and re-try addition
1717 			 * TODO: better multipath&pinned support
1718 			 */
1719 			struct sockaddr *info_dst = info->rti_info[RTAX_DST];
1720 			info->rti_info[RTAX_DST] = ndst;
1721 			/* Do not delete existing PINNED(interface) routes */
1722 			info->rti_flags &= ~RTF_PINNED;
1723 			rt_old = rt_unlinkrte(rnh, info, &error);
1724 			info->rti_flags |= RTF_PINNED;
1725 			info->rti_info[RTAX_DST] = info_dst;
1726 			if (rt_old != NULL)
1727 				rn = rnh->rnh_addaddr(ndst, netmask, rnh,
1728 				    rt->rt_nodes);
1729 		}
1730 		RADIX_NODE_HEAD_UNLOCK(rnh);
1731 
1732 		if (rt_old != NULL)
1733 			RT_UNLOCK(rt_old);
1734 
1735 		/*
1736 		 * If it still failed to go into the tree,
1737 		 * then un-make it (this should be a function)
1738 		 */
1739 		if (rn == NULL) {
1740 			ifa_free(rt->rt_ifa);
1741 			R_Free(rt_key(rt));
1742 			uma_zfree(V_rtzone, rt);
1743 #ifdef FLOWTABLE
1744 			if (rt0 != NULL)
1745 				RTFREE(rt0);
1746 #endif
1747 			return (EEXIST);
1748 		}
1749 #ifdef FLOWTABLE
1750 		else if (rt0 != NULL) {
1751 			flowtable_route_flush(dst->sa_family, rt0);
1752 			RTFREE(rt0);
1753 		}
1754 #endif
1755 
1756 		if (rt_old != NULL) {
1757 			rt_notifydelete(rt_old, info);
1758 			RTFREE(rt_old);
1759 		}
1760 
1761 		/*
1762 		 * If this protocol has something to add to this then
1763 		 * allow it to do that as well.
1764 		 */
1765 		if (ifa->ifa_rtrequest)
1766 			ifa->ifa_rtrequest(req, rt, info);
1767 
1768 		/*
1769 		 * actually return a resultant rtentry and
1770 		 * give the caller a single reference.
1771 		 */
1772 		if (ret_nrt) {
1773 			*ret_nrt = rt;
1774 			RT_ADDREF(rt);
1775 		}
1776 		RT_UNLOCK(rt);
1777 		break;
1778 	case RTM_CHANGE:
1779 		RADIX_NODE_HEAD_LOCK(rnh);
1780 		error = rtrequest1_fib_change(rnh, info, ret_nrt, fibnum);
1781 		RADIX_NODE_HEAD_UNLOCK(rnh);
1782 		break;
1783 	default:
1784 		error = EOPNOTSUPP;
1785 	}
1786 
1787 	return (error);
1788 }
1789 
1790 #undef dst
1791 #undef gateway
1792 #undef netmask
1793 #undef ifaaddr
1794 #undef ifpaddr
1795 #undef flags
1796 
1797 static int
1798 rtrequest1_fib_change(struct radix_node_head *rnh, struct rt_addrinfo *info,
1799     struct rtentry **ret_nrt, u_int fibnum)
1800 {
1801 	struct rtentry *rt = NULL;
1802 	int error = 0;
1803 	int free_ifa = 0;
1804 	int family, mtu;
1805 	struct if_mtuinfo ifmtu;
1806 
1807 	rt = (struct rtentry *)rnh->rnh_lookup(info->rti_info[RTAX_DST],
1808 	    info->rti_info[RTAX_NETMASK], rnh);
1809 
1810 	if (rt == NULL)
1811 		return (ESRCH);
1812 
1813 #ifdef RADIX_MPATH
1814 	/*
1815 	 * If we got multipath routes,
1816 	 * we require users to specify a matching RTAX_GATEWAY.
1817 	 */
1818 	if (rn_mpath_capable(rnh)) {
1819 		rt = rt_mpath_matchgate(rt, info->rti_info[RTAX_GATEWAY]);
1820 		if (rt == NULL)
1821 			return (ESRCH);
1822 	}
1823 #endif
1824 
1825 	RT_LOCK(rt);
1826 
1827 	rt_setmetrics(info, rt);
1828 
1829 	/*
1830 	 * New gateway could require new ifaddr, ifp;
1831 	 * flags may also be different; ifp may be specified
1832 	 * by ll sockaddr when protocol address is ambiguous
1833 	 */
1834 	if (((rt->rt_flags & RTF_GATEWAY) &&
1835 	    info->rti_info[RTAX_GATEWAY] != NULL) ||
1836 	    info->rti_info[RTAX_IFP] != NULL ||
1837 	    (info->rti_info[RTAX_IFA] != NULL &&
1838 	     !sa_equal(info->rti_info[RTAX_IFA], rt->rt_ifa->ifa_addr))) {
1839 
1840 		error = rt_getifa_fib(info, fibnum);
1841 		if (info->rti_ifa != NULL)
1842 			free_ifa = 1;
1843 
1844 		if (error != 0)
1845 			goto bad;
1846 	}
1847 
1848 	/* Check if outgoing interface has changed */
1849 	if (info->rti_ifa != NULL && info->rti_ifa != rt->rt_ifa &&
1850 	    rt->rt_ifa != NULL && rt->rt_ifa->ifa_rtrequest != NULL) {
1851 		rt->rt_ifa->ifa_rtrequest(RTM_DELETE, rt, info);
1852 		ifa_free(rt->rt_ifa);
1853 	}
1854 	/* Update gateway address */
1855 	if (info->rti_info[RTAX_GATEWAY] != NULL) {
1856 		error = rt_setgate(rt, rt_key(rt), info->rti_info[RTAX_GATEWAY]);
1857 		if (error != 0)
1858 			goto bad;
1859 
1860 		rt->rt_flags &= ~RTF_GATEWAY;
1861 		rt->rt_flags |= (RTF_GATEWAY & info->rti_flags);
1862 	}
1863 
1864 	if (info->rti_ifa != NULL && info->rti_ifa != rt->rt_ifa) {
1865 		ifa_ref(info->rti_ifa);
1866 		rt->rt_ifa = info->rti_ifa;
1867 		rt->rt_ifp = info->rti_ifp;
1868 	}
1869 	/* Allow some flags to be toggled on change. */
1870 	rt->rt_flags &= ~RTF_FMASK;
1871 	rt->rt_flags |= info->rti_flags & RTF_FMASK;
1872 
1873 	if (rt->rt_ifa && rt->rt_ifa->ifa_rtrequest != NULL)
1874 	       rt->rt_ifa->ifa_rtrequest(RTM_ADD, rt, info);
1875 
1876 	/* Alter route MTU if necessary */
1877 	if (rt->rt_ifp != NULL) {
1878 		family = info->rti_info[RTAX_DST]->sa_family;
1879 		mtu = if_getmtu_family(rt->rt_ifp, family);
1880 		/* Set default MTU */
1881 		if (rt->rt_mtu == 0)
1882 			rt->rt_mtu = mtu;
1883 		if (rt->rt_mtu != mtu) {
1884 			/* Check if we really need to update */
1885 			ifmtu.ifp = rt->rt_ifp;
1886 			ifmtu.mtu = mtu;
1887 			if_updatemtu_cb(rt->rt_nodes, &ifmtu);
1888 		}
1889 	}
1890 
1891 	if (ret_nrt) {
1892 		*ret_nrt = rt;
1893 		RT_ADDREF(rt);
1894 	}
1895 bad:
1896 	RT_UNLOCK(rt);
1897 	if (free_ifa != 0)
1898 		ifa_free(info->rti_ifa);
1899 	return (error);
1900 }
1901 
1902 static void
1903 rt_setmetrics(const struct rt_addrinfo *info, struct rtentry *rt)
1904 {
1905 
1906 	if (info->rti_mflags & RTV_MTU) {
1907 		if (info->rti_rmx->rmx_mtu != 0) {
1908 
1909 			/*
1910 			 * MTU was explicitly provided by user.
1911 			 * Keep it.
1912 			 */
1913 			rt->rt_flags |= RTF_FIXEDMTU;
1914 		} else {
1915 
1916 			/*
1917 			 * User explicitly sets MTU to 0.
1918 			 * Assume rollback to default.
1919 			 */
1920 			rt->rt_flags &= ~RTF_FIXEDMTU;
1921 		}
1922 		rt->rt_mtu = info->rti_rmx->rmx_mtu;
1923 	}
1924 	if (info->rti_mflags & RTV_WEIGHT)
1925 		rt->rt_weight = info->rti_rmx->rmx_weight;
1926 	/* Kernel -> userland timebase conversion. */
1927 	if (info->rti_mflags & RTV_EXPIRE)
1928 		rt->rt_expire = info->rti_rmx->rmx_expire ?
1929 		    info->rti_rmx->rmx_expire - time_second + time_uptime : 0;
1930 }
1931 
1932 int
1933 rt_setgate(struct rtentry *rt, struct sockaddr *dst, struct sockaddr *gate)
1934 {
1935 	/* XXX dst may be overwritten, can we move this to below */
1936 	int dlen = SA_SIZE(dst), glen = SA_SIZE(gate);
1937 
1938 	/*
1939 	 * Prepare to store the gateway in rt->rt_gateway.
1940 	 * Both dst and gateway are stored one after the other in the same
1941 	 * malloc'd chunk. If we have room, we can reuse the old buffer,
1942 	 * rt_gateway already points to the right place.
1943 	 * Otherwise, malloc a new block and update the 'dst' address.
1944 	 */
1945 	if (rt->rt_gateway == NULL || glen > SA_SIZE(rt->rt_gateway)) {
1946 		caddr_t new;
1947 
1948 		R_Malloc(new, caddr_t, dlen + glen);
1949 		if (new == NULL)
1950 			return ENOBUFS;
1951 		/*
1952 		 * XXX note, we copy from *dst and not *rt_key(rt) because
1953 		 * rt_setgate() can be called to initialize a newly
1954 		 * allocated route entry, in which case rt_key(rt) == NULL
1955 		 * (and also rt->rt_gateway == NULL).
1956 		 * Free()/free() handle a NULL argument just fine.
1957 		 */
1958 		bcopy(dst, new, dlen);
1959 		R_Free(rt_key(rt));	/* free old block, if any */
1960 		rt_key(rt) = (struct sockaddr *)new;
1961 		rt->rt_gateway = (struct sockaddr *)(new + dlen);
1962 	}
1963 
1964 	/*
1965 	 * Copy the new gateway value into the memory chunk.
1966 	 */
1967 	bcopy(gate, rt->rt_gateway, glen);
1968 
1969 	return (0);
1970 }
1971 
1972 void
1973 rt_maskedcopy(struct sockaddr *src, struct sockaddr *dst, struct sockaddr *netmask)
1974 {
1975 	u_char *cp1 = (u_char *)src;
1976 	u_char *cp2 = (u_char *)dst;
1977 	u_char *cp3 = (u_char *)netmask;
1978 	u_char *cplim = cp2 + *cp3;
1979 	u_char *cplim2 = cp2 + *cp1;
1980 
1981 	*cp2++ = *cp1++; *cp2++ = *cp1++; /* copies sa_len & sa_family */
1982 	cp3 += 2;
1983 	if (cplim > cplim2)
1984 		cplim = cplim2;
1985 	while (cp2 < cplim)
1986 		*cp2++ = *cp1++ & *cp3++;
1987 	if (cp2 < cplim2)
1988 		bzero((caddr_t)cp2, (unsigned)(cplim2 - cp2));
1989 }
1990 
1991 /*
1992  * Set up a routing table entry, normally
1993  * for an interface.
1994  */
1995 #define _SOCKADDR_TMPSIZE 128 /* Not too big.. kernel stack size is limited */
1996 static inline  int
1997 rtinit1(struct ifaddr *ifa, int cmd, int flags, int fibnum)
1998 {
1999 	struct sockaddr *dst;
2000 	struct sockaddr *netmask;
2001 	struct rtentry *rt = NULL;
2002 	struct rt_addrinfo info;
2003 	int error = 0;
2004 	int startfib, endfib;
2005 	char tempbuf[_SOCKADDR_TMPSIZE];
2006 	int didwork = 0;
2007 	int a_failure = 0;
2008 	static struct sockaddr_dl null_sdl = {sizeof(null_sdl), AF_LINK};
2009 	struct radix_node_head *rnh;
2010 
2011 	if (flags & RTF_HOST) {
2012 		dst = ifa->ifa_dstaddr;
2013 		netmask = NULL;
2014 	} else {
2015 		dst = ifa->ifa_addr;
2016 		netmask = ifa->ifa_netmask;
2017 	}
2018 	if (dst->sa_len == 0)
2019 		return(EINVAL);
2020 	switch (dst->sa_family) {
2021 	case AF_INET6:
2022 	case AF_INET:
2023 		/* We support multiple FIBs. */
2024 		break;
2025 	default:
2026 		fibnum = RT_DEFAULT_FIB;
2027 		break;
2028 	}
2029 	if (fibnum == RT_ALL_FIBS) {
2030 		if (V_rt_add_addr_allfibs == 0 && cmd == (int)RTM_ADD)
2031 			startfib = endfib = ifa->ifa_ifp->if_fib;
2032 		else {
2033 			startfib = 0;
2034 			endfib = rt_numfibs - 1;
2035 		}
2036 	} else {
2037 		KASSERT((fibnum < rt_numfibs), ("rtinit1: bad fibnum"));
2038 		startfib = fibnum;
2039 		endfib = fibnum;
2040 	}
2041 
2042 	/*
2043 	 * If it's a delete, check that if it exists,
2044 	 * it's on the correct interface or we might scrub
2045 	 * a route to another ifa which would
2046 	 * be confusing at best and possibly worse.
2047 	 */
2048 	if (cmd == RTM_DELETE) {
2049 		/*
2050 		 * It's a delete, so it should already exist..
2051 		 * If it's a net, mask off the host bits
2052 		 * (Assuming we have a mask)
2053 		 * XXX this is kinda inet specific..
2054 		 */
2055 		if (netmask != NULL) {
2056 			rt_maskedcopy(dst, (struct sockaddr *)tempbuf, netmask);
2057 			dst = (struct sockaddr *)tempbuf;
2058 		}
2059 	}
2060 	/*
2061 	 * Now go through all the requested tables (fibs) and do the
2062 	 * requested action. Realistically, this will either be fib 0
2063 	 * for protocols that don't do multiple tables or all the
2064 	 * tables for those that do.
2065 	 */
2066 	for ( fibnum = startfib; fibnum <= endfib; fibnum++) {
2067 		if (cmd == RTM_DELETE) {
2068 			struct radix_node *rn;
2069 			/*
2070 			 * Look up an rtentry that is in the routing tree and
2071 			 * contains the correct info.
2072 			 */
2073 			rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
2074 			if (rnh == NULL)
2075 				/* this table doesn't exist but others might */
2076 				continue;
2077 			RADIX_NODE_HEAD_RLOCK(rnh);
2078 			rn = rnh->rnh_lookup(dst, netmask, rnh);
2079 #ifdef RADIX_MPATH
2080 			if (rn_mpath_capable(rnh)) {
2081 
2082 				if (rn == NULL)
2083 					error = ESRCH;
2084 				else {
2085 					rt = RNTORT(rn);
2086 					/*
2087 					 * for interface route the
2088 					 * rt->rt_gateway is sockaddr_intf
2089 					 * for cloning ARP entries, so
2090 					 * rt_mpath_matchgate must use the
2091 					 * interface address
2092 					 */
2093 					rt = rt_mpath_matchgate(rt,
2094 					    ifa->ifa_addr);
2095 					if (rt == NULL)
2096 						error = ESRCH;
2097 				}
2098 			}
2099 #endif
2100 			error = (rn == NULL ||
2101 			    (rn->rn_flags & RNF_ROOT) ||
2102 			    RNTORT(rn)->rt_ifa != ifa);
2103 			RADIX_NODE_HEAD_RUNLOCK(rnh);
2104 			if (error) {
2105 				/* this is only an error if bad on ALL tables */
2106 				continue;
2107 			}
2108 		}
2109 		/*
2110 		 * Do the actual request
2111 		 */
2112 		bzero((caddr_t)&info, sizeof(info));
2113 		info.rti_ifa = ifa;
2114 		info.rti_flags = flags |
2115 		    (ifa->ifa_flags & ~IFA_RTSELF) | RTF_PINNED;
2116 		info.rti_info[RTAX_DST] = dst;
2117 		/*
2118 		 * doing this for compatibility reasons
2119 		 */
2120 		if (cmd == RTM_ADD)
2121 			info.rti_info[RTAX_GATEWAY] =
2122 			    (struct sockaddr *)&null_sdl;
2123 		else
2124 			info.rti_info[RTAX_GATEWAY] = ifa->ifa_addr;
2125 		info.rti_info[RTAX_NETMASK] = netmask;
2126 		error = rtrequest1_fib(cmd, &info, &rt, fibnum);
2127 
2128 		if (error == 0 && rt != NULL) {
2129 			/*
2130 			 * notify any listening routing agents of the change
2131 			 */
2132 			RT_LOCK(rt);
2133 #ifdef RADIX_MPATH
2134 			/*
2135 			 * in case address alias finds the first address
2136 			 * e.g. ifconfig bge0 192.0.2.246/24
2137 			 * e.g. ifconfig bge0 192.0.2.247/24
2138 			 * the address set in the route is 192.0.2.246
2139 			 * so we need to replace it with 192.0.2.247
2140 			 */
2141 			if (memcmp(rt->rt_ifa->ifa_addr,
2142 			    ifa->ifa_addr, ifa->ifa_addr->sa_len)) {
2143 				ifa_free(rt->rt_ifa);
2144 				ifa_ref(ifa);
2145 				rt->rt_ifp = ifa->ifa_ifp;
2146 				rt->rt_ifa = ifa;
2147 			}
2148 #endif
2149 			/*
2150 			 * doing this for compatibility reasons
2151 			 */
2152 			if (cmd == RTM_ADD) {
2153 			    ((struct sockaddr_dl *)rt->rt_gateway)->sdl_type  =
2154 				rt->rt_ifp->if_type;
2155 			    ((struct sockaddr_dl *)rt->rt_gateway)->sdl_index =
2156 				rt->rt_ifp->if_index;
2157 			}
2158 			RT_ADDREF(rt);
2159 			RT_UNLOCK(rt);
2160 			rt_newaddrmsg_fib(cmd, ifa, error, rt, fibnum);
2161 			RT_LOCK(rt);
2162 			RT_REMREF(rt);
2163 			if (cmd == RTM_DELETE) {
2164 				/*
2165 				 * If we are deleting, and we found an entry,
2166 				 * then it's been removed from the tree..
2167 				 * now throw it away.
2168 				 */
2169 				RTFREE_LOCKED(rt);
2170 			} else {
2171 				if (cmd == RTM_ADD) {
2172 					/*
2173 					 * We just wanted to add it..
2174 					 * we don't actually need a reference.
2175 					 */
2176 					RT_REMREF(rt);
2177 				}
2178 				RT_UNLOCK(rt);
2179 			}
2180 			didwork = 1;
2181 		}
2182 		if (error)
2183 			a_failure = error;
2184 	}
2185 	if (cmd == RTM_DELETE) {
2186 		if (didwork) {
2187 			error = 0;
2188 		} else {
2189 			/* we only give an error if it wasn't in any table */
2190 			error = ((flags & RTF_HOST) ?
2191 			    EHOSTUNREACH : ENETUNREACH);
2192 		}
2193 	} else {
2194 		if (a_failure) {
2195 			/* return an error if any of them failed */
2196 			error = a_failure;
2197 		}
2198 	}
2199 	return (error);
2200 }
2201 
2202 /*
2203  * Set up a routing table entry, normally
2204  * for an interface.
2205  */
2206 int
2207 rtinit(struct ifaddr *ifa, int cmd, int flags)
2208 {
2209 	struct sockaddr *dst;
2210 	int fib = RT_DEFAULT_FIB;
2211 
2212 	if (flags & RTF_HOST) {
2213 		dst = ifa->ifa_dstaddr;
2214 	} else {
2215 		dst = ifa->ifa_addr;
2216 	}
2217 
2218 	switch (dst->sa_family) {
2219 	case AF_INET6:
2220 	case AF_INET:
2221 		/* We do support multiple FIBs. */
2222 		fib = RT_ALL_FIBS;
2223 		break;
2224 	}
2225 	return (rtinit1(ifa, cmd, flags, fib));
2226 }
2227 
2228 /*
2229  * Announce interface address arrival/withdraw
2230  * Returns 0 on success.
2231  */
2232 int
2233 rt_addrmsg(int cmd, struct ifaddr *ifa, int fibnum)
2234 {
2235 
2236 	KASSERT(cmd == RTM_ADD || cmd == RTM_DELETE,
2237 	    ("unexpected cmd %d", cmd));
2238 
2239 	KASSERT(fibnum == RT_ALL_FIBS || (fibnum >= 0 && fibnum < rt_numfibs),
2240 	    ("%s: fib out of range 0 <=%d<%d", __func__, fibnum, rt_numfibs));
2241 
2242 #if defined(INET) || defined(INET6)
2243 #ifdef SCTP
2244 	/*
2245 	 * notify the SCTP stack
2246 	 * this will only get called when an address is added/deleted
2247 	 * XXX pass the ifaddr struct instead if ifa->ifa_addr...
2248 	 */
2249 	sctp_addr_change(ifa, cmd);
2250 #endif /* SCTP */
2251 #endif
2252 	return (rtsock_addrmsg(cmd, ifa, fibnum));
2253 }
2254 
2255 /*
2256  * Announce route addition/removal.
2257  * Users of this function MUST validate input data BEFORE calling.
2258  * However we have to be able to handle invalid data:
2259  * if some userland app sends us "invalid" route message (invalid mask,
2260  * no dst, wrong address families, etc...) we need to pass it back
2261  * to app (and any other rtsock consumers) with rtm_errno field set to
2262  * non-zero value.
2263  * Returns 0 on success.
2264  */
2265 int
2266 rt_routemsg(int cmd, struct ifnet *ifp, int error, struct rtentry *rt,
2267     int fibnum)
2268 {
2269 
2270 	KASSERT(cmd == RTM_ADD || cmd == RTM_DELETE,
2271 	    ("unexpected cmd %d", cmd));
2272 
2273 	KASSERT(fibnum == RT_ALL_FIBS || (fibnum >= 0 && fibnum < rt_numfibs),
2274 	    ("%s: fib out of range 0 <=%d<%d", __func__, fibnum, rt_numfibs));
2275 
2276 	KASSERT(rt_key(rt) != NULL, (":%s: rt_key must be supplied", __func__));
2277 
2278 	return (rtsock_routemsg(cmd, ifp, error, rt, fibnum));
2279 }
2280 
2281 void
2282 rt_newaddrmsg(int cmd, struct ifaddr *ifa, int error, struct rtentry *rt)
2283 {
2284 
2285 	rt_newaddrmsg_fib(cmd, ifa, error, rt, RT_ALL_FIBS);
2286 }
2287 
2288 /*
2289  * This is called to generate messages from the routing socket
2290  * indicating a network interface has had addresses associated with it.
2291  */
2292 void
2293 rt_newaddrmsg_fib(int cmd, struct ifaddr *ifa, int error, struct rtentry *rt,
2294     int fibnum)
2295 {
2296 
2297 	KASSERT(cmd == RTM_ADD || cmd == RTM_DELETE,
2298 		("unexpected cmd %u", cmd));
2299 	KASSERT(fibnum == RT_ALL_FIBS || (fibnum >= 0 && fibnum < rt_numfibs),
2300 	    ("%s: fib out of range 0 <=%d<%d", __func__, fibnum, rt_numfibs));
2301 
2302 	if (cmd == RTM_ADD) {
2303 		rt_addrmsg(cmd, ifa, fibnum);
2304 		if (rt != NULL)
2305 			rt_routemsg(cmd, ifa->ifa_ifp, error, rt, fibnum);
2306 	} else {
2307 		if (rt != NULL)
2308 			rt_routemsg(cmd, ifa->ifa_ifp, error, rt, fibnum);
2309 		rt_addrmsg(cmd, ifa, fibnum);
2310 	}
2311 }
2312 
2313