xref: /freebsd/sys/net/route.c (revision f18976136625a7d016e97bfd9eabddf640b3e06d)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1980, 1986, 1991, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  * 3. Neither the name of the University nor the names of its contributors
16  *    may be used to endorse or promote products derived from this software
17  *    without specific prior written permission.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29  * SUCH DAMAGE.
30  *
31  *	@(#)route.c	8.3.1.1 (Berkeley) 2/23/95
32  * $FreeBSD$
33  */
34 /************************************************************************
35  * Note: In this file a 'fib' is a "forwarding information base"	*
36  * Which is the new name for an in kernel routing (next hop) table.	*
37  ***********************************************************************/
38 
39 #include "opt_inet.h"
40 #include "opt_inet6.h"
41 #include "opt_route.h"
42 #include "opt_sctp.h"
43 #include "opt_mrouting.h"
44 #include "opt_mpath.h"
45 
46 #include <sys/param.h>
47 #include <sys/systm.h>
48 #include <sys/malloc.h>
49 #include <sys/mbuf.h>
50 #include <sys/socket.h>
51 #include <sys/sysctl.h>
52 #include <sys/syslog.h>
53 #include <sys/sysproto.h>
54 #include <sys/proc.h>
55 #include <sys/domain.h>
56 #include <sys/kernel.h>
57 #include <sys/lock.h>
58 #include <sys/rmlock.h>
59 
60 #include <net/if.h>
61 #include <net/if_var.h>
62 #include <net/if_dl.h>
63 #include <net/route.h>
64 #include <net/route_var.h>
65 #include <net/vnet.h>
66 
67 #ifdef RADIX_MPATH
68 #include <net/radix_mpath.h>
69 #endif
70 
71 #include <netinet/in.h>
72 #include <netinet/ip_mroute.h>
73 
74 #include <vm/uma.h>
75 
76 #define	RT_MAXFIBS	UINT16_MAX
77 
78 /* Kernel config default option. */
79 #ifdef ROUTETABLES
80 #if ROUTETABLES <= 0
81 #error "ROUTETABLES defined too low"
82 #endif
83 #if ROUTETABLES > RT_MAXFIBS
84 #error "ROUTETABLES defined too big"
85 #endif
86 #define	RT_NUMFIBS	ROUTETABLES
87 #endif /* ROUTETABLES */
88 /* Initialize to default if not otherwise set. */
89 #ifndef	RT_NUMFIBS
90 #define	RT_NUMFIBS	1
91 #endif
92 
93 #if defined(INET) || defined(INET6)
94 #ifdef SCTP
95 extern void sctp_addr_change(struct ifaddr *ifa, int cmd);
96 #endif /* SCTP */
97 #endif
98 
99 
100 /* This is read-only.. */
101 u_int rt_numfibs = RT_NUMFIBS;
102 SYSCTL_UINT(_net, OID_AUTO, fibs, CTLFLAG_RDTUN, &rt_numfibs, 0, "");
103 
104 /*
105  * By default add routes to all fibs for new interfaces.
106  * Once this is set to 0 then only allocate routes on interface
107  * changes for the FIB of the caller when adding a new set of addresses
108  * to an interface.  XXX this is a shotgun aproach to a problem that needs
109  * a more fine grained solution.. that will come.
110  * XXX also has the problems getting the FIB from curthread which will not
111  * always work given the fib can be overridden and prefixes can be added
112  * from the network stack context.
113  */
114 VNET_DEFINE(u_int, rt_add_addr_allfibs) = 1;
115 SYSCTL_UINT(_net, OID_AUTO, add_addr_allfibs, CTLFLAG_RWTUN | CTLFLAG_VNET,
116     &VNET_NAME(rt_add_addr_allfibs), 0, "");
117 
118 VNET_DEFINE(struct rtstat, rtstat);
119 #define	V_rtstat	VNET(rtstat)
120 
121 VNET_DEFINE(struct rib_head *, rt_tables);
122 #define	V_rt_tables	VNET(rt_tables)
123 
124 VNET_DEFINE(int, rttrash);		/* routes not in table but not freed */
125 #define	V_rttrash	VNET(rttrash)
126 
127 
128 /*
129  * Convert a 'struct radix_node *' to a 'struct rtentry *'.
130  * The operation can be done safely (in this code) because a
131  * 'struct rtentry' starts with two 'struct radix_node''s, the first
132  * one representing leaf nodes in the routing tree, which is
133  * what the code in radix.c passes us as a 'struct radix_node'.
134  *
135  * But because there are a lot of assumptions in this conversion,
136  * do not cast explicitly, but always use the macro below.
137  */
138 #define RNTORT(p)	((struct rtentry *)(p))
139 
140 VNET_DEFINE_STATIC(uma_zone_t, rtzone);		/* Routing table UMA zone. */
141 #define	V_rtzone	VNET(rtzone)
142 
143 static int rtrequest1_fib_change(struct rib_head *, struct rt_addrinfo *,
144     struct rtentry **, u_int);
145 static void rt_setmetrics(const struct rt_addrinfo *, struct rtentry *);
146 static int rt_ifdelroute(const struct rtentry *rt, void *arg);
147 static struct rtentry *rt_unlinkrte(struct rib_head *rnh,
148     struct rt_addrinfo *info, int *perror);
149 static void rt_notifydelete(struct rtentry *rt, struct rt_addrinfo *info);
150 #ifdef RADIX_MPATH
151 static struct radix_node *rt_mpath_unlink(struct rib_head *rnh,
152     struct rt_addrinfo *info, struct rtentry *rto, int *perror);
153 #endif
154 static int rt_exportinfo(struct rtentry *rt, struct rt_addrinfo *info,
155     int flags);
156 
157 struct if_mtuinfo
158 {
159 	struct ifnet	*ifp;
160 	int		mtu;
161 };
162 
163 static int	if_updatemtu_cb(struct radix_node *, void *);
164 
165 /*
166  * handler for net.my_fibnum
167  */
168 static int
169 sysctl_my_fibnum(SYSCTL_HANDLER_ARGS)
170 {
171         int fibnum;
172         int error;
173 
174         fibnum = curthread->td_proc->p_fibnum;
175         error = sysctl_handle_int(oidp, &fibnum, 0, req);
176         return (error);
177 }
178 
179 SYSCTL_PROC(_net, OID_AUTO, my_fibnum, CTLTYPE_INT|CTLFLAG_RD,
180             NULL, 0, &sysctl_my_fibnum, "I", "default FIB of caller");
181 
182 static __inline struct rib_head **
183 rt_tables_get_rnh_ptr(int table, int fam)
184 {
185 	struct rib_head **rnh;
186 
187 	KASSERT(table >= 0 && table < rt_numfibs, ("%s: table out of bounds.",
188 	    __func__));
189 	KASSERT(fam >= 0 && fam < (AF_MAX+1), ("%s: fam out of bounds.",
190 	    __func__));
191 
192 	/* rnh is [fib=0][af=0]. */
193 	rnh = (struct rib_head **)V_rt_tables;
194 	/* Get the offset to the requested table and fam. */
195 	rnh += table * (AF_MAX+1) + fam;
196 
197 	return (rnh);
198 }
199 
200 struct rib_head *
201 rt_tables_get_rnh(int table, int fam)
202 {
203 
204 	return (*rt_tables_get_rnh_ptr(table, fam));
205 }
206 
207 u_int
208 rt_tables_get_gen(int table, int fam)
209 {
210 	struct rib_head *rnh;
211 
212 	rnh = *rt_tables_get_rnh_ptr(table, fam);
213 	KASSERT(rnh != NULL, ("%s: NULL rib_head pointer table %d fam %d",
214 	    __func__, table, fam));
215 	return (rnh->rnh_gen);
216 }
217 
218 
219 /*
220  * route initialization must occur before ip6_init2(), which happenas at
221  * SI_ORDER_MIDDLE.
222  */
223 static void
224 route_init(void)
225 {
226 
227 	/* whack the tunable ints into  line. */
228 	if (rt_numfibs > RT_MAXFIBS)
229 		rt_numfibs = RT_MAXFIBS;
230 	if (rt_numfibs == 0)
231 		rt_numfibs = 1;
232 }
233 SYSINIT(route_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD, route_init, NULL);
234 
235 static int
236 rtentry_zinit(void *mem, int size, int how)
237 {
238 	struct rtentry *rt = mem;
239 
240 	rt->rt_pksent = counter_u64_alloc(how);
241 	if (rt->rt_pksent == NULL)
242 		return (ENOMEM);
243 
244 	RT_LOCK_INIT(rt);
245 
246 	return (0);
247 }
248 
249 static void
250 rtentry_zfini(void *mem, int size)
251 {
252 	struct rtentry *rt = mem;
253 
254 	RT_LOCK_DESTROY(rt);
255 	counter_u64_free(rt->rt_pksent);
256 }
257 
258 static int
259 rtentry_ctor(void *mem, int size, void *arg, int how)
260 {
261 	struct rtentry *rt = mem;
262 
263 	bzero(rt, offsetof(struct rtentry, rt_endzero));
264 	counter_u64_zero(rt->rt_pksent);
265 	rt->rt_chain = NULL;
266 
267 	return (0);
268 }
269 
270 static void
271 rtentry_dtor(void *mem, int size, void *arg)
272 {
273 	struct rtentry *rt = mem;
274 
275 	RT_UNLOCK_COND(rt);
276 }
277 
278 static void
279 vnet_route_init(const void *unused __unused)
280 {
281 	struct domain *dom;
282 	struct rib_head **rnh;
283 	int table;
284 	int fam;
285 
286 	V_rt_tables = malloc(rt_numfibs * (AF_MAX+1) *
287 	    sizeof(struct rib_head *), M_RTABLE, M_WAITOK|M_ZERO);
288 
289 	V_rtzone = uma_zcreate("rtentry", sizeof(struct rtentry),
290 	    rtentry_ctor, rtentry_dtor,
291 	    rtentry_zinit, rtentry_zfini, UMA_ALIGN_PTR, 0);
292 	for (dom = domains; dom; dom = dom->dom_next) {
293 		if (dom->dom_rtattach == NULL)
294 			continue;
295 
296 		for  (table = 0; table < rt_numfibs; table++) {
297 			fam = dom->dom_family;
298 			if (table != 0 && fam != AF_INET6 && fam != AF_INET)
299 				break;
300 
301 			rnh = rt_tables_get_rnh_ptr(table, fam);
302 			if (rnh == NULL)
303 				panic("%s: rnh NULL", __func__);
304 			dom->dom_rtattach((void **)rnh, 0);
305 		}
306 	}
307 }
308 VNET_SYSINIT(vnet_route_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH,
309     vnet_route_init, 0);
310 
311 #ifdef VIMAGE
312 static void
313 vnet_route_uninit(const void *unused __unused)
314 {
315 	int table;
316 	int fam;
317 	struct domain *dom;
318 	struct rib_head **rnh;
319 
320 	for (dom = domains; dom; dom = dom->dom_next) {
321 		if (dom->dom_rtdetach == NULL)
322 			continue;
323 
324 		for (table = 0; table < rt_numfibs; table++) {
325 			fam = dom->dom_family;
326 
327 			if (table != 0 && fam != AF_INET6 && fam != AF_INET)
328 				break;
329 
330 			rnh = rt_tables_get_rnh_ptr(table, fam);
331 			if (rnh == NULL)
332 				panic("%s: rnh NULL", __func__);
333 			dom->dom_rtdetach((void **)rnh, 0);
334 		}
335 	}
336 
337 	free(V_rt_tables, M_RTABLE);
338 	uma_zdestroy(V_rtzone);
339 }
340 VNET_SYSUNINIT(vnet_route_uninit, SI_SUB_PROTO_DOMAIN, SI_ORDER_FIRST,
341     vnet_route_uninit, 0);
342 #endif
343 
344 struct rib_head *
345 rt_table_init(int offset)
346 {
347 	struct rib_head *rh;
348 
349 	rh = malloc(sizeof(struct rib_head), M_RTABLE, M_WAITOK | M_ZERO);
350 
351 	/* TODO: These details should be hidded inside radix.c */
352 	/* Init masks tree */
353 	rn_inithead_internal(&rh->head, rh->rnh_nodes, offset);
354 	rn_inithead_internal(&rh->rmhead.head, rh->rmhead.mask_nodes, 0);
355 	rh->head.rnh_masks = &rh->rmhead;
356 
357 	/* Init locks */
358 	RIB_LOCK_INIT(rh);
359 
360 	/* Finally, set base callbacks */
361 	rh->rnh_addaddr = rn_addroute;
362 	rh->rnh_deladdr = rn_delete;
363 	rh->rnh_matchaddr = rn_match;
364 	rh->rnh_lookup = rn_lookup;
365 	rh->rnh_walktree = rn_walktree;
366 	rh->rnh_walktree_from = rn_walktree_from;
367 
368 	return (rh);
369 }
370 
371 static int
372 rt_freeentry(struct radix_node *rn, void *arg)
373 {
374 	struct radix_head * const rnh = arg;
375 	struct radix_node *x;
376 
377 	x = (struct radix_node *)rn_delete(rn + 2, NULL, rnh);
378 	if (x != NULL)
379 		R_Free(x);
380 	return (0);
381 }
382 
383 void
384 rt_table_destroy(struct rib_head *rh)
385 {
386 
387 	rn_walktree(&rh->rmhead.head, rt_freeentry, &rh->rmhead.head);
388 
389 	/* Assume table is already empty */
390 	RIB_LOCK_DESTROY(rh);
391 	free(rh, M_RTABLE);
392 }
393 
394 
395 #ifndef _SYS_SYSPROTO_H_
396 struct setfib_args {
397 	int     fibnum;
398 };
399 #endif
400 int
401 sys_setfib(struct thread *td, struct setfib_args *uap)
402 {
403 	if (uap->fibnum < 0 || uap->fibnum >= rt_numfibs)
404 		return EINVAL;
405 	td->td_proc->p_fibnum = uap->fibnum;
406 	return (0);
407 }
408 
409 /*
410  * Packet routing routines.
411  */
412 void
413 rtalloc_ign_fib(struct route *ro, u_long ignore, u_int fibnum)
414 {
415 	struct rtentry *rt;
416 
417 	if ((rt = ro->ro_rt) != NULL) {
418 		if (rt->rt_ifp != NULL && rt->rt_flags & RTF_UP)
419 			return;
420 		RTFREE(rt);
421 		ro->ro_rt = NULL;
422 	}
423 	ro->ro_rt = rtalloc1_fib(&ro->ro_dst, 1, ignore, fibnum);
424 	if (ro->ro_rt)
425 		RT_UNLOCK(ro->ro_rt);
426 }
427 
428 /*
429  * Look up the route that matches the address given
430  * Or, at least try.. Create a cloned route if needed.
431  *
432  * The returned route, if any, is locked.
433  */
434 struct rtentry *
435 rtalloc1(struct sockaddr *dst, int report, u_long ignflags)
436 {
437 
438 	return (rtalloc1_fib(dst, report, ignflags, RT_DEFAULT_FIB));
439 }
440 
441 struct rtentry *
442 rtalloc1_fib(struct sockaddr *dst, int report, u_long ignflags,
443 		    u_int fibnum)
444 {
445 	RIB_RLOCK_TRACKER;
446 	struct rib_head *rh;
447 	struct radix_node *rn;
448 	struct rtentry *newrt;
449 	struct rt_addrinfo info;
450 	int err = 0, msgtype = RTM_MISS;
451 
452 	KASSERT((fibnum < rt_numfibs), ("rtalloc1_fib: bad fibnum"));
453 	rh = rt_tables_get_rnh(fibnum, dst->sa_family);
454 	newrt = NULL;
455 	if (rh == NULL)
456 		goto miss;
457 
458 	/*
459 	 * Look up the address in the table for that Address Family
460 	 */
461 	if ((ignflags & RTF_RNH_LOCKED) == 0)
462 		RIB_RLOCK(rh);
463 #ifdef INVARIANTS
464 	else
465 		RIB_LOCK_ASSERT(rh);
466 #endif
467 	rn = rh->rnh_matchaddr(dst, &rh->head);
468 	if (rn && ((rn->rn_flags & RNF_ROOT) == 0)) {
469 		newrt = RNTORT(rn);
470 		RT_LOCK(newrt);
471 		RT_ADDREF(newrt);
472 		if ((ignflags & RTF_RNH_LOCKED) == 0)
473 			RIB_RUNLOCK(rh);
474 		return (newrt);
475 
476 	} else if ((ignflags & RTF_RNH_LOCKED) == 0)
477 		RIB_RUNLOCK(rh);
478 	/*
479 	 * Either we hit the root or could not find any match,
480 	 * which basically means: "cannot get there from here".
481 	 */
482 miss:
483 	V_rtstat.rts_unreach++;
484 
485 	if (report) {
486 		/*
487 		 * If required, report the failure to the supervising
488 		 * Authorities.
489 		 * For a delete, this is not an error. (report == 0)
490 		 */
491 		bzero(&info, sizeof(info));
492 		info.rti_info[RTAX_DST] = dst;
493 		rt_missmsg_fib(msgtype, &info, 0, err, fibnum);
494 	}
495 	return (newrt);
496 }
497 
498 /*
499  * Remove a reference count from an rtentry.
500  * If the count gets low enough, take it out of the routing table
501  */
502 void
503 rtfree(struct rtentry *rt)
504 {
505 	struct rib_head *rnh;
506 
507 	KASSERT(rt != NULL,("%s: NULL rt", __func__));
508 	rnh = rt_tables_get_rnh(rt->rt_fibnum, rt_key(rt)->sa_family);
509 	KASSERT(rnh != NULL,("%s: NULL rnh", __func__));
510 
511 	RT_LOCK_ASSERT(rt);
512 
513 	/*
514 	 * The callers should use RTFREE_LOCKED() or RTFREE(), so
515 	 * we should come here exactly with the last reference.
516 	 */
517 	RT_REMREF(rt);
518 	if (rt->rt_refcnt > 0) {
519 		log(LOG_DEBUG, "%s: %p has %d refs\n", __func__, rt, rt->rt_refcnt);
520 		goto done;
521 	}
522 
523 	/*
524 	 * On last reference give the "close method" a chance
525 	 * to cleanup private state.  This also permits (for
526 	 * IPv4 and IPv6) a chance to decide if the routing table
527 	 * entry should be purged immediately or at a later time.
528 	 * When an immediate purge is to happen the close routine
529 	 * typically calls rtexpunge which clears the RTF_UP flag
530 	 * on the entry so that the code below reclaims the storage.
531 	 */
532 	if (rt->rt_refcnt == 0 && rnh->rnh_close)
533 		rnh->rnh_close((struct radix_node *)rt, &rnh->head);
534 
535 	/*
536 	 * If we are no longer "up" (and ref == 0)
537 	 * then we can free the resources associated
538 	 * with the route.
539 	 */
540 	if ((rt->rt_flags & RTF_UP) == 0) {
541 		if (rt->rt_nodes->rn_flags & (RNF_ACTIVE | RNF_ROOT))
542 			panic("rtfree 2");
543 		/*
544 		 * the rtentry must have been removed from the routing table
545 		 * so it is represented in rttrash.. remove that now.
546 		 */
547 		V_rttrash--;
548 #ifdef	DIAGNOSTIC
549 		if (rt->rt_refcnt < 0) {
550 			printf("rtfree: %p not freed (neg refs)\n", rt);
551 			goto done;
552 		}
553 #endif
554 		/*
555 		 * release references on items we hold them on..
556 		 * e.g other routes and ifaddrs.
557 		 */
558 		if (rt->rt_ifa)
559 			ifa_free(rt->rt_ifa);
560 		/*
561 		 * The key is separatly alloc'd so free it (see rt_setgate()).
562 		 * This also frees the gateway, as they are always malloc'd
563 		 * together.
564 		 */
565 		R_Free(rt_key(rt));
566 
567 		/*
568 		 * and the rtentry itself of course
569 		 */
570 		uma_zfree(V_rtzone, rt);
571 		return;
572 	}
573 done:
574 	RT_UNLOCK(rt);
575 }
576 
577 
578 /*
579  * Force a routing table entry to the specified
580  * destination to go through the given gateway.
581  * Normally called as a result of a routing redirect
582  * message from the network layer.
583  */
584 void
585 rtredirect_fib(struct sockaddr *dst,
586 	struct sockaddr *gateway,
587 	struct sockaddr *netmask,
588 	int flags,
589 	struct sockaddr *src,
590 	u_int fibnum)
591 {
592 	struct rtentry *rt;
593 	int error = 0;
594 	short *stat = NULL;
595 	struct rt_addrinfo info;
596 	struct ifaddr *ifa;
597 	struct rib_head *rnh;
598 
599 	NET_EPOCH_ASSERT();
600 
601 	ifa = NULL;
602 	rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
603 	if (rnh == NULL) {
604 		error = EAFNOSUPPORT;
605 		goto out;
606 	}
607 	/* verify the gateway is directly reachable */
608 	if ((ifa = ifa_ifwithnet(gateway, 0, fibnum)) == NULL) {
609 		error = ENETUNREACH;
610 		goto out;
611 	}
612 	rt = rtalloc1_fib(dst, 0, 0UL, fibnum);	/* NB: rt is locked */
613 	/*
614 	 * If the redirect isn't from our current router for this dst,
615 	 * it's either old or wrong.  If it redirects us to ourselves,
616 	 * we have a routing loop, perhaps as a result of an interface
617 	 * going down recently.
618 	 */
619 	if (!(flags & RTF_DONE) && rt) {
620 		if (!sa_equal(src, rt->rt_gateway)) {
621 			error = EINVAL;
622 			goto done;
623 		}
624 		if (rt->rt_ifa != ifa && ifa->ifa_addr->sa_family != AF_LINK) {
625 			error = EINVAL;
626 			goto done;
627 		}
628 	}
629 	if ((flags & RTF_GATEWAY) && ifa_ifwithaddr_check(gateway)) {
630 		error = EHOSTUNREACH;
631 		goto done;
632 	}
633 	/*
634 	 * Create a new entry if we just got back a wildcard entry
635 	 * or the lookup failed.  This is necessary for hosts
636 	 * which use routing redirects generated by smart gateways
637 	 * to dynamically build the routing tables.
638 	 */
639 	if (rt == NULL || (rt_mask(rt) && rt_mask(rt)->sa_len < 2))
640 		goto create;
641 	/*
642 	 * Don't listen to the redirect if it's
643 	 * for a route to an interface.
644 	 */
645 	if (rt->rt_flags & RTF_GATEWAY) {
646 		if (((rt->rt_flags & RTF_HOST) == 0) && (flags & RTF_HOST)) {
647 			/*
648 			 * Changing from route to net => route to host.
649 			 * Create new route, rather than smashing route to net.
650 			 */
651 		create:
652 			if (rt != NULL)
653 				RTFREE_LOCKED(rt);
654 
655 			flags |= RTF_DYNAMIC;
656 			bzero((caddr_t)&info, sizeof(info));
657 			info.rti_info[RTAX_DST] = dst;
658 			info.rti_info[RTAX_GATEWAY] = gateway;
659 			info.rti_info[RTAX_NETMASK] = netmask;
660 			ifa_ref(ifa);
661 			info.rti_ifa = ifa;
662 			info.rti_flags = flags;
663 			error = rtrequest1_fib(RTM_ADD, &info, &rt, fibnum);
664 			if (rt != NULL) {
665 				RT_LOCK(rt);
666 				flags = rt->rt_flags;
667 			}
668 
669 			stat = &V_rtstat.rts_dynamic;
670 		} else {
671 
672 			/*
673 			 * Smash the current notion of the gateway to
674 			 * this destination.  Should check about netmask!!!
675 			 */
676 			if ((flags & RTF_GATEWAY) == 0)
677 				rt->rt_flags &= ~RTF_GATEWAY;
678 			rt->rt_flags |= RTF_MODIFIED;
679 			flags |= RTF_MODIFIED;
680 			stat = &V_rtstat.rts_newgateway;
681 			/*
682 			 * add the key and gateway (in one malloc'd chunk).
683 			 */
684 			RT_UNLOCK(rt);
685 			RIB_WLOCK(rnh);
686 			RT_LOCK(rt);
687 			rt_setgate(rt, rt_key(rt), gateway);
688 			RIB_WUNLOCK(rnh);
689 		}
690 	} else
691 		error = EHOSTUNREACH;
692 done:
693 	if (rt)
694 		RTFREE_LOCKED(rt);
695  out:
696 	if (error)
697 		V_rtstat.rts_badredirect++;
698 	else if (stat != NULL)
699 		(*stat)++;
700 	bzero((caddr_t)&info, sizeof(info));
701 	info.rti_info[RTAX_DST] = dst;
702 	info.rti_info[RTAX_GATEWAY] = gateway;
703 	info.rti_info[RTAX_NETMASK] = netmask;
704 	info.rti_info[RTAX_AUTHOR] = src;
705 	rt_missmsg_fib(RTM_REDIRECT, &info, flags, error, fibnum);
706 }
707 
708 /*
709  * Routing table ioctl interface.
710  */
711 int
712 rtioctl_fib(u_long req, caddr_t data, u_int fibnum)
713 {
714 
715 	/*
716 	 * If more ioctl commands are added here, make sure the proper
717 	 * super-user checks are being performed because it is possible for
718 	 * prison-root to make it this far if raw sockets have been enabled
719 	 * in jails.
720 	 */
721 #ifdef INET
722 	/* Multicast goop, grrr... */
723 	return mrt_ioctl ? mrt_ioctl(req, data, fibnum) : EOPNOTSUPP;
724 #else /* INET */
725 	return ENXIO;
726 #endif /* INET */
727 }
728 
729 struct ifaddr *
730 ifa_ifwithroute(int flags, const struct sockaddr *dst, struct sockaddr *gateway,
731 				u_int fibnum)
732 {
733 	struct ifaddr *ifa;
734 	int not_found = 0;
735 
736 	MPASS(in_epoch(net_epoch_preempt));
737 	if ((flags & RTF_GATEWAY) == 0) {
738 		/*
739 		 * If we are adding a route to an interface,
740 		 * and the interface is a pt to pt link
741 		 * we should search for the destination
742 		 * as our clue to the interface.  Otherwise
743 		 * we can use the local address.
744 		 */
745 		ifa = NULL;
746 		if (flags & RTF_HOST)
747 			ifa = ifa_ifwithdstaddr(dst, fibnum);
748 		if (ifa == NULL)
749 			ifa = ifa_ifwithaddr(gateway);
750 	} else {
751 		/*
752 		 * If we are adding a route to a remote net
753 		 * or host, the gateway may still be on the
754 		 * other end of a pt to pt link.
755 		 */
756 		ifa = ifa_ifwithdstaddr(gateway, fibnum);
757 	}
758 	if (ifa == NULL)
759 		ifa = ifa_ifwithnet(gateway, 0, fibnum);
760 	if (ifa == NULL) {
761 		struct rtentry *rt;
762 
763 		rt = rtalloc1_fib(gateway, 0, flags, fibnum);
764 		if (rt == NULL)
765 			goto out;
766 		/*
767 		 * dismiss a gateway that is reachable only
768 		 * through the default router
769 		 */
770 		switch (gateway->sa_family) {
771 		case AF_INET:
772 			if (satosin(rt_key(rt))->sin_addr.s_addr == INADDR_ANY)
773 				not_found = 1;
774 			break;
775 		case AF_INET6:
776 			if (IN6_IS_ADDR_UNSPECIFIED(&satosin6(rt_key(rt))->sin6_addr))
777 				not_found = 1;
778 			break;
779 		default:
780 			break;
781 		}
782 		if (!not_found && rt->rt_ifa != NULL) {
783 			ifa = rt->rt_ifa;
784 		}
785 		RT_REMREF(rt);
786 		RT_UNLOCK(rt);
787 		if (not_found || ifa == NULL)
788 			goto out;
789 	}
790 	if (ifa->ifa_addr->sa_family != dst->sa_family) {
791 		struct ifaddr *oifa = ifa;
792 		ifa = ifaof_ifpforaddr(dst, ifa->ifa_ifp);
793 		if (ifa == NULL)
794 			ifa = oifa;
795 	}
796  out:
797 	return (ifa);
798 }
799 
800 /*
801  * Do appropriate manipulations of a routing tree given
802  * all the bits of info needed
803  */
804 int
805 rtrequest_fib(int req,
806 	struct sockaddr *dst,
807 	struct sockaddr *gateway,
808 	struct sockaddr *netmask,
809 	int flags,
810 	struct rtentry **ret_nrt,
811 	u_int fibnum)
812 {
813 	struct rt_addrinfo info;
814 
815 	if (dst->sa_len == 0)
816 		return(EINVAL);
817 
818 	bzero((caddr_t)&info, sizeof(info));
819 	info.rti_flags = flags;
820 	info.rti_info[RTAX_DST] = dst;
821 	info.rti_info[RTAX_GATEWAY] = gateway;
822 	info.rti_info[RTAX_NETMASK] = netmask;
823 	return rtrequest1_fib(req, &info, ret_nrt, fibnum);
824 }
825 
826 
827 /*
828  * Copy most of @rt data into @info.
829  *
830  * If @flags contains NHR_COPY, copies dst,netmask and gw to the
831  * pointers specified by @info structure. Assume such pointers
832  * are zeroed sockaddr-like structures with sa_len field initialized
833  * to reflect size of the provided buffer. if no NHR_COPY is specified,
834  * point dst,netmask and gw @info fields to appropriate @rt values.
835  *
836  * if @flags contains NHR_REF, do refcouting on rt_ifp.
837  *
838  * Returns 0 on success.
839  */
840 int
841 rt_exportinfo(struct rtentry *rt, struct rt_addrinfo *info, int flags)
842 {
843 	struct rt_metrics *rmx;
844 	struct sockaddr *src, *dst;
845 	int sa_len;
846 
847 	if (flags & NHR_COPY) {
848 		/* Copy destination if dst is non-zero */
849 		src = rt_key(rt);
850 		dst = info->rti_info[RTAX_DST];
851 		sa_len = src->sa_len;
852 		if (dst != NULL) {
853 			if (src->sa_len > dst->sa_len)
854 				return (ENOMEM);
855 			memcpy(dst, src, src->sa_len);
856 			info->rti_addrs |= RTA_DST;
857 		}
858 
859 		/* Copy mask if set && dst is non-zero */
860 		src = rt_mask(rt);
861 		dst = info->rti_info[RTAX_NETMASK];
862 		if (src != NULL && dst != NULL) {
863 
864 			/*
865 			 * Radix stores different value in sa_len,
866 			 * assume rt_mask() to have the same length
867 			 * as rt_key()
868 			 */
869 			if (sa_len > dst->sa_len)
870 				return (ENOMEM);
871 			memcpy(dst, src, src->sa_len);
872 			info->rti_addrs |= RTA_NETMASK;
873 		}
874 
875 		/* Copy gateway is set && dst is non-zero */
876 		src = rt->rt_gateway;
877 		dst = info->rti_info[RTAX_GATEWAY];
878 		if ((rt->rt_flags & RTF_GATEWAY) && src != NULL && dst != NULL){
879 			if (src->sa_len > dst->sa_len)
880 				return (ENOMEM);
881 			memcpy(dst, src, src->sa_len);
882 			info->rti_addrs |= RTA_GATEWAY;
883 		}
884 	} else {
885 		info->rti_info[RTAX_DST] = rt_key(rt);
886 		info->rti_addrs |= RTA_DST;
887 		if (rt_mask(rt) != NULL) {
888 			info->rti_info[RTAX_NETMASK] = rt_mask(rt);
889 			info->rti_addrs |= RTA_NETMASK;
890 		}
891 		if (rt->rt_flags & RTF_GATEWAY) {
892 			info->rti_info[RTAX_GATEWAY] = rt->rt_gateway;
893 			info->rti_addrs |= RTA_GATEWAY;
894 		}
895 	}
896 
897 	rmx = info->rti_rmx;
898 	if (rmx != NULL) {
899 		info->rti_mflags |= RTV_MTU;
900 		rmx->rmx_mtu = rt->rt_mtu;
901 	}
902 
903 	info->rti_flags = rt->rt_flags;
904 	info->rti_ifp = rt->rt_ifp;
905 	info->rti_ifa = rt->rt_ifa;
906 	ifa_ref(info->rti_ifa);
907 	if (flags & NHR_REF) {
908 		/* Do 'traditional' refcouting */
909 		if_ref(info->rti_ifp);
910 	}
911 
912 	return (0);
913 }
914 
915 /*
916  * Lookups up route entry for @dst in RIB database for fib @fibnum.
917  * Exports entry data to @info using rt_exportinfo().
918  *
919  * if @flags contains NHR_REF, refcouting is performed on rt_ifp.
920  *   All references can be released later by calling rib_free_info()
921  *
922  * Returns 0 on success.
923  * Returns ENOENT for lookup failure, ENOMEM for export failure.
924  */
925 int
926 rib_lookup_info(uint32_t fibnum, const struct sockaddr *dst, uint32_t flags,
927     uint32_t flowid, struct rt_addrinfo *info)
928 {
929 	RIB_RLOCK_TRACKER;
930 	struct rib_head *rh;
931 	struct radix_node *rn;
932 	struct rtentry *rt;
933 	int error;
934 
935 	KASSERT((fibnum < rt_numfibs), ("rib_lookup_rte: bad fibnum"));
936 	rh = rt_tables_get_rnh(fibnum, dst->sa_family);
937 	if (rh == NULL)
938 		return (ENOENT);
939 
940 	RIB_RLOCK(rh);
941 	rn = rh->rnh_matchaddr(__DECONST(void *, dst), &rh->head);
942 	if (rn != NULL && ((rn->rn_flags & RNF_ROOT) == 0)) {
943 		rt = RNTORT(rn);
944 		/* Ensure route & ifp is UP */
945 		if (RT_LINK_IS_UP(rt->rt_ifp)) {
946 			flags = (flags & NHR_REF) | NHR_COPY;
947 			error = rt_exportinfo(rt, info, flags);
948 			RIB_RUNLOCK(rh);
949 
950 			return (error);
951 		}
952 	}
953 	RIB_RUNLOCK(rh);
954 
955 	return (ENOENT);
956 }
957 
958 /*
959  * Releases all references acquired by rib_lookup_info() when
960  * called with NHR_REF flags.
961  */
962 void
963 rib_free_info(struct rt_addrinfo *info)
964 {
965 
966 	if_rele(info->rti_ifp);
967 }
968 
969 /*
970  * Iterates over all existing fibs in system calling
971  *  @setwa_f function prior to traversing each fib.
972  *  Calls @wa_f function for each element in current fib.
973  * If af is not AF_UNSPEC, iterates over fibs in particular
974  * address family.
975  */
976 void
977 rt_foreach_fib_walk(int af, rt_setwarg_t *setwa_f, rt_walktree_f_t *wa_f,
978     void *arg)
979 {
980 	struct rib_head *rnh;
981 	uint32_t fibnum;
982 	int i;
983 
984 	for (fibnum = 0; fibnum < rt_numfibs; fibnum++) {
985 		/* Do we want some specific family? */
986 		if (af != AF_UNSPEC) {
987 			rnh = rt_tables_get_rnh(fibnum, af);
988 			if (rnh == NULL)
989 				continue;
990 			if (setwa_f != NULL)
991 				setwa_f(rnh, fibnum, af, arg);
992 
993 			RIB_WLOCK(rnh);
994 			rnh->rnh_walktree(&rnh->head, (walktree_f_t *)wa_f,arg);
995 			RIB_WUNLOCK(rnh);
996 			continue;
997 		}
998 
999 		for (i = 1; i <= AF_MAX; i++) {
1000 			rnh = rt_tables_get_rnh(fibnum, i);
1001 			if (rnh == NULL)
1002 				continue;
1003 			if (setwa_f != NULL)
1004 				setwa_f(rnh, fibnum, i, arg);
1005 
1006 			RIB_WLOCK(rnh);
1007 			rnh->rnh_walktree(&rnh->head, (walktree_f_t *)wa_f,arg);
1008 			RIB_WUNLOCK(rnh);
1009 		}
1010 	}
1011 }
1012 
1013 struct rt_delinfo
1014 {
1015 	struct rt_addrinfo info;
1016 	struct rib_head *rnh;
1017 	struct rtentry *head;
1018 };
1019 
1020 /*
1021  * Conditionally unlinks @rn from radix tree based
1022  * on info data passed in @arg.
1023  */
1024 static int
1025 rt_checkdelroute(struct radix_node *rn, void *arg)
1026 {
1027 	struct rt_delinfo *di;
1028 	struct rt_addrinfo *info;
1029 	struct rtentry *rt;
1030 	int error;
1031 
1032 	di = (struct rt_delinfo *)arg;
1033 	rt = (struct rtentry *)rn;
1034 	info = &di->info;
1035 	error = 0;
1036 
1037 	info->rti_info[RTAX_DST] = rt_key(rt);
1038 	info->rti_info[RTAX_NETMASK] = rt_mask(rt);
1039 	info->rti_info[RTAX_GATEWAY] = rt->rt_gateway;
1040 
1041 	rt = rt_unlinkrte(di->rnh, info, &error);
1042 	if (rt == NULL) {
1043 		/* Either not allowed or not matched. Skip entry */
1044 		return (0);
1045 	}
1046 
1047 	/* Entry was unlinked. Add to the list and return */
1048 	rt->rt_chain = di->head;
1049 	di->head = rt;
1050 
1051 	return (0);
1052 }
1053 
1054 /*
1055  * Iterates over all existing fibs in system.
1056  * Deletes each element for which @filter_f function returned
1057  * non-zero value.
1058  * If @af is not AF_UNSPEC, iterates over fibs in particular
1059  * address family.
1060  */
1061 void
1062 rt_foreach_fib_walk_del(int af, rt_filter_f_t *filter_f, void *arg)
1063 {
1064 	struct rib_head *rnh;
1065 	struct rt_delinfo di;
1066 	struct rtentry *rt;
1067 	uint32_t fibnum;
1068 	int i, start, end;
1069 
1070 	bzero(&di, sizeof(di));
1071 	di.info.rti_filter = filter_f;
1072 	di.info.rti_filterdata = arg;
1073 
1074 	for (fibnum = 0; fibnum < rt_numfibs; fibnum++) {
1075 		/* Do we want some specific family? */
1076 		if (af != AF_UNSPEC) {
1077 			start = af;
1078 			end = af;
1079 		} else {
1080 			start = 1;
1081 			end = AF_MAX;
1082 		}
1083 
1084 		for (i = start; i <= end; i++) {
1085 			rnh = rt_tables_get_rnh(fibnum, i);
1086 			if (rnh == NULL)
1087 				continue;
1088 			di.rnh = rnh;
1089 
1090 			RIB_WLOCK(rnh);
1091 			rnh->rnh_walktree(&rnh->head, rt_checkdelroute, &di);
1092 			RIB_WUNLOCK(rnh);
1093 
1094 			if (di.head == NULL)
1095 				continue;
1096 
1097 			/* We might have something to reclaim */
1098 			while (di.head != NULL) {
1099 				rt = di.head;
1100 				di.head = rt->rt_chain;
1101 				rt->rt_chain = NULL;
1102 
1103 				/* TODO std rt -> rt_addrinfo export */
1104 				di.info.rti_info[RTAX_DST] = rt_key(rt);
1105 				di.info.rti_info[RTAX_NETMASK] = rt_mask(rt);
1106 
1107 				rt_notifydelete(rt, &di.info);
1108 				RTFREE_LOCKED(rt);
1109 			}
1110 
1111 		}
1112 	}
1113 }
1114 
1115 /*
1116  * Delete Routes for a Network Interface
1117  *
1118  * Called for each routing entry via the rnh->rnh_walktree() call above
1119  * to delete all route entries referencing a detaching network interface.
1120  *
1121  * Arguments:
1122  *	rt	pointer to rtentry
1123  *	arg	argument passed to rnh->rnh_walktree() - detaching interface
1124  *
1125  * Returns:
1126  *	0	successful
1127  *	errno	failed - reason indicated
1128  */
1129 static int
1130 rt_ifdelroute(const struct rtentry *rt, void *arg)
1131 {
1132 	struct ifnet	*ifp = arg;
1133 
1134 	if (rt->rt_ifp != ifp)
1135 		return (0);
1136 
1137 	/*
1138 	 * Protect (sorta) against walktree recursion problems
1139 	 * with cloned routes
1140 	 */
1141 	if ((rt->rt_flags & RTF_UP) == 0)
1142 		return (0);
1143 
1144 	return (1);
1145 }
1146 
1147 /*
1148  * Delete all remaining routes using this interface
1149  * Unfortuneatly the only way to do this is to slog through
1150  * the entire routing table looking for routes which point
1151  * to this interface...oh well...
1152  */
1153 void
1154 rt_flushifroutes_af(struct ifnet *ifp, int af)
1155 {
1156 	KASSERT((af >= 1 && af <= AF_MAX), ("%s: af %d not >= 1 and <= %d",
1157 	    __func__, af, AF_MAX));
1158 
1159 	rt_foreach_fib_walk_del(af, rt_ifdelroute, ifp);
1160 }
1161 
1162 void
1163 rt_flushifroutes(struct ifnet *ifp)
1164 {
1165 
1166 	rt_foreach_fib_walk_del(AF_UNSPEC, rt_ifdelroute, ifp);
1167 }
1168 
1169 /*
1170  * Conditionally unlinks rtentry matching data inside @info from @rnh.
1171  * Returns unlinked, locked and referenced @rtentry on success,
1172  * Returns NULL and sets @perror to:
1173  * ESRCH - if prefix was not found,
1174  * EADDRINUSE - if trying to delete PINNED route without appropriate flag.
1175  * ENOENT - if supplied filter function returned 0 (not matched).
1176  */
1177 static struct rtentry *
1178 rt_unlinkrte(struct rib_head *rnh, struct rt_addrinfo *info, int *perror)
1179 {
1180 	struct sockaddr *dst, *netmask;
1181 	struct rtentry *rt;
1182 	struct radix_node *rn;
1183 
1184 	dst = info->rti_info[RTAX_DST];
1185 	netmask = info->rti_info[RTAX_NETMASK];
1186 
1187 	rt = (struct rtentry *)rnh->rnh_lookup(dst, netmask, &rnh->head);
1188 	if (rt == NULL) {
1189 		*perror = ESRCH;
1190 		return (NULL);
1191 	}
1192 
1193 	if ((info->rti_flags & RTF_PINNED) == 0) {
1194 		/* Check if target route can be deleted */
1195 		if (rt->rt_flags & RTF_PINNED) {
1196 			*perror = EADDRINUSE;
1197 			return (NULL);
1198 		}
1199 	}
1200 
1201 	if (info->rti_filter != NULL) {
1202 		if (info->rti_filter(rt, info->rti_filterdata) == 0) {
1203 			/* Not matched */
1204 			*perror = ENOENT;
1205 			return (NULL);
1206 		}
1207 
1208 		/*
1209 		 * Filter function requested rte deletion.
1210 		 * Ease the caller work by filling in remaining info
1211 		 * from that particular entry.
1212 		 */
1213 		info->rti_info[RTAX_GATEWAY] = rt->rt_gateway;
1214 	}
1215 
1216 	/*
1217 	 * Remove the item from the tree and return it.
1218 	 * Complain if it is not there and do no more processing.
1219 	 */
1220 	*perror = ESRCH;
1221 #ifdef RADIX_MPATH
1222 	if (rt_mpath_capable(rnh))
1223 		rn = rt_mpath_unlink(rnh, info, rt, perror);
1224 	else
1225 #endif
1226 	rn = rnh->rnh_deladdr(dst, netmask, &rnh->head);
1227 	if (rn == NULL)
1228 		return (NULL);
1229 
1230 	if (rn->rn_flags & (RNF_ACTIVE | RNF_ROOT))
1231 		panic ("rtrequest delete");
1232 
1233 	rt = RNTORT(rn);
1234 	RT_LOCK(rt);
1235 	RT_ADDREF(rt);
1236 	rt->rt_flags &= ~RTF_UP;
1237 
1238 	*perror = 0;
1239 
1240 	return (rt);
1241 }
1242 
1243 static void
1244 rt_notifydelete(struct rtentry *rt, struct rt_addrinfo *info)
1245 {
1246 	struct ifaddr *ifa;
1247 
1248 	/*
1249 	 * give the protocol a chance to keep things in sync.
1250 	 */
1251 	ifa = rt->rt_ifa;
1252 	if (ifa != NULL && ifa->ifa_rtrequest != NULL)
1253 		ifa->ifa_rtrequest(RTM_DELETE, rt, info);
1254 
1255 	/*
1256 	 * One more rtentry floating around that is not
1257 	 * linked to the routing table. rttrash will be decremented
1258 	 * when RTFREE(rt) is eventually called.
1259 	 */
1260 	V_rttrash++;
1261 }
1262 
1263 
1264 /*
1265  * These (questionable) definitions of apparent local variables apply
1266  * to the next two functions.  XXXXXX!!!
1267  */
1268 #define	dst	info->rti_info[RTAX_DST]
1269 #define	gateway	info->rti_info[RTAX_GATEWAY]
1270 #define	netmask	info->rti_info[RTAX_NETMASK]
1271 #define	ifaaddr	info->rti_info[RTAX_IFA]
1272 #define	ifpaddr	info->rti_info[RTAX_IFP]
1273 #define	flags	info->rti_flags
1274 
1275 /*
1276  * Look up rt_addrinfo for a specific fib.  Note that if rti_ifa is defined,
1277  * it will be referenced so the caller must free it.
1278  *
1279  * Assume basic consistency checks are executed by callers:
1280  * RTAX_DST exists, if RTF_GATEWAY is set, RTAX_GATEWAY exists as well.
1281  */
1282 int
1283 rt_getifa_fib(struct rt_addrinfo *info, u_int fibnum)
1284 {
1285 	struct epoch_tracker et;
1286 	int needref, error;
1287 
1288 	/*
1289 	 * ifp may be specified by sockaddr_dl
1290 	 * when protocol address is ambiguous.
1291 	 */
1292 	error = 0;
1293 	needref = (info->rti_ifa == NULL);
1294 	NET_EPOCH_ENTER(et);
1295 
1296 	/* If we have interface specified by the ifindex in the address, use it */
1297 	if (info->rti_ifp == NULL && ifpaddr != NULL &&
1298 	    ifpaddr->sa_family == AF_LINK) {
1299 	    const struct sockaddr_dl *sdl = (const struct sockaddr_dl *)ifpaddr;
1300 	    if (sdl->sdl_index != 0)
1301 		    info->rti_ifp = ifnet_byindex_locked(sdl->sdl_index);
1302 	}
1303 	/*
1304 	 * If we have source address specified, try to find it
1305 	 * TODO: avoid enumerating all ifas on all interfaces.
1306 	 */
1307 	if (info->rti_ifa == NULL && ifaaddr != NULL)
1308 		info->rti_ifa = ifa_ifwithaddr(ifaaddr);
1309 	if (info->rti_ifa == NULL) {
1310 		struct sockaddr *sa;
1311 
1312 		/*
1313 		 * Most common use case for the userland-supplied routes.
1314 		 *
1315 		 * Choose sockaddr to select ifa.
1316 		 * -- if ifp is set --
1317 		 * Order of preference:
1318 		 * 1) IFA address
1319 		 * 2) gateway address
1320 		 *   Note: for interface routes link-level gateway address
1321 		 *     is specified to indicate the interface index without
1322 		 *     specifying RTF_GATEWAY. In this case, ignore gateway
1323 		 *   Note: gateway AF may be different from dst AF. In this case,
1324 		 *   ignore gateway
1325 		 * 3) final destination.
1326 		 * 4) if all of these fails, try to get at least link-level ifa.
1327 		 * -- else --
1328 		 * try to lookup gateway or dst in the routing table to get ifa
1329 		 */
1330 		if (info->rti_info[RTAX_IFA] != NULL)
1331 			sa = info->rti_info[RTAX_IFA];
1332 		else if ((info->rti_flags & RTF_GATEWAY) != 0 &&
1333 		    gateway->sa_family == dst->sa_family)
1334 			sa = gateway;
1335 		else
1336 			sa = dst;
1337 		if (info->rti_ifp != NULL) {
1338 			info->rti_ifa = ifaof_ifpforaddr(sa, info->rti_ifp);
1339 			/* Case 4 */
1340 			if (info->rti_ifa == NULL && gateway != NULL)
1341 				info->rti_ifa = ifaof_ifpforaddr(gateway, info->rti_ifp);
1342 		} else if (dst != NULL && gateway != NULL)
1343 			info->rti_ifa = ifa_ifwithroute(flags, dst, gateway,
1344 							fibnum);
1345 		else if (sa != NULL)
1346 			info->rti_ifa = ifa_ifwithroute(flags, sa, sa,
1347 							fibnum);
1348 	}
1349 	if (needref && info->rti_ifa != NULL) {
1350 		if (info->rti_ifp == NULL)
1351 			info->rti_ifp = info->rti_ifa->ifa_ifp;
1352 		ifa_ref(info->rti_ifa);
1353 	} else
1354 		error = ENETUNREACH;
1355 	NET_EPOCH_EXIT(et);
1356 	return (error);
1357 }
1358 
1359 static int
1360 if_updatemtu_cb(struct radix_node *rn, void *arg)
1361 {
1362 	struct rtentry *rt;
1363 	struct if_mtuinfo *ifmtu;
1364 
1365 	rt = (struct rtentry *)rn;
1366 	ifmtu = (struct if_mtuinfo *)arg;
1367 
1368 	if (rt->rt_ifp != ifmtu->ifp)
1369 		return (0);
1370 
1371 	if (rt->rt_mtu >= ifmtu->mtu) {
1372 		/* We have to decrease mtu regardless of flags */
1373 		rt->rt_mtu = ifmtu->mtu;
1374 		return (0);
1375 	}
1376 
1377 	/*
1378 	 * New MTU is bigger. Check if are allowed to alter it
1379 	 */
1380 	if ((rt->rt_flags & (RTF_FIXEDMTU | RTF_GATEWAY | RTF_HOST)) != 0) {
1381 
1382 		/*
1383 		 * Skip routes with user-supplied MTU and
1384 		 * non-interface routes
1385 		 */
1386 		return (0);
1387 	}
1388 
1389 	/* We are safe to update route MTU */
1390 	rt->rt_mtu = ifmtu->mtu;
1391 
1392 	return (0);
1393 }
1394 
1395 void
1396 rt_updatemtu(struct ifnet *ifp)
1397 {
1398 	struct if_mtuinfo ifmtu;
1399 	struct rib_head *rnh;
1400 	int i, j;
1401 
1402 	ifmtu.ifp = ifp;
1403 
1404 	/*
1405 	 * Try to update rt_mtu for all routes using this interface
1406 	 * Unfortunately the only way to do this is to traverse all
1407 	 * routing tables in all fibs/domains.
1408 	 */
1409 	for (i = 1; i <= AF_MAX; i++) {
1410 		ifmtu.mtu = if_getmtu_family(ifp, i);
1411 		for (j = 0; j < rt_numfibs; j++) {
1412 			rnh = rt_tables_get_rnh(j, i);
1413 			if (rnh == NULL)
1414 				continue;
1415 			RIB_WLOCK(rnh);
1416 			rnh->rnh_walktree(&rnh->head, if_updatemtu_cb, &ifmtu);
1417 			RIB_WUNLOCK(rnh);
1418 		}
1419 	}
1420 }
1421 
1422 
1423 #if 0
1424 int p_sockaddr(char *buf, int buflen, struct sockaddr *s);
1425 int rt_print(char *buf, int buflen, struct rtentry *rt);
1426 
1427 int
1428 p_sockaddr(char *buf, int buflen, struct sockaddr *s)
1429 {
1430 	void *paddr = NULL;
1431 
1432 	switch (s->sa_family) {
1433 	case AF_INET:
1434 		paddr = &((struct sockaddr_in *)s)->sin_addr;
1435 		break;
1436 	case AF_INET6:
1437 		paddr = &((struct sockaddr_in6 *)s)->sin6_addr;
1438 		break;
1439 	}
1440 
1441 	if (paddr == NULL)
1442 		return (0);
1443 
1444 	if (inet_ntop(s->sa_family, paddr, buf, buflen) == NULL)
1445 		return (0);
1446 
1447 	return (strlen(buf));
1448 }
1449 
1450 int
1451 rt_print(char *buf, int buflen, struct rtentry *rt)
1452 {
1453 	struct sockaddr *addr, *mask;
1454 	int i = 0;
1455 
1456 	addr = rt_key(rt);
1457 	mask = rt_mask(rt);
1458 
1459 	i = p_sockaddr(buf, buflen, addr);
1460 	if (!(rt->rt_flags & RTF_HOST)) {
1461 		buf[i++] = '/';
1462 		i += p_sockaddr(buf + i, buflen - i, mask);
1463 	}
1464 
1465 	if (rt->rt_flags & RTF_GATEWAY) {
1466 		buf[i++] = '>';
1467 		i += p_sockaddr(buf + i, buflen - i, rt->rt_gateway);
1468 	}
1469 
1470 	return (i);
1471 }
1472 #endif
1473 
1474 #ifdef RADIX_MPATH
1475 /*
1476  * Deletes key for single-path routes, unlinks rtentry with
1477  * gateway specified in @info from multi-path routes.
1478  *
1479  * Returnes unlinked entry. In case of failure, returns NULL
1480  * and sets @perror to ESRCH.
1481  */
1482 static struct radix_node *
1483 rt_mpath_unlink(struct rib_head *rnh, struct rt_addrinfo *info,
1484     struct rtentry *rto, int *perror)
1485 {
1486 	/*
1487 	 * if we got multipath routes, we require users to specify
1488 	 * a matching RTAX_GATEWAY.
1489 	 */
1490 	struct rtentry *rt; // *rto = NULL;
1491 	struct radix_node *rn;
1492 	struct sockaddr *gw;
1493 
1494 	gw = info->rti_info[RTAX_GATEWAY];
1495 	rt = rt_mpath_matchgate(rto, gw);
1496 	if (rt == NULL) {
1497 		*perror = ESRCH;
1498 		return (NULL);
1499 	}
1500 
1501 	/*
1502 	 * this is the first entry in the chain
1503 	 */
1504 	if (rto == rt) {
1505 		rn = rn_mpath_next((struct radix_node *)rt);
1506 		/*
1507 		 * there is another entry, now it's active
1508 		 */
1509 		if (rn) {
1510 			rto = RNTORT(rn);
1511 			RT_LOCK(rto);
1512 			rto->rt_flags |= RTF_UP;
1513 			RT_UNLOCK(rto);
1514 		} else if (rt->rt_flags & RTF_GATEWAY) {
1515 			/*
1516 			 * For gateway routes, we need to
1517 			 * make sure that we we are deleting
1518 			 * the correct gateway.
1519 			 * rt_mpath_matchgate() does not
1520 			 * check the case when there is only
1521 			 * one route in the chain.
1522 			 */
1523 			if (gw &&
1524 			    (rt->rt_gateway->sa_len != gw->sa_len ||
1525 				memcmp(rt->rt_gateway, gw, gw->sa_len))) {
1526 				*perror = ESRCH;
1527 				return (NULL);
1528 			}
1529 		}
1530 
1531 		/*
1532 		 * use the normal delete code to remove
1533 		 * the first entry
1534 		 */
1535 		rn = rnh->rnh_deladdr(dst, netmask, &rnh->head);
1536 		*perror = 0;
1537 		return (rn);
1538 	}
1539 
1540 	/*
1541 	 * if the entry is 2nd and on up
1542 	 */
1543 	if (rt_mpath_deldup(rto, rt) == 0)
1544 		panic ("rtrequest1: rt_mpath_deldup");
1545 	*perror = 0;
1546 	rn = (struct radix_node *)rt;
1547 	return (rn);
1548 }
1549 #endif
1550 
1551 int
1552 rtrequest1_fib(int req, struct rt_addrinfo *info, struct rtentry **ret_nrt,
1553 				u_int fibnum)
1554 {
1555 	int error = 0;
1556 	struct rtentry *rt, *rt_old;
1557 	struct radix_node *rn;
1558 	struct rib_head *rnh;
1559 	struct ifaddr *ifa;
1560 	struct sockaddr *ndst;
1561 	struct sockaddr_storage mdst;
1562 
1563 	KASSERT((fibnum < rt_numfibs), ("rtrequest1_fib: bad fibnum"));
1564 	KASSERT((flags & RTF_RNH_LOCKED) == 0, ("rtrequest1_fib: locked"));
1565 	switch (dst->sa_family) {
1566 	case AF_INET6:
1567 	case AF_INET:
1568 		/* We support multiple FIBs. */
1569 		break;
1570 	default:
1571 		fibnum = RT_DEFAULT_FIB;
1572 		break;
1573 	}
1574 
1575 	/*
1576 	 * Find the correct routing tree to use for this Address Family
1577 	 */
1578 	rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
1579 	if (rnh == NULL)
1580 		return (EAFNOSUPPORT);
1581 
1582 	/*
1583 	 * If we are adding a host route then we don't want to put
1584 	 * a netmask in the tree, nor do we want to clone it.
1585 	 */
1586 	if (flags & RTF_HOST)
1587 		netmask = NULL;
1588 
1589 	switch (req) {
1590 	case RTM_DELETE:
1591 		if (netmask) {
1592 			if (dst->sa_len > sizeof(mdst))
1593 				return (EINVAL);
1594 			rt_maskedcopy(dst, (struct sockaddr *)&mdst, netmask);
1595 			dst = (struct sockaddr *)&mdst;
1596 		}
1597 
1598 		RIB_WLOCK(rnh);
1599 		rt = rt_unlinkrte(rnh, info, &error);
1600 		RIB_WUNLOCK(rnh);
1601 		if (error != 0)
1602 			return (error);
1603 
1604 		rt_notifydelete(rt, info);
1605 
1606 		/*
1607 		 * If the caller wants it, then it can have it,
1608 		 * but it's up to it to free the rtentry as we won't be
1609 		 * doing it.
1610 		 */
1611 		if (ret_nrt) {
1612 			*ret_nrt = rt;
1613 			RT_UNLOCK(rt);
1614 		} else
1615 			RTFREE_LOCKED(rt);
1616 		break;
1617 	case RTM_RESOLVE:
1618 		/*
1619 		 * resolve was only used for route cloning
1620 		 * here for compat
1621 		 */
1622 		break;
1623 	case RTM_ADD:
1624 		if ((flags & RTF_GATEWAY) && !gateway)
1625 			return (EINVAL);
1626 		if (dst && gateway && (dst->sa_family != gateway->sa_family) &&
1627 		    (gateway->sa_family != AF_UNSPEC) && (gateway->sa_family != AF_LINK))
1628 			return (EINVAL);
1629 
1630 		if (info->rti_ifa == NULL) {
1631 			error = rt_getifa_fib(info, fibnum);
1632 			if (error)
1633 				return (error);
1634 		}
1635 		rt = uma_zalloc(V_rtzone, M_NOWAIT);
1636 		if (rt == NULL) {
1637 			return (ENOBUFS);
1638 		}
1639 		rt->rt_flags = RTF_UP | flags;
1640 		rt->rt_fibnum = fibnum;
1641 		/*
1642 		 * Add the gateway. Possibly re-malloc-ing the storage for it.
1643 		 */
1644 		if ((error = rt_setgate(rt, dst, gateway)) != 0) {
1645 			uma_zfree(V_rtzone, rt);
1646 			return (error);
1647 		}
1648 
1649 		/*
1650 		 * point to the (possibly newly malloc'd) dest address.
1651 		 */
1652 		ndst = (struct sockaddr *)rt_key(rt);
1653 
1654 		/*
1655 		 * make sure it contains the value we want (masked if needed).
1656 		 */
1657 		if (netmask) {
1658 			rt_maskedcopy(dst, ndst, netmask);
1659 		} else
1660 			bcopy(dst, ndst, dst->sa_len);
1661 
1662 		/*
1663 		 * We use the ifa reference returned by rt_getifa_fib().
1664 		 * This moved from below so that rnh->rnh_addaddr() can
1665 		 * examine the ifa and  ifa->ifa_ifp if it so desires.
1666 		 */
1667 		ifa = info->rti_ifa;
1668 		ifa_ref(ifa);
1669 		rt->rt_ifa = ifa;
1670 		rt->rt_ifp = ifa->ifa_ifp;
1671 		rt->rt_weight = 1;
1672 
1673 		rt_setmetrics(info, rt);
1674 
1675 		RIB_WLOCK(rnh);
1676 		RT_LOCK(rt);
1677 #ifdef RADIX_MPATH
1678 		/* do not permit exactly the same dst/mask/gw pair */
1679 		if (rt_mpath_capable(rnh) &&
1680 			rt_mpath_conflict(rnh, rt, netmask)) {
1681 			RIB_WUNLOCK(rnh);
1682 
1683 			ifa_free(rt->rt_ifa);
1684 			R_Free(rt_key(rt));
1685 			uma_zfree(V_rtzone, rt);
1686 			return (EEXIST);
1687 		}
1688 #endif
1689 
1690 		/* XXX mtu manipulation will be done in rnh_addaddr -- itojun */
1691 		rn = rnh->rnh_addaddr(ndst, netmask, &rnh->head, rt->rt_nodes);
1692 
1693 		rt_old = NULL;
1694 		if (rn == NULL && (info->rti_flags & RTF_PINNED) != 0) {
1695 
1696 			/*
1697 			 * Force removal and re-try addition
1698 			 * TODO: better multipath&pinned support
1699 			 */
1700 			struct sockaddr *info_dst = info->rti_info[RTAX_DST];
1701 			info->rti_info[RTAX_DST] = ndst;
1702 			/* Do not delete existing PINNED(interface) routes */
1703 			info->rti_flags &= ~RTF_PINNED;
1704 			rt_old = rt_unlinkrte(rnh, info, &error);
1705 			info->rti_flags |= RTF_PINNED;
1706 			info->rti_info[RTAX_DST] = info_dst;
1707 			if (rt_old != NULL)
1708 				rn = rnh->rnh_addaddr(ndst, netmask, &rnh->head,
1709 				    rt->rt_nodes);
1710 		}
1711 		RIB_WUNLOCK(rnh);
1712 
1713 		if (rt_old != NULL)
1714 			RT_UNLOCK(rt_old);
1715 
1716 		/*
1717 		 * If it still failed to go into the tree,
1718 		 * then un-make it (this should be a function)
1719 		 */
1720 		if (rn == NULL) {
1721 			ifa_free(rt->rt_ifa);
1722 			R_Free(rt_key(rt));
1723 			uma_zfree(V_rtzone, rt);
1724 			return (EEXIST);
1725 		}
1726 
1727 		if (rt_old != NULL) {
1728 			rt_notifydelete(rt_old, info);
1729 			RTFREE(rt_old);
1730 		}
1731 
1732 		/*
1733 		 * If this protocol has something to add to this then
1734 		 * allow it to do that as well.
1735 		 */
1736 		if (ifa->ifa_rtrequest)
1737 			ifa->ifa_rtrequest(req, rt, info);
1738 
1739 		/*
1740 		 * actually return a resultant rtentry and
1741 		 * give the caller a single reference.
1742 		 */
1743 		if (ret_nrt) {
1744 			*ret_nrt = rt;
1745 			RT_ADDREF(rt);
1746 		}
1747 		rnh->rnh_gen++;		/* Routing table updated */
1748 		RT_UNLOCK(rt);
1749 		break;
1750 	case RTM_CHANGE:
1751 		RIB_WLOCK(rnh);
1752 		error = rtrequest1_fib_change(rnh, info, ret_nrt, fibnum);
1753 		RIB_WUNLOCK(rnh);
1754 		break;
1755 	default:
1756 		error = EOPNOTSUPP;
1757 	}
1758 
1759 	return (error);
1760 }
1761 
1762 #undef dst
1763 #undef gateway
1764 #undef netmask
1765 #undef ifaaddr
1766 #undef ifpaddr
1767 #undef flags
1768 
1769 static int
1770 rtrequest1_fib_change(struct rib_head *rnh, struct rt_addrinfo *info,
1771     struct rtentry **ret_nrt, u_int fibnum)
1772 {
1773 	struct rtentry *rt = NULL;
1774 	int error = 0;
1775 	int free_ifa = 0;
1776 	int family, mtu;
1777 	struct if_mtuinfo ifmtu;
1778 
1779 	RIB_WLOCK_ASSERT(rnh);
1780 
1781 	rt = (struct rtentry *)rnh->rnh_lookup(info->rti_info[RTAX_DST],
1782 	    info->rti_info[RTAX_NETMASK], &rnh->head);
1783 
1784 	if (rt == NULL)
1785 		return (ESRCH);
1786 
1787 #ifdef RADIX_MPATH
1788 	/*
1789 	 * If we got multipath routes,
1790 	 * we require users to specify a matching RTAX_GATEWAY.
1791 	 */
1792 	if (rt_mpath_capable(rnh)) {
1793 		rt = rt_mpath_matchgate(rt, info->rti_info[RTAX_GATEWAY]);
1794 		if (rt == NULL)
1795 			return (ESRCH);
1796 	}
1797 #endif
1798 
1799 	RT_LOCK(rt);
1800 
1801 	rt_setmetrics(info, rt);
1802 
1803 	/*
1804 	 * New gateway could require new ifaddr, ifp;
1805 	 * flags may also be different; ifp may be specified
1806 	 * by ll sockaddr when protocol address is ambiguous
1807 	 */
1808 	if (((rt->rt_flags & RTF_GATEWAY) &&
1809 	    info->rti_info[RTAX_GATEWAY] != NULL) ||
1810 	    info->rti_info[RTAX_IFP] != NULL ||
1811 	    (info->rti_info[RTAX_IFA] != NULL &&
1812 	     !sa_equal(info->rti_info[RTAX_IFA], rt->rt_ifa->ifa_addr))) {
1813 		/*
1814 		 * XXX: Temporarily set RTF_RNH_LOCKED flag in the rti_flags
1815 		 *	to avoid rlock in the ifa_ifwithroute().
1816 		 */
1817 		info->rti_flags |= RTF_RNH_LOCKED;
1818 		error = rt_getifa_fib(info, fibnum);
1819 		info->rti_flags &= ~RTF_RNH_LOCKED;
1820 		if (info->rti_ifa != NULL)
1821 			free_ifa = 1;
1822 
1823 		if (error != 0)
1824 			goto bad;
1825 	}
1826 
1827 	/* Check if outgoing interface has changed */
1828 	if (info->rti_ifa != NULL && info->rti_ifa != rt->rt_ifa &&
1829 	    rt->rt_ifa != NULL) {
1830 		if (rt->rt_ifa->ifa_rtrequest != NULL)
1831 			rt->rt_ifa->ifa_rtrequest(RTM_DELETE, rt, info);
1832 		ifa_free(rt->rt_ifa);
1833 		rt->rt_ifa = NULL;
1834 	}
1835 	/* Update gateway address */
1836 	if (info->rti_info[RTAX_GATEWAY] != NULL) {
1837 		error = rt_setgate(rt, rt_key(rt), info->rti_info[RTAX_GATEWAY]);
1838 		if (error != 0)
1839 			goto bad;
1840 
1841 		rt->rt_flags &= ~RTF_GATEWAY;
1842 		rt->rt_flags |= (RTF_GATEWAY & info->rti_flags);
1843 	}
1844 
1845 	if (info->rti_ifa != NULL && info->rti_ifa != rt->rt_ifa) {
1846 		ifa_ref(info->rti_ifa);
1847 		rt->rt_ifa = info->rti_ifa;
1848 		rt->rt_ifp = info->rti_ifp;
1849 	}
1850 	/* Allow some flags to be toggled on change. */
1851 	rt->rt_flags &= ~RTF_FMASK;
1852 	rt->rt_flags |= info->rti_flags & RTF_FMASK;
1853 
1854 	if (rt->rt_ifa && rt->rt_ifa->ifa_rtrequest != NULL)
1855 	       rt->rt_ifa->ifa_rtrequest(RTM_ADD, rt, info);
1856 
1857 	/* Alter route MTU if necessary */
1858 	if (rt->rt_ifp != NULL) {
1859 		family = info->rti_info[RTAX_DST]->sa_family;
1860 		mtu = if_getmtu_family(rt->rt_ifp, family);
1861 		/* Set default MTU */
1862 		if (rt->rt_mtu == 0)
1863 			rt->rt_mtu = mtu;
1864 		if (rt->rt_mtu != mtu) {
1865 			/* Check if we really need to update */
1866 			ifmtu.ifp = rt->rt_ifp;
1867 			ifmtu.mtu = mtu;
1868 			if_updatemtu_cb(rt->rt_nodes, &ifmtu);
1869 		}
1870 	}
1871 
1872 	/*
1873 	 * This route change may have modified the route's gateway.  In that
1874 	 * case, any inpcbs that have cached this route need to invalidate their
1875 	 * llentry cache.
1876 	 */
1877 	rnh->rnh_gen++;
1878 
1879 	if (ret_nrt) {
1880 		*ret_nrt = rt;
1881 		RT_ADDREF(rt);
1882 	}
1883 bad:
1884 	RT_UNLOCK(rt);
1885 	if (free_ifa != 0) {
1886 		ifa_free(info->rti_ifa);
1887 		info->rti_ifa = NULL;
1888 	}
1889 	return (error);
1890 }
1891 
1892 static void
1893 rt_setmetrics(const struct rt_addrinfo *info, struct rtentry *rt)
1894 {
1895 
1896 	if (info->rti_mflags & RTV_MTU) {
1897 		if (info->rti_rmx->rmx_mtu != 0) {
1898 
1899 			/*
1900 			 * MTU was explicitly provided by user.
1901 			 * Keep it.
1902 			 */
1903 			rt->rt_flags |= RTF_FIXEDMTU;
1904 		} else {
1905 
1906 			/*
1907 			 * User explicitly sets MTU to 0.
1908 			 * Assume rollback to default.
1909 			 */
1910 			rt->rt_flags &= ~RTF_FIXEDMTU;
1911 		}
1912 		rt->rt_mtu = info->rti_rmx->rmx_mtu;
1913 	}
1914 	if (info->rti_mflags & RTV_WEIGHT)
1915 		rt->rt_weight = info->rti_rmx->rmx_weight;
1916 	/* Kernel -> userland timebase conversion. */
1917 	if (info->rti_mflags & RTV_EXPIRE)
1918 		rt->rt_expire = info->rti_rmx->rmx_expire ?
1919 		    info->rti_rmx->rmx_expire - time_second + time_uptime : 0;
1920 }
1921 
1922 int
1923 rt_setgate(struct rtentry *rt, struct sockaddr *dst, struct sockaddr *gate)
1924 {
1925 	/* XXX dst may be overwritten, can we move this to below */
1926 	int dlen = SA_SIZE(dst), glen = SA_SIZE(gate);
1927 
1928 	/*
1929 	 * Prepare to store the gateway in rt->rt_gateway.
1930 	 * Both dst and gateway are stored one after the other in the same
1931 	 * malloc'd chunk. If we have room, we can reuse the old buffer,
1932 	 * rt_gateway already points to the right place.
1933 	 * Otherwise, malloc a new block and update the 'dst' address.
1934 	 */
1935 	if (rt->rt_gateway == NULL || glen > SA_SIZE(rt->rt_gateway)) {
1936 		caddr_t new;
1937 
1938 		R_Malloc(new, caddr_t, dlen + glen);
1939 		if (new == NULL)
1940 			return ENOBUFS;
1941 		/*
1942 		 * XXX note, we copy from *dst and not *rt_key(rt) because
1943 		 * rt_setgate() can be called to initialize a newly
1944 		 * allocated route entry, in which case rt_key(rt) == NULL
1945 		 * (and also rt->rt_gateway == NULL).
1946 		 * Free()/free() handle a NULL argument just fine.
1947 		 */
1948 		bcopy(dst, new, dlen);
1949 		R_Free(rt_key(rt));	/* free old block, if any */
1950 		rt_key(rt) = (struct sockaddr *)new;
1951 		rt->rt_gateway = (struct sockaddr *)(new + dlen);
1952 	}
1953 
1954 	/*
1955 	 * Copy the new gateway value into the memory chunk.
1956 	 */
1957 	bcopy(gate, rt->rt_gateway, glen);
1958 
1959 	return (0);
1960 }
1961 
1962 void
1963 rt_maskedcopy(struct sockaddr *src, struct sockaddr *dst, struct sockaddr *netmask)
1964 {
1965 	u_char *cp1 = (u_char *)src;
1966 	u_char *cp2 = (u_char *)dst;
1967 	u_char *cp3 = (u_char *)netmask;
1968 	u_char *cplim = cp2 + *cp3;
1969 	u_char *cplim2 = cp2 + *cp1;
1970 
1971 	*cp2++ = *cp1++; *cp2++ = *cp1++; /* copies sa_len & sa_family */
1972 	cp3 += 2;
1973 	if (cplim > cplim2)
1974 		cplim = cplim2;
1975 	while (cp2 < cplim)
1976 		*cp2++ = *cp1++ & *cp3++;
1977 	if (cp2 < cplim2)
1978 		bzero((caddr_t)cp2, (unsigned)(cplim2 - cp2));
1979 }
1980 
1981 /*
1982  * Set up a routing table entry, normally
1983  * for an interface.
1984  */
1985 #define _SOCKADDR_TMPSIZE 128 /* Not too big.. kernel stack size is limited */
1986 static inline  int
1987 rtinit1(struct ifaddr *ifa, int cmd, int flags, int fibnum)
1988 {
1989 	RIB_RLOCK_TRACKER;
1990 	struct sockaddr *dst;
1991 	struct sockaddr *netmask;
1992 	struct rtentry *rt = NULL;
1993 	struct rt_addrinfo info;
1994 	int error = 0;
1995 	int startfib, endfib;
1996 	char tempbuf[_SOCKADDR_TMPSIZE];
1997 	int didwork = 0;
1998 	int a_failure = 0;
1999 	struct sockaddr_dl *sdl = NULL;
2000 	struct rib_head *rnh;
2001 
2002 	if (flags & RTF_HOST) {
2003 		dst = ifa->ifa_dstaddr;
2004 		netmask = NULL;
2005 	} else {
2006 		dst = ifa->ifa_addr;
2007 		netmask = ifa->ifa_netmask;
2008 	}
2009 	if (dst->sa_len == 0)
2010 		return(EINVAL);
2011 	switch (dst->sa_family) {
2012 	case AF_INET6:
2013 	case AF_INET:
2014 		/* We support multiple FIBs. */
2015 		break;
2016 	default:
2017 		fibnum = RT_DEFAULT_FIB;
2018 		break;
2019 	}
2020 	if (fibnum == RT_ALL_FIBS) {
2021 		if (V_rt_add_addr_allfibs == 0 && cmd == (int)RTM_ADD)
2022 			startfib = endfib = ifa->ifa_ifp->if_fib;
2023 		else {
2024 			startfib = 0;
2025 			endfib = rt_numfibs - 1;
2026 		}
2027 	} else {
2028 		KASSERT((fibnum < rt_numfibs), ("rtinit1: bad fibnum"));
2029 		startfib = fibnum;
2030 		endfib = fibnum;
2031 	}
2032 
2033 	/*
2034 	 * If it's a delete, check that if it exists,
2035 	 * it's on the correct interface or we might scrub
2036 	 * a route to another ifa which would
2037 	 * be confusing at best and possibly worse.
2038 	 */
2039 	if (cmd == RTM_DELETE) {
2040 		/*
2041 		 * It's a delete, so it should already exist..
2042 		 * If it's a net, mask off the host bits
2043 		 * (Assuming we have a mask)
2044 		 * XXX this is kinda inet specific..
2045 		 */
2046 		if (netmask != NULL) {
2047 			rt_maskedcopy(dst, (struct sockaddr *)tempbuf, netmask);
2048 			dst = (struct sockaddr *)tempbuf;
2049 		}
2050 	} else if (cmd == RTM_ADD) {
2051 		sdl = (struct sockaddr_dl *)tempbuf;
2052 		bzero(sdl, sizeof(struct sockaddr_dl));
2053 		sdl->sdl_family = AF_LINK;
2054 		sdl->sdl_len = sizeof(struct sockaddr_dl);
2055 		sdl->sdl_type = ifa->ifa_ifp->if_type;
2056 		sdl->sdl_index = ifa->ifa_ifp->if_index;
2057         }
2058 	/*
2059 	 * Now go through all the requested tables (fibs) and do the
2060 	 * requested action. Realistically, this will either be fib 0
2061 	 * for protocols that don't do multiple tables or all the
2062 	 * tables for those that do.
2063 	 */
2064 	for ( fibnum = startfib; fibnum <= endfib; fibnum++) {
2065 		if (cmd == RTM_DELETE) {
2066 			struct radix_node *rn;
2067 			/*
2068 			 * Look up an rtentry that is in the routing tree and
2069 			 * contains the correct info.
2070 			 */
2071 			rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
2072 			if (rnh == NULL)
2073 				/* this table doesn't exist but others might */
2074 				continue;
2075 			RIB_RLOCK(rnh);
2076 			rn = rnh->rnh_lookup(dst, netmask, &rnh->head);
2077 #ifdef RADIX_MPATH
2078 			if (rt_mpath_capable(rnh)) {
2079 
2080 				if (rn == NULL)
2081 					error = ESRCH;
2082 				else {
2083 					rt = RNTORT(rn);
2084 					/*
2085 					 * for interface route the
2086 					 * rt->rt_gateway is sockaddr_intf
2087 					 * for cloning ARP entries, so
2088 					 * rt_mpath_matchgate must use the
2089 					 * interface address
2090 					 */
2091 					rt = rt_mpath_matchgate(rt,
2092 					    ifa->ifa_addr);
2093 					if (rt == NULL)
2094 						error = ESRCH;
2095 				}
2096 			}
2097 #endif
2098 			error = (rn == NULL ||
2099 			    (rn->rn_flags & RNF_ROOT) ||
2100 			    RNTORT(rn)->rt_ifa != ifa);
2101 			RIB_RUNLOCK(rnh);
2102 			if (error) {
2103 				/* this is only an error if bad on ALL tables */
2104 				continue;
2105 			}
2106 		}
2107 		/*
2108 		 * Do the actual request
2109 		 */
2110 		bzero((caddr_t)&info, sizeof(info));
2111 		ifa_ref(ifa);
2112 		info.rti_ifa = ifa;
2113 		info.rti_flags = flags |
2114 		    (ifa->ifa_flags & ~IFA_RTSELF) | RTF_PINNED;
2115 		info.rti_info[RTAX_DST] = dst;
2116 		/*
2117 		 * doing this for compatibility reasons
2118 		 */
2119 		if (cmd == RTM_ADD)
2120 			info.rti_info[RTAX_GATEWAY] = (struct sockaddr *)sdl;
2121 		else
2122 			info.rti_info[RTAX_GATEWAY] = ifa->ifa_addr;
2123 		info.rti_info[RTAX_NETMASK] = netmask;
2124 		error = rtrequest1_fib(cmd, &info, &rt, fibnum);
2125 
2126 		if (error == 0 && rt != NULL) {
2127 			/*
2128 			 * notify any listening routing agents of the change
2129 			 */
2130 			RT_LOCK(rt);
2131 #ifdef RADIX_MPATH
2132 			/*
2133 			 * in case address alias finds the first address
2134 			 * e.g. ifconfig bge0 192.0.2.246/24
2135 			 * e.g. ifconfig bge0 192.0.2.247/24
2136 			 * the address set in the route is 192.0.2.246
2137 			 * so we need to replace it with 192.0.2.247
2138 			 */
2139 			if (memcmp(rt->rt_ifa->ifa_addr,
2140 			    ifa->ifa_addr, ifa->ifa_addr->sa_len)) {
2141 				ifa_free(rt->rt_ifa);
2142 				ifa_ref(ifa);
2143 				rt->rt_ifp = ifa->ifa_ifp;
2144 				rt->rt_ifa = ifa;
2145 			}
2146 #endif
2147 			RT_ADDREF(rt);
2148 			RT_UNLOCK(rt);
2149 			rt_newaddrmsg_fib(cmd, ifa, error, rt, fibnum);
2150 			RT_LOCK(rt);
2151 			RT_REMREF(rt);
2152 			if (cmd == RTM_DELETE) {
2153 				/*
2154 				 * If we are deleting, and we found an entry,
2155 				 * then it's been removed from the tree..
2156 				 * now throw it away.
2157 				 */
2158 				RTFREE_LOCKED(rt);
2159 			} else {
2160 				if (cmd == RTM_ADD) {
2161 					/*
2162 					 * We just wanted to add it..
2163 					 * we don't actually need a reference.
2164 					 */
2165 					RT_REMREF(rt);
2166 				}
2167 				RT_UNLOCK(rt);
2168 			}
2169 			didwork = 1;
2170 		}
2171 		if (error)
2172 			a_failure = error;
2173 	}
2174 	if (cmd == RTM_DELETE) {
2175 		if (didwork) {
2176 			error = 0;
2177 		} else {
2178 			/* we only give an error if it wasn't in any table */
2179 			error = ((flags & RTF_HOST) ?
2180 			    EHOSTUNREACH : ENETUNREACH);
2181 		}
2182 	} else {
2183 		if (a_failure) {
2184 			/* return an error if any of them failed */
2185 			error = a_failure;
2186 		}
2187 	}
2188 	return (error);
2189 }
2190 
2191 /*
2192  * Set up a routing table entry, normally
2193  * for an interface.
2194  */
2195 int
2196 rtinit(struct ifaddr *ifa, int cmd, int flags)
2197 {
2198 	struct sockaddr *dst;
2199 	int fib = RT_DEFAULT_FIB;
2200 
2201 	if (flags & RTF_HOST) {
2202 		dst = ifa->ifa_dstaddr;
2203 	} else {
2204 		dst = ifa->ifa_addr;
2205 	}
2206 
2207 	switch (dst->sa_family) {
2208 	case AF_INET6:
2209 	case AF_INET:
2210 		/* We do support multiple FIBs. */
2211 		fib = RT_ALL_FIBS;
2212 		break;
2213 	}
2214 	return (rtinit1(ifa, cmd, flags, fib));
2215 }
2216 
2217 /*
2218  * Announce interface address arrival/withdraw
2219  * Returns 0 on success.
2220  */
2221 int
2222 rt_addrmsg(int cmd, struct ifaddr *ifa, int fibnum)
2223 {
2224 
2225 	KASSERT(cmd == RTM_ADD || cmd == RTM_DELETE,
2226 	    ("unexpected cmd %d", cmd));
2227 
2228 	KASSERT(fibnum == RT_ALL_FIBS || (fibnum >= 0 && fibnum < rt_numfibs),
2229 	    ("%s: fib out of range 0 <=%d<%d", __func__, fibnum, rt_numfibs));
2230 
2231 #if defined(INET) || defined(INET6)
2232 #ifdef SCTP
2233 	/*
2234 	 * notify the SCTP stack
2235 	 * this will only get called when an address is added/deleted
2236 	 * XXX pass the ifaddr struct instead if ifa->ifa_addr...
2237 	 */
2238 	sctp_addr_change(ifa, cmd);
2239 #endif /* SCTP */
2240 #endif
2241 	return (rtsock_addrmsg(cmd, ifa, fibnum));
2242 }
2243 
2244 /*
2245  * Announce route addition/removal.
2246  * Users of this function MUST validate input data BEFORE calling.
2247  * However we have to be able to handle invalid data:
2248  * if some userland app sends us "invalid" route message (invalid mask,
2249  * no dst, wrong address families, etc...) we need to pass it back
2250  * to app (and any other rtsock consumers) with rtm_errno field set to
2251  * non-zero value.
2252  * Returns 0 on success.
2253  */
2254 int
2255 rt_routemsg(int cmd, struct ifnet *ifp, int error, struct rtentry *rt,
2256     int fibnum)
2257 {
2258 
2259 	KASSERT(cmd == RTM_ADD || cmd == RTM_DELETE,
2260 	    ("unexpected cmd %d", cmd));
2261 
2262 	KASSERT(fibnum == RT_ALL_FIBS || (fibnum >= 0 && fibnum < rt_numfibs),
2263 	    ("%s: fib out of range 0 <=%d<%d", __func__, fibnum, rt_numfibs));
2264 
2265 	KASSERT(rt_key(rt) != NULL, (":%s: rt_key must be supplied", __func__));
2266 
2267 	return (rtsock_routemsg(cmd, ifp, error, rt, fibnum));
2268 }
2269 
2270 void
2271 rt_newaddrmsg(int cmd, struct ifaddr *ifa, int error, struct rtentry *rt)
2272 {
2273 
2274 	rt_newaddrmsg_fib(cmd, ifa, error, rt, RT_ALL_FIBS);
2275 }
2276 
2277 /*
2278  * This is called to generate messages from the routing socket
2279  * indicating a network interface has had addresses associated with it.
2280  */
2281 void
2282 rt_newaddrmsg_fib(int cmd, struct ifaddr *ifa, int error, struct rtentry *rt,
2283     int fibnum)
2284 {
2285 
2286 	KASSERT(cmd == RTM_ADD || cmd == RTM_DELETE,
2287 		("unexpected cmd %u", cmd));
2288 	KASSERT(fibnum == RT_ALL_FIBS || (fibnum >= 0 && fibnum < rt_numfibs),
2289 	    ("%s: fib out of range 0 <=%d<%d", __func__, fibnum, rt_numfibs));
2290 
2291 	if (cmd == RTM_ADD) {
2292 		rt_addrmsg(cmd, ifa, fibnum);
2293 		if (rt != NULL)
2294 			rt_routemsg(cmd, ifa->ifa_ifp, error, rt, fibnum);
2295 	} else {
2296 		if (rt != NULL)
2297 			rt_routemsg(cmd, ifa->ifa_ifp, error, rt, fibnum);
2298 		rt_addrmsg(cmd, ifa, fibnum);
2299 	}
2300 }
2301 
2302