xref: /freebsd/sys/net/route.c (revision 6966ac055c3b7a39266fb982493330df7a097997)
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_mrouting.h"
42 #include "opt_mpath.h"
43 #include "opt_route.h"
44 
45 #include <sys/param.h>
46 #include <sys/systm.h>
47 #include <sys/malloc.h>
48 #include <sys/mbuf.h>
49 #include <sys/socket.h>
50 #include <sys/sysctl.h>
51 #include <sys/syslog.h>
52 #include <sys/sysproto.h>
53 #include <sys/proc.h>
54 #include <sys/domain.h>
55 #include <sys/eventhandler.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 /* This is read-only.. */
94 u_int rt_numfibs = RT_NUMFIBS;
95 SYSCTL_UINT(_net, OID_AUTO, fibs, CTLFLAG_RDTUN, &rt_numfibs, 0, "");
96 
97 /*
98  * By default add routes to all fibs for new interfaces.
99  * Once this is set to 0 then only allocate routes on interface
100  * changes for the FIB of the caller when adding a new set of addresses
101  * to an interface.  XXX this is a shotgun aproach to a problem that needs
102  * a more fine grained solution.. that will come.
103  * XXX also has the problems getting the FIB from curthread which will not
104  * always work given the fib can be overridden and prefixes can be added
105  * from the network stack context.
106  */
107 VNET_DEFINE(u_int, rt_add_addr_allfibs) = 1;
108 SYSCTL_UINT(_net, OID_AUTO, add_addr_allfibs, CTLFLAG_RWTUN | CTLFLAG_VNET,
109     &VNET_NAME(rt_add_addr_allfibs), 0, "");
110 
111 VNET_PCPUSTAT_DEFINE_STATIC(struct rtstat, rtstat);
112 #define	RTSTAT_ADD(name, val)	\
113 	VNET_PCPUSTAT_ADD(struct rtstat, rtstat, name, (val))
114 #define	RTSTAT_INC(name)	RTSTAT_ADD(name, 1)
115 
116 VNET_PCPUSTAT_SYSINIT(rtstat);
117 #ifdef VIMAGE
118 VNET_PCPUSTAT_SYSUNINIT(rtstat);
119 #endif
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 EVENTHANDLER_LIST_DEFINE(rt_addrmsg);
144 
145 static int rt_getifa_fib(struct rt_addrinfo *, u_int);
146 static int rtrequest1_fib_change(struct rib_head *, struct rt_addrinfo *,
147     struct rtentry **, u_int);
148 static void rt_setmetrics(const struct rt_addrinfo *, struct rtentry *);
149 static int rt_ifdelroute(const struct rtentry *rt, void *arg);
150 static struct rtentry *rt_unlinkrte(struct rib_head *rnh,
151     struct rt_addrinfo *info, int *perror);
152 static void rt_notifydelete(struct rtentry *rt, struct rt_addrinfo *info);
153 #ifdef RADIX_MPATH
154 static struct radix_node *rt_mpath_unlink(struct rib_head *rnh,
155     struct rt_addrinfo *info, struct rtentry *rto, int *perror);
156 #endif
157 static int rt_exportinfo(struct rtentry *rt, struct rt_addrinfo *info,
158     int flags);
159 
160 struct if_mtuinfo
161 {
162 	struct ifnet	*ifp;
163 	int		mtu;
164 };
165 
166 static int	if_updatemtu_cb(struct radix_node *, void *);
167 
168 /*
169  * handler for net.my_fibnum
170  */
171 static int
172 sysctl_my_fibnum(SYSCTL_HANDLER_ARGS)
173 {
174         int fibnum;
175         int error;
176 
177         fibnum = curthread->td_proc->p_fibnum;
178         error = sysctl_handle_int(oidp, &fibnum, 0, req);
179         return (error);
180 }
181 
182 SYSCTL_PROC(_net, OID_AUTO, my_fibnum, CTLTYPE_INT|CTLFLAG_RD,
183             NULL, 0, &sysctl_my_fibnum, "I", "default FIB of caller");
184 
185 static __inline struct rib_head **
186 rt_tables_get_rnh_ptr(int table, int fam)
187 {
188 	struct rib_head **rnh;
189 
190 	KASSERT(table >= 0 && table < rt_numfibs,
191 	    ("%s: table out of bounds (0 <= %d < %d)", __func__, table,
192 	     rt_numfibs));
193 	KASSERT(fam >= 0 && fam < (AF_MAX + 1),
194 	    ("%s: fam out of bounds (0 <= %d < %d)", __func__, fam, AF_MAX+1));
195 
196 	/* rnh is [fib=0][af=0]. */
197 	rnh = (struct rib_head **)V_rt_tables;
198 	/* Get the offset to the requested table and fam. */
199 	rnh += table * (AF_MAX+1) + fam;
200 
201 	return (rnh);
202 }
203 
204 struct rib_head *
205 rt_tables_get_rnh(int table, int fam)
206 {
207 
208 	return (*rt_tables_get_rnh_ptr(table, fam));
209 }
210 
211 u_int
212 rt_tables_get_gen(int table, int fam)
213 {
214 	struct rib_head *rnh;
215 
216 	rnh = *rt_tables_get_rnh_ptr(table, fam);
217 	KASSERT(rnh != NULL, ("%s: NULL rib_head pointer table %d fam %d",
218 	    __func__, table, fam));
219 	return (rnh->rnh_gen);
220 }
221 
222 
223 /*
224  * route initialization must occur before ip6_init2(), which happenas at
225  * SI_ORDER_MIDDLE.
226  */
227 static void
228 route_init(void)
229 {
230 
231 	/* whack the tunable ints into  line. */
232 	if (rt_numfibs > RT_MAXFIBS)
233 		rt_numfibs = RT_MAXFIBS;
234 	if (rt_numfibs == 0)
235 		rt_numfibs = 1;
236 }
237 SYSINIT(route_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD, route_init, NULL);
238 
239 static int
240 rtentry_zinit(void *mem, int size, int how)
241 {
242 	struct rtentry *rt = mem;
243 
244 	rt->rt_pksent = counter_u64_alloc(how);
245 	if (rt->rt_pksent == NULL)
246 		return (ENOMEM);
247 
248 	RT_LOCK_INIT(rt);
249 
250 	return (0);
251 }
252 
253 static void
254 rtentry_zfini(void *mem, int size)
255 {
256 	struct rtentry *rt = mem;
257 
258 	RT_LOCK_DESTROY(rt);
259 	counter_u64_free(rt->rt_pksent);
260 }
261 
262 static int
263 rtentry_ctor(void *mem, int size, void *arg, int how)
264 {
265 	struct rtentry *rt = mem;
266 
267 	bzero(rt, offsetof(struct rtentry, rt_endzero));
268 	counter_u64_zero(rt->rt_pksent);
269 	rt->rt_chain = NULL;
270 
271 	return (0);
272 }
273 
274 static void
275 rtentry_dtor(void *mem, int size, void *arg)
276 {
277 	struct rtentry *rt = mem;
278 
279 	RT_UNLOCK_COND(rt);
280 }
281 
282 static void
283 vnet_route_init(const void *unused __unused)
284 {
285 	struct domain *dom;
286 	struct rib_head **rnh;
287 	int table;
288 	int fam;
289 
290 	V_rt_tables = malloc(rt_numfibs * (AF_MAX+1) *
291 	    sizeof(struct rib_head *), M_RTABLE, M_WAITOK|M_ZERO);
292 
293 	V_rtzone = uma_zcreate("rtentry", sizeof(struct rtentry),
294 	    rtentry_ctor, rtentry_dtor,
295 	    rtentry_zinit, rtentry_zfini, UMA_ALIGN_PTR, 0);
296 	for (dom = domains; dom; dom = dom->dom_next) {
297 		if (dom->dom_rtattach == NULL)
298 			continue;
299 
300 		for  (table = 0; table < rt_numfibs; table++) {
301 			fam = dom->dom_family;
302 			if (table != 0 && fam != AF_INET6 && fam != AF_INET)
303 				break;
304 
305 			rnh = rt_tables_get_rnh_ptr(table, fam);
306 			if (rnh == NULL)
307 				panic("%s: rnh NULL", __func__);
308 			dom->dom_rtattach((void **)rnh, 0, table);
309 		}
310 	}
311 }
312 VNET_SYSINIT(vnet_route_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH,
313     vnet_route_init, 0);
314 
315 #ifdef VIMAGE
316 static void
317 vnet_route_uninit(const void *unused __unused)
318 {
319 	int table;
320 	int fam;
321 	struct domain *dom;
322 	struct rib_head **rnh;
323 
324 	for (dom = domains; dom; dom = dom->dom_next) {
325 		if (dom->dom_rtdetach == NULL)
326 			continue;
327 
328 		for (table = 0; table < rt_numfibs; table++) {
329 			fam = dom->dom_family;
330 
331 			if (table != 0 && fam != AF_INET6 && fam != AF_INET)
332 				break;
333 
334 			rnh = rt_tables_get_rnh_ptr(table, fam);
335 			if (rnh == NULL)
336 				panic("%s: rnh NULL", __func__);
337 			dom->dom_rtdetach((void **)rnh, 0);
338 		}
339 	}
340 
341 	free(V_rt_tables, M_RTABLE);
342 	uma_zdestroy(V_rtzone);
343 }
344 VNET_SYSUNINIT(vnet_route_uninit, SI_SUB_PROTO_DOMAIN, SI_ORDER_FIRST,
345     vnet_route_uninit, 0);
346 #endif
347 
348 struct rib_head *
349 rt_table_init(int offset, int family, u_int fibnum)
350 {
351 	struct rib_head *rh;
352 
353 	rh = malloc(sizeof(struct rib_head), M_RTABLE, M_WAITOK | M_ZERO);
354 
355 	/* TODO: These details should be hidded inside radix.c */
356 	/* Init masks tree */
357 	rn_inithead_internal(&rh->head, rh->rnh_nodes, offset);
358 	rn_inithead_internal(&rh->rmhead.head, rh->rmhead.mask_nodes, 0);
359 	rh->head.rnh_masks = &rh->rmhead;
360 
361 	/* Save metadata associated with this routing table. */
362 	rh->rib_family = family;
363 	rh->rib_fibnum = fibnum;
364 #ifdef VIMAGE
365 	rh->rib_vnet = curvnet;
366 #endif
367 
368 	tmproutes_init(rh);
369 
370 	/* Init locks */
371 	RIB_LOCK_INIT(rh);
372 
373 	/* Finally, set base callbacks */
374 	rh->rnh_addaddr = rn_addroute;
375 	rh->rnh_deladdr = rn_delete;
376 	rh->rnh_matchaddr = rn_match;
377 	rh->rnh_lookup = rn_lookup;
378 	rh->rnh_walktree = rn_walktree;
379 	rh->rnh_walktree_from = rn_walktree_from;
380 
381 	return (rh);
382 }
383 
384 static int
385 rt_freeentry(struct radix_node *rn, void *arg)
386 {
387 	struct radix_head * const rnh = arg;
388 	struct radix_node *x;
389 
390 	x = (struct radix_node *)rn_delete(rn + 2, NULL, rnh);
391 	if (x != NULL)
392 		R_Free(x);
393 	return (0);
394 }
395 
396 void
397 rt_table_destroy(struct rib_head *rh)
398 {
399 
400 	tmproutes_destroy(rh);
401 
402 	rn_walktree(&rh->rmhead.head, rt_freeentry, &rh->rmhead.head);
403 
404 	/* Assume table is already empty */
405 	RIB_LOCK_DESTROY(rh);
406 	free(rh, M_RTABLE);
407 }
408 
409 
410 #ifndef _SYS_SYSPROTO_H_
411 struct setfib_args {
412 	int     fibnum;
413 };
414 #endif
415 int
416 sys_setfib(struct thread *td, struct setfib_args *uap)
417 {
418 	if (uap->fibnum < 0 || uap->fibnum >= rt_numfibs)
419 		return EINVAL;
420 	td->td_proc->p_fibnum = uap->fibnum;
421 	return (0);
422 }
423 
424 /*
425  * Packet routing routines.
426  */
427 void
428 rtalloc_ign_fib(struct route *ro, u_long ignore, u_int fibnum)
429 {
430 	struct rtentry *rt;
431 
432 	if ((rt = ro->ro_rt) != NULL) {
433 		if (rt->rt_ifp != NULL && rt->rt_flags & RTF_UP)
434 			return;
435 		RTFREE(rt);
436 		ro->ro_rt = NULL;
437 	}
438 	ro->ro_rt = rtalloc1_fib(&ro->ro_dst, 1, ignore, fibnum);
439 	if (ro->ro_rt)
440 		RT_UNLOCK(ro->ro_rt);
441 }
442 
443 /*
444  * Look up the route that matches the address given
445  * Or, at least try.. Create a cloned route if needed.
446  *
447  * The returned route, if any, is locked.
448  */
449 struct rtentry *
450 rtalloc1(struct sockaddr *dst, int report, u_long ignflags)
451 {
452 
453 	return (rtalloc1_fib(dst, report, ignflags, RT_DEFAULT_FIB));
454 }
455 
456 struct rtentry *
457 rtalloc1_fib(struct sockaddr *dst, int report, u_long ignflags,
458 		    u_int fibnum)
459 {
460 	RIB_RLOCK_TRACKER;
461 	struct rib_head *rh;
462 	struct radix_node *rn;
463 	struct rtentry *newrt;
464 	struct rt_addrinfo info;
465 	int err = 0, msgtype = RTM_MISS;
466 
467 	KASSERT((fibnum < rt_numfibs), ("rtalloc1_fib: bad fibnum"));
468 	rh = rt_tables_get_rnh(fibnum, dst->sa_family);
469 	newrt = NULL;
470 	if (rh == NULL)
471 		goto miss;
472 
473 	/*
474 	 * Look up the address in the table for that Address Family
475 	 */
476 	if ((ignflags & RTF_RNH_LOCKED) == 0)
477 		RIB_RLOCK(rh);
478 #ifdef INVARIANTS
479 	else
480 		RIB_LOCK_ASSERT(rh);
481 #endif
482 	rn = rh->rnh_matchaddr(dst, &rh->head);
483 	if (rn && ((rn->rn_flags & RNF_ROOT) == 0)) {
484 		newrt = RNTORT(rn);
485 		RT_LOCK(newrt);
486 		RT_ADDREF(newrt);
487 		if ((ignflags & RTF_RNH_LOCKED) == 0)
488 			RIB_RUNLOCK(rh);
489 		return (newrt);
490 
491 	} else if ((ignflags & RTF_RNH_LOCKED) == 0)
492 		RIB_RUNLOCK(rh);
493 	/*
494 	 * Either we hit the root or could not find any match,
495 	 * which basically means: "cannot get there from here".
496 	 */
497 miss:
498 	RTSTAT_INC(rts_unreach);
499 
500 	if (report) {
501 		/*
502 		 * If required, report the failure to the supervising
503 		 * Authorities.
504 		 * For a delete, this is not an error. (report == 0)
505 		 */
506 		bzero(&info, sizeof(info));
507 		info.rti_info[RTAX_DST] = dst;
508 		rt_missmsg_fib(msgtype, &info, 0, err, fibnum);
509 	}
510 	return (newrt);
511 }
512 
513 /*
514  * Remove a reference count from an rtentry.
515  * If the count gets low enough, take it out of the routing table
516  */
517 void
518 rtfree(struct rtentry *rt)
519 {
520 	struct rib_head *rnh;
521 
522 	KASSERT(rt != NULL,("%s: NULL rt", __func__));
523 	rnh = rt_tables_get_rnh(rt->rt_fibnum, rt_key(rt)->sa_family);
524 	KASSERT(rnh != NULL,("%s: NULL rnh", __func__));
525 
526 	RT_LOCK_ASSERT(rt);
527 
528 	/*
529 	 * The callers should use RTFREE_LOCKED() or RTFREE(), so
530 	 * we should come here exactly with the last reference.
531 	 */
532 	RT_REMREF(rt);
533 	if (rt->rt_refcnt > 0) {
534 		log(LOG_DEBUG, "%s: %p has %d refs\n", __func__, rt, rt->rt_refcnt);
535 		goto done;
536 	}
537 
538 	/*
539 	 * On last reference give the "close method" a chance
540 	 * to cleanup private state.  This also permits (for
541 	 * IPv4 and IPv6) a chance to decide if the routing table
542 	 * entry should be purged immediately or at a later time.
543 	 * When an immediate purge is to happen the close routine
544 	 * typically calls rtexpunge which clears the RTF_UP flag
545 	 * on the entry so that the code below reclaims the storage.
546 	 */
547 	if (rt->rt_refcnt == 0 && rnh->rnh_close)
548 		rnh->rnh_close((struct radix_node *)rt, &rnh->head);
549 
550 	/*
551 	 * If we are no longer "up" (and ref == 0)
552 	 * then we can free the resources associated
553 	 * with the route.
554 	 */
555 	if ((rt->rt_flags & RTF_UP) == 0) {
556 		if (rt->rt_nodes->rn_flags & (RNF_ACTIVE | RNF_ROOT))
557 			panic("rtfree 2");
558 		/*
559 		 * the rtentry must have been removed from the routing table
560 		 * so it is represented in rttrash.. remove that now.
561 		 */
562 		V_rttrash--;
563 #ifdef	DIAGNOSTIC
564 		if (rt->rt_refcnt < 0) {
565 			printf("rtfree: %p not freed (neg refs)\n", rt);
566 			goto done;
567 		}
568 #endif
569 		/*
570 		 * release references on items we hold them on..
571 		 * e.g other routes and ifaddrs.
572 		 */
573 		if (rt->rt_ifa)
574 			ifa_free(rt->rt_ifa);
575 		/*
576 		 * The key is separatly alloc'd so free it (see rt_setgate()).
577 		 * This also frees the gateway, as they are always malloc'd
578 		 * together.
579 		 */
580 		R_Free(rt_key(rt));
581 
582 		/*
583 		 * and the rtentry itself of course
584 		 */
585 		uma_zfree(V_rtzone, rt);
586 		return;
587 	}
588 done:
589 	RT_UNLOCK(rt);
590 }
591 
592 /*
593  * Adds a temporal redirect entry to the routing table.
594  * @fibnum: fib number
595  * @dst: destination to install redirect to
596  * @gateway: gateway to go via
597  * @author: sockaddr of originating router, can be NULL
598  * @ifp: interface to use for the redirected route
599  * @flags: set of flags to add. Allowed: RTF_GATEWAY
600  * @lifetime_sec: time in seconds to expire this redirect.
601  *
602  * Retuns 0 on success, errno otherwise.
603  */
604 int
605 rib_add_redirect(u_int fibnum, struct sockaddr *dst, struct sockaddr *gateway,
606     struct sockaddr *author, struct ifnet *ifp, int flags, int lifetime_sec)
607 {
608 	struct rtentry *rt;
609 	int error;
610 	struct rt_addrinfo info;
611 	struct rt_metrics rti_rmx;
612 	struct ifaddr *ifa;
613 
614 	NET_EPOCH_ASSERT();
615 
616 	if (rt_tables_get_rnh(fibnum, dst->sa_family) == NULL)
617 		return (EAFNOSUPPORT);
618 
619 	/* Verify the allowed flag mask. */
620 	KASSERT(((flags & ~(RTF_GATEWAY)) == 0),
621 	    ("invalid redirect flags: %x", flags));
622 
623 	/* Get the best ifa for the given interface and gateway. */
624 	if ((ifa = ifaof_ifpforaddr(gateway, ifp)) == NULL)
625 		return (ENETUNREACH);
626 	ifa_ref(ifa);
627 
628 	bzero(&info, sizeof(info));
629 	info.rti_info[RTAX_DST] = dst;
630 	info.rti_info[RTAX_GATEWAY] = gateway;
631 	info.rti_ifa = ifa;
632 	info.rti_ifp = ifp;
633 	info.rti_flags = flags | RTF_DYNAMIC;
634 
635 	/* Setup route metrics to define expire time. */
636 	bzero(&rti_rmx, sizeof(rti_rmx));
637 	/* Set expire time as absolute. */
638 	rti_rmx.rmx_expire = lifetime_sec + time_second;
639 	info.rti_mflags |= RTV_EXPIRE;
640 	info.rti_rmx = &rti_rmx;
641 
642 	error = rtrequest1_fib(RTM_ADD, &info, &rt, fibnum);
643 	ifa_free(ifa);
644 
645 	if (error != 0) {
646 		/* TODO: add per-fib redirect stats. */
647 		return (error);
648 	}
649 
650 	RT_LOCK(rt);
651 	flags = rt->rt_flags;
652 	RTFREE_LOCKED(rt);
653 
654 	RTSTAT_INC(rts_dynamic);
655 
656 	/* Send notification of a route addition to userland. */
657 	bzero(&info, sizeof(info));
658 	info.rti_info[RTAX_DST] = dst;
659 	info.rti_info[RTAX_GATEWAY] = gateway;
660 	info.rti_info[RTAX_AUTHOR] = author;
661 	rt_missmsg_fib(RTM_REDIRECT, &info, flags, error, fibnum);
662 
663 	return (0);
664 }
665 
666 /*
667  * Routing table ioctl interface.
668  */
669 int
670 rtioctl_fib(u_long req, caddr_t data, u_int fibnum)
671 {
672 
673 	/*
674 	 * If more ioctl commands are added here, make sure the proper
675 	 * super-user checks are being performed because it is possible for
676 	 * prison-root to make it this far if raw sockets have been enabled
677 	 * in jails.
678 	 */
679 #ifdef INET
680 	/* Multicast goop, grrr... */
681 	return mrt_ioctl ? mrt_ioctl(req, data, fibnum) : EOPNOTSUPP;
682 #else /* INET */
683 	return ENXIO;
684 #endif /* INET */
685 }
686 
687 struct ifaddr *
688 ifa_ifwithroute(int flags, const struct sockaddr *dst, struct sockaddr *gateway,
689 				u_int fibnum)
690 {
691 	struct ifaddr *ifa;
692 	int not_found = 0;
693 
694 	NET_EPOCH_ASSERT();
695 	if ((flags & RTF_GATEWAY) == 0) {
696 		/*
697 		 * If we are adding a route to an interface,
698 		 * and the interface is a pt to pt link
699 		 * we should search for the destination
700 		 * as our clue to the interface.  Otherwise
701 		 * we can use the local address.
702 		 */
703 		ifa = NULL;
704 		if (flags & RTF_HOST)
705 			ifa = ifa_ifwithdstaddr(dst, fibnum);
706 		if (ifa == NULL)
707 			ifa = ifa_ifwithaddr(gateway);
708 	} else {
709 		/*
710 		 * If we are adding a route to a remote net
711 		 * or host, the gateway may still be on the
712 		 * other end of a pt to pt link.
713 		 */
714 		ifa = ifa_ifwithdstaddr(gateway, fibnum);
715 	}
716 	if (ifa == NULL)
717 		ifa = ifa_ifwithnet(gateway, 0, fibnum);
718 	if (ifa == NULL) {
719 		struct rtentry *rt;
720 
721 		rt = rtalloc1_fib(gateway, 0, flags, fibnum);
722 		if (rt == NULL)
723 			goto out;
724 		/*
725 		 * dismiss a gateway that is reachable only
726 		 * through the default router
727 		 */
728 		switch (gateway->sa_family) {
729 		case AF_INET:
730 			if (satosin(rt_key(rt))->sin_addr.s_addr == INADDR_ANY)
731 				not_found = 1;
732 			break;
733 		case AF_INET6:
734 			if (IN6_IS_ADDR_UNSPECIFIED(&satosin6(rt_key(rt))->sin6_addr))
735 				not_found = 1;
736 			break;
737 		default:
738 			break;
739 		}
740 		if (!not_found && rt->rt_ifa != NULL) {
741 			ifa = rt->rt_ifa;
742 		}
743 		RT_REMREF(rt);
744 		RT_UNLOCK(rt);
745 		if (not_found || ifa == NULL)
746 			goto out;
747 	}
748 	if (ifa->ifa_addr->sa_family != dst->sa_family) {
749 		struct ifaddr *oifa = ifa;
750 		ifa = ifaof_ifpforaddr(dst, ifa->ifa_ifp);
751 		if (ifa == NULL)
752 			ifa = oifa;
753 	}
754  out:
755 	return (ifa);
756 }
757 
758 /*
759  * Do appropriate manipulations of a routing tree given
760  * all the bits of info needed
761  */
762 int
763 rtrequest_fib(int req,
764 	struct sockaddr *dst,
765 	struct sockaddr *gateway,
766 	struct sockaddr *netmask,
767 	int flags,
768 	struct rtentry **ret_nrt,
769 	u_int fibnum)
770 {
771 	struct rt_addrinfo info;
772 
773 	if (dst->sa_len == 0)
774 		return(EINVAL);
775 
776 	bzero((caddr_t)&info, sizeof(info));
777 	info.rti_flags = flags;
778 	info.rti_info[RTAX_DST] = dst;
779 	info.rti_info[RTAX_GATEWAY] = gateway;
780 	info.rti_info[RTAX_NETMASK] = netmask;
781 	return rtrequest1_fib(req, &info, ret_nrt, fibnum);
782 }
783 
784 
785 /*
786  * Copy most of @rt data into @info.
787  *
788  * If @flags contains NHR_COPY, copies dst,netmask and gw to the
789  * pointers specified by @info structure. Assume such pointers
790  * are zeroed sockaddr-like structures with sa_len field initialized
791  * to reflect size of the provided buffer. if no NHR_COPY is specified,
792  * point dst,netmask and gw @info fields to appropriate @rt values.
793  *
794  * if @flags contains NHR_REF, do refcouting on rt_ifp and rt_ifa.
795  *
796  * Returns 0 on success.
797  */
798 int
799 rt_exportinfo(struct rtentry *rt, struct rt_addrinfo *info, int flags)
800 {
801 	struct rt_metrics *rmx;
802 	struct sockaddr *src, *dst;
803 	int sa_len;
804 
805 	if (flags & NHR_COPY) {
806 		/* Copy destination if dst is non-zero */
807 		src = rt_key(rt);
808 		dst = info->rti_info[RTAX_DST];
809 		sa_len = src->sa_len;
810 		if (dst != NULL) {
811 			if (src->sa_len > dst->sa_len)
812 				return (ENOMEM);
813 			memcpy(dst, src, src->sa_len);
814 			info->rti_addrs |= RTA_DST;
815 		}
816 
817 		/* Copy mask if set && dst is non-zero */
818 		src = rt_mask(rt);
819 		dst = info->rti_info[RTAX_NETMASK];
820 		if (src != NULL && dst != NULL) {
821 
822 			/*
823 			 * Radix stores different value in sa_len,
824 			 * assume rt_mask() to have the same length
825 			 * as rt_key()
826 			 */
827 			if (sa_len > dst->sa_len)
828 				return (ENOMEM);
829 			memcpy(dst, src, src->sa_len);
830 			info->rti_addrs |= RTA_NETMASK;
831 		}
832 
833 		/* Copy gateway is set && dst is non-zero */
834 		src = rt->rt_gateway;
835 		dst = info->rti_info[RTAX_GATEWAY];
836 		if ((rt->rt_flags & RTF_GATEWAY) && src != NULL && dst != NULL){
837 			if (src->sa_len > dst->sa_len)
838 				return (ENOMEM);
839 			memcpy(dst, src, src->sa_len);
840 			info->rti_addrs |= RTA_GATEWAY;
841 		}
842 	} else {
843 		info->rti_info[RTAX_DST] = rt_key(rt);
844 		info->rti_addrs |= RTA_DST;
845 		if (rt_mask(rt) != NULL) {
846 			info->rti_info[RTAX_NETMASK] = rt_mask(rt);
847 			info->rti_addrs |= RTA_NETMASK;
848 		}
849 		if (rt->rt_flags & RTF_GATEWAY) {
850 			info->rti_info[RTAX_GATEWAY] = rt->rt_gateway;
851 			info->rti_addrs |= RTA_GATEWAY;
852 		}
853 	}
854 
855 	rmx = info->rti_rmx;
856 	if (rmx != NULL) {
857 		info->rti_mflags |= RTV_MTU;
858 		rmx->rmx_mtu = rt->rt_mtu;
859 	}
860 
861 	info->rti_flags = rt->rt_flags;
862 	info->rti_ifp = rt->rt_ifp;
863 	info->rti_ifa = rt->rt_ifa;
864 	if (flags & NHR_REF) {
865 		if_ref(info->rti_ifp);
866 		ifa_ref(info->rti_ifa);
867 	}
868 
869 	return (0);
870 }
871 
872 /*
873  * Lookups up route entry for @dst in RIB database for fib @fibnum.
874  * Exports entry data to @info using rt_exportinfo().
875  *
876  * If @flags contains NHR_REF, refcouting is performed on rt_ifp and rt_ifa.
877  * All references can be released later by calling rib_free_info().
878  *
879  * Returns 0 on success.
880  * Returns ENOENT for lookup failure, ENOMEM for export failure.
881  */
882 int
883 rib_lookup_info(uint32_t fibnum, const struct sockaddr *dst, uint32_t flags,
884     uint32_t flowid, struct rt_addrinfo *info)
885 {
886 	RIB_RLOCK_TRACKER;
887 	struct rib_head *rh;
888 	struct radix_node *rn;
889 	struct rtentry *rt;
890 	int error;
891 
892 	KASSERT((fibnum < rt_numfibs), ("rib_lookup_rte: bad fibnum"));
893 	rh = rt_tables_get_rnh(fibnum, dst->sa_family);
894 	if (rh == NULL)
895 		return (ENOENT);
896 
897 	RIB_RLOCK(rh);
898 	rn = rh->rnh_matchaddr(__DECONST(void *, dst), &rh->head);
899 	if (rn != NULL && ((rn->rn_flags & RNF_ROOT) == 0)) {
900 		rt = RNTORT(rn);
901 		/* Ensure route & ifp is UP */
902 		if (RT_LINK_IS_UP(rt->rt_ifp)) {
903 			flags = (flags & NHR_REF) | NHR_COPY;
904 			error = rt_exportinfo(rt, info, flags);
905 			RIB_RUNLOCK(rh);
906 
907 			return (error);
908 		}
909 	}
910 	RIB_RUNLOCK(rh);
911 
912 	return (ENOENT);
913 }
914 
915 /*
916  * Releases all references acquired by rib_lookup_info() when
917  * called with NHR_REF flags.
918  */
919 void
920 rib_free_info(struct rt_addrinfo *info)
921 {
922 
923 	ifa_free(info->rti_ifa);
924 	if_rele(info->rti_ifp);
925 }
926 
927 /*
928  * Iterates over all existing fibs in system calling
929  *  @setwa_f function prior to traversing each fib.
930  *  Calls @wa_f function for each element in current fib.
931  * If af is not AF_UNSPEC, iterates over fibs in particular
932  * address family.
933  */
934 void
935 rt_foreach_fib_walk(int af, rt_setwarg_t *setwa_f, rt_walktree_f_t *wa_f,
936     void *arg)
937 {
938 	struct rib_head *rnh;
939 	uint32_t fibnum;
940 	int i;
941 
942 	for (fibnum = 0; fibnum < rt_numfibs; fibnum++) {
943 		/* Do we want some specific family? */
944 		if (af != AF_UNSPEC) {
945 			rnh = rt_tables_get_rnh(fibnum, af);
946 			if (rnh == NULL)
947 				continue;
948 			if (setwa_f != NULL)
949 				setwa_f(rnh, fibnum, af, arg);
950 
951 			RIB_WLOCK(rnh);
952 			rnh->rnh_walktree(&rnh->head, (walktree_f_t *)wa_f,arg);
953 			RIB_WUNLOCK(rnh);
954 			continue;
955 		}
956 
957 		for (i = 1; i <= AF_MAX; i++) {
958 			rnh = rt_tables_get_rnh(fibnum, i);
959 			if (rnh == NULL)
960 				continue;
961 			if (setwa_f != NULL)
962 				setwa_f(rnh, fibnum, i, arg);
963 
964 			RIB_WLOCK(rnh);
965 			rnh->rnh_walktree(&rnh->head, (walktree_f_t *)wa_f,arg);
966 			RIB_WUNLOCK(rnh);
967 		}
968 	}
969 }
970 
971 struct rt_delinfo
972 {
973 	struct rt_addrinfo info;
974 	struct rib_head *rnh;
975 	struct rtentry *head;
976 };
977 
978 /*
979  * Conditionally unlinks @rn from radix tree based
980  * on info data passed in @arg.
981  */
982 static int
983 rt_checkdelroute(struct radix_node *rn, void *arg)
984 {
985 	struct rt_delinfo *di;
986 	struct rt_addrinfo *info;
987 	struct rtentry *rt;
988 	int error;
989 
990 	di = (struct rt_delinfo *)arg;
991 	rt = (struct rtentry *)rn;
992 	info = &di->info;
993 	error = 0;
994 
995 	info->rti_info[RTAX_DST] = rt_key(rt);
996 	info->rti_info[RTAX_NETMASK] = rt_mask(rt);
997 	info->rti_info[RTAX_GATEWAY] = rt->rt_gateway;
998 
999 	rt = rt_unlinkrte(di->rnh, info, &error);
1000 	if (rt == NULL) {
1001 		/* Either not allowed or not matched. Skip entry */
1002 		return (0);
1003 	}
1004 
1005 	/* Entry was unlinked. Add to the list and return */
1006 	rt->rt_chain = di->head;
1007 	di->head = rt;
1008 
1009 	return (0);
1010 }
1011 
1012 /*
1013  * Iterates over a routing table specified by @fibnum and @family and
1014  *  deletes elements marked by @filter_f.
1015  * @fibnum: rtable id
1016  * @family: AF_ address family
1017  * @filter_f: function returning non-zero value for items to delete
1018  * @arg: data to pass to the @filter_f function
1019  * @report: true if rtsock notification is needed.
1020  */
1021 void
1022 rib_walk_del(u_int fibnum, int family, rt_filter_f_t *filter_f, void *arg, bool report)
1023 {
1024 	struct rib_head *rnh;
1025 	struct rt_delinfo di;
1026 	struct rtentry *rt;
1027 
1028 	rnh = rt_tables_get_rnh(fibnum, family);
1029 	if (rnh == NULL)
1030 		return;
1031 
1032 	bzero(&di, sizeof(di));
1033 	di.info.rti_filter = filter_f;
1034 	di.info.rti_filterdata = arg;
1035 	di.rnh = rnh;
1036 
1037 	RIB_WLOCK(rnh);
1038 	rnh->rnh_walktree(&rnh->head, rt_checkdelroute, &di);
1039 	RIB_WUNLOCK(rnh);
1040 
1041 	if (di.head == NULL)
1042 		return;
1043 
1044 	/* We might have something to reclaim. */
1045 	while (di.head != NULL) {
1046 		rt = di.head;
1047 		di.head = rt->rt_chain;
1048 		rt->rt_chain = NULL;
1049 
1050 		/* TODO std rt -> rt_addrinfo export */
1051 		di.info.rti_info[RTAX_DST] = rt_key(rt);
1052 		di.info.rti_info[RTAX_NETMASK] = rt_mask(rt);
1053 
1054 		rt_notifydelete(rt, &di.info);
1055 
1056 		if (report)
1057 			rt_routemsg(RTM_DELETE, rt, rt->rt_ifp, 0, fibnum);
1058 		RTFREE_LOCKED(rt);
1059 	}
1060 }
1061 
1062 /*
1063  * Iterates over all existing fibs in system and deletes each element
1064  *  for which @filter_f function returns non-zero value.
1065  * If @family is not AF_UNSPEC, iterates over fibs in particular
1066  * address family.
1067  */
1068 void
1069 rt_foreach_fib_walk_del(int family, rt_filter_f_t *filter_f, void *arg)
1070 {
1071 	u_int fibnum;
1072 	int i, start, end;
1073 
1074 	for (fibnum = 0; fibnum < rt_numfibs; fibnum++) {
1075 		/* Do we want some specific family? */
1076 		if (family != AF_UNSPEC) {
1077 			start = family;
1078 			end = family;
1079 		} else {
1080 			start = 1;
1081 			end = AF_MAX;
1082 		}
1083 
1084 		for (i = start; i <= end; i++) {
1085 			if (rt_tables_get_rnh(fibnum, i) == NULL)
1086 				continue;
1087 
1088 			rib_walk_del(fibnum, i, filter_f, arg, 0);
1089 		}
1090 	}
1091 }
1092 
1093 /*
1094  * Delete Routes for a Network Interface
1095  *
1096  * Called for each routing entry via the rnh->rnh_walktree() call above
1097  * to delete all route entries referencing a detaching network interface.
1098  *
1099  * Arguments:
1100  *	rt	pointer to rtentry
1101  *	arg	argument passed to rnh->rnh_walktree() - detaching interface
1102  *
1103  * Returns:
1104  *	0	successful
1105  *	errno	failed - reason indicated
1106  */
1107 static int
1108 rt_ifdelroute(const struct rtentry *rt, void *arg)
1109 {
1110 	struct ifnet	*ifp = arg;
1111 
1112 	if (rt->rt_ifp != ifp)
1113 		return (0);
1114 
1115 	/*
1116 	 * Protect (sorta) against walktree recursion problems
1117 	 * with cloned routes
1118 	 */
1119 	if ((rt->rt_flags & RTF_UP) == 0)
1120 		return (0);
1121 
1122 	return (1);
1123 }
1124 
1125 /*
1126  * Delete all remaining routes using this interface
1127  * Unfortuneatly the only way to do this is to slog through
1128  * the entire routing table looking for routes which point
1129  * to this interface...oh well...
1130  */
1131 void
1132 rt_flushifroutes_af(struct ifnet *ifp, int af)
1133 {
1134 	KASSERT((af >= 1 && af <= AF_MAX), ("%s: af %d not >= 1 and <= %d",
1135 	    __func__, af, AF_MAX));
1136 
1137 	rt_foreach_fib_walk_del(af, rt_ifdelroute, ifp);
1138 }
1139 
1140 void
1141 rt_flushifroutes(struct ifnet *ifp)
1142 {
1143 
1144 	rt_foreach_fib_walk_del(AF_UNSPEC, rt_ifdelroute, ifp);
1145 }
1146 
1147 /*
1148  * Conditionally unlinks rtentry matching data inside @info from @rnh.
1149  * Returns unlinked, locked and referenced @rtentry on success,
1150  * Returns NULL and sets @perror to:
1151  * ESRCH - if prefix was not found,
1152  * EADDRINUSE - if trying to delete PINNED route without appropriate flag.
1153  * ENOENT - if supplied filter function returned 0 (not matched).
1154  */
1155 static struct rtentry *
1156 rt_unlinkrte(struct rib_head *rnh, struct rt_addrinfo *info, int *perror)
1157 {
1158 	struct sockaddr *dst, *netmask;
1159 	struct rtentry *rt;
1160 	struct radix_node *rn;
1161 
1162 	dst = info->rti_info[RTAX_DST];
1163 	netmask = info->rti_info[RTAX_NETMASK];
1164 
1165 	rt = (struct rtentry *)rnh->rnh_lookup(dst, netmask, &rnh->head);
1166 	if (rt == NULL) {
1167 		*perror = ESRCH;
1168 		return (NULL);
1169 	}
1170 
1171 	if ((info->rti_flags & RTF_PINNED) == 0) {
1172 		/* Check if target route can be deleted */
1173 		if (rt->rt_flags & RTF_PINNED) {
1174 			*perror = EADDRINUSE;
1175 			return (NULL);
1176 		}
1177 	}
1178 
1179 	if (info->rti_filter != NULL) {
1180 		if (info->rti_filter(rt, info->rti_filterdata) == 0) {
1181 			/* Not matched */
1182 			*perror = ENOENT;
1183 			return (NULL);
1184 		}
1185 
1186 		/*
1187 		 * Filter function requested rte deletion.
1188 		 * Ease the caller work by filling in remaining info
1189 		 * from that particular entry.
1190 		 */
1191 		info->rti_info[RTAX_GATEWAY] = rt->rt_gateway;
1192 	}
1193 
1194 	/*
1195 	 * Remove the item from the tree and return it.
1196 	 * Complain if it is not there and do no more processing.
1197 	 */
1198 	*perror = ESRCH;
1199 #ifdef RADIX_MPATH
1200 	if (rt_mpath_capable(rnh))
1201 		rn = rt_mpath_unlink(rnh, info, rt, perror);
1202 	else
1203 #endif
1204 	rn = rnh->rnh_deladdr(dst, netmask, &rnh->head);
1205 	if (rn == NULL)
1206 		return (NULL);
1207 
1208 	if (rn->rn_flags & (RNF_ACTIVE | RNF_ROOT))
1209 		panic ("rtrequest delete");
1210 
1211 	rt = RNTORT(rn);
1212 	RT_LOCK(rt);
1213 	RT_ADDREF(rt);
1214 	rt->rt_flags &= ~RTF_UP;
1215 
1216 	*perror = 0;
1217 
1218 	return (rt);
1219 }
1220 
1221 static void
1222 rt_notifydelete(struct rtentry *rt, struct rt_addrinfo *info)
1223 {
1224 	struct ifaddr *ifa;
1225 
1226 	/*
1227 	 * give the protocol a chance to keep things in sync.
1228 	 */
1229 	ifa = rt->rt_ifa;
1230 	if (ifa != NULL && ifa->ifa_rtrequest != NULL)
1231 		ifa->ifa_rtrequest(RTM_DELETE, rt, info);
1232 
1233 	/*
1234 	 * One more rtentry floating around that is not
1235 	 * linked to the routing table. rttrash will be decremented
1236 	 * when RTFREE(rt) is eventually called.
1237 	 */
1238 	V_rttrash++;
1239 }
1240 
1241 
1242 /*
1243  * These (questionable) definitions of apparent local variables apply
1244  * to the next two functions.  XXXXXX!!!
1245  */
1246 #define	dst	info->rti_info[RTAX_DST]
1247 #define	gateway	info->rti_info[RTAX_GATEWAY]
1248 #define	netmask	info->rti_info[RTAX_NETMASK]
1249 #define	ifaaddr	info->rti_info[RTAX_IFA]
1250 #define	ifpaddr	info->rti_info[RTAX_IFP]
1251 #define	flags	info->rti_flags
1252 
1253 /*
1254  * Look up rt_addrinfo for a specific fib.  Note that if rti_ifa is defined,
1255  * it will be referenced so the caller must free it.
1256  *
1257  * Assume basic consistency checks are executed by callers:
1258  * RTAX_DST exists, if RTF_GATEWAY is set, RTAX_GATEWAY exists as well.
1259  */
1260 int
1261 rt_getifa_fib(struct rt_addrinfo *info, u_int fibnum)
1262 {
1263 	struct epoch_tracker et;
1264 	int needref, error;
1265 
1266 	/*
1267 	 * ifp may be specified by sockaddr_dl
1268 	 * when protocol address is ambiguous.
1269 	 */
1270 	error = 0;
1271 	needref = (info->rti_ifa == NULL);
1272 	NET_EPOCH_ENTER(et);
1273 
1274 	/* If we have interface specified by the ifindex in the address, use it */
1275 	if (info->rti_ifp == NULL && ifpaddr != NULL &&
1276 	    ifpaddr->sa_family == AF_LINK) {
1277 	    const struct sockaddr_dl *sdl = (const struct sockaddr_dl *)ifpaddr;
1278 	    if (sdl->sdl_index != 0)
1279 		    info->rti_ifp = ifnet_byindex(sdl->sdl_index);
1280 	}
1281 	/*
1282 	 * If we have source address specified, try to find it
1283 	 * TODO: avoid enumerating all ifas on all interfaces.
1284 	 */
1285 	if (info->rti_ifa == NULL && ifaaddr != NULL)
1286 		info->rti_ifa = ifa_ifwithaddr(ifaaddr);
1287 	if (info->rti_ifa == NULL) {
1288 		struct sockaddr *sa;
1289 
1290 		/*
1291 		 * Most common use case for the userland-supplied routes.
1292 		 *
1293 		 * Choose sockaddr to select ifa.
1294 		 * -- if ifp is set --
1295 		 * Order of preference:
1296 		 * 1) IFA address
1297 		 * 2) gateway address
1298 		 *   Note: for interface routes link-level gateway address
1299 		 *     is specified to indicate the interface index without
1300 		 *     specifying RTF_GATEWAY. In this case, ignore gateway
1301 		 *   Note: gateway AF may be different from dst AF. In this case,
1302 		 *   ignore gateway
1303 		 * 3) final destination.
1304 		 * 4) if all of these fails, try to get at least link-level ifa.
1305 		 * -- else --
1306 		 * try to lookup gateway or dst in the routing table to get ifa
1307 		 */
1308 		if (info->rti_info[RTAX_IFA] != NULL)
1309 			sa = info->rti_info[RTAX_IFA];
1310 		else if ((info->rti_flags & RTF_GATEWAY) != 0 &&
1311 		    gateway->sa_family == dst->sa_family)
1312 			sa = gateway;
1313 		else
1314 			sa = dst;
1315 		if (info->rti_ifp != NULL) {
1316 			info->rti_ifa = ifaof_ifpforaddr(sa, info->rti_ifp);
1317 			/* Case 4 */
1318 			if (info->rti_ifa == NULL && gateway != NULL)
1319 				info->rti_ifa = ifaof_ifpforaddr(gateway, info->rti_ifp);
1320 		} else if (dst != NULL && gateway != NULL)
1321 			info->rti_ifa = ifa_ifwithroute(flags, dst, gateway,
1322 							fibnum);
1323 		else if (sa != NULL)
1324 			info->rti_ifa = ifa_ifwithroute(flags, sa, sa,
1325 							fibnum);
1326 	}
1327 	if (needref && info->rti_ifa != NULL) {
1328 		if (info->rti_ifp == NULL)
1329 			info->rti_ifp = info->rti_ifa->ifa_ifp;
1330 		ifa_ref(info->rti_ifa);
1331 	} else
1332 		error = ENETUNREACH;
1333 	NET_EPOCH_EXIT(et);
1334 	return (error);
1335 }
1336 
1337 static int
1338 if_updatemtu_cb(struct radix_node *rn, void *arg)
1339 {
1340 	struct rtentry *rt;
1341 	struct if_mtuinfo *ifmtu;
1342 
1343 	rt = (struct rtentry *)rn;
1344 	ifmtu = (struct if_mtuinfo *)arg;
1345 
1346 	if (rt->rt_ifp != ifmtu->ifp)
1347 		return (0);
1348 
1349 	if (rt->rt_mtu >= ifmtu->mtu) {
1350 		/* We have to decrease mtu regardless of flags */
1351 		rt->rt_mtu = ifmtu->mtu;
1352 		return (0);
1353 	}
1354 
1355 	/*
1356 	 * New MTU is bigger. Check if are allowed to alter it
1357 	 */
1358 	if ((rt->rt_flags & (RTF_FIXEDMTU | RTF_GATEWAY | RTF_HOST)) != 0) {
1359 
1360 		/*
1361 		 * Skip routes with user-supplied MTU and
1362 		 * non-interface routes
1363 		 */
1364 		return (0);
1365 	}
1366 
1367 	/* We are safe to update route MTU */
1368 	rt->rt_mtu = ifmtu->mtu;
1369 
1370 	return (0);
1371 }
1372 
1373 void
1374 rt_updatemtu(struct ifnet *ifp)
1375 {
1376 	struct if_mtuinfo ifmtu;
1377 	struct rib_head *rnh;
1378 	int i, j;
1379 
1380 	ifmtu.ifp = ifp;
1381 
1382 	/*
1383 	 * Try to update rt_mtu for all routes using this interface
1384 	 * Unfortunately the only way to do this is to traverse all
1385 	 * routing tables in all fibs/domains.
1386 	 */
1387 	for (i = 1; i <= AF_MAX; i++) {
1388 		ifmtu.mtu = if_getmtu_family(ifp, i);
1389 		for (j = 0; j < rt_numfibs; j++) {
1390 			rnh = rt_tables_get_rnh(j, i);
1391 			if (rnh == NULL)
1392 				continue;
1393 			RIB_WLOCK(rnh);
1394 			rnh->rnh_walktree(&rnh->head, if_updatemtu_cb, &ifmtu);
1395 			RIB_WUNLOCK(rnh);
1396 		}
1397 	}
1398 }
1399 
1400 
1401 #if 0
1402 int p_sockaddr(char *buf, int buflen, struct sockaddr *s);
1403 int rt_print(char *buf, int buflen, struct rtentry *rt);
1404 
1405 int
1406 p_sockaddr(char *buf, int buflen, struct sockaddr *s)
1407 {
1408 	void *paddr = NULL;
1409 
1410 	switch (s->sa_family) {
1411 	case AF_INET:
1412 		paddr = &((struct sockaddr_in *)s)->sin_addr;
1413 		break;
1414 	case AF_INET6:
1415 		paddr = &((struct sockaddr_in6 *)s)->sin6_addr;
1416 		break;
1417 	}
1418 
1419 	if (paddr == NULL)
1420 		return (0);
1421 
1422 	if (inet_ntop(s->sa_family, paddr, buf, buflen) == NULL)
1423 		return (0);
1424 
1425 	return (strlen(buf));
1426 }
1427 
1428 int
1429 rt_print(char *buf, int buflen, struct rtentry *rt)
1430 {
1431 	struct sockaddr *addr, *mask;
1432 	int i = 0;
1433 
1434 	addr = rt_key(rt);
1435 	mask = rt_mask(rt);
1436 
1437 	i = p_sockaddr(buf, buflen, addr);
1438 	if (!(rt->rt_flags & RTF_HOST)) {
1439 		buf[i++] = '/';
1440 		i += p_sockaddr(buf + i, buflen - i, mask);
1441 	}
1442 
1443 	if (rt->rt_flags & RTF_GATEWAY) {
1444 		buf[i++] = '>';
1445 		i += p_sockaddr(buf + i, buflen - i, rt->rt_gateway);
1446 	}
1447 
1448 	return (i);
1449 }
1450 #endif
1451 
1452 #ifdef RADIX_MPATH
1453 /*
1454  * Deletes key for single-path routes, unlinks rtentry with
1455  * gateway specified in @info from multi-path routes.
1456  *
1457  * Returnes unlinked entry. In case of failure, returns NULL
1458  * and sets @perror to ESRCH.
1459  */
1460 static struct radix_node *
1461 rt_mpath_unlink(struct rib_head *rnh, struct rt_addrinfo *info,
1462     struct rtentry *rto, int *perror)
1463 {
1464 	/*
1465 	 * if we got multipath routes, we require users to specify
1466 	 * a matching RTAX_GATEWAY.
1467 	 */
1468 	struct rtentry *rt; // *rto = NULL;
1469 	struct radix_node *rn;
1470 	struct sockaddr *gw;
1471 
1472 	gw = info->rti_info[RTAX_GATEWAY];
1473 	rt = rt_mpath_matchgate(rto, gw);
1474 	if (rt == NULL) {
1475 		*perror = ESRCH;
1476 		return (NULL);
1477 	}
1478 
1479 	/*
1480 	 * this is the first entry in the chain
1481 	 */
1482 	if (rto == rt) {
1483 		rn = rn_mpath_next((struct radix_node *)rt);
1484 		/*
1485 		 * there is another entry, now it's active
1486 		 */
1487 		if (rn) {
1488 			rto = RNTORT(rn);
1489 			RT_LOCK(rto);
1490 			rto->rt_flags |= RTF_UP;
1491 			RT_UNLOCK(rto);
1492 		} else if (rt->rt_flags & RTF_GATEWAY) {
1493 			/*
1494 			 * For gateway routes, we need to
1495 			 * make sure that we we are deleting
1496 			 * the correct gateway.
1497 			 * rt_mpath_matchgate() does not
1498 			 * check the case when there is only
1499 			 * one route in the chain.
1500 			 */
1501 			if (gw &&
1502 			    (rt->rt_gateway->sa_len != gw->sa_len ||
1503 				memcmp(rt->rt_gateway, gw, gw->sa_len))) {
1504 				*perror = ESRCH;
1505 				return (NULL);
1506 			}
1507 		}
1508 
1509 		/*
1510 		 * use the normal delete code to remove
1511 		 * the first entry
1512 		 */
1513 		rn = rnh->rnh_deladdr(dst, netmask, &rnh->head);
1514 		*perror = 0;
1515 		return (rn);
1516 	}
1517 
1518 	/*
1519 	 * if the entry is 2nd and on up
1520 	 */
1521 	if (rt_mpath_deldup(rto, rt) == 0)
1522 		panic ("rtrequest1: rt_mpath_deldup");
1523 	*perror = 0;
1524 	rn = (struct radix_node *)rt;
1525 	return (rn);
1526 }
1527 #endif
1528 
1529 int
1530 rtrequest1_fib(int req, struct rt_addrinfo *info, struct rtentry **ret_nrt,
1531 				u_int fibnum)
1532 {
1533 	int error = 0;
1534 	struct rtentry *rt, *rt_old;
1535 	struct radix_node *rn;
1536 	struct rib_head *rnh;
1537 	struct ifaddr *ifa;
1538 	struct sockaddr *ndst;
1539 	struct sockaddr_storage mdst;
1540 
1541 	KASSERT((fibnum < rt_numfibs), ("rtrequest1_fib: bad fibnum"));
1542 	KASSERT((flags & RTF_RNH_LOCKED) == 0, ("rtrequest1_fib: locked"));
1543 	switch (dst->sa_family) {
1544 	case AF_INET6:
1545 	case AF_INET:
1546 		/* We support multiple FIBs. */
1547 		break;
1548 	default:
1549 		fibnum = RT_DEFAULT_FIB;
1550 		break;
1551 	}
1552 
1553 	/*
1554 	 * Find the correct routing tree to use for this Address Family
1555 	 */
1556 	rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
1557 	if (rnh == NULL)
1558 		return (EAFNOSUPPORT);
1559 
1560 	/*
1561 	 * If we are adding a host route then we don't want to put
1562 	 * a netmask in the tree, nor do we want to clone it.
1563 	 */
1564 	if (flags & RTF_HOST)
1565 		netmask = NULL;
1566 
1567 	switch (req) {
1568 	case RTM_DELETE:
1569 		if (netmask) {
1570 			if (dst->sa_len > sizeof(mdst))
1571 				return (EINVAL);
1572 			rt_maskedcopy(dst, (struct sockaddr *)&mdst, netmask);
1573 			dst = (struct sockaddr *)&mdst;
1574 		}
1575 
1576 		RIB_WLOCK(rnh);
1577 		rt = rt_unlinkrte(rnh, info, &error);
1578 		RIB_WUNLOCK(rnh);
1579 		if (error != 0)
1580 			return (error);
1581 
1582 		rt_notifydelete(rt, info);
1583 
1584 		/*
1585 		 * If the caller wants it, then it can have it,
1586 		 * but it's up to it to free the rtentry as we won't be
1587 		 * doing it.
1588 		 */
1589 		if (ret_nrt) {
1590 			*ret_nrt = rt;
1591 			RT_UNLOCK(rt);
1592 		} else
1593 			RTFREE_LOCKED(rt);
1594 		break;
1595 	case RTM_RESOLVE:
1596 		/*
1597 		 * resolve was only used for route cloning
1598 		 * here for compat
1599 		 */
1600 		break;
1601 	case RTM_ADD:
1602 		if ((flags & RTF_GATEWAY) && !gateway)
1603 			return (EINVAL);
1604 		if (dst && gateway && (dst->sa_family != gateway->sa_family) &&
1605 		    (gateway->sa_family != AF_UNSPEC) && (gateway->sa_family != AF_LINK))
1606 			return (EINVAL);
1607 
1608 		if (info->rti_ifa == NULL) {
1609 			error = rt_getifa_fib(info, fibnum);
1610 			if (error)
1611 				return (error);
1612 		} else {
1613 			ifa_ref(info->rti_ifa);
1614 		}
1615 		rt = uma_zalloc(V_rtzone, M_NOWAIT);
1616 		if (rt == NULL) {
1617 			ifa_free(info->rti_ifa);
1618 			return (ENOBUFS);
1619 		}
1620 		rt->rt_flags = RTF_UP | flags;
1621 		rt->rt_fibnum = fibnum;
1622 		/*
1623 		 * Add the gateway. Possibly re-malloc-ing the storage for it.
1624 		 */
1625 		if ((error = rt_setgate(rt, dst, gateway)) != 0) {
1626 			ifa_free(info->rti_ifa);
1627 			uma_zfree(V_rtzone, rt);
1628 			return (error);
1629 		}
1630 
1631 		/*
1632 		 * point to the (possibly newly malloc'd) dest address.
1633 		 */
1634 		ndst = (struct sockaddr *)rt_key(rt);
1635 
1636 		/*
1637 		 * make sure it contains the value we want (masked if needed).
1638 		 */
1639 		if (netmask) {
1640 			rt_maskedcopy(dst, ndst, netmask);
1641 		} else
1642 			bcopy(dst, ndst, dst->sa_len);
1643 
1644 		/*
1645 		 * We use the ifa reference returned by rt_getifa_fib().
1646 		 * This moved from below so that rnh->rnh_addaddr() can
1647 		 * examine the ifa and  ifa->ifa_ifp if it so desires.
1648 		 */
1649 		ifa = info->rti_ifa;
1650 		rt->rt_ifa = ifa;
1651 		rt->rt_ifp = ifa->ifa_ifp;
1652 		rt->rt_weight = 1;
1653 
1654 		rt_setmetrics(info, rt);
1655 
1656 		RIB_WLOCK(rnh);
1657 		RT_LOCK(rt);
1658 #ifdef RADIX_MPATH
1659 		/* do not permit exactly the same dst/mask/gw pair */
1660 		if (rt_mpath_capable(rnh) &&
1661 			rt_mpath_conflict(rnh, rt, netmask)) {
1662 			RIB_WUNLOCK(rnh);
1663 
1664 			ifa_free(rt->rt_ifa);
1665 			R_Free(rt_key(rt));
1666 			uma_zfree(V_rtzone, rt);
1667 			return (EEXIST);
1668 		}
1669 #endif
1670 
1671 		/* XXX mtu manipulation will be done in rnh_addaddr -- itojun */
1672 		rn = rnh->rnh_addaddr(ndst, netmask, &rnh->head, rt->rt_nodes);
1673 
1674 		if (rn != NULL && rt->rt_expire > 0)
1675 			tmproutes_update(rnh, rt);
1676 
1677 		rt_old = NULL;
1678 		if (rn == NULL && (info->rti_flags & RTF_PINNED) != 0) {
1679 
1680 			/*
1681 			 * Force removal and re-try addition
1682 			 * TODO: better multipath&pinned support
1683 			 */
1684 			struct sockaddr *info_dst = info->rti_info[RTAX_DST];
1685 			info->rti_info[RTAX_DST] = ndst;
1686 			/* Do not delete existing PINNED(interface) routes */
1687 			info->rti_flags &= ~RTF_PINNED;
1688 			rt_old = rt_unlinkrte(rnh, info, &error);
1689 			info->rti_flags |= RTF_PINNED;
1690 			info->rti_info[RTAX_DST] = info_dst;
1691 			if (rt_old != NULL)
1692 				rn = rnh->rnh_addaddr(ndst, netmask, &rnh->head,
1693 				    rt->rt_nodes);
1694 		}
1695 		RIB_WUNLOCK(rnh);
1696 
1697 		if (rt_old != NULL)
1698 			RT_UNLOCK(rt_old);
1699 
1700 		/*
1701 		 * If it still failed to go into the tree,
1702 		 * then un-make it (this should be a function)
1703 		 */
1704 		if (rn == NULL) {
1705 			ifa_free(rt->rt_ifa);
1706 			R_Free(rt_key(rt));
1707 			uma_zfree(V_rtzone, rt);
1708 			return (EEXIST);
1709 		}
1710 
1711 		if (rt_old != NULL) {
1712 			rt_notifydelete(rt_old, info);
1713 			RTFREE(rt_old);
1714 		}
1715 
1716 		/*
1717 		 * If this protocol has something to add to this then
1718 		 * allow it to do that as well.
1719 		 */
1720 		if (ifa->ifa_rtrequest)
1721 			ifa->ifa_rtrequest(req, rt, info);
1722 
1723 		/*
1724 		 * actually return a resultant rtentry and
1725 		 * give the caller a single reference.
1726 		 */
1727 		if (ret_nrt) {
1728 			*ret_nrt = rt;
1729 			RT_ADDREF(rt);
1730 		}
1731 		rnh->rnh_gen++;		/* Routing table updated */
1732 		RT_UNLOCK(rt);
1733 		break;
1734 	case RTM_CHANGE:
1735 		RIB_WLOCK(rnh);
1736 		error = rtrequest1_fib_change(rnh, info, ret_nrt, fibnum);
1737 		RIB_WUNLOCK(rnh);
1738 		break;
1739 	default:
1740 		error = EOPNOTSUPP;
1741 	}
1742 
1743 	return (error);
1744 }
1745 
1746 #undef dst
1747 #undef gateway
1748 #undef netmask
1749 #undef ifaaddr
1750 #undef ifpaddr
1751 #undef flags
1752 
1753 static int
1754 rtrequest1_fib_change(struct rib_head *rnh, struct rt_addrinfo *info,
1755     struct rtentry **ret_nrt, u_int fibnum)
1756 {
1757 	struct rtentry *rt = NULL;
1758 	int error = 0;
1759 	int free_ifa = 0;
1760 	int family, mtu;
1761 	struct if_mtuinfo ifmtu;
1762 
1763 	RIB_WLOCK_ASSERT(rnh);
1764 
1765 	rt = (struct rtentry *)rnh->rnh_lookup(info->rti_info[RTAX_DST],
1766 	    info->rti_info[RTAX_NETMASK], &rnh->head);
1767 
1768 	if (rt == NULL)
1769 		return (ESRCH);
1770 
1771 #ifdef RADIX_MPATH
1772 	/*
1773 	 * If we got multipath routes,
1774 	 * we require users to specify a matching RTAX_GATEWAY.
1775 	 */
1776 	if (rt_mpath_capable(rnh)) {
1777 		rt = rt_mpath_matchgate(rt, info->rti_info[RTAX_GATEWAY]);
1778 		if (rt == NULL)
1779 			return (ESRCH);
1780 	}
1781 #endif
1782 
1783 	RT_LOCK(rt);
1784 
1785 	rt_setmetrics(info, rt);
1786 
1787 	/*
1788 	 * New gateway could require new ifaddr, ifp;
1789 	 * flags may also be different; ifp may be specified
1790 	 * by ll sockaddr when protocol address is ambiguous
1791 	 */
1792 	if (((rt->rt_flags & RTF_GATEWAY) &&
1793 	    info->rti_info[RTAX_GATEWAY] != NULL) ||
1794 	    info->rti_info[RTAX_IFP] != NULL ||
1795 	    (info->rti_info[RTAX_IFA] != NULL &&
1796 	     !sa_equal(info->rti_info[RTAX_IFA], rt->rt_ifa->ifa_addr))) {
1797 		/*
1798 		 * XXX: Temporarily set RTF_RNH_LOCKED flag in the rti_flags
1799 		 *	to avoid rlock in the ifa_ifwithroute().
1800 		 */
1801 		info->rti_flags |= RTF_RNH_LOCKED;
1802 		error = rt_getifa_fib(info, fibnum);
1803 		info->rti_flags &= ~RTF_RNH_LOCKED;
1804 		if (info->rti_ifa != NULL)
1805 			free_ifa = 1;
1806 
1807 		if (error != 0)
1808 			goto bad;
1809 	}
1810 
1811 	/* Check if outgoing interface has changed */
1812 	if (info->rti_ifa != NULL && info->rti_ifa != rt->rt_ifa &&
1813 	    rt->rt_ifa != NULL) {
1814 		if (rt->rt_ifa->ifa_rtrequest != NULL)
1815 			rt->rt_ifa->ifa_rtrequest(RTM_DELETE, rt, info);
1816 		ifa_free(rt->rt_ifa);
1817 		rt->rt_ifa = NULL;
1818 	}
1819 	/* Update gateway address */
1820 	if (info->rti_info[RTAX_GATEWAY] != NULL) {
1821 		error = rt_setgate(rt, rt_key(rt), info->rti_info[RTAX_GATEWAY]);
1822 		if (error != 0)
1823 			goto bad;
1824 
1825 		rt->rt_flags &= ~RTF_GATEWAY;
1826 		rt->rt_flags |= (RTF_GATEWAY & info->rti_flags);
1827 	}
1828 
1829 	if (info->rti_ifa != NULL && info->rti_ifa != rt->rt_ifa) {
1830 		ifa_ref(info->rti_ifa);
1831 		rt->rt_ifa = info->rti_ifa;
1832 		rt->rt_ifp = info->rti_ifp;
1833 	}
1834 	/* Allow some flags to be toggled on change. */
1835 	rt->rt_flags &= ~RTF_FMASK;
1836 	rt->rt_flags |= info->rti_flags & RTF_FMASK;
1837 
1838 	if (rt->rt_ifa && rt->rt_ifa->ifa_rtrequest != NULL)
1839 	       rt->rt_ifa->ifa_rtrequest(RTM_ADD, rt, info);
1840 
1841 	/* Alter route MTU if necessary */
1842 	if (rt->rt_ifp != NULL) {
1843 		family = info->rti_info[RTAX_DST]->sa_family;
1844 		mtu = if_getmtu_family(rt->rt_ifp, family);
1845 		/* Set default MTU */
1846 		if (rt->rt_mtu == 0)
1847 			rt->rt_mtu = mtu;
1848 		if (rt->rt_mtu != mtu) {
1849 			/* Check if we really need to update */
1850 			ifmtu.ifp = rt->rt_ifp;
1851 			ifmtu.mtu = mtu;
1852 			if_updatemtu_cb(rt->rt_nodes, &ifmtu);
1853 		}
1854 	}
1855 
1856 	/*
1857 	 * This route change may have modified the route's gateway.  In that
1858 	 * case, any inpcbs that have cached this route need to invalidate their
1859 	 * llentry cache.
1860 	 */
1861 	rnh->rnh_gen++;
1862 
1863 	if (ret_nrt) {
1864 		*ret_nrt = rt;
1865 		RT_ADDREF(rt);
1866 	}
1867 bad:
1868 	RT_UNLOCK(rt);
1869 	if (free_ifa != 0) {
1870 		ifa_free(info->rti_ifa);
1871 		info->rti_ifa = NULL;
1872 	}
1873 	return (error);
1874 }
1875 
1876 static void
1877 rt_setmetrics(const struct rt_addrinfo *info, struct rtentry *rt)
1878 {
1879 
1880 	if (info->rti_mflags & RTV_MTU) {
1881 		if (info->rti_rmx->rmx_mtu != 0) {
1882 
1883 			/*
1884 			 * MTU was explicitly provided by user.
1885 			 * Keep it.
1886 			 */
1887 			rt->rt_flags |= RTF_FIXEDMTU;
1888 		} else {
1889 
1890 			/*
1891 			 * User explicitly sets MTU to 0.
1892 			 * Assume rollback to default.
1893 			 */
1894 			rt->rt_flags &= ~RTF_FIXEDMTU;
1895 		}
1896 		rt->rt_mtu = info->rti_rmx->rmx_mtu;
1897 	}
1898 	if (info->rti_mflags & RTV_WEIGHT)
1899 		rt->rt_weight = info->rti_rmx->rmx_weight;
1900 	/* Kernel -> userland timebase conversion. */
1901 	if (info->rti_mflags & RTV_EXPIRE)
1902 		rt->rt_expire = info->rti_rmx->rmx_expire ?
1903 		    info->rti_rmx->rmx_expire - time_second + time_uptime : 0;
1904 }
1905 
1906 int
1907 rt_setgate(struct rtentry *rt, struct sockaddr *dst, struct sockaddr *gate)
1908 {
1909 	/* XXX dst may be overwritten, can we move this to below */
1910 	int dlen = SA_SIZE(dst), glen = SA_SIZE(gate);
1911 
1912 	/*
1913 	 * Prepare to store the gateway in rt->rt_gateway.
1914 	 * Both dst and gateway are stored one after the other in the same
1915 	 * malloc'd chunk. If we have room, we can reuse the old buffer,
1916 	 * rt_gateway already points to the right place.
1917 	 * Otherwise, malloc a new block and update the 'dst' address.
1918 	 */
1919 	if (rt->rt_gateway == NULL || glen > SA_SIZE(rt->rt_gateway)) {
1920 		caddr_t new;
1921 
1922 		R_Malloc(new, caddr_t, dlen + glen);
1923 		if (new == NULL)
1924 			return ENOBUFS;
1925 		/*
1926 		 * XXX note, we copy from *dst and not *rt_key(rt) because
1927 		 * rt_setgate() can be called to initialize a newly
1928 		 * allocated route entry, in which case rt_key(rt) == NULL
1929 		 * (and also rt->rt_gateway == NULL).
1930 		 * Free()/free() handle a NULL argument just fine.
1931 		 */
1932 		bcopy(dst, new, dlen);
1933 		R_Free(rt_key(rt));	/* free old block, if any */
1934 		rt_key(rt) = (struct sockaddr *)new;
1935 		rt->rt_gateway = (struct sockaddr *)(new + dlen);
1936 	}
1937 
1938 	/*
1939 	 * Copy the new gateway value into the memory chunk.
1940 	 */
1941 	bcopy(gate, rt->rt_gateway, glen);
1942 
1943 	return (0);
1944 }
1945 
1946 void
1947 rt_maskedcopy(struct sockaddr *src, struct sockaddr *dst, struct sockaddr *netmask)
1948 {
1949 	u_char *cp1 = (u_char *)src;
1950 	u_char *cp2 = (u_char *)dst;
1951 	u_char *cp3 = (u_char *)netmask;
1952 	u_char *cplim = cp2 + *cp3;
1953 	u_char *cplim2 = cp2 + *cp1;
1954 
1955 	*cp2++ = *cp1++; *cp2++ = *cp1++; /* copies sa_len & sa_family */
1956 	cp3 += 2;
1957 	if (cplim > cplim2)
1958 		cplim = cplim2;
1959 	while (cp2 < cplim)
1960 		*cp2++ = *cp1++ & *cp3++;
1961 	if (cp2 < cplim2)
1962 		bzero((caddr_t)cp2, (unsigned)(cplim2 - cp2));
1963 }
1964 
1965 /*
1966  * Set up a routing table entry, normally
1967  * for an interface.
1968  */
1969 #define _SOCKADDR_TMPSIZE 128 /* Not too big.. kernel stack size is limited */
1970 static inline  int
1971 rtinit1(struct ifaddr *ifa, int cmd, int flags, int fibnum)
1972 {
1973 	RIB_RLOCK_TRACKER;
1974 	struct sockaddr *dst;
1975 	struct sockaddr *netmask;
1976 	struct rtentry *rt = NULL;
1977 	struct rt_addrinfo info;
1978 	int error = 0;
1979 	int startfib, endfib;
1980 	char tempbuf[_SOCKADDR_TMPSIZE];
1981 	int didwork = 0;
1982 	int a_failure = 0;
1983 	struct sockaddr_dl *sdl = NULL;
1984 	struct rib_head *rnh;
1985 
1986 	if (flags & RTF_HOST) {
1987 		dst = ifa->ifa_dstaddr;
1988 		netmask = NULL;
1989 	} else {
1990 		dst = ifa->ifa_addr;
1991 		netmask = ifa->ifa_netmask;
1992 	}
1993 	if (dst->sa_len == 0)
1994 		return(EINVAL);
1995 	switch (dst->sa_family) {
1996 	case AF_INET6:
1997 	case AF_INET:
1998 		/* We support multiple FIBs. */
1999 		break;
2000 	default:
2001 		fibnum = RT_DEFAULT_FIB;
2002 		break;
2003 	}
2004 	if (fibnum == RT_ALL_FIBS) {
2005 		if (V_rt_add_addr_allfibs == 0 && cmd == (int)RTM_ADD)
2006 			startfib = endfib = ifa->ifa_ifp->if_fib;
2007 		else {
2008 			startfib = 0;
2009 			endfib = rt_numfibs - 1;
2010 		}
2011 	} else {
2012 		KASSERT((fibnum < rt_numfibs), ("rtinit1: bad fibnum"));
2013 		startfib = fibnum;
2014 		endfib = fibnum;
2015 	}
2016 
2017 	/*
2018 	 * If it's a delete, check that if it exists,
2019 	 * it's on the correct interface or we might scrub
2020 	 * a route to another ifa which would
2021 	 * be confusing at best and possibly worse.
2022 	 */
2023 	if (cmd == RTM_DELETE) {
2024 		/*
2025 		 * It's a delete, so it should already exist..
2026 		 * If it's a net, mask off the host bits
2027 		 * (Assuming we have a mask)
2028 		 * XXX this is kinda inet specific..
2029 		 */
2030 		if (netmask != NULL) {
2031 			rt_maskedcopy(dst, (struct sockaddr *)tempbuf, netmask);
2032 			dst = (struct sockaddr *)tempbuf;
2033 		}
2034 	} else if (cmd == RTM_ADD) {
2035 		sdl = (struct sockaddr_dl *)tempbuf;
2036 		bzero(sdl, sizeof(struct sockaddr_dl));
2037 		sdl->sdl_family = AF_LINK;
2038 		sdl->sdl_len = sizeof(struct sockaddr_dl);
2039 		sdl->sdl_type = ifa->ifa_ifp->if_type;
2040 		sdl->sdl_index = ifa->ifa_ifp->if_index;
2041         }
2042 	/*
2043 	 * Now go through all the requested tables (fibs) and do the
2044 	 * requested action. Realistically, this will either be fib 0
2045 	 * for protocols that don't do multiple tables or all the
2046 	 * tables for those that do.
2047 	 */
2048 	for ( fibnum = startfib; fibnum <= endfib; fibnum++) {
2049 		if (cmd == RTM_DELETE) {
2050 			struct radix_node *rn;
2051 			/*
2052 			 * Look up an rtentry that is in the routing tree and
2053 			 * contains the correct info.
2054 			 */
2055 			rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
2056 			if (rnh == NULL)
2057 				/* this table doesn't exist but others might */
2058 				continue;
2059 			RIB_RLOCK(rnh);
2060 			rn = rnh->rnh_lookup(dst, netmask, &rnh->head);
2061 #ifdef RADIX_MPATH
2062 			if (rt_mpath_capable(rnh)) {
2063 
2064 				if (rn == NULL)
2065 					error = ESRCH;
2066 				else {
2067 					rt = RNTORT(rn);
2068 					/*
2069 					 * for interface route the
2070 					 * rt->rt_gateway is sockaddr_intf
2071 					 * for cloning ARP entries, so
2072 					 * rt_mpath_matchgate must use the
2073 					 * interface address
2074 					 */
2075 					rt = rt_mpath_matchgate(rt,
2076 					    ifa->ifa_addr);
2077 					if (rt == NULL)
2078 						error = ESRCH;
2079 				}
2080 			}
2081 #endif
2082 			error = (rn == NULL ||
2083 			    (rn->rn_flags & RNF_ROOT) ||
2084 			    RNTORT(rn)->rt_ifa != ifa);
2085 			RIB_RUNLOCK(rnh);
2086 			if (error) {
2087 				/* this is only an error if bad on ALL tables */
2088 				continue;
2089 			}
2090 		}
2091 		/*
2092 		 * Do the actual request
2093 		 */
2094 		bzero((caddr_t)&info, sizeof(info));
2095 		info.rti_ifa = ifa;
2096 		info.rti_flags = flags |
2097 		    (ifa->ifa_flags & ~IFA_RTSELF) | RTF_PINNED;
2098 		info.rti_info[RTAX_DST] = dst;
2099 		/*
2100 		 * doing this for compatibility reasons
2101 		 */
2102 		if (cmd == RTM_ADD)
2103 			info.rti_info[RTAX_GATEWAY] = (struct sockaddr *)sdl;
2104 		else
2105 			info.rti_info[RTAX_GATEWAY] = ifa->ifa_addr;
2106 		info.rti_info[RTAX_NETMASK] = netmask;
2107 		error = rtrequest1_fib(cmd, &info, &rt, fibnum);
2108 		if (error == 0 && rt != NULL) {
2109 			/*
2110 			 * notify any listening routing agents of the change
2111 			 */
2112 			RT_LOCK(rt);
2113 #ifdef RADIX_MPATH
2114 			/*
2115 			 * in case address alias finds the first address
2116 			 * e.g. ifconfig bge0 192.0.2.246/24
2117 			 * e.g. ifconfig bge0 192.0.2.247/24
2118 			 * the address set in the route is 192.0.2.246
2119 			 * so we need to replace it with 192.0.2.247
2120 			 */
2121 			if (memcmp(rt->rt_ifa->ifa_addr,
2122 			    ifa->ifa_addr, ifa->ifa_addr->sa_len)) {
2123 				ifa_free(rt->rt_ifa);
2124 				ifa_ref(ifa);
2125 				rt->rt_ifp = ifa->ifa_ifp;
2126 				rt->rt_ifa = ifa;
2127 			}
2128 #endif
2129 			RT_ADDREF(rt);
2130 			RT_UNLOCK(rt);
2131 			rt_newaddrmsg_fib(cmd, ifa, rt, fibnum);
2132 			RT_LOCK(rt);
2133 			RT_REMREF(rt);
2134 			if (cmd == RTM_DELETE) {
2135 				/*
2136 				 * If we are deleting, and we found an entry,
2137 				 * then it's been removed from the tree..
2138 				 * now throw it away.
2139 				 */
2140 				RTFREE_LOCKED(rt);
2141 			} else {
2142 				if (cmd == RTM_ADD) {
2143 					/*
2144 					 * We just wanted to add it..
2145 					 * we don't actually need a reference.
2146 					 */
2147 					RT_REMREF(rt);
2148 				}
2149 				RT_UNLOCK(rt);
2150 			}
2151 			didwork = 1;
2152 		}
2153 		if (error)
2154 			a_failure = error;
2155 	}
2156 	if (cmd == RTM_DELETE) {
2157 		if (didwork) {
2158 			error = 0;
2159 		} else {
2160 			/* we only give an error if it wasn't in any table */
2161 			error = ((flags & RTF_HOST) ?
2162 			    EHOSTUNREACH : ENETUNREACH);
2163 		}
2164 	} else {
2165 		if (a_failure) {
2166 			/* return an error if any of them failed */
2167 			error = a_failure;
2168 		}
2169 	}
2170 	return (error);
2171 }
2172 
2173 /*
2174  * Set up a routing table entry, normally
2175  * for an interface.
2176  */
2177 int
2178 rtinit(struct ifaddr *ifa, int cmd, int flags)
2179 {
2180 	struct sockaddr *dst;
2181 	int fib = RT_DEFAULT_FIB;
2182 
2183 	if (flags & RTF_HOST) {
2184 		dst = ifa->ifa_dstaddr;
2185 	} else {
2186 		dst = ifa->ifa_addr;
2187 	}
2188 
2189 	switch (dst->sa_family) {
2190 	case AF_INET6:
2191 	case AF_INET:
2192 		/* We do support multiple FIBs. */
2193 		fib = RT_ALL_FIBS;
2194 		break;
2195 	}
2196 	return (rtinit1(ifa, cmd, flags, fib));
2197 }
2198 
2199 /*
2200  * Announce interface address arrival/withdraw
2201  * Returns 0 on success.
2202  */
2203 int
2204 rt_addrmsg(int cmd, struct ifaddr *ifa, int fibnum)
2205 {
2206 
2207 	KASSERT(cmd == RTM_ADD || cmd == RTM_DELETE,
2208 	    ("unexpected cmd %d", cmd));
2209 	KASSERT(fibnum == RT_ALL_FIBS || (fibnum >= 0 && fibnum < rt_numfibs),
2210 	    ("%s: fib out of range 0 <=%d<%d", __func__, fibnum, rt_numfibs));
2211 
2212 	EVENTHANDLER_DIRECT_INVOKE(rt_addrmsg, ifa, cmd);
2213 	return (rtsock_addrmsg(cmd, ifa, fibnum));
2214 }
2215 
2216 /*
2217  * Announce kernel-originated route addition/removal to rtsock based on @rt data.
2218  * cmd: RTM_ cmd
2219  * @rt: valid rtentry
2220  * @ifp: target route interface
2221  * @fibnum: fib id or RT_ALL_FIBS
2222  *
2223  * Returns 0 on success.
2224  */
2225 int
2226 rt_routemsg(int cmd, struct rtentry *rt, struct ifnet *ifp, int rti_addrs,
2227     int fibnum)
2228 {
2229 
2230 	KASSERT(cmd == RTM_ADD || cmd == RTM_DELETE,
2231 	    ("unexpected cmd %d", cmd));
2232 
2233 	KASSERT(fibnum == RT_ALL_FIBS || (fibnum >= 0 && fibnum < rt_numfibs),
2234 	    ("%s: fib out of range 0 <=%d<%d", __func__, fibnum, rt_numfibs));
2235 
2236 	KASSERT(rt_key(rt) != NULL, (":%s: rt_key must be supplied", __func__));
2237 
2238 	return (rtsock_routemsg(cmd, rt, ifp, 0, fibnum));
2239 }
2240 
2241 /*
2242  * Announce kernel-originated route addition/removal to rtsock based on @rt data.
2243  * cmd: RTM_ cmd
2244  * @info: addrinfo structure with valid data.
2245  * @fibnum: fib id or RT_ALL_FIBS
2246  *
2247  * Returns 0 on success.
2248  */
2249 int
2250 rt_routemsg_info(int cmd, struct rt_addrinfo *info, int fibnum)
2251 {
2252 
2253 	KASSERT(cmd == RTM_ADD || cmd == RTM_DELETE || cmd == RTM_CHANGE,
2254 	    ("unexpected cmd %d", cmd));
2255 
2256 	KASSERT(fibnum == RT_ALL_FIBS || (fibnum >= 0 && fibnum < rt_numfibs),
2257 	    ("%s: fib out of range 0 <=%d<%d", __func__, fibnum, rt_numfibs));
2258 
2259 	KASSERT(info->rti_info[RTAX_DST] != NULL, (":%s: RTAX_DST must be supplied", __func__));
2260 
2261 	return (rtsock_routemsg_info(cmd, info, fibnum));
2262 }
2263 
2264 
2265 /*
2266  * This is called to generate messages from the routing socket
2267  * indicating a network interface has had addresses associated with it.
2268  */
2269 void
2270 rt_newaddrmsg_fib(int cmd, struct ifaddr *ifa, struct rtentry *rt, int fibnum)
2271 {
2272 
2273 	KASSERT(cmd == RTM_ADD || cmd == RTM_DELETE,
2274 		("unexpected cmd %u", cmd));
2275 	KASSERT(fibnum == RT_ALL_FIBS || (fibnum >= 0 && fibnum < rt_numfibs),
2276 	    ("%s: fib out of range 0 <=%d<%d", __func__, fibnum, rt_numfibs));
2277 
2278 	if (cmd == RTM_ADD) {
2279 		rt_addrmsg(cmd, ifa, fibnum);
2280 		if (rt != NULL)
2281 			rt_routemsg(cmd, rt, ifa->ifa_ifp, 0, fibnum);
2282 	} else {
2283 		if (rt != NULL)
2284 			rt_routemsg(cmd, rt, ifa->ifa_ifp, 0, fibnum);
2285 		rt_addrmsg(cmd, ifa, fibnum);
2286 	}
2287 }
2288 
2289