xref: /freebsd/sys/net/route.c (revision 71d104536b513298902be65342afe6f3792f29e4)
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/route_var.h>
65 #include <net/route/nhop.h>
66 #include <net/route/shared.h>
67 #include <net/vnet.h>
68 
69 #ifdef RADIX_MPATH
70 #include <net/radix_mpath.h>
71 #endif
72 
73 #include <netinet/in.h>
74 #include <netinet/ip_mroute.h>
75 
76 #include <vm/uma.h>
77 
78 #define	RT_MAXFIBS	UINT16_MAX
79 
80 /* Kernel config default option. */
81 #ifdef ROUTETABLES
82 #if ROUTETABLES <= 0
83 #error "ROUTETABLES defined too low"
84 #endif
85 #if ROUTETABLES > RT_MAXFIBS
86 #error "ROUTETABLES defined too big"
87 #endif
88 #define	RT_NUMFIBS	ROUTETABLES
89 #endif /* ROUTETABLES */
90 /* Initialize to default if not otherwise set. */
91 #ifndef	RT_NUMFIBS
92 #define	RT_NUMFIBS	1
93 #endif
94 
95 /* This is read-only.. */
96 u_int rt_numfibs = RT_NUMFIBS;
97 SYSCTL_UINT(_net, OID_AUTO, fibs, CTLFLAG_RDTUN, &rt_numfibs, 0, "");
98 
99 /*
100  * By default add routes to all fibs for new interfaces.
101  * Once this is set to 0 then only allocate routes on interface
102  * changes for the FIB of the caller when adding a new set of addresses
103  * to an interface.  XXX this is a shotgun aproach to a problem that needs
104  * a more fine grained solution.. that will come.
105  * XXX also has the problems getting the FIB from curthread which will not
106  * always work given the fib can be overridden and prefixes can be added
107  * from the network stack context.
108  */
109 VNET_DEFINE(u_int, rt_add_addr_allfibs) = 1;
110 SYSCTL_UINT(_net, OID_AUTO, add_addr_allfibs, CTLFLAG_RWTUN | CTLFLAG_VNET,
111     &VNET_NAME(rt_add_addr_allfibs), 0, "");
112 
113 VNET_PCPUSTAT_DEFINE(struct rtstat, rtstat);
114 
115 VNET_PCPUSTAT_SYSINIT(rtstat);
116 #ifdef VIMAGE
117 VNET_PCPUSTAT_SYSUNINIT(rtstat);
118 #endif
119 
120 VNET_DEFINE(struct rib_head *, rt_tables);
121 #define	V_rt_tables	VNET(rt_tables)
122 
123 
124 VNET_DEFINE(uma_zone_t, rtzone);		/* Routing table UMA zone. */
125 #define	V_rtzone	VNET(rtzone)
126 
127 EVENTHANDLER_LIST_DEFINE(rt_addrmsg);
128 
129 static int rt_ifdelroute(const struct rtentry *rt, const struct nhop_object *,
130     void *arg);
131 static void destroy_rtentry_epoch(epoch_context_t ctx);
132 static int rt_exportinfo(struct rtentry *rt, struct rt_addrinfo *info,
133     int flags);
134 
135 /*
136  * handler for net.my_fibnum
137  */
138 static int
139 sysctl_my_fibnum(SYSCTL_HANDLER_ARGS)
140 {
141         int fibnum;
142         int error;
143 
144         fibnum = curthread->td_proc->p_fibnum;
145         error = sysctl_handle_int(oidp, &fibnum, 0, req);
146         return (error);
147 }
148 
149 SYSCTL_PROC(_net, OID_AUTO, my_fibnum,
150     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
151     &sysctl_my_fibnum, "I",
152     "default FIB of caller");
153 
154 static __inline struct rib_head **
155 rt_tables_get_rnh_ptr(int table, int fam)
156 {
157 	struct rib_head **rnh;
158 
159 	KASSERT(table >= 0 && table < rt_numfibs,
160 	    ("%s: table out of bounds (0 <= %d < %d)", __func__, table,
161 	     rt_numfibs));
162 	KASSERT(fam >= 0 && fam < (AF_MAX + 1),
163 	    ("%s: fam out of bounds (0 <= %d < %d)", __func__, fam, AF_MAX+1));
164 
165 	/* rnh is [fib=0][af=0]. */
166 	rnh = (struct rib_head **)V_rt_tables;
167 	/* Get the offset to the requested table and fam. */
168 	rnh += table * (AF_MAX+1) + fam;
169 
170 	return (rnh);
171 }
172 
173 struct rib_head *
174 rt_tables_get_rnh(int table, int fam)
175 {
176 
177 	return (*rt_tables_get_rnh_ptr(table, fam));
178 }
179 
180 u_int
181 rt_tables_get_gen(int table, int fam)
182 {
183 	struct rib_head *rnh;
184 
185 	rnh = *rt_tables_get_rnh_ptr(table, fam);
186 	KASSERT(rnh != NULL, ("%s: NULL rib_head pointer table %d fam %d",
187 	    __func__, table, fam));
188 	return (rnh->rnh_gen);
189 }
190 
191 
192 /*
193  * route initialization must occur before ip6_init2(), which happenas at
194  * SI_ORDER_MIDDLE.
195  */
196 static void
197 route_init(void)
198 {
199 
200 	/* whack the tunable ints into  line. */
201 	if (rt_numfibs > RT_MAXFIBS)
202 		rt_numfibs = RT_MAXFIBS;
203 	if (rt_numfibs == 0)
204 		rt_numfibs = 1;
205 	nhops_init();
206 }
207 SYSINIT(route_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD, route_init, NULL);
208 
209 static int
210 rtentry_zinit(void *mem, int size, int how)
211 {
212 	struct rtentry *rt = mem;
213 
214 	RT_LOCK_INIT(rt);
215 
216 	return (0);
217 }
218 
219 static void
220 rtentry_zfini(void *mem, int size)
221 {
222 	struct rtentry *rt = mem;
223 
224 	RT_LOCK_DESTROY(rt);
225 }
226 
227 static int
228 rtentry_ctor(void *mem, int size, void *arg, int how)
229 {
230 	struct rtentry *rt = mem;
231 
232 	bzero(rt, offsetof(struct rtentry, rt_endzero));
233 	rt->rt_chain = NULL;
234 
235 	return (0);
236 }
237 
238 static void
239 rtentry_dtor(void *mem, int size, void *arg)
240 {
241 	struct rtentry *rt = mem;
242 
243 	RT_UNLOCK_COND(rt);
244 }
245 
246 static void
247 vnet_route_init(const void *unused __unused)
248 {
249 	struct domain *dom;
250 	struct rib_head **rnh;
251 	int table;
252 	int fam;
253 
254 	V_rt_tables = malloc(rt_numfibs * (AF_MAX+1) *
255 	    sizeof(struct rib_head *), M_RTABLE, M_WAITOK|M_ZERO);
256 
257 	V_rtzone = uma_zcreate("rtentry", sizeof(struct rtentry),
258 	    rtentry_ctor, rtentry_dtor,
259 	    rtentry_zinit, rtentry_zfini, UMA_ALIGN_PTR, 0);
260 	for (dom = domains; dom; dom = dom->dom_next) {
261 		if (dom->dom_rtattach == NULL)
262 			continue;
263 
264 		for  (table = 0; table < rt_numfibs; table++) {
265 			fam = dom->dom_family;
266 			if (table != 0 && fam != AF_INET6 && fam != AF_INET)
267 				break;
268 
269 			rnh = rt_tables_get_rnh_ptr(table, fam);
270 			if (rnh == NULL)
271 				panic("%s: rnh NULL", __func__);
272 			dom->dom_rtattach((void **)rnh, 0, table);
273 		}
274 	}
275 }
276 VNET_SYSINIT(vnet_route_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH,
277     vnet_route_init, 0);
278 
279 #ifdef VIMAGE
280 static void
281 vnet_route_uninit(const void *unused __unused)
282 {
283 	int table;
284 	int fam;
285 	struct domain *dom;
286 	struct rib_head **rnh;
287 
288 	for (dom = domains; dom; dom = dom->dom_next) {
289 		if (dom->dom_rtdetach == NULL)
290 			continue;
291 
292 		for (table = 0; table < rt_numfibs; table++) {
293 			fam = dom->dom_family;
294 
295 			if (table != 0 && fam != AF_INET6 && fam != AF_INET)
296 				break;
297 
298 			rnh = rt_tables_get_rnh_ptr(table, fam);
299 			if (rnh == NULL)
300 				panic("%s: rnh NULL", __func__);
301 			dom->dom_rtdetach((void **)rnh, 0);
302 		}
303 	}
304 
305 	/*
306 	 * dom_rtdetach calls rt_table_destroy(), which
307 	 *  schedules deletion for all rtentries, nexthops and control
308 	 *  structures. Wait for the destruction callbacks to fire.
309 	 * Note that this should result in freeing all rtentries, but
310 	 *  nexthops deletions will be scheduled for the next epoch run
311 	 *  and will be completed after vnet teardown.
312 	 */
313 	epoch_drain_callbacks(net_epoch_preempt);
314 
315 	free(V_rt_tables, M_RTABLE);
316 	uma_zdestroy(V_rtzone);
317 }
318 VNET_SYSUNINIT(vnet_route_uninit, SI_SUB_PROTO_DOMAIN, SI_ORDER_FIRST,
319     vnet_route_uninit, 0);
320 #endif
321 
322 struct rib_head *
323 rt_table_init(int offset, int family, u_int fibnum)
324 {
325 	struct rib_head *rh;
326 
327 	rh = malloc(sizeof(struct rib_head), M_RTABLE, M_WAITOK | M_ZERO);
328 
329 	/* TODO: These details should be hidded inside radix.c */
330 	/* Init masks tree */
331 	rn_inithead_internal(&rh->head, rh->rnh_nodes, offset);
332 	rn_inithead_internal(&rh->rmhead.head, rh->rmhead.mask_nodes, 0);
333 	rh->head.rnh_masks = &rh->rmhead;
334 
335 	/* Save metadata associated with this routing table. */
336 	rh->rib_family = family;
337 	rh->rib_fibnum = fibnum;
338 #ifdef VIMAGE
339 	rh->rib_vnet = curvnet;
340 #endif
341 
342 	tmproutes_init(rh);
343 
344 	/* Init locks */
345 	RIB_LOCK_INIT(rh);
346 
347 	nhops_init_rib(rh);
348 
349 	/* Finally, set base callbacks */
350 	rh->rnh_addaddr = rn_addroute;
351 	rh->rnh_deladdr = rn_delete;
352 	rh->rnh_matchaddr = rn_match;
353 	rh->rnh_lookup = rn_lookup;
354 	rh->rnh_walktree = rn_walktree;
355 	rh->rnh_walktree_from = rn_walktree_from;
356 
357 	return (rh);
358 }
359 
360 static int
361 rt_freeentry(struct radix_node *rn, void *arg)
362 {
363 	struct radix_head * const rnh = arg;
364 	struct radix_node *x;
365 
366 	x = (struct radix_node *)rn_delete(rn + 2, NULL, rnh);
367 	if (x != NULL)
368 		R_Free(x);
369 	return (0);
370 }
371 
372 void
373 rt_table_destroy(struct rib_head *rh)
374 {
375 
376 	tmproutes_destroy(rh);
377 
378 	rn_walktree(&rh->rmhead.head, rt_freeentry, &rh->rmhead.head);
379 
380 	nhops_destroy_rib(rh);
381 
382 	/* Assume table is already empty */
383 	RIB_LOCK_DESTROY(rh);
384 	free(rh, M_RTABLE);
385 }
386 
387 
388 #ifndef _SYS_SYSPROTO_H_
389 struct setfib_args {
390 	int     fibnum;
391 };
392 #endif
393 int
394 sys_setfib(struct thread *td, struct setfib_args *uap)
395 {
396 	if (uap->fibnum < 0 || uap->fibnum >= rt_numfibs)
397 		return EINVAL;
398 	td->td_proc->p_fibnum = uap->fibnum;
399 	return (0);
400 }
401 
402 /*
403  * Remove a reference count from an rtentry.
404  * If the count gets low enough, take it out of the routing table
405  */
406 void
407 rtfree(struct rtentry *rt)
408 {
409 
410 	KASSERT(rt != NULL,("%s: NULL rt", __func__));
411 
412 	RT_LOCK_ASSERT(rt);
413 
414 	epoch_call(net_epoch_preempt, destroy_rtentry_epoch,
415 	    &rt->rt_epoch_ctx);
416 	RT_UNLOCK(rt);
417 }
418 
419 static void
420 destroy_rtentry(struct rtentry *rt)
421 {
422 
423 	/*
424 	 * At this moment rnh, nh_control may be already freed.
425 	 * nhop interface may have been migrated to a different vnet.
426 	 * Use vnet stored in the nexthop to delete the entry.
427 	 */
428 	CURVNET_SET(nhop_get_vnet(rt->rt_nhop));
429 
430 	/* Unreference nexthop */
431 	nhop_free(rt->rt_nhop);
432 
433 	uma_zfree(V_rtzone, rt);
434 
435 	CURVNET_RESTORE();
436 }
437 
438 /*
439  * Epoch callback indicating rtentry is safe to destroy
440  */
441 static void
442 destroy_rtentry_epoch(epoch_context_t ctx)
443 {
444 	struct rtentry *rt;
445 
446 	rt = __containerof(ctx, struct rtentry, rt_epoch_ctx);
447 
448 	destroy_rtentry(rt);
449 }
450 
451 /*
452  * Adds a temporal redirect entry to the routing table.
453  * @fibnum: fib number
454  * @dst: destination to install redirect to
455  * @gateway: gateway to go via
456  * @author: sockaddr of originating router, can be NULL
457  * @ifp: interface to use for the redirected route
458  * @flags: set of flags to add. Allowed: RTF_GATEWAY
459  * @lifetime_sec: time in seconds to expire this redirect.
460  *
461  * Retuns 0 on success, errno otherwise.
462  */
463 int
464 rib_add_redirect(u_int fibnum, struct sockaddr *dst, struct sockaddr *gateway,
465     struct sockaddr *author, struct ifnet *ifp, int flags, int lifetime_sec)
466 {
467 	struct rtentry *rt;
468 	int error;
469 	struct rt_addrinfo info;
470 	struct rt_metrics rti_rmx;
471 	struct ifaddr *ifa;
472 
473 	NET_EPOCH_ASSERT();
474 
475 	if (rt_tables_get_rnh(fibnum, dst->sa_family) == NULL)
476 		return (EAFNOSUPPORT);
477 
478 	/* Verify the allowed flag mask. */
479 	KASSERT(((flags & ~(RTF_GATEWAY)) == 0),
480 	    ("invalid redirect flags: %x", flags));
481 
482 	/* Get the best ifa for the given interface and gateway. */
483 	if ((ifa = ifaof_ifpforaddr(gateway, ifp)) == NULL)
484 		return (ENETUNREACH);
485 	ifa_ref(ifa);
486 
487 	bzero(&info, sizeof(info));
488 	info.rti_info[RTAX_DST] = dst;
489 	info.rti_info[RTAX_GATEWAY] = gateway;
490 	info.rti_ifa = ifa;
491 	info.rti_ifp = ifp;
492 	info.rti_flags = flags | RTF_HOST | RTF_DYNAMIC;
493 
494 	/* Setup route metrics to define expire time. */
495 	bzero(&rti_rmx, sizeof(rti_rmx));
496 	/* Set expire time as absolute. */
497 	rti_rmx.rmx_expire = lifetime_sec + time_second;
498 	info.rti_mflags |= RTV_EXPIRE;
499 	info.rti_rmx = &rti_rmx;
500 
501 	error = rtrequest1_fib(RTM_ADD, &info, &rt, fibnum);
502 	ifa_free(ifa);
503 
504 	if (error != 0) {
505 		/* TODO: add per-fib redirect stats. */
506 		return (error);
507 	}
508 
509 	RT_LOCK(rt);
510 	flags = rt->rt_flags;
511 	RT_UNLOCK(rt);
512 
513 	RTSTAT_INC(rts_dynamic);
514 
515 	/* Send notification of a route addition to userland. */
516 	bzero(&info, sizeof(info));
517 	info.rti_info[RTAX_DST] = dst;
518 	info.rti_info[RTAX_GATEWAY] = gateway;
519 	info.rti_info[RTAX_AUTHOR] = author;
520 	rt_missmsg_fib(RTM_REDIRECT, &info, flags, error, fibnum);
521 
522 	return (0);
523 }
524 
525 /*
526  * Routing table ioctl interface.
527  */
528 int
529 rtioctl_fib(u_long req, caddr_t data, u_int fibnum)
530 {
531 
532 	/*
533 	 * If more ioctl commands are added here, make sure the proper
534 	 * super-user checks are being performed because it is possible for
535 	 * prison-root to make it this far if raw sockets have been enabled
536 	 * in jails.
537 	 */
538 #ifdef INET
539 	/* Multicast goop, grrr... */
540 	return mrt_ioctl ? mrt_ioctl(req, data, fibnum) : EOPNOTSUPP;
541 #else /* INET */
542 	return ENXIO;
543 #endif /* INET */
544 }
545 
546 struct ifaddr *
547 ifa_ifwithroute(int flags, const struct sockaddr *dst,
548     const struct sockaddr *gateway, u_int fibnum)
549 {
550 	struct ifaddr *ifa;
551 
552 	NET_EPOCH_ASSERT();
553 	if ((flags & RTF_GATEWAY) == 0) {
554 		/*
555 		 * If we are adding a route to an interface,
556 		 * and the interface is a pt to pt link
557 		 * we should search for the destination
558 		 * as our clue to the interface.  Otherwise
559 		 * we can use the local address.
560 		 */
561 		ifa = NULL;
562 		if (flags & RTF_HOST)
563 			ifa = ifa_ifwithdstaddr(dst, fibnum);
564 		if (ifa == NULL)
565 			ifa = ifa_ifwithaddr(gateway);
566 	} else {
567 		/*
568 		 * If we are adding a route to a remote net
569 		 * or host, the gateway may still be on the
570 		 * other end of a pt to pt link.
571 		 */
572 		ifa = ifa_ifwithdstaddr(gateway, fibnum);
573 	}
574 	if (ifa == NULL)
575 		ifa = ifa_ifwithnet(gateway, 0, fibnum);
576 	if (ifa == NULL) {
577 		struct nhop_object *nh;
578 
579 		nh = rib_lookup(fibnum, gateway, NHR_NONE, 0);
580 
581 		/*
582 		 * dismiss a gateway that is reachable only
583 		 * through the default router
584 		 */
585 		if ((nh == NULL) || (nh->nh_flags & NHF_DEFAULT))
586 			return (NULL);
587 		ifa = nh->nh_ifa;
588 	}
589 	if (ifa->ifa_addr->sa_family != dst->sa_family) {
590 		struct ifaddr *oifa = ifa;
591 		ifa = ifaof_ifpforaddr(dst, ifa->ifa_ifp);
592 		if (ifa == NULL)
593 			ifa = oifa;
594 	}
595 
596 	return (ifa);
597 }
598 
599 /*
600  * Do appropriate manipulations of a routing tree given
601  * all the bits of info needed
602  */
603 int
604 rtrequest_fib(int req,
605 	struct sockaddr *dst,
606 	struct sockaddr *gateway,
607 	struct sockaddr *netmask,
608 	int flags,
609 	struct rtentry **ret_nrt,
610 	u_int fibnum)
611 {
612 	struct rt_addrinfo info;
613 
614 	if (dst->sa_len == 0)
615 		return(EINVAL);
616 
617 	bzero((caddr_t)&info, sizeof(info));
618 	info.rti_flags = flags;
619 	info.rti_info[RTAX_DST] = dst;
620 	info.rti_info[RTAX_GATEWAY] = gateway;
621 	info.rti_info[RTAX_NETMASK] = netmask;
622 	return rtrequest1_fib(req, &info, ret_nrt, fibnum);
623 }
624 
625 
626 /*
627  * Copy most of @rt data into @info.
628  *
629  * If @flags contains NHR_COPY, copies dst,netmask and gw to the
630  * pointers specified by @info structure. Assume such pointers
631  * are zeroed sockaddr-like structures with sa_len field initialized
632  * to reflect size of the provided buffer. if no NHR_COPY is specified,
633  * point dst,netmask and gw @info fields to appropriate @rt values.
634  *
635  * if @flags contains NHR_REF, do refcouting on rt_ifp and rt_ifa.
636  *
637  * Returns 0 on success.
638  */
639 int
640 rt_exportinfo(struct rtentry *rt, struct rt_addrinfo *info, int flags)
641 {
642 	struct rt_metrics *rmx;
643 	struct sockaddr *src, *dst;
644 	struct nhop_object *nh;
645 	int sa_len;
646 
647 	if (flags & NHR_COPY) {
648 		/* Copy destination if dst is non-zero */
649 		src = rt_key(rt);
650 		dst = info->rti_info[RTAX_DST];
651 		sa_len = src->sa_len;
652 		if (dst != NULL) {
653 			if (src->sa_len > dst->sa_len)
654 				return (ENOMEM);
655 			memcpy(dst, src, src->sa_len);
656 			info->rti_addrs |= RTA_DST;
657 		}
658 
659 		/* Copy mask if set && dst is non-zero */
660 		src = rt_mask(rt);
661 		dst = info->rti_info[RTAX_NETMASK];
662 		if (src != NULL && dst != NULL) {
663 
664 			/*
665 			 * Radix stores different value in sa_len,
666 			 * assume rt_mask() to have the same length
667 			 * as rt_key()
668 			 */
669 			if (sa_len > dst->sa_len)
670 				return (ENOMEM);
671 			memcpy(dst, src, src->sa_len);
672 			info->rti_addrs |= RTA_NETMASK;
673 		}
674 
675 		/* Copy gateway is set && dst is non-zero */
676 		src = &rt->rt_nhop->gw_sa;
677 		dst = info->rti_info[RTAX_GATEWAY];
678 		if ((rt->rt_flags & RTF_GATEWAY) && src != NULL && dst != NULL){
679 			if (src->sa_len > dst->sa_len)
680 				return (ENOMEM);
681 			memcpy(dst, src, src->sa_len);
682 			info->rti_addrs |= RTA_GATEWAY;
683 		}
684 	} else {
685 		info->rti_info[RTAX_DST] = rt_key(rt);
686 		info->rti_addrs |= RTA_DST;
687 		if (rt_mask(rt) != NULL) {
688 			info->rti_info[RTAX_NETMASK] = rt_mask(rt);
689 			info->rti_addrs |= RTA_NETMASK;
690 		}
691 		if (rt->rt_flags & RTF_GATEWAY) {
692 			info->rti_info[RTAX_GATEWAY] = &rt->rt_nhop->gw_sa;
693 			info->rti_addrs |= RTA_GATEWAY;
694 		}
695 	}
696 
697 	nh = rt->rt_nhop;
698 	rmx = info->rti_rmx;
699 	if (rmx != NULL) {
700 		info->rti_mflags |= RTV_MTU;
701 		rmx->rmx_mtu = nh->nh_mtu;
702 	}
703 
704 	info->rti_flags = rt->rt_flags | nhop_get_rtflags(nh);
705 	info->rti_ifp = nh->nh_ifp;
706 	info->rti_ifa = nh->nh_ifa;
707 	if (flags & NHR_REF) {
708 		if_ref(info->rti_ifp);
709 		ifa_ref(info->rti_ifa);
710 	}
711 
712 	return (0);
713 }
714 
715 /*
716  * Lookups up route entry for @dst in RIB database for fib @fibnum.
717  * Exports entry data to @info using rt_exportinfo().
718  *
719  * If @flags contains NHR_REF, refcouting is performed on rt_ifp and rt_ifa.
720  * All references can be released later by calling rib_free_info().
721  *
722  * Returns 0 on success.
723  * Returns ENOENT for lookup failure, ENOMEM for export failure.
724  */
725 int
726 rib_lookup_info(uint32_t fibnum, const struct sockaddr *dst, uint32_t flags,
727     uint32_t flowid, struct rt_addrinfo *info)
728 {
729 	RIB_RLOCK_TRACKER;
730 	struct rib_head *rh;
731 	struct radix_node *rn;
732 	struct rtentry *rt;
733 	int error;
734 
735 	KASSERT((fibnum < rt_numfibs), ("rib_lookup_rte: bad fibnum"));
736 	rh = rt_tables_get_rnh(fibnum, dst->sa_family);
737 	if (rh == NULL)
738 		return (ENOENT);
739 
740 	RIB_RLOCK(rh);
741 	rn = rh->rnh_matchaddr(__DECONST(void *, dst), &rh->head);
742 	if (rn != NULL && ((rn->rn_flags & RNF_ROOT) == 0)) {
743 		rt = RNTORT(rn);
744 		/* Ensure route & ifp is UP */
745 		if (RT_LINK_IS_UP(rt->rt_nhop->nh_ifp)) {
746 			flags = (flags & NHR_REF) | NHR_COPY;
747 			error = rt_exportinfo(rt, info, flags);
748 			RIB_RUNLOCK(rh);
749 
750 			return (error);
751 		}
752 	}
753 	RIB_RUNLOCK(rh);
754 
755 	return (ENOENT);
756 }
757 
758 /*
759  * Releases all references acquired by rib_lookup_info() when
760  * called with NHR_REF flags.
761  */
762 void
763 rib_free_info(struct rt_addrinfo *info)
764 {
765 
766 	ifa_free(info->rti_ifa);
767 	if_rele(info->rti_ifp);
768 }
769 
770 /*
771  * Iterates over all existing fibs in system calling
772  *  @setwa_f function prior to traversing each fib.
773  *  Calls @wa_f function for each element in current fib.
774  * If af is not AF_UNSPEC, iterates over fibs in particular
775  * address family.
776  */
777 void
778 rt_foreach_fib_walk(int af, rt_setwarg_t *setwa_f, rt_walktree_f_t *wa_f,
779     void *arg)
780 {
781 	struct rib_head *rnh;
782 	uint32_t fibnum;
783 	int i;
784 
785 	for (fibnum = 0; fibnum < rt_numfibs; fibnum++) {
786 		/* Do we want some specific family? */
787 		if (af != AF_UNSPEC) {
788 			rnh = rt_tables_get_rnh(fibnum, af);
789 			if (rnh == NULL)
790 				continue;
791 			if (setwa_f != NULL)
792 				setwa_f(rnh, fibnum, af, arg);
793 
794 			RIB_WLOCK(rnh);
795 			rnh->rnh_walktree(&rnh->head, (walktree_f_t *)wa_f,arg);
796 			RIB_WUNLOCK(rnh);
797 			continue;
798 		}
799 
800 		for (i = 1; i <= AF_MAX; i++) {
801 			rnh = rt_tables_get_rnh(fibnum, i);
802 			if (rnh == NULL)
803 				continue;
804 			if (setwa_f != NULL)
805 				setwa_f(rnh, fibnum, i, arg);
806 
807 			RIB_WLOCK(rnh);
808 			rnh->rnh_walktree(&rnh->head, (walktree_f_t *)wa_f,arg);
809 			RIB_WUNLOCK(rnh);
810 		}
811 	}
812 }
813 
814 /*
815  * Iterates over all existing fibs in system and deletes each element
816  *  for which @filter_f function returns non-zero value.
817  * If @family is not AF_UNSPEC, iterates over fibs in particular
818  * address family.
819  */
820 void
821 rt_foreach_fib_walk_del(int family, rt_filter_f_t *filter_f, void *arg)
822 {
823 	u_int fibnum;
824 	int i, start, end;
825 
826 	for (fibnum = 0; fibnum < rt_numfibs; fibnum++) {
827 		/* Do we want some specific family? */
828 		if (family != AF_UNSPEC) {
829 			start = family;
830 			end = family;
831 		} else {
832 			start = 1;
833 			end = AF_MAX;
834 		}
835 
836 		for (i = start; i <= end; i++) {
837 			if (rt_tables_get_rnh(fibnum, i) == NULL)
838 				continue;
839 
840 			rib_walk_del(fibnum, i, filter_f, arg, 0);
841 		}
842 	}
843 }
844 
845 /*
846  * Delete Routes for a Network Interface
847  *
848  * Called for each routing entry via the rnh->rnh_walktree() call above
849  * to delete all route entries referencing a detaching network interface.
850  *
851  * Arguments:
852  *	rt	pointer to rtentry
853  *	nh	pointer to nhop
854  *	arg	argument passed to rnh->rnh_walktree() - detaching interface
855  *
856  * Returns:
857  *	0	successful
858  *	errno	failed - reason indicated
859  */
860 static int
861 rt_ifdelroute(const struct rtentry *rt, const struct nhop_object *nh, void *arg)
862 {
863 	struct ifnet	*ifp = arg;
864 
865 	if (nh->nh_ifp != ifp)
866 		return (0);
867 
868 	/*
869 	 * Protect (sorta) against walktree recursion problems
870 	 * with cloned routes
871 	 */
872 	if ((rt->rt_flags & RTF_UP) == 0)
873 		return (0);
874 
875 	return (1);
876 }
877 
878 /*
879  * Delete all remaining routes using this interface
880  * Unfortuneatly the only way to do this is to slog through
881  * the entire routing table looking for routes which point
882  * to this interface...oh well...
883  */
884 void
885 rt_flushifroutes_af(struct ifnet *ifp, int af)
886 {
887 	KASSERT((af >= 1 && af <= AF_MAX), ("%s: af %d not >= 1 and <= %d",
888 	    __func__, af, AF_MAX));
889 
890 	rt_foreach_fib_walk_del(af, rt_ifdelroute, ifp);
891 }
892 
893 void
894 rt_flushifroutes(struct ifnet *ifp)
895 {
896 
897 	rt_foreach_fib_walk_del(AF_UNSPEC, rt_ifdelroute, ifp);
898 }
899 
900 /*
901  * Look up rt_addrinfo for a specific fib.  Note that if rti_ifa is defined,
902  * it will be referenced so the caller must free it.
903  *
904  * Assume basic consistency checks are executed by callers:
905  * RTAX_DST exists, if RTF_GATEWAY is set, RTAX_GATEWAY exists as well.
906  */
907 int
908 rt_getifa_fib(struct rt_addrinfo *info, u_int fibnum)
909 {
910 	const struct sockaddr *dst, *gateway, *ifpaddr, *ifaaddr;
911 	struct epoch_tracker et;
912 	int needref, error, flags;
913 
914 	dst = info->rti_info[RTAX_DST];
915 	gateway = info->rti_info[RTAX_GATEWAY];
916 	ifpaddr = info->rti_info[RTAX_IFP];
917 	ifaaddr = info->rti_info[RTAX_IFA];
918 	flags = info->rti_flags;
919 
920 	/*
921 	 * ifp may be specified by sockaddr_dl
922 	 * when protocol address is ambiguous.
923 	 */
924 	error = 0;
925 	needref = (info->rti_ifa == NULL);
926 	NET_EPOCH_ENTER(et);
927 
928 	/* If we have interface specified by the ifindex in the address, use it */
929 	if (info->rti_ifp == NULL && ifpaddr != NULL &&
930 	    ifpaddr->sa_family == AF_LINK) {
931 	    const struct sockaddr_dl *sdl = (const struct sockaddr_dl *)ifpaddr;
932 	    if (sdl->sdl_index != 0)
933 		    info->rti_ifp = ifnet_byindex(sdl->sdl_index);
934 	}
935 	/*
936 	 * If we have source address specified, try to find it
937 	 * TODO: avoid enumerating all ifas on all interfaces.
938 	 */
939 	if (info->rti_ifa == NULL && ifaaddr != NULL)
940 		info->rti_ifa = ifa_ifwithaddr(ifaaddr);
941 	if (info->rti_ifa == NULL) {
942 		const struct sockaddr *sa;
943 
944 		/*
945 		 * Most common use case for the userland-supplied routes.
946 		 *
947 		 * Choose sockaddr to select ifa.
948 		 * -- if ifp is set --
949 		 * Order of preference:
950 		 * 1) IFA address
951 		 * 2) gateway address
952 		 *   Note: for interface routes link-level gateway address
953 		 *     is specified to indicate the interface index without
954 		 *     specifying RTF_GATEWAY. In this case, ignore gateway
955 		 *   Note: gateway AF may be different from dst AF. In this case,
956 		 *   ignore gateway
957 		 * 3) final destination.
958 		 * 4) if all of these fails, try to get at least link-level ifa.
959 		 * -- else --
960 		 * try to lookup gateway or dst in the routing table to get ifa
961 		 */
962 		if (info->rti_info[RTAX_IFA] != NULL)
963 			sa = info->rti_info[RTAX_IFA];
964 		else if ((info->rti_flags & RTF_GATEWAY) != 0 &&
965 		    gateway->sa_family == dst->sa_family)
966 			sa = gateway;
967 		else
968 			sa = dst;
969 		if (info->rti_ifp != NULL) {
970 			info->rti_ifa = ifaof_ifpforaddr(sa, info->rti_ifp);
971 			/* Case 4 */
972 			if (info->rti_ifa == NULL && gateway != NULL)
973 				info->rti_ifa = ifaof_ifpforaddr(gateway, info->rti_ifp);
974 		} else if (dst != NULL && gateway != NULL)
975 			info->rti_ifa = ifa_ifwithroute(flags, dst, gateway,
976 							fibnum);
977 		else if (sa != NULL)
978 			info->rti_ifa = ifa_ifwithroute(flags, sa, sa,
979 							fibnum);
980 	}
981 	if (needref && info->rti_ifa != NULL) {
982 		if (info->rti_ifp == NULL)
983 			info->rti_ifp = info->rti_ifa->ifa_ifp;
984 		ifa_ref(info->rti_ifa);
985 	} else
986 		error = ENETUNREACH;
987 	NET_EPOCH_EXIT(et);
988 	return (error);
989 }
990 
991 void
992 rt_updatemtu(struct ifnet *ifp)
993 {
994 	struct rib_head *rnh;
995 	int mtu;
996 	int i, j;
997 
998 	/*
999 	 * Try to update rt_mtu for all routes using this interface
1000 	 * Unfortunately the only way to do this is to traverse all
1001 	 * routing tables in all fibs/domains.
1002 	 */
1003 	for (i = 1; i <= AF_MAX; i++) {
1004 		mtu = if_getmtu_family(ifp, i);
1005 		for (j = 0; j < rt_numfibs; j++) {
1006 			rnh = rt_tables_get_rnh(j, i);
1007 			if (rnh == NULL)
1008 				continue;
1009 			nhops_update_ifmtu(rnh, ifp, mtu);
1010 		}
1011 	}
1012 }
1013 
1014 
1015 #if 0
1016 int p_sockaddr(char *buf, int buflen, struct sockaddr *s);
1017 int rt_print(char *buf, int buflen, struct rtentry *rt);
1018 
1019 int
1020 p_sockaddr(char *buf, int buflen, struct sockaddr *s)
1021 {
1022 	void *paddr = NULL;
1023 
1024 	switch (s->sa_family) {
1025 	case AF_INET:
1026 		paddr = &((struct sockaddr_in *)s)->sin_addr;
1027 		break;
1028 	case AF_INET6:
1029 		paddr = &((struct sockaddr_in6 *)s)->sin6_addr;
1030 		break;
1031 	}
1032 
1033 	if (paddr == NULL)
1034 		return (0);
1035 
1036 	if (inet_ntop(s->sa_family, paddr, buf, buflen) == NULL)
1037 		return (0);
1038 
1039 	return (strlen(buf));
1040 }
1041 
1042 int
1043 rt_print(char *buf, int buflen, struct rtentry *rt)
1044 {
1045 	struct sockaddr *addr, *mask;
1046 	int i = 0;
1047 
1048 	addr = rt_key(rt);
1049 	mask = rt_mask(rt);
1050 
1051 	i = p_sockaddr(buf, buflen, addr);
1052 	if (!(rt->rt_flags & RTF_HOST)) {
1053 		buf[i++] = '/';
1054 		i += p_sockaddr(buf + i, buflen - i, mask);
1055 	}
1056 
1057 	if (rt->rt_flags & RTF_GATEWAY) {
1058 		buf[i++] = '>';
1059 		i += p_sockaddr(buf + i, buflen - i, &rt->rt_nhop->gw_sa);
1060 	}
1061 
1062 	return (i);
1063 }
1064 #endif
1065 
1066 #ifdef RADIX_MPATH
1067 /*
1068  * Deletes key for single-path routes, unlinks rtentry with
1069  * gateway specified in @info from multi-path routes.
1070  *
1071  * Returnes unlinked entry. In case of failure, returns NULL
1072  * and sets @perror to ESRCH.
1073  */
1074 struct radix_node *
1075 rt_mpath_unlink(struct rib_head *rnh, struct rt_addrinfo *info,
1076     struct rtentry *rto, int *perror)
1077 {
1078 	/*
1079 	 * if we got multipath routes, we require users to specify
1080 	 * a matching RTAX_GATEWAY.
1081 	 */
1082 	struct rtentry *rt; // *rto = NULL;
1083 	struct radix_node *rn;
1084 	struct sockaddr *gw;
1085 
1086 	gw = info->rti_info[RTAX_GATEWAY];
1087 	rt = rt_mpath_matchgate(rto, gw);
1088 	if (rt == NULL) {
1089 		*perror = ESRCH;
1090 		return (NULL);
1091 	}
1092 
1093 	/*
1094 	 * this is the first entry in the chain
1095 	 */
1096 	if (rto == rt) {
1097 		rn = rn_mpath_next((struct radix_node *)rt);
1098 		/*
1099 		 * there is another entry, now it's active
1100 		 */
1101 		if (rn) {
1102 			rto = RNTORT(rn);
1103 			RT_LOCK(rto);
1104 			rto->rt_flags |= RTF_UP;
1105 			RT_UNLOCK(rto);
1106 		} else if (rt->rt_flags & RTF_GATEWAY) {
1107 			/*
1108 			 * For gateway routes, we need to
1109 			 * make sure that we we are deleting
1110 			 * the correct gateway.
1111 			 * rt_mpath_matchgate() does not
1112 			 * check the case when there is only
1113 			 * one route in the chain.
1114 			 */
1115 			if (gw &&
1116 			    (rt->rt_nhop->gw_sa.sa_len != gw->sa_len ||
1117 				memcmp(&rt->rt_nhop->gw_sa, gw, gw->sa_len))) {
1118 				*perror = ESRCH;
1119 				return (NULL);
1120 			}
1121 		}
1122 
1123 		/*
1124 		 * use the normal delete code to remove
1125 		 * the first entry
1126 		 */
1127 		rn = rnh->rnh_deladdr(info->rti_info[RTAX_DST],
1128 					info->rti_info[RTAX_NETMASK],
1129 					&rnh->head);
1130 		*perror = 0;
1131 		return (rn);
1132 	}
1133 
1134 	/*
1135 	 * if the entry is 2nd and on up
1136 	 */
1137 	if (rt_mpath_deldup(rto, rt) == 0)
1138 		panic ("rtrequest1: rt_mpath_deldup");
1139 	*perror = 0;
1140 	rn = (struct radix_node *)rt;
1141 	return (rn);
1142 }
1143 #endif
1144 
1145 int
1146 rtrequest1_fib(int req, struct rt_addrinfo *info, struct rtentry **ret_nrt,
1147 				u_int fibnum)
1148 {
1149 	const struct sockaddr *dst;
1150 	struct rib_head *rnh;
1151 	int error;
1152 
1153 	KASSERT((fibnum < rt_numfibs), ("rtrequest1_fib: bad fibnum"));
1154 	KASSERT((info->rti_flags & RTF_RNH_LOCKED) == 0, ("rtrequest1_fib: locked"));
1155 	NET_EPOCH_ASSERT();
1156 
1157 	dst = info->rti_info[RTAX_DST];
1158 
1159 	switch (dst->sa_family) {
1160 	case AF_INET6:
1161 	case AF_INET:
1162 		/* We support multiple FIBs. */
1163 		break;
1164 	default:
1165 		fibnum = RT_DEFAULT_FIB;
1166 		break;
1167 	}
1168 
1169 	/*
1170 	 * Find the correct routing tree to use for this Address Family
1171 	 */
1172 	rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
1173 	if (rnh == NULL)
1174 		return (EAFNOSUPPORT);
1175 
1176 	/*
1177 	 * If we are adding a host route then we don't want to put
1178 	 * a netmask in the tree, nor do we want to clone it.
1179 	 */
1180 	if (info->rti_flags & RTF_HOST)
1181 		info->rti_info[RTAX_NETMASK] = NULL;
1182 
1183 	error = 0;
1184 	switch (req) {
1185 	case RTM_DELETE:
1186 		error = del_route(rnh, info, ret_nrt);
1187 		break;
1188 	case RTM_RESOLVE:
1189 		/*
1190 		 * resolve was only used for route cloning
1191 		 * here for compat
1192 		 */
1193 		break;
1194 	case RTM_ADD:
1195 		error = add_route(rnh, info, ret_nrt);
1196 		break;
1197 	case RTM_CHANGE:
1198 		error = change_route(rnh, info, ret_nrt);
1199 		break;
1200 	default:
1201 		error = EOPNOTSUPP;
1202 	}
1203 
1204 	return (error);
1205 }
1206 
1207 void
1208 rt_setmetrics(const struct rt_addrinfo *info, struct rtentry *rt)
1209 {
1210 
1211 	if (info->rti_mflags & RTV_WEIGHT)
1212 		rt->rt_weight = info->rti_rmx->rmx_weight;
1213 	/* Kernel -> userland timebase conversion. */
1214 	if (info->rti_mflags & RTV_EXPIRE)
1215 		rt->rt_expire = info->rti_rmx->rmx_expire ?
1216 		    info->rti_rmx->rmx_expire - time_second + time_uptime : 0;
1217 }
1218 
1219 void
1220 rt_maskedcopy(struct sockaddr *src, struct sockaddr *dst, struct sockaddr *netmask)
1221 {
1222 	u_char *cp1 = (u_char *)src;
1223 	u_char *cp2 = (u_char *)dst;
1224 	u_char *cp3 = (u_char *)netmask;
1225 	u_char *cplim = cp2 + *cp3;
1226 	u_char *cplim2 = cp2 + *cp1;
1227 
1228 	*cp2++ = *cp1++; *cp2++ = *cp1++; /* copies sa_len & sa_family */
1229 	cp3 += 2;
1230 	if (cplim > cplim2)
1231 		cplim = cplim2;
1232 	while (cp2 < cplim)
1233 		*cp2++ = *cp1++ & *cp3++;
1234 	if (cp2 < cplim2)
1235 		bzero((caddr_t)cp2, (unsigned)(cplim2 - cp2));
1236 }
1237 
1238 /*
1239  * Set up a routing table entry, normally
1240  * for an interface.
1241  */
1242 #define _SOCKADDR_TMPSIZE 128 /* Not too big.. kernel stack size is limited */
1243 static inline  int
1244 rtinit1(struct ifaddr *ifa, int cmd, int flags, int fibnum)
1245 {
1246 	RIB_RLOCK_TRACKER;
1247 	struct epoch_tracker et;
1248 	struct sockaddr *dst;
1249 	struct sockaddr *netmask;
1250 	struct rtentry *rt = NULL;
1251 	struct rt_addrinfo info;
1252 	int error = 0;
1253 	int startfib, endfib;
1254 	char tempbuf[_SOCKADDR_TMPSIZE];
1255 	int didwork = 0;
1256 	int a_failure = 0;
1257 	struct sockaddr_dl_short *sdl = NULL;
1258 	struct rib_head *rnh;
1259 
1260 	if (flags & RTF_HOST) {
1261 		dst = ifa->ifa_dstaddr;
1262 		netmask = NULL;
1263 	} else {
1264 		dst = ifa->ifa_addr;
1265 		netmask = ifa->ifa_netmask;
1266 	}
1267 	if (dst->sa_len == 0)
1268 		return(EINVAL);
1269 	switch (dst->sa_family) {
1270 	case AF_INET6:
1271 	case AF_INET:
1272 		/* We support multiple FIBs. */
1273 		break;
1274 	default:
1275 		fibnum = RT_DEFAULT_FIB;
1276 		break;
1277 	}
1278 	if (fibnum == RT_ALL_FIBS) {
1279 		if (V_rt_add_addr_allfibs == 0 && cmd == (int)RTM_ADD)
1280 			startfib = endfib = ifa->ifa_ifp->if_fib;
1281 		else {
1282 			startfib = 0;
1283 			endfib = rt_numfibs - 1;
1284 		}
1285 	} else {
1286 		KASSERT((fibnum < rt_numfibs), ("rtinit1: bad fibnum"));
1287 		startfib = fibnum;
1288 		endfib = fibnum;
1289 	}
1290 
1291 	/*
1292 	 * If it's a delete, check that if it exists,
1293 	 * it's on the correct interface or we might scrub
1294 	 * a route to another ifa which would
1295 	 * be confusing at best and possibly worse.
1296 	 */
1297 	if (cmd == RTM_DELETE) {
1298 		/*
1299 		 * It's a delete, so it should already exist..
1300 		 * If it's a net, mask off the host bits
1301 		 * (Assuming we have a mask)
1302 		 * XXX this is kinda inet specific..
1303 		 */
1304 		if (netmask != NULL) {
1305 			rt_maskedcopy(dst, (struct sockaddr *)tempbuf, netmask);
1306 			dst = (struct sockaddr *)tempbuf;
1307 		}
1308 	} else if (cmd == RTM_ADD) {
1309 		sdl = (struct sockaddr_dl_short *)tempbuf;
1310 		bzero(sdl, sizeof(struct sockaddr_dl_short));
1311 		sdl->sdl_family = AF_LINK;
1312 		sdl->sdl_len = sizeof(struct sockaddr_dl_short);
1313 		sdl->sdl_type = ifa->ifa_ifp->if_type;
1314 		sdl->sdl_index = ifa->ifa_ifp->if_index;
1315         }
1316 	/*
1317 	 * Now go through all the requested tables (fibs) and do the
1318 	 * requested action. Realistically, this will either be fib 0
1319 	 * for protocols that don't do multiple tables or all the
1320 	 * tables for those that do.
1321 	 */
1322 	for ( fibnum = startfib; fibnum <= endfib; fibnum++) {
1323 		if (cmd == RTM_DELETE) {
1324 			struct radix_node *rn;
1325 			/*
1326 			 * Look up an rtentry that is in the routing tree and
1327 			 * contains the correct info.
1328 			 */
1329 			rnh = rt_tables_get_rnh(fibnum, dst->sa_family);
1330 			if (rnh == NULL)
1331 				/* this table doesn't exist but others might */
1332 				continue;
1333 			RIB_RLOCK(rnh);
1334 			rn = rnh->rnh_lookup(dst, netmask, &rnh->head);
1335 #ifdef RADIX_MPATH
1336 			if (rt_mpath_capable(rnh)) {
1337 
1338 				if (rn == NULL)
1339 					error = ESRCH;
1340 				else {
1341 					rt = RNTORT(rn);
1342 					/*
1343 					 * for interface route the gateway
1344 					 * gateway is sockaddr_dl, so
1345 					 * rt_mpath_matchgate must use the
1346 					 * interface address
1347 					 */
1348 					rt = rt_mpath_matchgate(rt,
1349 					    ifa->ifa_addr);
1350 					if (rt == NULL)
1351 						error = ESRCH;
1352 				}
1353 			}
1354 #endif
1355 			error = (rn == NULL ||
1356 			    (rn->rn_flags & RNF_ROOT) ||
1357 			    RNTORT(rn)->rt_nhop->nh_ifa != ifa);
1358 			RIB_RUNLOCK(rnh);
1359 			if (error) {
1360 				/* this is only an error if bad on ALL tables */
1361 				continue;
1362 			}
1363 		}
1364 		/*
1365 		 * Do the actual request
1366 		 */
1367 		bzero((caddr_t)&info, sizeof(info));
1368 		info.rti_ifa = ifa;
1369 		info.rti_flags = flags |
1370 		    (ifa->ifa_flags & ~IFA_RTSELF) | RTF_PINNED;
1371 		info.rti_info[RTAX_DST] = dst;
1372 		/*
1373 		 * doing this for compatibility reasons
1374 		 */
1375 		if (cmd == RTM_ADD)
1376 			info.rti_info[RTAX_GATEWAY] = (struct sockaddr *)sdl;
1377 		else
1378 			info.rti_info[RTAX_GATEWAY] = ifa->ifa_addr;
1379 		info.rti_info[RTAX_NETMASK] = netmask;
1380 		NET_EPOCH_ENTER(et);
1381 		error = rtrequest1_fib(cmd, &info, &rt, fibnum);
1382 		if (error == 0 && rt != NULL) {
1383 			/*
1384 			 * notify any listening routing agents of the change
1385 			 */
1386 
1387 			/* TODO: interface routes/aliases */
1388 			rt_newaddrmsg_fib(cmd, ifa, rt, fibnum);
1389 			didwork = 1;
1390 		}
1391 		NET_EPOCH_EXIT(et);
1392 		if (error)
1393 			a_failure = error;
1394 	}
1395 	if (cmd == RTM_DELETE) {
1396 		if (didwork) {
1397 			error = 0;
1398 		} else {
1399 			/* we only give an error if it wasn't in any table */
1400 			error = ((flags & RTF_HOST) ?
1401 			    EHOSTUNREACH : ENETUNREACH);
1402 		}
1403 	} else {
1404 		if (a_failure) {
1405 			/* return an error if any of them failed */
1406 			error = a_failure;
1407 		}
1408 	}
1409 	return (error);
1410 }
1411 
1412 /*
1413  * Set up a routing table entry, normally
1414  * for an interface.
1415  */
1416 int
1417 rtinit(struct ifaddr *ifa, int cmd, int flags)
1418 {
1419 	struct sockaddr *dst;
1420 	int fib = RT_DEFAULT_FIB;
1421 
1422 	if (flags & RTF_HOST) {
1423 		dst = ifa->ifa_dstaddr;
1424 	} else {
1425 		dst = ifa->ifa_addr;
1426 	}
1427 
1428 	switch (dst->sa_family) {
1429 	case AF_INET6:
1430 	case AF_INET:
1431 		/* We do support multiple FIBs. */
1432 		fib = RT_ALL_FIBS;
1433 		break;
1434 	}
1435 	return (rtinit1(ifa, cmd, flags, fib));
1436 }
1437 
1438 /*
1439  * Announce interface address arrival/withdraw
1440  * Returns 0 on success.
1441  */
1442 int
1443 rt_addrmsg(int cmd, struct ifaddr *ifa, int fibnum)
1444 {
1445 
1446 	KASSERT(cmd == RTM_ADD || cmd == RTM_DELETE,
1447 	    ("unexpected cmd %d", cmd));
1448 	KASSERT(fibnum == RT_ALL_FIBS || (fibnum >= 0 && fibnum < rt_numfibs),
1449 	    ("%s: fib out of range 0 <=%d<%d", __func__, fibnum, rt_numfibs));
1450 
1451 	EVENTHANDLER_DIRECT_INVOKE(rt_addrmsg, ifa, cmd);
1452 	return (rtsock_addrmsg(cmd, ifa, fibnum));
1453 }
1454 
1455 /*
1456  * Announce kernel-originated route addition/removal to rtsock based on @rt data.
1457  * cmd: RTM_ cmd
1458  * @rt: valid rtentry
1459  * @ifp: target route interface
1460  * @fibnum: fib id or RT_ALL_FIBS
1461  *
1462  * Returns 0 on success.
1463  */
1464 int
1465 rt_routemsg(int cmd, struct rtentry *rt, struct ifnet *ifp, int rti_addrs,
1466     int fibnum)
1467 {
1468 
1469 	KASSERT(cmd == RTM_ADD || cmd == RTM_DELETE,
1470 	    ("unexpected cmd %d", cmd));
1471 
1472 	KASSERT(fibnum == RT_ALL_FIBS || (fibnum >= 0 && fibnum < rt_numfibs),
1473 	    ("%s: fib out of range 0 <=%d<%d", __func__, fibnum, rt_numfibs));
1474 
1475 	KASSERT(rt_key(rt) != NULL, (":%s: rt_key must be supplied", __func__));
1476 
1477 	return (rtsock_routemsg(cmd, rt, ifp, 0, fibnum));
1478 }
1479 
1480 /*
1481  * Announce kernel-originated route addition/removal to rtsock based on @rt data.
1482  * cmd: RTM_ cmd
1483  * @info: addrinfo structure with valid data.
1484  * @fibnum: fib id or RT_ALL_FIBS
1485  *
1486  * Returns 0 on success.
1487  */
1488 int
1489 rt_routemsg_info(int cmd, struct rt_addrinfo *info, int fibnum)
1490 {
1491 
1492 	KASSERT(cmd == RTM_ADD || cmd == RTM_DELETE || cmd == RTM_CHANGE,
1493 	    ("unexpected cmd %d", cmd));
1494 
1495 	KASSERT(fibnum == RT_ALL_FIBS || (fibnum >= 0 && fibnum < rt_numfibs),
1496 	    ("%s: fib out of range 0 <=%d<%d", __func__, fibnum, rt_numfibs));
1497 
1498 	KASSERT(info->rti_info[RTAX_DST] != NULL, (":%s: RTAX_DST must be supplied", __func__));
1499 
1500 	return (rtsock_routemsg_info(cmd, info, fibnum));
1501 }
1502 
1503 
1504 /*
1505  * This is called to generate messages from the routing socket
1506  * indicating a network interface has had addresses associated with it.
1507  */
1508 void
1509 rt_newaddrmsg_fib(int cmd, struct ifaddr *ifa, struct rtentry *rt, int fibnum)
1510 {
1511 
1512 	KASSERT(cmd == RTM_ADD || cmd == RTM_DELETE,
1513 		("unexpected cmd %u", cmd));
1514 	KASSERT(fibnum == RT_ALL_FIBS || (fibnum >= 0 && fibnum < rt_numfibs),
1515 	    ("%s: fib out of range 0 <=%d<%d", __func__, fibnum, rt_numfibs));
1516 
1517 	if (cmd == RTM_ADD) {
1518 		rt_addrmsg(cmd, ifa, fibnum);
1519 		if (rt != NULL)
1520 			rt_routemsg(cmd, rt, ifa->ifa_ifp, 0, fibnum);
1521 	} else {
1522 		if (rt != NULL)
1523 			rt_routemsg(cmd, rt, ifa->ifa_ifp, 0, fibnum);
1524 		rt_addrmsg(cmd, ifa, fibnum);
1525 	}
1526 }
1527 
1528